From 635c725b30319d9c829912032213c91b23aa28b3 Mon Sep 17 00:00:00 2001 From: skc7 Date: Mon, 10 Apr 2023 18:37:17 +0530 Subject: [PATCH] [AMDGPU][NFC] Regenerate test checks for merge-tbuffer.mir --- llvm/test/CodeGen/AMDGPU/merge-tbuffer.mir | 6763 +++++++++++++++++++++++----- 1 file changed, 5694 insertions(+), 1069 deletions(-) diff --git a/llvm/test/CodeGen/AMDGPU/merge-tbuffer.mir b/llvm/test/CodeGen/AMDGPU/merge-tbuffer.mir index 852ed14..a3c012d 100644 --- a/llvm/test/CodeGen/AMDGPU/merge-tbuffer.mir +++ b/llvm/test/CodeGen/AMDGPU/merge-tbuffer.mir @@ -1,3 +1,4 @@ +# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2 # RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs -run-pass si-load-store-opt -o - %s | FileCheck -check-prefix=GFX9 %s # RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs -run-pass si-load-store-opt -o - %s | FileCheck -check-prefix=GFX10 %s # RUN: llc -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs -run-pass si-load-store-opt -o - %s | FileCheck -check-prefix=GFX11 %s @@ -6,10 +7,6 @@ # GFX9 tests # -# GFX9-LABEL: name: gfx9_tbuffer_load_x_xyz -# GFX9: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET %4, 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY %7.sub0 -# GFX9: %{{[0-9]+}}:vreg_96 = COPY killed %7.sub1_sub2_sub3 name: gfx9_tbuffer_load_x_xyz body: | bb.0.entry: @@ -23,13 +20,34 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_load_xyz_x -# GFX9: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET %4, 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vreg_96 = COPY %7.sub0_sub1_sub2 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %7.sub3 name: gfx9_tbuffer_load_xyz_x body: | bb.0.entry: + ; GFX9-LABEL: name: gfx9_tbuffer_load_xyz_x + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_96 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1_sub2 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub3 + ; GFX10-LABEL: name: gfx9_tbuffer_load_xyz_x + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_xyz_x + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -40,13 +58,34 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_load_xy_xy -# GFX9: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET %4, 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vreg_64 = COPY %7.sub0_sub1 -# GFX9: %{{[0-9]+}}:vreg_64 = COPY killed %7.sub2_sub3 name: gfx9_tbuffer_load_xy_xy body: | bb.0.entry: + ; GFX9-LABEL: name: gfx9_tbuffer_load_xy_xy + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub2_sub3 + ; GFX10-LABEL: name: gfx9_tbuffer_load_xy_xy + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_xy_xy + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -57,13 +96,34 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_load_x_xy -# GFX9: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET %4, 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY %7.sub0 -# GFX9: %{{[0-9]+}}:vreg_64 = COPY killed %7.sub1_sub2 name: gfx9_tbuffer_load_x_xy body: | bb.0.entry: + ; GFX9-LABEL: name: gfx9_tbuffer_load_x_xy + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub1_sub2 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_xy + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 8, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_xy + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 8, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -74,13 +134,34 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_load_xy_x -# GFX9: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET %4, 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vreg_64 = COPY %7.sub0_sub1 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %7.sub2 name: gfx9_tbuffer_load_xy_x body: | bb.0.entry: + ; GFX9-LABEL: name: gfx9_tbuffer_load_xy_x + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0_sub1 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub2 + ; GFX10-LABEL: name: gfx9_tbuffer_load_xy_x + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 12, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_xy_x + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 12, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -92,14 +173,34 @@ body: | --- -# GFX9-LABEL: name: gfx9_tbuffer_load_x_x -# GFX9: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET %4, 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY %7.sub0 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %7.sub1 - name: gfx9_tbuffer_load_x_x body: | bb.0.entry: + ; GFX9-LABEL: name: gfx9_tbuffer_load_x_x + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_x + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_x + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -110,14 +211,34 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_load_x_x_format_32_32_32_32 -# GFX9: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET %4, 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY %7.sub0 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %7.sub1 - name: gfx9_tbuffer_load_x_x_format_32_32_32_32 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx9_tbuffer_load_x_x_format_32_32_32_32 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_x_format_32_32_32_32 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 126, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_x_format_32_32_32_32 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 126, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -128,27 +249,60 @@ body: | ... --- - -# GFX9-LABEL: name: gfx9_tbuffer_load_float_32 -# GFX9: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET %4, 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY %14.sub0 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %14.sub1 -# GFX9: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET %4, 0, 16, 126, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vreg_96 = COPY %17.sub0_sub1_sub2 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %17.sub3 -# GFX9: %{{[0-9]+}}:vreg_64 = COPY %16.sub0_sub1 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %16.sub2 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY %15.sub0 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %15.sub1 -# GFX9: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET %4, 0, 36, 125, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vreg_64 = COPY %19.sub0_sub1 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %19.sub2 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY %18.sub0 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %18.sub1 - name: gfx9_tbuffer_load_float_32 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx9_tbuffer_load_float_32 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 16, 126, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1_sub2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub3 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vreg_64 = COPY [[COPY6]].sub0_sub1 + ; GFX9-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY killed [[COPY6]].sub2 + ; GFX9-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[COPY8]].sub0 + ; GFX9-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY killed [[COPY8]].sub1 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 36, 125, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY12:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0_sub1 + ; GFX9-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub2 + ; GFX9-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[COPY12]].sub0 + ; GFX9-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY killed [[COPY12]].sub1 + ; GFX10-LABEL: name: gfx9_tbuffer_load_float_32 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_float_32 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -166,26 +320,60 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_load_sint_32 -# GFX9: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET %4, 0, 4, 91, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY %14.sub0 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %14.sub1 -# GFX9: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET %4, 0, 16, 94, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vreg_96 = COPY %17.sub0_sub1_sub2 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %17.sub3 -# GFX9: %{{[0-9]+}}:vreg_64 = COPY %16.sub0_sub1 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %16.sub2 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY %15.sub0 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %15.sub1 -# GFX9: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET %4, 0, 36, 93, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vreg_64 = COPY %19.sub0_sub1 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %19.sub2 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY %18.sub0 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %18.sub1 - name: gfx9_tbuffer_load_sint_32 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx9_tbuffer_load_sint_32 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 91, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 16, 94, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1_sub2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub3 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vreg_64 = COPY [[COPY6]].sub0_sub1 + ; GFX9-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY killed [[COPY6]].sub2 + ; GFX9-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[COPY8]].sub0 + ; GFX9-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY killed [[COPY8]].sub1 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 36, 93, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY12:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0_sub1 + ; GFX9-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub2 + ; GFX9-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[COPY12]].sub0 + ; GFX9-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY killed [[COPY12]].sub1 + ; GFX10-LABEL: name: gfx9_tbuffer_load_sint_32 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_sint_32 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -203,26 +391,60 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_load_uint_32 -# GFX9: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET %4, 0, 4, 75, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY %14.sub0 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %14.sub1 -# GFX9: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET %4, 0, 16, 78, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vreg_96 = COPY %17.sub0_sub1_sub2 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %17.sub3 -# GFX9: %{{[0-9]+}}:vreg_64 = COPY %16.sub0_sub1 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %16.sub2 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY %15.sub0 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %15.sub1 -# GFX9: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET %4, 0, 36, 77, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vreg_64 = COPY %19.sub0_sub1 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %19.sub2 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY %18.sub0 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %18.sub1 - name: gfx9_tbuffer_load_uint_32 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx9_tbuffer_load_uint_32 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 75, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 16, 78, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1_sub2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub3 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vreg_64 = COPY [[COPY6]].sub0_sub1 + ; GFX9-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY killed [[COPY6]].sub2 + ; GFX9-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[COPY8]].sub0 + ; GFX9-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY killed [[COPY8]].sub1 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 36, 77, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY12:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0_sub1 + ; GFX9-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub2 + ; GFX9-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[COPY12]].sub0 + ; GFX9-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY killed [[COPY12]].sub1 + ; GFX10-LABEL: name: gfx9_tbuffer_load_uint_32 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 68, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 68, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 68, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 68, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 68, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 68, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 68, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 68, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 68, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_uint_32 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 68, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 68, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 68, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 68, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 68, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 68, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 68, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 68, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 68, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -240,26 +462,54 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_load_not_merged_data_format_mismatch -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX9: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %0, %subreg.sub0, %1, %subreg.sub1, %2, %subreg.sub2, %3, %subreg.sub3 -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 8, 114, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 20, 114, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 28, 114, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 40, 114, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - - name: gfx9_tbuffer_load_not_merged_data_format_mismatch body: | bb.0.entry: + ; GFX9-LABEL: name: gfx9_tbuffer_load_not_merged_data_format_mismatch + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 114, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 114, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 114, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 114, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_load_not_merged_data_format_mismatch + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 114, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 114, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 114, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 114, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_not_merged_data_format_mismatch + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 114, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 114, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 114, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 114, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -277,24 +527,54 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_load_not_merged_num_format_mismatch -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX9: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %0, %subreg.sub0, %1, %subreg.sub1, %2, %subreg.sub2, %3, %subreg.sub3 -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 8, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 20, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 28, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 40, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) name: gfx9_tbuffer_load_not_merged_num_format_mismatch body: | bb.0.entry: + ; GFX9-LABEL: name: gfx9_tbuffer_load_not_merged_num_format_mismatch + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_load_not_merged_num_format_mismatch + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_not_merged_num_format_mismatch + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 84, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -312,14 +592,55 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_store_x_xyz -# GFX9: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE %3, %subreg.sub0, %2, %subreg.sub1, %1, %subreg.sub2 -# GFX9: %{{[0-9]+}}:vreg_128 = REG_SEQUENCE %0, %subreg.sub0, %9, %subreg.sub1_sub2_sub3 -# GFX9: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed %10, %8, 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) name: gfx9_tbuffer_store_x_xyz body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx9_tbuffer_store_x_xyz + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2 + ; GFX9-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[REG_SEQUENCE1]], %subreg.sub1_sub2_sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_store_x_xyz + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 8, 125, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_store_x_xyz + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 8, 125, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -335,16 +656,55 @@ body: | ... --- - - -# GFX9-LABEL: name: gfx9_tbuffer_store_xyz_x -# GFX9: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE %3, %subreg.sub0, %2, %subreg.sub1, %1, %subreg.sub2 -# GFX9: %{{[0-9]+}}:vreg_128 = REG_SEQUENCE %9, %subreg.sub0_sub1_sub2, %0, %subreg.sub3 -# GFX9: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed %10, %8, 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) name: gfx9_tbuffer_store_xyz_x body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx9_tbuffer_store_xyz_x + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2 + ; GFX9-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[REG_SEQUENCE1]], %subreg.sub0_sub1_sub2, [[COPY]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_store_xyz_x + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_store_xyz_x + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -360,15 +720,58 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_store_xy_xy -# GFX9: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %3, %subreg.sub0, %2, %subreg.sub1 -# GFX9: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %1, %subreg.sub0, %0, %subreg.sub1 -# GFX9: %{{[0-9]+}}:vreg_128 = REG_SEQUENCE %9, %subreg.sub0_sub1, %10, %subreg.sub2_sub3 -# GFX9: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed %11, %8, 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) name: gfx9_tbuffer_store_xy_xy body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx9_tbuffer_store_xy_xy + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX9-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX9-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[REG_SEQUENCE1]], %subreg.sub0_sub1, [[REG_SEQUENCE2]], %subreg.sub2_sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE3]], [[REG_SEQUENCE]], 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_store_xy_xy + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX10-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_store_xy_xy + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX11-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -385,14 +788,55 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_store_x_xy -# GFX9: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %3, %subreg.sub0, %2, %subreg.sub1 -# GFX9: %{{[0-9]+}}:vreg_64, %subreg.sub1_sub2 -# GFX9: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed %11, %8, 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) name: gfx9_tbuffer_store_x_xy body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx9_tbuffer_store_x_xy + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX9-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY]], %subreg.sub0, %10:vreg_64, %subreg.sub1_sub2 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_store_x_xy + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact %10:vreg_64, [[REG_SEQUENCE]], 0, 8, 123, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_store_x_xy + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact %10:vreg_64, [[REG_SEQUENCE]], 0, 8, 123, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -408,15 +852,56 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_store_xy_x -# GFX9: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %3, %subreg.sub0, %2, %subreg.sub1 -# GFX9: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE %9, %subreg.sub0_sub1, %0, %subreg.sub2 -# GFX9: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed %10, %8, 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) name: gfx9_tbuffer_store_xy_x body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx9_tbuffer_store_xy_x + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX9-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[REG_SEQUENCE1]], %subreg.sub0_sub1, [[COPY]], %subreg.sub2 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_store_xy_x + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 12, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_store_xy_x + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 12, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -432,14 +917,52 @@ body: | ... --- - -# GFX9-LABEL: name: gfx9_tbuffer_store_x_x -# GFX9: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %1, %subreg.sub0, %0, %subreg.sub1 -# GFX9: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed %9, %8, 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) name: gfx9_tbuffer_store_x_x body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx9_tbuffer_store_x_x + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_store_x_x + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_store_x_x + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -454,13 +977,52 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_store_x_x_format_32_32_32_32 -# GFX9: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %1, %subreg.sub0, %0, %subreg.sub1 -# GFX9: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed %9, %8, 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) name: gfx9_tbuffer_store_x_x_format_32_32_32_32 body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx9_tbuffer_store_x_x_format_32_32_32_32 + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_store_x_x_format_32_32_32_32 + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 8, 126, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_store_x_x_format_32_32_32_32 + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 8, 126, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -475,34 +1037,88 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_store_float32 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr8 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr7 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr6 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr5 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr4 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr3 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr2 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr1 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr0 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX9: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %12, %subreg.sub0, %11, %subreg.sub1, %10, %subreg.sub2, %9, %subreg.sub3 -# GFX9: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %8, %subreg.sub0, %7, %subreg.sub1 -# GFX9: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed %14, %13, 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %6, %subreg.sub0, %5, %subreg.sub1 -# GFX9: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE killed %15, %subreg.sub0_sub1, %4, %subreg.sub2 -# GFX9: %{{[0-9]+}}:vreg_128 = REG_SEQUENCE killed %16, %subreg.sub0_sub1_sub2, %3, %subreg.sub3 -# GFX9: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed %17, %13, 0, 16, 126, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %2, %subreg.sub0, %1, %subreg.sub1 -# GFX9: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE killed %18, %subreg.sub0_sub1, %0, %subreg.sub2 -# GFX9: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed %19, %13, 0, 36, 125, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) name: gfx9_tbuffer_store_float32 body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-LABEL: name: gfx9_tbuffer_store_float32 + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1 + ; GFX9-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE2]], %subreg.sub0_sub1, [[COPY4]], %subreg.sub2 + ; GFX9-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[REG_SEQUENCE3]], %subreg.sub0_sub1_sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE4]], [[REG_SEQUENCE]], 0, 16, 126, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX9-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1 + ; GFX9-NEXT: [[REG_SEQUENCE6:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE5]], %subreg.sub0_sub1, [[COPY]], %subreg.sub2 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE6]], [[REG_SEQUENCE]], 0, 36, 125, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_store_float32 + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_store_float32 + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) %12:vgpr_32 = COPY $vgpr8 %11:vgpr_32 = COPY $vgpr7 %10:vgpr_32 = COPY $vgpr6 @@ -529,34 +1145,88 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_store_sint32 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr8 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr7 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr6 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr5 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr4 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr3 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr2 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr1 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr0 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX9: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %12, %subreg.sub0, %11, %subreg.sub1, %10, %subreg.sub2, %9, %subreg.sub3 -# GFX9: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %8, %subreg.sub0, %7, %subreg.sub1 -# GFX9: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed %14, %13, 0, 4, 91, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %6, %subreg.sub0, %5, %subreg.sub1 -# GFX9: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE killed %15, %subreg.sub0_sub1, %4, %subreg.sub2 -# GFX9: %{{[0-9]+}}:vreg_128 = REG_SEQUENCE killed %16, %subreg.sub0_sub1_sub2, %3, %subreg.sub3 -# GFX9: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed %17, %13, 0, 16, 94, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %2, %subreg.sub0, %1, %subreg.sub1 -# GFX9: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE killed %18, %subreg.sub0_sub1, %0, %subreg.sub2 -# GFX9: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed %19, %13, 0, 36, 93, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) name: gfx9_tbuffer_store_sint32 body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-LABEL: name: gfx9_tbuffer_store_sint32 + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 91, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1 + ; GFX9-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE2]], %subreg.sub0_sub1, [[COPY4]], %subreg.sub2 + ; GFX9-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[REG_SEQUENCE3]], %subreg.sub0_sub1_sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE4]], [[REG_SEQUENCE]], 0, 16, 94, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX9-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1 + ; GFX9-NEXT: [[REG_SEQUENCE6:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE5]], %subreg.sub0_sub1, [[COPY]], %subreg.sub2 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE6]], [[REG_SEQUENCE]], 0, 36, 93, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_store_sint32 + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_store_sint32 + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) %12:vgpr_32 = COPY $vgpr8 %11:vgpr_32 = COPY $vgpr7 %10:vgpr_32 = COPY $vgpr6 @@ -583,34 +1253,88 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_store_uint32 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr8 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr7 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr6 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr5 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr4 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr3 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr2 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr1 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr0 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX9: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %12, %subreg.sub0, %11, %subreg.sub1, %10, %subreg.sub2, %9, %subreg.sub3 -# GFX9: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %8, %subreg.sub0, %7, %subreg.sub1 -# GFX9: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed %14, %13, 0, 4, 75, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %6, %subreg.sub0, %5, %subreg.sub1 -# GFX9: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE killed %15, %subreg.sub0_sub1, %4, %subreg.sub2 -# GFX9: %{{[0-9]+}}:vreg_128 = REG_SEQUENCE killed %16, %subreg.sub0_sub1_sub2, %3, %subreg.sub3 -# GFX9: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed %17, %13, 0, 16, 78, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %2, %subreg.sub0, %1, %subreg.sub1 -# GFX9: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE killed %18, %subreg.sub0_sub1, %0, %subreg.sub2 -# GFX9: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed %19, %13, 0, 36, 77, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) name: gfx9_tbuffer_store_uint32 body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-LABEL: name: gfx9_tbuffer_store_uint32 + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 75, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1 + ; GFX9-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE2]], %subreg.sub0_sub1, [[COPY4]], %subreg.sub2 + ; GFX9-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[REG_SEQUENCE3]], %subreg.sub0_sub1_sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE4]], [[REG_SEQUENCE]], 0, 16, 78, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX9-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1 + ; GFX9-NEXT: [[REG_SEQUENCE6:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE5]], %subreg.sub0_sub1, [[COPY]], %subreg.sub2 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE6]], [[REG_SEQUENCE]], 0, 36, 77, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_store_uint32 + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 68, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 68, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 68, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 68, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 68, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 68, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 68, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 68, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 68, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_store_uint32 + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 68, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 68, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 68, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 68, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 68, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 68, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 68, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 68, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 68, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) %12:vgpr_32 = COPY $vgpr8 %11:vgpr_32 = COPY $vgpr7 %10:vgpr_32 = COPY $vgpr6 @@ -637,34 +1361,88 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_store_not_merged_data_format_mismatch -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr8 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr7 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr6 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr5 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr4 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr3 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr2 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr1 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr0 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX9: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %12, %subreg.sub0, %11, %subreg.sub1, %10, %subreg.sub2, %9, %subreg.sub3 -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %8, %13, 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %7, %13, 0, 8, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %6, %13, 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %5, %13, 0, 20, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %4, %13, 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %3, %13, 0, 28, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %2, %13, 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %1, %13, 0, 40, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %0, %13, 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) name: gfx9_tbuffer_store_not_merged_data_format_mismatch body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-LABEL: name: gfx9_tbuffer_store_not_merged_data_format_mismatch + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_store_not_merged_data_format_mismatch + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_store_not_merged_data_format_mismatch + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 84, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) %12:vgpr_32 = COPY $vgpr8 %11:vgpr_32 = COPY $vgpr7 %10:vgpr_32 = COPY $vgpr6 @@ -691,34 +1469,88 @@ body: | ... --- -# GFX9-LABEL: name: gfx9_tbuffer_store_not_merged_num_format_mismatch -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr8 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr7 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr6 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr5 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr4 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr3 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr2 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr1 -# GFX9: %{{[0-9]+}}:vgpr_32 = COPY $vgpr0 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX9: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX9: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %12, %subreg.sub0, %11, %subreg.sub1, %10, %subreg.sub2, %9, %subreg.sub3 -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %8, %13, 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %7, %13, 0, 8, 114, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %6, %13, 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %5, %13, 0, 20, 114, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %4, %13, 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %3, %13, 0, 28, 114, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %2, %13, 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %1, %13, 0, 40, 114, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %0, %13, 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) name: gfx9_tbuffer_store_not_merged_num_format_mismatch body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-LABEL: name: gfx9_tbuffer_store_not_merged_num_format_mismatch + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 114, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 114, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 114, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 114, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_store_not_merged_num_format_mismatch + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 114, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 114, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 114, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 114, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_store_not_merged_num_format_mismatch + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 114, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 114, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 114, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 114, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 116, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) %12:vgpr_32 = COPY $vgpr8 %11:vgpr_32 = COPY $vgpr7 %10:vgpr_32 = COPY $vgpr6 @@ -745,13 +1577,33 @@ body: | ... --- - -# GFX9-LABEL: name: gfx9_tbuffer_load_not_merged_swizzled_0 -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 4, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) name: gfx9_tbuffer_load_not_merged_swizzled_0 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx9_tbuffer_load_not_merged_swizzled_0 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_load_not_merged_swizzled_0 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_not_merged_swizzled_0 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -762,13 +1614,33 @@ body: | ... --- - -# GFX9-LABEL: name: gfx9_tbuffer_load_not_merged_swizzled_1 -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 8, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) name: gfx9_tbuffer_load_not_merged_swizzled_1 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx9_tbuffer_load_not_merged_swizzled_1 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_load_not_merged_swizzled_1 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_not_merged_swizzled_1 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -779,13 +1651,37 @@ body: | ... --- - -# GFX9-LABEL: name: gfx9_tbuffer_load_merge_across_swizzle -# GFX9-DAG: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 12, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9-DAG: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET %4, 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) name: gfx9_tbuffer_load_merge_across_swizzle body: | bb.0.entry: + ; GFX9-LABEL: name: gfx9_tbuffer_load_merge_across_swizzle + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 12, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_load_merge_across_swizzle + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 12, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_merge_across_swizzle + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 12, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -797,14 +1693,39 @@ body: | ... --- - -# GFX9-LABEL: name: gfx9_tbuffer_load_merge_across_swizzled_store -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %5, 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX9: TBUFFER_STORE_FORMAT_X_OFFSET_exact %4, %5, 0, 6, 116, 0, 1, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %5, 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) name: gfx9_tbuffer_load_merge_across_swizzled_store body: | bb.0.entry: + ; GFX9-LABEL: name: gfx9_tbuffer_load_merge_across_swizzled_store + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 6, 116, 0, 1, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_load_merge_across_swizzled_store + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 6, 116, 0, 1, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_merge_across_swizzled_store + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 6, 116, 0, 1, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -821,9 +1742,33 @@ name: gfx9_tbuffer_load_x_x_idxen body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_x_x_idxen - ; GFX9: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN %4, %5, 0, 0, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %8.sub1 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_IDXEN]].sub0 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_IDXEN]].sub1 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_x_idxen + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_x_idxen + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -839,9 +1784,33 @@ name: gfx9_tbuffer_load_x_xy_idxen body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_x_xy_idxen - ; GFX9: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN %4, %5, 0, 0, 125, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX9: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub1_sub2 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 125, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN]].sub0 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN]].sub1_sub2 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_xy_idxen + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_xy_idxen + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -857,9 +1826,33 @@ name: gfx9_tbuffer_load_xy_xy_idxen body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_xy_xy_idxen - ; GFX9: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN %4, %5, 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vreg_64 = COPY %8.sub0_sub1 - ; GFX9: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub2_sub3 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN]].sub0_sub1 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN]].sub2_sub3 + ; GFX10-LABEL: name: gfx9_tbuffer_load_xy_xy_idxen + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_xy_xy_idxen + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -875,9 +1868,33 @@ name: gfx9_tbuffer_load_x_xyz_idxen body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_x_xyz_idxen - ; GFX9: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN %4, %5, 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX9: %{{[0-9]+}}:vreg_96 = COPY killed %8.sub1_sub2_sub3 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN]].sub0 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN]].sub1_sub2_sub3 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_xyz_idxen + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 8, 125, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_xyz_idxen + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 8, 125, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -893,9 +1910,33 @@ name: gfx9_tbuffer_load_x_x_bothen body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_x_x_bothen - ; GFX9: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN %4, %5, 0, 0, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %8.sub1 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN]].sub0 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_BOTHEN]].sub1 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_x_bothen + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_x_bothen + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -911,9 +1952,33 @@ name: gfx9_tbuffer_load_x_xy_bothen body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_x_xy_bothen - ; GFX9: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN %4, %5, 0, 0, 125, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX9: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub1_sub2 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 125, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN]].sub0 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN]].sub1_sub2 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_xy_bothen + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_xy_bothen + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -929,9 +1994,33 @@ name: gfx9_tbuffer_load_xy_xy_bothen body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_xy_xy_bothen - ; GFX9: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN %4, %5, 0, 0, 126, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vreg_64 = COPY %8.sub0_sub1 - ; GFX9: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub2_sub3 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 126, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub0_sub1 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub2_sub3 + ; GFX10-LABEL: name: gfx9_tbuffer_load_xy_xy_bothen + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 8, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_xy_xy_bothen + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 8, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -947,9 +2036,33 @@ name: gfx9_tbuffer_load_x_xyz_bothen body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_x_xyz_bothen - ; GFX9: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN %4, %5, 0, 0, 126, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX9: %{{[0-9]+}}:vreg_96 = COPY killed %8.sub1_sub2_sub3 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 126, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub0 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub1_sub2_sub3 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_xyz_bothen + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_xyz_bothen + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -965,9 +2078,33 @@ name: gfx9_tbuffer_load_x_x_idxen_exact body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_x_x_idxen_exact - ; GFX9: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact %4, %5, 0, 0, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %8.sub1 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub0 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub1 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_x_idxen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_x_idxen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -983,9 +2120,33 @@ name: gfx9_tbuffer_load_x_xy_idxen_exact body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_x_xy_idxen_exact - ; GFX9: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact %4, %5, 0, 0, 125, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX9: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub1_sub2 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 125, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact]].sub0 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact]].sub1_sub2 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_xy_idxen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_xy_idxen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1001,9 +2162,33 @@ name: gfx9_tbuffer_load_xy_xy_idxen_exact body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_xy_xy_idxen_exact - ; GFX9: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact %4, %5, 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vreg_64 = COPY %8.sub0_sub1 - ; GFX9: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub2_sub3 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact]].sub0_sub1 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact]].sub2_sub3 + ; GFX10-LABEL: name: gfx9_tbuffer_load_xy_xy_idxen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_xy_xy_idxen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1019,9 +2204,33 @@ name: gfx9_tbuffer_load_x_xyz_idxen_exact body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_x_xyz_idxen_exact - ; GFX9: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact %4, %5, 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX9: %{{[0-9]+}}:vreg_96 = COPY killed %8.sub1_sub2_sub3 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 126, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact]].sub0 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact]].sub1_sub2_sub3 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_xyz_idxen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 125, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_xyz_idxen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 125, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1037,11 +2246,37 @@ name: gfx9_tbuffer_load_x_x_x_idxen_exact body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_x_x_x_idxen_exact - ; GFX9: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact %4, %5, 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vreg_64 = COPY %10.sub0_sub1 - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %10.sub2 - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY %9.sub0 - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %9.sub1 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact]].sub0_sub1 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact]].sub2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY5]].sub0 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY killed [[COPY5]].sub1 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_x_x_idxen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_x_x_idxen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1058,10 +2293,36 @@ name: gfx9_tbuffer_load_x_x_x_idxen_exact_swizzled_0 body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_x_x_x_idxen_exact_swizzled_0 - ; GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact %4, %5, 0, 4, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact %4, %5, 0, 8, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY %9.sub0 - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %9.sub1 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub0 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub1 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_x_x_idxen_exact_swizzled_0 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_x_x_idxen_exact_swizzled_0 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1078,9 +2339,33 @@ name: gfx9_tbuffer_load_x_x_bothen_exact body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_x_x_bothen_exact - ; GFX9: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact %4, %5, 0, 0, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %8.sub1 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub0 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub1 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_x_bothen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_x_bothen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1096,9 +2381,33 @@ name: gfx9_tbuffer_load_x_xy_bothen_exact body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_x_xy_bothen_exact - ; GFX9: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact %4, %5, 0, 0, 125, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX9: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub1_sub2 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 125, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact]].sub0 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact]].sub1_sub2 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_xy_bothen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_xy_bothen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1113,11 +2422,35 @@ body: | name: gfx9_tbuffer_load_xy_xy_bothen_exact body: | bb.0.entry: - ; GFX9-LABEL: name: gfx9_tbuffer_load_xy_xy_bothen_exact - ; GFX9: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact %4, %5, 0, 0, 126, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vreg_64 = COPY %8.sub0_sub1 - ; GFX9: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub2_sub3 + ; GFX9-LABEL: name: gfx9_tbuffer_load_xy_xy_bothen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 126, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact]].sub0_sub1 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact]].sub2_sub3 + ; GFX10-LABEL: name: gfx9_tbuffer_load_xy_xy_bothen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_xy_xy_bothen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1132,11 +2465,35 @@ body: | name: gfx9_tbuffer_load_x_xyz_bothen_exact body: | bb.0.entry: - ; GFX9-LABEL: name: gfx9_tbuffer_load_x_xyz_bothen_exact - ; GFX9: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact %4, %5, 0, 0, 126, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX9: %{{[0-9]+}}:vreg_96 = COPY killed %8.sub1_sub2_sub3 + ; GFX9-LABEL: name: gfx9_tbuffer_load_x_xyz_bothen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 126, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact]].sub0 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact]].sub1_sub2_sub3 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_xyz_bothen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_xyz_bothen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1152,11 +2509,37 @@ name: gfx9_tbuffer_load_x_x_x_bothen_exact body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_x_x_x_bothen_exact - ; GFX9: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact %4, %5, 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vreg_64 = COPY %10.sub0_sub1 - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %10.sub2 - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY %9.sub0 - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %9.sub1 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 125, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact]].sub0_sub1 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact]].sub2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY5]].sub0 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY killed [[COPY5]].sub1 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_x_x_bothen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_x_x_bothen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1173,10 +2556,36 @@ name: gfx9_tbuffer_load_x_x_x_bothen_exact_swizzled_0 body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_x_x_x_bothen_exact_swizzled_0 - ; GFX9: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact %4, %5, 0, 4, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact %4, %5, 0, 8, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY %9.sub0 - ; GFX9: %{{[0-9]+}}:vgpr_32 = COPY killed %9.sub1 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 123, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub0 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub1 + ; GFX10-LABEL: name: gfx9_tbuffer_load_x_x_x_bothen_exact_swizzled_0 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_x_x_x_bothen_exact_swizzled_0 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 116, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 116, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1193,8 +2602,35 @@ name: gfx9_tbuffer_load_xy_xy_bothen_exact_diff_vaddr body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_xy_xy_bothen_exact_diff_vaddr - ; GFX9: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact %4, %6, 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact %5, %6, 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr1 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_load_xy_xy_bothen_exact_diff_vaddr + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr1 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_xy_xy_bothen_exact_diff_vaddr + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr1 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1211,8 +2647,38 @@ name: gfx9_tbuffer_load_xy_xy_bothen_exact_diff_srsrc body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_xy_xy_bothen_exact_diff_srsrc - ; GFX9: {{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact %5, %6, 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - ; GFX9: {{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact %5, %7, 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE1]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_load_xy_xy_bothen_exact_diff_srsrc + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE1]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_xy_xy_bothen_exact_diff_srsrc + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE1]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1230,8 +2696,35 @@ name: gfx9_tbuffer_load_xy_xy_idxen_exact_diff_vaddr body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_xy_xy_idxen_exact_diff_vaddr - ; GFX9: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact %4, %6, 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - ; GFX9: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact %5, %6, 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_load_xy_xy_idxen_exact_diff_vaddr + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_xy_xy_idxen_exact_diff_vaddr + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1248,8 +2741,38 @@ name: gfx9_tbuffer_load_xy_xy_idxen_exact_diff_srsrc body: | bb.0.entry: ; GFX9-LABEL: name: gfx9_tbuffer_load_xy_xy_idxen_exact_diff_srsrc - ; GFX9: {{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact %5, %6, 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - ; GFX9: {{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact %5, %7, 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE1]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx9_tbuffer_load_xy_xy_idxen_exact_diff_srsrc + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE1]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx9_tbuffer_load_xy_xy_idxen_exact_diff_srsrc + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 4, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE1]], 0, 12, 123, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1267,13 +2790,34 @@ body: | # GFX10 tests # -# GFX10-LABEL: name: gfx10_tbuffer_load_x_xyz -# GFX10: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET %4, 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY %7.sub0 -# GFX10: %{{[0-9]+}}:vreg_96 = COPY killed %7.sub1_sub2_sub3 name: gfx10_tbuffer_load_x_xyz body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_xyz + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 8, 74, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_load_x_xyz + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vreg_96 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub1_sub2_sub3 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_xyz + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 8, 74, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1284,13 +2828,34 @@ body: | ... --- -# GFX10-LABEL: name: gfx10_tbuffer_load_xyz_x -# GFX10: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET %4, 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vreg_96 = COPY %7.sub0_sub1_sub2 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %7.sub3 name: gfx10_tbuffer_load_xyz_x body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_xyz_x + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_load_xyz_x + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_96 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1_sub2 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub3 + ; GFX11-LABEL: name: gfx10_tbuffer_load_xyz_x + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1301,13 +2866,34 @@ body: | ... --- -# GFX10-LABEL: name: gfx10_tbuffer_load_xy_xy -# GFX10: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET %4, 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vreg_64 = COPY %7.sub0_sub1 -# GFX10: %{{[0-9]+}}:vreg_64 = COPY killed %7.sub2_sub3 name: gfx10_tbuffer_load_xy_xy body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_xy_xy + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_load_xy_xy + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub2_sub3 + ; GFX11-LABEL: name: gfx10_tbuffer_load_xy_xy + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1318,13 +2904,34 @@ body: | ... --- -# GFX10-LABEL: name: gfx10_tbuffer_load_x_xy -# GFX10: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET %4, 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY %7.sub0 -# GFX10: %{{[0-9]+}}:vreg_64 = COPY killed %7.sub1_sub2 name: gfx10_tbuffer_load_x_xy body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_xy + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 8, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_load_x_xy + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub1_sub2 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_xy + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 8, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1335,13 +2942,34 @@ body: | ... --- -# GFX10-LABEL: name: gfx10_tbuffer_load_xy_x -# GFX10: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET %4, 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vreg_64 = COPY %7.sub0_sub1 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %7.sub2 name: gfx10_tbuffer_load_xy_x body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_xy_x + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 12, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_load_xy_x + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0_sub1 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub2 + ; GFX11-LABEL: name: gfx10_tbuffer_load_xy_x + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 12, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1352,15 +2980,35 @@ body: | ... --- - -# GFX10-LABEL: name: gfx10_tbuffer_load_x_x -# GFX10: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET %4, 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY %7.sub0 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %7.sub1 - name: gfx10_tbuffer_load_x_x body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_x + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_load_x_x + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_x + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1371,14 +3019,35 @@ body: | ... --- -# GFX10-LABEL: name: gfx10_tbuffer_load_x_x_format_32_32_32_32 -# GFX10: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET %4, 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY %7.sub0 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %7.sub1 - name: gfx10_tbuffer_load_x_x_format_32_32_32_32 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_x_format_32_32_32_32 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 75, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX10-LABEL: name: gfx10_tbuffer_load_x_x_format_32_32_32_32 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_x_format_32_32_32_32 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 77, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1389,27 +3058,66 @@ body: | ... --- - -# GFX10-LABEL: name: gfx10_tbuffer_load_float_32 -# GFX10: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET %4, 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY %14.sub0 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %14.sub1 -# GFX10: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET %4, 0, 16, 77, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vreg_96 = COPY %17.sub0_sub1_sub2 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %17.sub3 -# GFX10: %{{[0-9]+}}:vreg_64 = COPY %16.sub0_sub1 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %16.sub2 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY %15.sub0 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %15.sub1 -# GFX10: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET %4, 0, 36, 74, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vreg_64 = COPY %19.sub0_sub1 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %19.sub2 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY %18.sub0 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %18.sub1 - name: gfx10_tbuffer_load_float_32 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_float_32 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_load_float_32 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 16, 77, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1_sub2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub3 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vreg_64 = COPY [[COPY6]].sub0_sub1 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY killed [[COPY6]].sub2 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[COPY8]].sub0 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY killed [[COPY8]].sub1 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 36, 74, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0_sub1 + ; GFX10-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub2 + ; GFX10-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[COPY12]].sub0 + ; GFX10-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY killed [[COPY12]].sub1 + ; GFX11-LABEL: name: gfx10_tbuffer_load_float_32 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 16, 63, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1_sub2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub3 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vreg_64 = COPY [[COPY6]].sub0_sub1 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY killed [[COPY6]].sub2 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[COPY8]].sub0 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY killed [[COPY8]].sub1 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 36, 60, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0_sub1 + ; GFX11-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub2 + ; GFX11-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[COPY12]].sub0 + ; GFX11-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY killed [[COPY12]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1427,26 +3135,66 @@ body: | ... --- -# GFX10-LABEL: name: gfx10_tbuffer_load_sint_32 -# GFX10: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET %4, 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY %14.sub0 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %14.sub1 -# GFX10: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET %4, 0, 16, 76, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vreg_96 = COPY %17.sub0_sub1_sub2 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %17.sub3 -# GFX10: %{{[0-9]+}}:vreg_64 = COPY %16.sub0_sub1 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %16.sub2 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY %15.sub0 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %15.sub1 -# GFX10: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET %4, 0, 36, 73, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vreg_64 = COPY %19.sub0_sub1 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %19.sub2 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY %18.sub0 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %18.sub1 - name: gfx10_tbuffer_load_sint_32 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_sint_32 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_load_sint_32 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 16, 76, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1_sub2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub3 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vreg_64 = COPY [[COPY6]].sub0_sub1 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY killed [[COPY6]].sub2 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[COPY8]].sub0 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY killed [[COPY8]].sub1 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 36, 73, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0_sub1 + ; GFX10-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub2 + ; GFX10-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[COPY12]].sub0 + ; GFX10-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY killed [[COPY12]].sub1 + ; GFX11-LABEL: name: gfx10_tbuffer_load_sint_32 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 49, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 16, 62, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1_sub2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub3 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vreg_64 = COPY [[COPY6]].sub0_sub1 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY killed [[COPY6]].sub2 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[COPY8]].sub0 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY killed [[COPY8]].sub1 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 36, 59, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0_sub1 + ; GFX11-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub2 + ; GFX11-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[COPY12]].sub0 + ; GFX11-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY killed [[COPY12]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1464,26 +3212,66 @@ body: | ... --- -# GFX10-LABEL: name: gfx10_tbuffer_load_uint_32 -# GFX10: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET %4, 0, 4, 62, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY %14.sub0 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %14.sub1 -# GFX10: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET %4, 0, 16, 75, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vreg_96 = COPY %17.sub0_sub1_sub2 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %17.sub3 -# GFX10: %{{[0-9]+}}:vreg_64 = COPY %16.sub0_sub1 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %16.sub2 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY %15.sub0 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %15.sub1 -# GFX10: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET %4, 0, 36, 72, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vreg_64 = COPY %19.sub0_sub1 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %19.sub2 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY %18.sub0 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %18.sub1 - name: gfx10_tbuffer_load_uint_32 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_uint_32 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 20, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 20, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 20, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 20, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 20, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 20, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 20, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 20, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 20, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_load_uint_32 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 62, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 16, 75, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1_sub2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub3 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vreg_64 = COPY [[COPY6]].sub0_sub1 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY killed [[COPY6]].sub2 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[COPY8]].sub0 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY killed [[COPY8]].sub1 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 36, 72, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0_sub1 + ; GFX10-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub2 + ; GFX10-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[COPY12]].sub0 + ; GFX10-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY killed [[COPY12]].sub1 + ; GFX11-LABEL: name: gfx10_tbuffer_load_uint_32 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 48, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 16, 61, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1_sub2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub3 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vreg_64 = COPY [[COPY6]].sub0_sub1 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY killed [[COPY6]].sub2 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[COPY8]].sub0 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY killed [[COPY8]].sub1 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 36, 58, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0_sub1 + ; GFX11-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub2 + ; GFX11-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[COPY12]].sub0 + ; GFX11-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY killed [[COPY12]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1501,26 +3289,54 @@ body: | ... --- -# GFX10-LABEL: name: gfx10_tbuffer_load_not_merged_data_format_mismatch -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX10: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %0, %subreg.sub0, %1, %subreg.sub1, %2, %subreg.sub2, %3, %subreg.sub3 -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 8, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 20, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 28, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 40, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - - name: gfx10_tbuffer_load_not_merged_data_format_mismatch body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_not_merged_data_format_mismatch + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_load_not_merged_data_format_mismatch + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_load_not_merged_data_format_mismatch + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1538,24 +3354,54 @@ body: | ... --- -# GFX10-LABEL: name: gfx10_tbuffer_load_not_merged_num_format_mismatch -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX10: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %0, %subreg.sub0, %1, %subreg.sub1, %2, %subreg.sub2, %3, %subreg.sub3 -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) name: gfx10_tbuffer_load_not_merged_num_format_mismatch body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_not_merged_num_format_mismatch + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_load_not_merged_num_format_mismatch + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_load_not_merged_num_format_mismatch + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -1573,16 +3419,55 @@ body: | ... --- - - -# GFX10-LABEL: name: gfx10_tbuffer_store_x_xyz -# GFX10: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE %3, %subreg.sub0, %2, %subreg.sub1, %1, %subreg.sub2 -# GFX10: %{{[0-9]+}}:vreg_128 = REG_SEQUENCE %0, %subreg.sub0, %9, %subreg.sub1_sub2_sub3 -# GFX10: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed %10, %8, 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) name: gfx10_tbuffer_store_x_xyz body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx10_tbuffer_store_x_xyz + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 8, 74, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_store_x_xyz + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2 + ; GFX10-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[REG_SEQUENCE1]], %subreg.sub1_sub2_sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_store_x_xyz + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 8, 74, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -1598,15 +3483,55 @@ body: | ... --- - -# GFX10-LABEL: name: gfx10_tbuffer_store_xyz_x -# GFX10: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE %3, %subreg.sub0, %2, %subreg.sub1, %1, %subreg.sub2 -# GFX10: %{{[0-9]+}}:vreg_128 = REG_SEQUENCE %9, %subreg.sub0_sub1_sub2, %0, %subreg.sub3 -# GFX10: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed %10, %8, 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) name: gfx10_tbuffer_store_xyz_x body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx10_tbuffer_store_xyz_x + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_store_xyz_x + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2 + ; GFX10-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[REG_SEQUENCE1]], %subreg.sub0_sub1_sub2, [[COPY]], %subreg.sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_store_xyz_x + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -1622,15 +3547,58 @@ body: | ... --- -# GFX10-LABEL: name: gfx10_tbuffer_store_xy_xy -# GFX10: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %3, %subreg.sub0, %2, %subreg.sub1 -# GFX10: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %1, %subreg.sub0, %0, %subreg.sub1 -# GFX10: %{{[0-9]+}}:vreg_128 = REG_SEQUENCE %9, %subreg.sub0_sub1, %10, %subreg.sub2_sub3 -# GFX10: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed %11, %8, 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) name: gfx10_tbuffer_store_xy_xy body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx10_tbuffer_store_xy_xy + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX9-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_store_xy_xy + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX10-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX10-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[REG_SEQUENCE1]], %subreg.sub0_sub1, [[REG_SEQUENCE2]], %subreg.sub2_sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE3]], [[REG_SEQUENCE]], 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_store_xy_xy + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX11-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -1647,14 +3615,55 @@ body: | ... --- -# GFX10-LABEL: name: gfx10_tbuffer_store_x_xy -# GFX10: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %3, %subreg.sub0, %2, %subreg.sub1 -# GFX10: %{{[0-9]+}}:vreg_64, %subreg.sub1_sub2 -# GFX10: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed %11, %8, 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) name: gfx10_tbuffer_store_x_xy body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx10_tbuffer_store_x_xy + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact %10:vreg_64, [[REG_SEQUENCE]], 0, 8, 64, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_store_x_xy + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX10-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY]], %subreg.sub0, %10:vreg_64, %subreg.sub1_sub2 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_store_x_xy + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact %10:vreg_64, [[REG_SEQUENCE]], 0, 8, 64, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -1670,15 +3679,56 @@ body: | ... --- -# GFX10-LABEL: name: gfx10_tbuffer_store_xy_x -# GFX10: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %3, %subreg.sub0, %2, %subreg.sub1 -# GFX10: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE %9, %subreg.sub0_sub1, %0, %subreg.sub2 -# GFX10: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed %10, %8, 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) name: gfx10_tbuffer_store_xy_x body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx10_tbuffer_store_xy_x + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 12, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_store_xy_x + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX10-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[REG_SEQUENCE1]], %subreg.sub0_sub1, [[COPY]], %subreg.sub2 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_store_xy_x + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 12, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -1694,14 +3744,52 @@ body: | ... --- - -# GFX10-LABEL: name: gfx10_tbuffer_store_x_x -# GFX10: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %1, %subreg.sub0, %0, %subreg.sub1 -# GFX10: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed %9, %8, 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) name: gfx10_tbuffer_store_x_x body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx10_tbuffer_store_x_x + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_store_x_x + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_store_x_x + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -1716,13 +3804,52 @@ body: | ... --- -# GFX10-LABEL: name: gfx10_tbuffer_store_x_x_format_32_32_32_32 -# GFX10: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %1, %subreg.sub0, %0, %subreg.sub1 -# GFX10: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed %9, %8, 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) name: gfx10_tbuffer_store_x_x_format_32_32_32_32 body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx10_tbuffer_store_x_x_format_32_32_32_32 + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 75, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_store_x_x_format_32_32_32_32 + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_store_x_x_format_32_32_32_32 + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 8, 77, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -1737,34 +3864,88 @@ body: | ... --- -# GFX10-LABEL: name: gfx10_tbuffer_store_float32 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr8 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr7 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr6 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr5 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr4 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr3 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr2 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr1 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr0 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX10: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %12, %subreg.sub0, %11, %subreg.sub1, %10, %subreg.sub2, %9, %subreg.sub3 -# GFX10: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %8, %subreg.sub0, %7, %subreg.sub1 -# GFX10: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed %14, %13, 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %6, %subreg.sub0, %5, %subreg.sub1 -# GFX10: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE killed %15, %subreg.sub0_sub1, %4, %subreg.sub2 -# GFX10: %{{[0-9]+}}:vreg_128 = REG_SEQUENCE killed %16, %subreg.sub0_sub1_sub2, %3, %subreg.sub3 -# GFX10: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed %17, %13, 0, 16, 77, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %2, %subreg.sub0, %1, %subreg.sub1 -# GFX10: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE killed %18, %subreg.sub0_sub1, %0, %subreg.sub2 -# GFX10: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed %19, %13, 0, 36, 74, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) name: gfx10_tbuffer_store_float32 body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-LABEL: name: gfx10_tbuffer_store_float32 + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_store_float32 + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1 + ; GFX10-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE2]], %subreg.sub0_sub1, [[COPY4]], %subreg.sub2 + ; GFX10-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[REG_SEQUENCE3]], %subreg.sub0_sub1_sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE4]], [[REG_SEQUENCE]], 0, 16, 77, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX10-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1 + ; GFX10-NEXT: [[REG_SEQUENCE6:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE5]], %subreg.sub0_sub1, [[COPY]], %subreg.sub2 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE6]], [[REG_SEQUENCE]], 0, 36, 74, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_store_float32 + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1 + ; GFX11-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE2]], %subreg.sub0_sub1, [[COPY4]], %subreg.sub2 + ; GFX11-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[REG_SEQUENCE3]], %subreg.sub0_sub1_sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE4]], [[REG_SEQUENCE]], 0, 16, 63, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX11-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1 + ; GFX11-NEXT: [[REG_SEQUENCE6:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE5]], %subreg.sub0_sub1, [[COPY]], %subreg.sub2 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE6]], [[REG_SEQUENCE]], 0, 36, 60, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) %12:vgpr_32 = COPY $vgpr8 %11:vgpr_32 = COPY $vgpr7 %10:vgpr_32 = COPY $vgpr6 @@ -1791,34 +3972,88 @@ body: | ... --- -# GFX10-LABEL: name: gfx10_tbuffer_store_sint32 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr8 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr7 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr6 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr5 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr4 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr3 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr2 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr1 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr0 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX10: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %12, %subreg.sub0, %11, %subreg.sub1, %10, %subreg.sub2, %9, %subreg.sub3 -# GFX10: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %8, %subreg.sub0, %7, %subreg.sub1 -# GFX10: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed %14, %13, 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %6, %subreg.sub0, %5, %subreg.sub1 -# GFX10: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE killed %15, %subreg.sub0_sub1, %4, %subreg.sub2 -# GFX10: %{{[0-9]+}}:vreg_128 = REG_SEQUENCE killed %16, %subreg.sub0_sub1_sub2, %3, %subreg.sub3 -# GFX10: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed %17, %13, 0, 16, 76, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %2, %subreg.sub0, %1, %subreg.sub1 -# GFX10: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE killed %18, %subreg.sub0_sub1, %0, %subreg.sub2 -# GFX10: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed %19, %13, 0, 36, 73, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) name: gfx10_tbuffer_store_sint32 body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-LABEL: name: gfx10_tbuffer_store_sint32 + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_store_sint32 + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1 + ; GFX10-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE2]], %subreg.sub0_sub1, [[COPY4]], %subreg.sub2 + ; GFX10-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[REG_SEQUENCE3]], %subreg.sub0_sub1_sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE4]], [[REG_SEQUENCE]], 0, 16, 76, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX10-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1 + ; GFX10-NEXT: [[REG_SEQUENCE6:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE5]], %subreg.sub0_sub1, [[COPY]], %subreg.sub2 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE6]], [[REG_SEQUENCE]], 0, 36, 73, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_store_sint32 + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 49, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1 + ; GFX11-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE2]], %subreg.sub0_sub1, [[COPY4]], %subreg.sub2 + ; GFX11-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[REG_SEQUENCE3]], %subreg.sub0_sub1_sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE4]], [[REG_SEQUENCE]], 0, 16, 62, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX11-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1 + ; GFX11-NEXT: [[REG_SEQUENCE6:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE5]], %subreg.sub0_sub1, [[COPY]], %subreg.sub2 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE6]], [[REG_SEQUENCE]], 0, 36, 59, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) %12:vgpr_32 = COPY $vgpr8 %11:vgpr_32 = COPY $vgpr7 %10:vgpr_32 = COPY $vgpr6 @@ -1845,34 +4080,88 @@ body: | ... --- -# GFX10-LABEL: name: gfx10_tbuffer_store_uint32 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr8 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr7 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr6 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr5 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr4 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr3 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr2 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr1 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr0 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX10: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %12, %subreg.sub0, %11, %subreg.sub1, %10, %subreg.sub2, %9, %subreg.sub3 -# GFX10: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %8, %subreg.sub0, %7, %subreg.sub1 -# GFX10: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed %14, %13, 0, 4, 62, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %6, %subreg.sub0, %5, %subreg.sub1 -# GFX10: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE killed %15, %subreg.sub0_sub1, %4, %subreg.sub2 -# GFX10: %{{[0-9]+}}:vreg_128 = REG_SEQUENCE killed %16, %subreg.sub0_sub1_sub2, %3, %subreg.sub3 -# GFX10: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed %17, %13, 0, 16, 75, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %2, %subreg.sub0, %1, %subreg.sub1 -# GFX10: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE killed %18, %subreg.sub0_sub1, %0, %subreg.sub2 -# GFX10: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed %19, %13, 0, 36, 72, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) name: gfx10_tbuffer_store_uint32 body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-LABEL: name: gfx10_tbuffer_store_uint32 + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 20, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 20, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 20, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 20, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 20, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 20, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 20, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 20, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 20, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_store_uint32 + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 62, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1 + ; GFX10-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE2]], %subreg.sub0_sub1, [[COPY4]], %subreg.sub2 + ; GFX10-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[REG_SEQUENCE3]], %subreg.sub0_sub1_sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE4]], [[REG_SEQUENCE]], 0, 16, 75, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX10-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1 + ; GFX10-NEXT: [[REG_SEQUENCE6:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE5]], %subreg.sub0_sub1, [[COPY]], %subreg.sub2 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE6]], [[REG_SEQUENCE]], 0, 36, 72, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_store_uint32 + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 48, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1 + ; GFX11-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE2]], %subreg.sub0_sub1, [[COPY4]], %subreg.sub2 + ; GFX11-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[REG_SEQUENCE3]], %subreg.sub0_sub1_sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE4]], [[REG_SEQUENCE]], 0, 16, 61, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX11-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1 + ; GFX11-NEXT: [[REG_SEQUENCE6:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE5]], %subreg.sub0_sub1, [[COPY]], %subreg.sub2 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE6]], [[REG_SEQUENCE]], 0, 36, 58, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) %12:vgpr_32 = COPY $vgpr8 %11:vgpr_32 = COPY $vgpr7 %10:vgpr_32 = COPY $vgpr6 @@ -1899,34 +4188,88 @@ body: | ... --- -# GFX10-LABEL: name: gfx10_tbuffer_store_not_merged_data_format_mismatch -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr8 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr7 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr6 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr5 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr4 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr3 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr2 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr1 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr0 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX10: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %12, %subreg.sub0, %11, %subreg.sub1, %10, %subreg.sub2, %9, %subreg.sub3 -# GFX10: TBUFFER_STORE_FORMAT_X_OFFSET_exact %8, %13, 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX10: TBUFFER_STORE_FORMAT_X_OFFSET_exact %7, %13, 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX10: TBUFFER_STORE_FORMAT_X_OFFSET_exact %6, %13, 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX10: TBUFFER_STORE_FORMAT_X_OFFSET_exact %5, %13, 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX10: TBUFFER_STORE_FORMAT_X_OFFSET_exact %4, %13, 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX10: TBUFFER_STORE_FORMAT_X_OFFSET_exact %3, %13, 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX10: TBUFFER_STORE_FORMAT_X_OFFSET_exact %2, %13, 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX10: TBUFFER_STORE_FORMAT_X_OFFSET_exact %1, %13, 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX10: TBUFFER_STORE_FORMAT_X_OFFSET_exact %0, %13, 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) name: gfx10_tbuffer_store_not_merged_data_format_mismatch body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-LABEL: name: gfx10_tbuffer_store_not_merged_data_format_mismatch + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_store_not_merged_data_format_mismatch + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_store_not_merged_data_format_mismatch + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) %12:vgpr_32 = COPY $vgpr8 %11:vgpr_32 = COPY $vgpr7 %10:vgpr_32 = COPY $vgpr6 @@ -1953,34 +4296,88 @@ body: | ... --- -# GFX10-LABEL: name: gfx10_tbuffer_store_not_merged_num_format_mismatch -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr8 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr7 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr6 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr5 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr4 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr3 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr2 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr1 -# GFX10: %{{[0-9]+}}:vgpr_32 = COPY $vgpr0 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX10: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX10: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %12, %subreg.sub0, %11, %subreg.sub1, %10, %subreg.sub2, %9, %subreg.sub3 -# GFX10: TBUFFER_STORE_FORMAT_X_OFFSET_exact %8, %13, 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX10: TBUFFER_STORE_FORMAT_X_OFFSET_exact %7, %13, 0, 8, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX10: TBUFFER_STORE_FORMAT_X_OFFSET_exact %6, %13, 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX10: TBUFFER_STORE_FORMAT_X_OFFSET_exact %5, %13, 0, 20, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX10: TBUFFER_STORE_FORMAT_X_OFFSET_exact %4, %13, 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX10: TBUFFER_STORE_FORMAT_X_OFFSET_exact %3, %13, 0, 28, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX10: TBUFFER_STORE_FORMAT_X_OFFSET_exact %2, %13, 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX10: TBUFFER_STORE_FORMAT_X_OFFSET_exact %1, %13, 0, 40, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX10: TBUFFER_STORE_FORMAT_X_OFFSET_exact %0, %13, 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) name: gfx10_tbuffer_store_not_merged_num_format_mismatch body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-LABEL: name: gfx10_tbuffer_store_not_merged_num_format_mismatch + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_store_not_merged_num_format_mismatch + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_store_not_merged_num_format_mismatch + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) %12:vgpr_32 = COPY $vgpr8 %11:vgpr_32 = COPY $vgpr7 %10:vgpr_32 = COPY $vgpr6 @@ -2007,13 +4404,33 @@ body: | ... --- - -# GFX10-LABEL: name: gfx10_tbuffer_load_not_merged_swizzled_0 -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) name: gfx10_tbuffer_load_not_merged_swizzled_0 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_not_merged_swizzled_0 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_load_not_merged_swizzled_0 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_load_not_merged_swizzled_0 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2024,13 +4441,33 @@ body: | ... --- - -# GFX10-LABEL: name: gfx10_tbuffer_load_not_merged_swizzled_1 -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 8, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) name: gfx10_tbuffer_load_not_merged_swizzled_1 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_not_merged_swizzled_1 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_load_not_merged_swizzled_1 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_load_not_merged_swizzled_1 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2041,13 +4478,38 @@ body: | ... --- - -# GFX10-LABEL: name: gfx10_tbuffer_load_merge_across_swizzle -# GFX10-DAG: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 12, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX10-DAG: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET %4, 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) name: gfx10_tbuffer_load_merge_across_swizzle body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_merge_across_swizzle + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 12, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_load_merge_across_swizzle + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 12, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_load_merge_across_swizzle + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 12, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2062,10 +4524,35 @@ body: | name: gfx10_tbuffer_load_x_x_idxen body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_x_idxen + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_x_x_idxen - ; GFX10: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN %4, %5, 0, 0, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %8.sub1 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_IDXEN]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_IDXEN]].sub1 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_x_idxen + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_IDXEN]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_IDXEN]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2080,10 +4567,34 @@ body: | name: gfx10_tbuffer_load_x_xy_idxen body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_xy_idxen + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_x_xy_idxen - ; GFX10: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN %4, %5, 0, 0, 74, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX10: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub1_sub2 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 74, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN]].sub1_sub2 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_xy_idxen + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2098,10 +4609,34 @@ body: | name: gfx10_tbuffer_load_xy_xy_idxen body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_xy_xy_idxen + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_xy_xy_idxen - ; GFX10: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN %4, %5, 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vreg_64 = COPY %8.sub0_sub1 - ; GFX10: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub2_sub3 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN]].sub0_sub1 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN]].sub2_sub3 + ; GFX11-LABEL: name: gfx10_tbuffer_load_xy_xy_idxen + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2116,10 +4651,34 @@ body: | name: gfx10_tbuffer_load_x_xyz_idxen body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_xyz_idxen + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 8, 74, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_x_xyz_idxen - ; GFX10: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN %4, %5, 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX10: %{{[0-9]+}}:vreg_96 = COPY killed %8.sub1_sub2_sub3 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN]].sub1_sub2_sub3 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_xyz_idxen + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 8, 74, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2134,12 +4693,40 @@ body: | name: gfx10_tbuffer_load_x_x_x_idxen_exact body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_x_x_idxen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_x_x_x_idxen_exact - ; GFX10: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact %4, %5, 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vreg_64 = COPY %10.sub0_sub1 - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %10.sub2 - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY %9.sub0 - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %9.sub1 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact]].sub0_sub1 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact]].sub2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY5]].sub0 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY killed [[COPY5]].sub1 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_x_x_idxen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact]].sub0_sub1 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact]].sub2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY5]].sub0 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY killed [[COPY5]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2155,11 +4742,38 @@ body: | name: gfx10_tbuffer_load_x_x_x_idxen_exact_swizzled_0 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_x_x_idxen_exact_swizzled_0 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_x_x_x_idxen_exact_swizzled_0 - ; GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact %4, %5, 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact %4, %5, 0, 8, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY %9.sub0 - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %9.sub1 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub1 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_x_x_idxen_exact_swizzled_0 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2175,10 +4789,35 @@ body: | name: gfx10_tbuffer_load_x_x_bothen body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_x_bothen + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_x_x_bothen - ; GFX10: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN %4, %5, 0, 0, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %8.sub1 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_BOTHEN]].sub1 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_x_bothen + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_BOTHEN]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2193,10 +4832,34 @@ body: | name: gfx10_tbuffer_load_x_xy_bothen body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_xy_bothen + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_x_xy_bothen - ; GFX10: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN %4, %5, 0, 0, 74, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX10: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub1_sub2 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 74, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN]].sub1_sub2 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_xy_bothen + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2211,10 +4874,34 @@ body: | name: gfx10_tbuffer_load_xy_xy_bothen body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_xy_xy_bothen + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 8, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_xy_xy_bothen - ; GFX10: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN %4, %5, 0, 0, 77, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vreg_64 = COPY %8.sub0_sub1 - ; GFX10: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub2_sub3 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 77, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub0_sub1 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub2_sub3 + ; GFX11-LABEL: name: gfx10_tbuffer_load_xy_xy_bothen + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 8, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2229,10 +4916,34 @@ body: | name: gfx10_tbuffer_load_x_xyz_bothen body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_xyz_bothen + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_x_xyz_bothen - ; GFX10: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN %4, %5, 0, 0, 77, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX10: %{{[0-9]+}}:vreg_96 = COPY killed %8.sub1_sub2_sub3 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 77, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub1_sub2_sub3 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_xyz_bothen + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2247,10 +4958,35 @@ body: | name: gfx10_tbuffer_load_x_x_idxen_exact body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_x_idxen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_x_x_idxen_exact - ; GFX10: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact %4, %5, 0, 0, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %8.sub1 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub1 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_x_idxen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2265,10 +5001,34 @@ body: | name: gfx10_tbuffer_load_x_xy_idxen_exact body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_xy_idxen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_x_xy_idxen_exact - ; GFX10: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact %4, %5, 0, 0, 74, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX10: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub1_sub2 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 74, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact]].sub1_sub2 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_xy_idxen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2283,10 +5043,34 @@ body: | name: gfx10_tbuffer_load_xy_xy_idxen_exact body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_xy_xy_idxen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_xy_xy_idxen_exact - ; GFX10: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact %4, %5, 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vreg_64 = COPY %8.sub0_sub1 - ; GFX10: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub2_sub3 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact]].sub0_sub1 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact]].sub2_sub3 + ; GFX11-LABEL: name: gfx10_tbuffer_load_xy_xy_idxen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2301,10 +5085,34 @@ body: | name: gfx10_tbuffer_load_x_xyz_idxen_exact body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_xyz_idxen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 74, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_x_xyz_idxen_exact - ; GFX10: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact %4, %5, 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX10: %{{[0-9]+}}:vreg_96 = COPY killed %8.sub1_sub2_sub3 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 77, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact]].sub1_sub2_sub3 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_xyz_idxen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 74, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2319,10 +5127,35 @@ body: | name: gfx10_tbuffer_load_x_x_bothen_exact body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_x_bothen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_x_x_bothen_exact - ; GFX10: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact %4, %5, 0, 0, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %8.sub1 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub1 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_x_bothen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2337,10 +5170,34 @@ body: | name: gfx10_tbuffer_load_x_xy_bothen_exact body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_xy_bothen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_x_xy_bothen_exact - ; GFX10: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact %4, %5, 0, 0, 74, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX10: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub1_sub2 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 74, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact]].sub1_sub2 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_xy_bothen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2355,11 +5212,35 @@ body: | name: gfx10_tbuffer_load_xy_xy_bothen_exact body: | bb.0.entry: - ; GFX10-LABEL: name: gfx10_tbuffer_load_xy_xy_bothen_exact - ; GFX10: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact %4, %5, 0, 0, 77, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vreg_64 = COPY %8.sub0_sub1 - ; GFX10: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub2_sub3 + ; GFX9-LABEL: name: gfx10_tbuffer_load_xy_xy_bothen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_load_xy_xy_bothen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 77, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact]].sub0_sub1 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact]].sub2_sub3 + ; GFX11-LABEL: name: gfx10_tbuffer_load_xy_xy_bothen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2374,11 +5255,35 @@ body: | name: gfx10_tbuffer_load_x_xyz_bothen_exact body: | bb.0.entry: - ; GFX10-LABEL: name: gfx10_tbuffer_load_x_xyz_bothen_exact - ; GFX10: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact %4, %5, 0, 0, 77, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX10: %{{[0-9]+}}:vreg_96 = COPY killed %8.sub1_sub2_sub3 + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_xyz_bothen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx10_tbuffer_load_x_xyz_bothen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 77, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact]].sub1_sub2_sub3 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_xyz_bothen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2393,12 +5298,40 @@ body: | name: gfx10_tbuffer_load_x_x_x_bothen_exact body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_x_x_bothen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_x_x_x_bothen_exact - ; GFX10: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact %4, %5, 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vreg_64 = COPY %10.sub0_sub1 - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %10.sub2 - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY %9.sub0 - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %9.sub1 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact]].sub0_sub1 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact]].sub2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY5]].sub0 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY killed [[COPY5]].sub1 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_x_x_bothen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact]].sub0_sub1 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact]].sub2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY5]].sub0 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY killed [[COPY5]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2414,11 +5347,38 @@ body: | name: gfx10_tbuffer_load_x_x_x_bothen_exact_swizzled_0 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_x_x_x_bothen_exact_swizzled_0 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_x_x_x_bothen_exact_swizzled_0 - ; GFX10: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact %4, %5, 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact %4, %5, 0, 8, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY %9.sub0 - ; GFX10: %{{[0-9]+}}:vgpr_32 = COPY killed %9.sub1 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub1 + ; GFX11-LABEL: name: gfx10_tbuffer_load_x_x_x_bothen_exact_swizzled_0 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2434,9 +5394,36 @@ body: | name: gfx10_tbuffer_load_xy_xy_bothen_exact_diff_vaddr body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_xy_xy_bothen_exact_diff_vaddr + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr1 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_xy_xy_bothen_exact_diff_vaddr - ; GFX10: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact %4, %6, 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact %5, %6, 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr1 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_load_xy_xy_bothen_exact_diff_vaddr + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr1 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2452,9 +5439,39 @@ body: | name: gfx10_tbuffer_load_xy_xy_bothen_exact_diff_srsrc body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_xy_xy_bothen_exact_diff_srsrc + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE1]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_xy_xy_bothen_exact_diff_srsrc - ; GFX10: {{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact %5, %6, 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - ; GFX10: {{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact %5, %7, 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE1]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_load_xy_xy_bothen_exact_diff_srsrc + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE1]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2471,9 +5488,36 @@ body: | name: gfx10_tbuffer_load_xy_xy_idxen_exact_diff_vaddr body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_xy_xy_idxen_exact_diff_vaddr + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_xy_xy_idxen_exact_diff_vaddr - ; GFX10: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact %4, %6, 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - ; GFX10: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact %5, %6, 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_load_xy_xy_idxen_exact_diff_vaddr + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2489,9 +5533,39 @@ body: | name: gfx10_tbuffer_load_xy_xy_idxen_exact_diff_srsrc body: | bb.0.entry: + ; GFX9-LABEL: name: gfx10_tbuffer_load_xy_xy_idxen_exact_diff_srsrc + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE1]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX10-LABEL: name: gfx10_tbuffer_load_xy_xy_idxen_exact_diff_srsrc - ; GFX10: {{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact %5, %6, 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - ; GFX10: {{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact %5, %7, 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE1]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx10_tbuffer_load_xy_xy_idxen_exact_diff_srsrc + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE1]], 0, 12, 64, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2510,13 +5584,34 @@ body: | # GFX11 tests # -# GFX11-LABEL: name: gfx11_tbuffer_load_x_xyz -# GFX11: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET %4, 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY %7.sub0 -# GFX11: %{{[0-9]+}}:vreg_96 = COPY killed %7.sub1_sub2_sub3 name: gfx11_tbuffer_load_x_xyz body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_xyz + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 8, 60, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_xyz + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 8, 60, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_load_x_xyz + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_96 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub1_sub2_sub3 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2527,13 +5622,34 @@ body: | ... --- -# GFX11-LABEL: name: gfx11_tbuffer_load_xyz_x -# GFX11: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET %4, 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vreg_96 = COPY %7.sub0_sub1_sub2 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %7.sub3 name: gfx11_tbuffer_load_xyz_x body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_xyz_x + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_xyz_x + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_load_xyz_x + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_96 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1_sub2 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub3 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2544,13 +5660,34 @@ body: | ... --- -# GFX11-LABEL: name: gfx11_tbuffer_load_xy_xy -# GFX11: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET %4, 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vreg_64 = COPY %7.sub0_sub1 -# GFX11: %{{[0-9]+}}:vreg_64 = COPY killed %7.sub2_sub3 name: gfx11_tbuffer_load_xy_xy body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_xy_xy + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_xy_xy + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_load_xy_xy + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub2_sub3 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2561,13 +5698,34 @@ body: | ... --- -# GFX11-LABEL: name: gfx11_tbuffer_load_x_xy -# GFX11: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET %4, 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY %7.sub0 -# GFX11: %{{[0-9]+}}:vreg_64 = COPY killed %7.sub1_sub2 name: gfx11_tbuffer_load_x_xy body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_xy + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 8, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_xy + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 8, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_load_x_xy + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub1_sub2 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2578,13 +5736,34 @@ body: | ... --- -# GFX11-LABEL: name: gfx11_tbuffer_load_xy_x -# GFX11: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET %4, 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vreg_64 = COPY %7.sub0_sub1 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %7.sub2 name: gfx11_tbuffer_load_xy_x body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_xy_x + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 12, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_xy_x + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 12, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_load_xy_x + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0_sub1 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub2 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2595,15 +5774,35 @@ body: | ... --- - -# GFX11-LABEL: name: gfx11_tbuffer_load_x_x -# GFX11: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET %4, 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY %7.sub0 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %7.sub1 - name: gfx11_tbuffer_load_x_x body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_x + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_x + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX11-LABEL: name: gfx11_tbuffer_load_x_x + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2614,14 +5813,35 @@ body: | ... --- -# GFX11-LABEL: name: gfx11_tbuffer_load_x_x_format_32_32_32_32 -# GFX11: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET %4, 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY %7.sub0 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %7.sub1 - name: gfx11_tbuffer_load_x_x_format_32_32_32_32 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_x_format_32_32_32_32 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 63, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_x_format_32_32_32_32 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX11-LABEL: name: gfx11_tbuffer_load_x_x_format_32_32_32_32 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2632,27 +5852,66 @@ body: | ... --- - -# GFX11-LABEL: name: gfx11_tbuffer_load_float_32 -# GFX11: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET %4, 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY %14.sub0 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %14.sub1 -# GFX11: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET %4, 0, 16, 63, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vreg_96 = COPY %17.sub0_sub1_sub2 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %17.sub3 -# GFX11: %{{[0-9]+}}:vreg_64 = COPY %16.sub0_sub1 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %16.sub2 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY %15.sub0 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %15.sub1 -# GFX11: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET %4, 0, 36, 60, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vreg_64 = COPY %19.sub0_sub1 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %19.sub2 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY %18.sub0 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %18.sub1 - name: gfx11_tbuffer_load_float_32 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_float_32 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_float_32 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 16, 77, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1_sub2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub3 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vreg_64 = COPY [[COPY6]].sub0_sub1 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY killed [[COPY6]].sub2 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[COPY8]].sub0 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY killed [[COPY8]].sub1 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 36, 74, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0_sub1 + ; GFX10-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub2 + ; GFX10-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[COPY12]].sub0 + ; GFX10-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY killed [[COPY12]].sub1 + ; GFX11-LABEL: name: gfx11_tbuffer_load_float_32 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 16, 63, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1_sub2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub3 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vreg_64 = COPY [[COPY6]].sub0_sub1 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY killed [[COPY6]].sub2 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[COPY8]].sub0 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY killed [[COPY8]].sub1 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 36, 60, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0_sub1 + ; GFX11-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub2 + ; GFX11-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[COPY12]].sub0 + ; GFX11-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY killed [[COPY12]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2670,26 +5929,66 @@ body: | ... --- -# GFX11-LABEL: name: gfx11_tbuffer_load_sint_32 -# GFX11: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET %4, 0, 4, 49, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY %14.sub0 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %14.sub1 -# GFX11: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET %4, 0, 16, 62, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vreg_96 = COPY %17.sub0_sub1_sub2 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %17.sub3 -# GFX11: %{{[0-9]+}}:vreg_64 = COPY %16.sub0_sub1 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %16.sub2 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY %15.sub0 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %15.sub1 -# GFX11: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET %4, 0, 36, 59, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vreg_64 = COPY %19.sub0_sub1 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %19.sub2 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY %18.sub0 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %18.sub1 - name: gfx11_tbuffer_load_sint_32 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_sint_32 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_sint_32 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 16, 76, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1_sub2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub3 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vreg_64 = COPY [[COPY6]].sub0_sub1 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY killed [[COPY6]].sub2 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[COPY8]].sub0 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY killed [[COPY8]].sub1 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 36, 73, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0_sub1 + ; GFX10-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub2 + ; GFX10-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[COPY12]].sub0 + ; GFX10-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY killed [[COPY12]].sub1 + ; GFX11-LABEL: name: gfx11_tbuffer_load_sint_32 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 49, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 16, 62, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1_sub2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub3 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vreg_64 = COPY [[COPY6]].sub0_sub1 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY killed [[COPY6]].sub2 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[COPY8]].sub0 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY killed [[COPY8]].sub1 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 36, 59, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0_sub1 + ; GFX11-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub2 + ; GFX11-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[COPY12]].sub0 + ; GFX11-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY killed [[COPY12]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2707,26 +6006,66 @@ body: | ... --- -# GFX11-LABEL: name: gfx11_tbuffer_load_uint_32 -# GFX11: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET %4, 0, 4, 48, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY %14.sub0 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %14.sub1 -# GFX11: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET %4, 0, 16, 61, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vreg_96 = COPY %17.sub0_sub1_sub2 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %17.sub3 -# GFX11: %{{[0-9]+}}:vreg_64 = COPY %16.sub0_sub1 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %16.sub2 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY %15.sub0 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %15.sub1 -# GFX11: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET %4, 0, 36, 58, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vreg_64 = COPY %19.sub0_sub1 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %19.sub2 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY %18.sub0 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %18.sub1 - name: gfx11_tbuffer_load_uint_32 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_uint_32 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 20, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 20, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 20, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 20, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 20, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 20, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 20, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 20, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 20, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_uint_32 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 62, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 16, 75, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1_sub2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub3 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vreg_64 = COPY [[COPY6]].sub0_sub1 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY killed [[COPY6]].sub2 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[COPY8]].sub0 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY killed [[COPY8]].sub1 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 36, 72, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0_sub1 + ; GFX10-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub2 + ; GFX10-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[COPY12]].sub0 + ; GFX10-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY killed [[COPY12]].sub1 + ; GFX11-LABEL: name: gfx11_tbuffer_load_uint_32 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 48, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFSET [[REG_SEQUENCE]], 0, 16, 61, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub0_sub1_sub2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_OFFSET]].sub3 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vreg_64 = COPY [[COPY6]].sub0_sub1 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY killed [[COPY6]].sub2 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[COPY8]].sub0 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY killed [[COPY8]].sub1 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFSET [[REG_SEQUENCE]], 0, 36, 58, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub0_sub1 + ; GFX11-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_OFFSET]].sub2 + ; GFX11-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[COPY12]].sub0 + ; GFX11-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY killed [[COPY12]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2744,26 +6083,54 @@ body: | ... --- -# GFX11-LABEL: name: gfx11_tbuffer_load_not_merged_data_format_mismatch -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX11: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %0, %subreg.sub0, %1, %subreg.sub1, %2, %subreg.sub2, %3, %subreg.sub3 -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 8, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 20, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 28, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 40, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - - name: gfx11_tbuffer_load_not_merged_data_format_mismatch body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_not_merged_data_format_mismatch + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_not_merged_data_format_mismatch + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_load_not_merged_data_format_mismatch + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 13, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2781,24 +6148,54 @@ body: | ... --- -# GFX11-LABEL: name: gfx11_tbuffer_load_not_merged_num_format_mismatch -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX11: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %0, %subreg.sub0, %1, %subreg.sub1, %2, %subreg.sub2, %3, %subreg.sub3 -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) name: gfx11_tbuffer_load_not_merged_num_format_mismatch body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_not_merged_num_format_mismatch + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_not_merged_num_format_mismatch + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_load_not_merged_num_format_mismatch + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET3:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET4:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET5:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET6:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET7:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET8:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -2816,16 +6213,55 @@ body: | ... --- - - -# GFX11-LABEL: name: gfx11_tbuffer_store_x_xyz -# GFX11: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE %3, %subreg.sub0, %2, %subreg.sub1, %1, %subreg.sub2 -# GFX11: %{{[0-9]+}}:vreg_128 = REG_SEQUENCE %0, %subreg.sub0, %9, %subreg.sub1_sub2_sub3 -# GFX11: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed %10, %8, 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) name: gfx11_tbuffer_store_x_xyz body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx11_tbuffer_store_x_xyz + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 8, 60, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_store_x_xyz + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 8, 60, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_store_x_xyz + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2 + ; GFX11-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[REG_SEQUENCE1]], %subreg.sub1_sub2_sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -2841,15 +6277,55 @@ body: | ... --- - -# GFX11-LABEL: name: gfx11_tbuffer_store_xyz_x -# GFX11: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE %3, %subreg.sub0, %2, %subreg.sub1, %1, %subreg.sub2 -# GFX11: %{{[0-9]+}}:vreg_128 = REG_SEQUENCE %9, %subreg.sub0_sub1_sub2, %0, %subreg.sub3 -# GFX11: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed %10, %8, 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) name: gfx11_tbuffer_store_xyz_x body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx11_tbuffer_store_xyz_x + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_store_xyz_x + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_store_xyz_x + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY1]], %subreg.sub2 + ; GFX11-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[REG_SEQUENCE1]], %subreg.sub0_sub1_sub2, [[COPY]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -2865,15 +6341,58 @@ body: | ... --- -# GFX11-LABEL: name: gfx11_tbuffer_store_xy_xy -# GFX11: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %3, %subreg.sub0, %2, %subreg.sub1 -# GFX11: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %1, %subreg.sub0, %0, %subreg.sub1 -# GFX11: %{{[0-9]+}}:vreg_128 = REG_SEQUENCE %9, %subreg.sub0_sub1, %10, %subreg.sub2_sub3 -# GFX11: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed %11, %8, 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) name: gfx11_tbuffer_store_xy_xy body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx11_tbuffer_store_xy_xy + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX9-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_store_xy_xy + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX10-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_store_xy_xy + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX11-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX11-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[REG_SEQUENCE1]], %subreg.sub0_sub1, [[REG_SEQUENCE2]], %subreg.sub2_sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE3]], [[REG_SEQUENCE]], 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -2890,14 +6409,55 @@ body: | ... --- -# GFX11-LABEL: name: gfx11_tbuffer_store_x_xy -# GFX11: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %3, %subreg.sub0, %2, %subreg.sub1 -# GFX11: %{{[0-9]+}}:vreg_64, %subreg.sub1_sub2 -# GFX11: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed %11, %8, 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) name: gfx11_tbuffer_store_x_xy body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx11_tbuffer_store_x_xy + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact %10:vreg_64, [[REG_SEQUENCE]], 0, 8, 50, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_store_x_xy + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact %10:vreg_64, [[REG_SEQUENCE]], 0, 8, 50, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_store_x_xy + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX11-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY]], %subreg.sub0, %10:vreg_64, %subreg.sub1_sub2 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -2913,15 +6473,56 @@ body: | ... --- -# GFX11-LABEL: name: gfx11_tbuffer_store_xy_x -# GFX11: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %3, %subreg.sub0, %2, %subreg.sub1 -# GFX11: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE %9, %subreg.sub0_sub1, %0, %subreg.sub2 -# GFX11: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed %10, %8, 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) name: gfx11_tbuffer_store_xy_x body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx11_tbuffer_store_xy_x + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 12, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_store_xy_x + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 12, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_store_xy_x + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1 + ; GFX11-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[REG_SEQUENCE1]], %subreg.sub0_sub1, [[COPY]], %subreg.sub2 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -2937,14 +6538,52 @@ body: | ... --- - -# GFX11-LABEL: name: gfx11_tbuffer_store_x_x -# GFX11: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %1, %subreg.sub0, %0, %subreg.sub1 -# GFX11: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed %9, %8, 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) name: gfx11_tbuffer_store_x_x body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx11_tbuffer_store_x_x + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_store_x_x + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_store_x_x + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -2959,13 +6598,52 @@ body: | ... --- -# GFX11-LABEL: name: gfx11_tbuffer_store_x_x_format_32_32_32_32 -# GFX11: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %1, %subreg.sub0, %0, %subreg.sub1 -# GFX11: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed %9, %8, 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) name: gfx11_tbuffer_store_x_x_format_32_32_32_32 body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-LABEL: name: gfx11_tbuffer_store_x_x_format_32_32_32_32 + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 8, 63, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_store_x_x_format_32_32_32_32 + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_store_x_x_format_32_32_32_32 + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) %7:vgpr_32 = COPY $vgpr3 %6:vgpr_32 = COPY $vgpr2 %5:vgpr_32 = COPY $vgpr1 @@ -2980,34 +6658,88 @@ body: | ... --- -# GFX11-LABEL: name: gfx11_tbuffer_store_float32 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr8 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr7 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr6 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr5 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr4 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr3 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr2 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr1 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr0 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX11: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %12, %subreg.sub0, %11, %subreg.sub1, %10, %subreg.sub2, %9, %subreg.sub3 -# GFX11: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %8, %subreg.sub0, %7, %subreg.sub1 -# GFX11: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed %14, %13, 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %6, %subreg.sub0, %5, %subreg.sub1 -# GFX11: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE killed %15, %subreg.sub0_sub1, %4, %subreg.sub2 -# GFX11: %{{[0-9]+}}:vreg_128 = REG_SEQUENCE killed %16, %subreg.sub0_sub1_sub2, %3, %subreg.sub3 -# GFX11: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed %17, %13, 0, 16, 63, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %2, %subreg.sub0, %1, %subreg.sub1 -# GFX11: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE killed %18, %subreg.sub0_sub1, %0, %subreg.sub2 -# GFX11: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed %19, %13, 0, 36, 60, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) name: gfx11_tbuffer_store_float32 body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-LABEL: name: gfx11_tbuffer_store_float32 + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_store_float32 + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1 + ; GFX10-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE2]], %subreg.sub0_sub1, [[COPY4]], %subreg.sub2 + ; GFX10-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[REG_SEQUENCE3]], %subreg.sub0_sub1_sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE4]], [[REG_SEQUENCE]], 0, 16, 77, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX10-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1 + ; GFX10-NEXT: [[REG_SEQUENCE6:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE5]], %subreg.sub0_sub1, [[COPY]], %subreg.sub2 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE6]], [[REG_SEQUENCE]], 0, 36, 74, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_store_float32 + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1 + ; GFX11-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE2]], %subreg.sub0_sub1, [[COPY4]], %subreg.sub2 + ; GFX11-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[REG_SEQUENCE3]], %subreg.sub0_sub1_sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE4]], [[REG_SEQUENCE]], 0, 16, 63, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX11-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1 + ; GFX11-NEXT: [[REG_SEQUENCE6:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE5]], %subreg.sub0_sub1, [[COPY]], %subreg.sub2 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE6]], [[REG_SEQUENCE]], 0, 36, 60, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) %12:vgpr_32 = COPY $vgpr8 %11:vgpr_32 = COPY $vgpr7 %10:vgpr_32 = COPY $vgpr6 @@ -3034,34 +6766,88 @@ body: | ... --- -# GFX11-LABEL: name: gfx11_tbuffer_store_sint32 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr8 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr7 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr6 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr5 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr4 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr3 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr2 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr1 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr0 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX11: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %12, %subreg.sub0, %11, %subreg.sub1, %10, %subreg.sub2, %9, %subreg.sub3 -# GFX11: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %8, %subreg.sub0, %7, %subreg.sub1 -# GFX11: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed %14, %13, 0, 4, 49, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %6, %subreg.sub0, %5, %subreg.sub1 -# GFX11: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE killed %15, %subreg.sub0_sub1, %4, %subreg.sub2 -# GFX11: %{{[0-9]+}}:vreg_128 = REG_SEQUENCE killed %16, %subreg.sub0_sub1_sub2, %3, %subreg.sub3 -# GFX11: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed %17, %13, 0, 16, 62, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %2, %subreg.sub0, %1, %subreg.sub1 -# GFX11: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE killed %18, %subreg.sub0_sub1, %0, %subreg.sub2 -# GFX11: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed %19, %13, 0, 36, 59, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) name: gfx11_tbuffer_store_sint32 body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-LABEL: name: gfx11_tbuffer_store_sint32 + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_store_sint32 + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1 + ; GFX10-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE2]], %subreg.sub0_sub1, [[COPY4]], %subreg.sub2 + ; GFX10-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[REG_SEQUENCE3]], %subreg.sub0_sub1_sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE4]], [[REG_SEQUENCE]], 0, 16, 76, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX10-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1 + ; GFX10-NEXT: [[REG_SEQUENCE6:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE5]], %subreg.sub0_sub1, [[COPY]], %subreg.sub2 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE6]], [[REG_SEQUENCE]], 0, 36, 73, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_store_sint32 + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 49, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1 + ; GFX11-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE2]], %subreg.sub0_sub1, [[COPY4]], %subreg.sub2 + ; GFX11-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[REG_SEQUENCE3]], %subreg.sub0_sub1_sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE4]], [[REG_SEQUENCE]], 0, 16, 62, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX11-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1 + ; GFX11-NEXT: [[REG_SEQUENCE6:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE5]], %subreg.sub0_sub1, [[COPY]], %subreg.sub2 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE6]], [[REG_SEQUENCE]], 0, 36, 59, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) %12:vgpr_32 = COPY $vgpr8 %11:vgpr_32 = COPY $vgpr7 %10:vgpr_32 = COPY $vgpr6 @@ -3088,34 +6874,88 @@ body: | ... --- -# GFX11-LABEL: name: gfx11_tbuffer_store_uint32 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr8 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr7 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr6 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr5 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr4 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr3 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr2 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr1 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr0 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX11: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %12, %subreg.sub0, %11, %subreg.sub1, %10, %subreg.sub2, %9, %subreg.sub3 -# GFX11: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %8, %subreg.sub0, %7, %subreg.sub1 -# GFX11: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed %14, %13, 0, 4, 48, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %6, %subreg.sub0, %5, %subreg.sub1 -# GFX11: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE killed %15, %subreg.sub0_sub1, %4, %subreg.sub2 -# GFX11: %{{[0-9]+}}:vreg_128 = REG_SEQUENCE killed %16, %subreg.sub0_sub1_sub2, %3, %subreg.sub3 -# GFX11: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed %17, %13, 0, 16, 61, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vreg_64 = REG_SEQUENCE %2, %subreg.sub0, %1, %subreg.sub1 -# GFX11: %{{[0-9]+}}:vreg_96 = REG_SEQUENCE killed %18, %subreg.sub0_sub1, %0, %subreg.sub2 -# GFX11: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed %19, %13, 0, 36, 58, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) name: gfx11_tbuffer_store_uint32 body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-LABEL: name: gfx11_tbuffer_store_uint32 + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 20, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 20, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 20, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 20, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 20, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 20, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 36, 20, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 20, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 44, 20, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_store_uint32 + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 62, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1 + ; GFX10-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE2]], %subreg.sub0_sub1, [[COPY4]], %subreg.sub2 + ; GFX10-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[REG_SEQUENCE3]], %subreg.sub0_sub1_sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE4]], [[REG_SEQUENCE]], 0, 16, 75, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX10-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1 + ; GFX10-NEXT: [[REG_SEQUENCE6:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE5]], %subreg.sub0_sub1, [[COPY]], %subreg.sub2 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE6]], [[REG_SEQUENCE]], 0, 36, 72, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_store_uint32 + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XY_OFFSET_exact killed [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 4, 48, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1 + ; GFX11-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE2]], %subreg.sub0_sub1, [[COPY4]], %subreg.sub2 + ; GFX11-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[REG_SEQUENCE3]], %subreg.sub0_sub1_sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZW_OFFSET_exact killed [[REG_SEQUENCE4]], [[REG_SEQUENCE]], 0, 16, 61, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 4) + ; GFX11-NEXT: [[REG_SEQUENCE5:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1 + ; GFX11-NEXT: [[REG_SEQUENCE6:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[REG_SEQUENCE5]], %subreg.sub0_sub1, [[COPY]], %subreg.sub2 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_XYZ_OFFSET_exact killed [[REG_SEQUENCE6]], [[REG_SEQUENCE]], 0, 36, 58, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 4) %12:vgpr_32 = COPY $vgpr8 %11:vgpr_32 = COPY $vgpr7 %10:vgpr_32 = COPY $vgpr6 @@ -3142,34 +6982,88 @@ body: | ... --- -# GFX11-LABEL: name: gfx11_tbuffer_store_not_merged_data_format_mismatch -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr8 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr7 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr6 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr5 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr4 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr3 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr2 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr1 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr0 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX11: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %12, %subreg.sub0, %11, %subreg.sub1, %10, %subreg.sub2, %9, %subreg.sub3 -# GFX11: TBUFFER_STORE_FORMAT_X_OFFSET_exact %8, %13, 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX11: TBUFFER_STORE_FORMAT_X_OFFSET_exact %7, %13, 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX11: TBUFFER_STORE_FORMAT_X_OFFSET_exact %6, %13, 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX11: TBUFFER_STORE_FORMAT_X_OFFSET_exact %5, %13, 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX11: TBUFFER_STORE_FORMAT_X_OFFSET_exact %4, %13, 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX11: TBUFFER_STORE_FORMAT_X_OFFSET_exact %3, %13, 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX11: TBUFFER_STORE_FORMAT_X_OFFSET_exact %2, %13, 0, 30, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX11: TBUFFER_STORE_FORMAT_X_OFFSET_exact %1, %13, 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX11: TBUFFER_STORE_FORMAT_X_OFFSET_exact %0, %13, 0, 32, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) name: gfx11_tbuffer_store_not_merged_data_format_mismatch body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-LABEL: name: gfx11_tbuffer_store_not_merged_data_format_mismatch + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 30, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 32, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_store_not_merged_data_format_mismatch + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 30, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 32, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_store_not_merged_data_format_mismatch + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 30, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 21, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 32, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) %12:vgpr_32 = COPY $vgpr8 %11:vgpr_32 = COPY $vgpr7 %10:vgpr_32 = COPY $vgpr6 @@ -3196,34 +7090,88 @@ body: | ... --- -# GFX11-LABEL: name: gfx11_tbuffer_store_not_merged_num_format_mismatch -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr8 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr7 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr6 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr5 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr4 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr3 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr2 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr1 -# GFX11: %{{[0-9]+}}:vgpr_32 = COPY $vgpr0 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr3 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr2 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr1 -# GFX11: %{{[0-9]+}}:sgpr_32 = COPY $sgpr0 -# GFX11: %{{[0-9]+}}:sgpr_128 = REG_SEQUENCE %12, %subreg.sub0, %11, %subreg.sub1, %10, %subreg.sub2, %9, %subreg.sub3 -# GFX11: TBUFFER_STORE_FORMAT_X_OFFSET_exact %8, %13, 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX11: TBUFFER_STORE_FORMAT_X_OFFSET_exact %7, %13, 0, 8, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX11: TBUFFER_STORE_FORMAT_X_OFFSET_exact %6, %13, 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX11: TBUFFER_STORE_FORMAT_X_OFFSET_exact %5, %13, 0, 20, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX11: TBUFFER_STORE_FORMAT_X_OFFSET_exact %4, %13, 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX11: TBUFFER_STORE_FORMAT_X_OFFSET_exact %3, %13, 0, 28, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX11: TBUFFER_STORE_FORMAT_X_OFFSET_exact %2, %13, 0, 30, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX11: TBUFFER_STORE_FORMAT_X_OFFSET_exact %1, %13, 0, 40, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) -# GFX11: TBUFFER_STORE_FORMAT_X_OFFSET_exact %0, %13, 0, 32, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) name: gfx11_tbuffer_store_not_merged_num_format_mismatch body: | bb.0.entry: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-LABEL: name: gfx11_tbuffer_store_not_merged_num_format_mismatch + ; GFX9: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX9-NEXT: {{ $}} + ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX9-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 30, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX9-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 32, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_store_not_merged_num_format_mismatch + ; GFX10: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX10-NEXT: {{ $}} + ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 30, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX10-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 32, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_store_not_merged_num_format_mismatch + ; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8 + ; GFX11-NEXT: {{ $}} + ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY9:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY10:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY11:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY12:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY12]], %subreg.sub0, [[COPY11]], %subreg.sub1, [[COPY10]], %subreg.sub2, [[COPY9]], %subreg.sub3 + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY8]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY7]], [[REG_SEQUENCE]], 0, 8, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY6]], [[REG_SEQUENCE]], 0, 16, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], [[REG_SEQUENCE]], 0, 20, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY4]], [[REG_SEQUENCE]], 0, 24, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY3]], [[REG_SEQUENCE]], 0, 28, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY2]], [[REG_SEQUENCE]], 0, 30, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], [[REG_SEQUENCE]], 0, 40, 13, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) + ; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY]], [[REG_SEQUENCE]], 0, 32, 22, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 4) %12:vgpr_32 = COPY $vgpr8 %11:vgpr_32 = COPY $vgpr7 %10:vgpr_32 = COPY $vgpr6 @@ -3250,13 +7198,33 @@ body: | ... --- - -# GFX11-LABEL: name: gfx11_tbuffer_load_not_merged_swizzled_0 -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) name: gfx11_tbuffer_load_not_merged_swizzled_0 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_not_merged_swizzled_0 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_not_merged_swizzled_0 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_load_not_merged_swizzled_0 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3267,13 +7235,33 @@ body: | ... --- - -# GFX11-LABEL: name: gfx11_tbuffer_load_not_merged_swizzled_1 -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 8, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) name: gfx11_tbuffer_load_not_merged_swizzled_1 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_not_merged_swizzled_1 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_not_merged_swizzled_1 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_load_not_merged_swizzled_1 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3284,13 +7272,38 @@ body: | ... --- - -# GFX11-LABEL: name: gfx11_tbuffer_load_merge_across_swizzle -# GFX11: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET %4, 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) -# GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET %4, 0, 12, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) name: gfx11_tbuffer_load_merge_across_swizzle body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_merge_across_swizzle + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 12, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_merge_across_swizzle + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 12, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_load_merge_across_swizzle + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFSET:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFSET [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_OFFSET]].sub1 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_OFFSET:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFSET [[REG_SEQUENCE]], 0, 12, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3305,10 +7318,35 @@ body: | name: gfx11_tbuffer_load_x_x_idxen body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_x_idxen + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_x_idxen + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_IDXEN]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_IDXEN]].sub1 ; GFX11-LABEL: name: gfx11_tbuffer_load_x_x_idxen - ; GFX11: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN %4, %5, 0, 0, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %8.sub1 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_IDXEN]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_IDXEN]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3323,10 +7361,34 @@ body: | name: gfx11_tbuffer_load_x_xy_idxen body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_xy_idxen + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_xy_idxen + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX11-LABEL: name: gfx11_tbuffer_load_x_xy_idxen - ; GFX11: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN %4, %5, 0, 0, 60, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX11: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub1_sub2 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 60, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN]].sub1_sub2 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3341,10 +7403,34 @@ body: | name: gfx11_tbuffer_load_xy_xy_idxen body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_xy_xy_idxen + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_xy_xy_idxen + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX11-LABEL: name: gfx11_tbuffer_load_xy_xy_idxen - ; GFX11: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN %4, %5, 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vreg_64 = COPY %8.sub0_sub1 - ; GFX11: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub2_sub3 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN]].sub0_sub1 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN]].sub2_sub3 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3359,10 +7445,34 @@ body: | name: gfx11_tbuffer_load_x_xyz_idxen body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_xyz_idxen + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 8, 60, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_xyz_idxen + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 8, 60, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX11-LABEL: name: gfx11_tbuffer_load_x_xyz_idxen - ; GFX11: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN %4, %5, 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX11: %{{[0-9]+}}:vreg_96 = COPY killed %8.sub1_sub2_sub3 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN]].sub1_sub2_sub3 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3377,10 +7487,35 @@ body: | name: gfx11_tbuffer_load_x_x_bothen body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_x_bothen + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_x_bothen + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_BOTHEN]].sub1 ; GFX11-LABEL: name: gfx11_tbuffer_load_x_x_bothen - ; GFX11: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN %4, %5, 0, 0, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %8.sub1 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_BOTHEN]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3395,10 +7530,34 @@ body: | name: gfx11_tbuffer_load_x_xy_bothen body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_xy_bothen + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_xy_bothen + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX11-LABEL: name: gfx11_tbuffer_load_x_xy_bothen - ; GFX11: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN %4, %5, 0, 0, 60, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX11: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub1_sub2 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 60, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN]].sub1_sub2 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3413,10 +7572,34 @@ body: | name: gfx11_tbuffer_load_xy_xy_bothen body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_xy_xy_bothen + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 8, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_xy_xy_bothen + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 8, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX11-LABEL: name: gfx11_tbuffer_load_xy_xy_bothen - ; GFX11: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN %4, %5, 0, 0, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vreg_64 = COPY %8.sub0_sub1 - ; GFX11: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub2_sub3 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub0_sub1 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub2_sub3 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3431,10 +7614,34 @@ body: | name: gfx11_tbuffer_load_x_xyz_bothen body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_xyz_bothen + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_xyz_bothen + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX11-LABEL: name: gfx11_tbuffer_load_x_xyz_bothen - ; GFX11: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN %4, %5, 0, 0, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX11: %{{[0-9]+}}:vreg_96 = COPY killed %8.sub1_sub2_sub3 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN [[COPY4]], [[REG_SEQUENCE]], 0, 0, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN]].sub1_sub2_sub3 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3449,10 +7656,35 @@ body: | name: gfx11_tbuffer_load_x_x_idxen_exact body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_x_idxen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_x_idxen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub1 ; GFX11-LABEL: name: gfx11_tbuffer_load_x_x_idxen_exact - ; GFX11: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact %4, %5, 0, 0, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %8.sub1 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3467,10 +7699,34 @@ body: | name: gfx11_tbuffer_load_x_xy_idxen_exact body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_xy_idxen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_xy_idxen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX11-LABEL: name: gfx11_tbuffer_load_x_xy_idxen_exact - ; GFX11: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact %4, %5, 0, 0, 60, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX11: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub1_sub2 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 60, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact]].sub1_sub2 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3485,10 +7741,34 @@ body: | name: gfx11_tbuffer_load_xy_xy_idxen_exact body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_xy_xy_idxen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_xy_xy_idxen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX11-LABEL: name: gfx11_tbuffer_load_xy_xy_idxen_exact - ; GFX11: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact %4, %5, 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vreg_64 = COPY %8.sub0_sub1 - ; GFX11: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub2_sub3 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact]].sub0_sub1 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact]].sub2_sub3 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3503,10 +7783,34 @@ body: | name: gfx11_tbuffer_load_x_xyz_idxen_exact body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_xyz_idxen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 60, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_xyz_idxen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 60, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX11-LABEL: name: gfx11_tbuffer_load_x_xyz_idxen_exact - ; GFX11: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact %4, %5, 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX11: %{{[0-9]+}}:vreg_96 = COPY killed %8.sub1_sub2_sub3 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_IDXEN_exact]].sub1_sub2_sub3 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3521,12 +7825,40 @@ body: | name: gfx11_tbuffer_load_x_x_x_idxen_exact body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_x_x_idxen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_x_x_idxen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact]].sub0_sub1 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact]].sub2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY5]].sub0 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY killed [[COPY5]].sub1 ; GFX11-LABEL: name: gfx11_tbuffer_load_x_x_x_idxen_exact - ; GFX11: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact %4, %5, 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vreg_64 = COPY %10.sub0_sub1 - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %10.sub2 - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY %9.sub0 - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %9.sub1 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact]].sub0_sub1 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_IDXEN_exact]].sub2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY5]].sub0 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY killed [[COPY5]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3542,11 +7874,38 @@ body: | name: gfx11_tbuffer_load_x_x_x_idxen_exact_swizzled_0 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_x_x_idxen_exact_swizzled_0 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_x_x_idxen_exact_swizzled_0 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub1 ; GFX11-LABEL: name: gfx11_tbuffer_load_x_x_x_idxen_exact_swizzled_0 - ; GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact %4, %5, 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact %4, %5, 0, 8, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY %9.sub0 - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %9.sub1 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_IDXEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3562,10 +7921,35 @@ body: | name: gfx11_tbuffer_load_x_x_bothen_exact body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_x_bothen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_x_bothen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub1 ; GFX11-LABEL: name: gfx11_tbuffer_load_x_x_bothen_exact - ; GFX11: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact %4, %5, 0, 0, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %8.sub1 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3580,10 +7964,34 @@ body: | name: gfx11_tbuffer_load_x_xy_bothen_exact body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_xy_bothen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_xy_bothen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) ; GFX11-LABEL: name: gfx11_tbuffer_load_x_xy_bothen_exact - ; GFX11: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact %4, %5, 0, 0, 60, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX11: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub1_sub2 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 60, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact]].sub1_sub2 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3598,11 +8006,35 @@ body: | name: gfx11_tbuffer_load_xy_xy_bothen_exact body: | bb.0.entry: - ; GFX11-LABEL: name: gfx11_tbuffer_load_xy_xy_bothen_exact - ; GFX11: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact %4, %5, 0, 0, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vreg_64 = COPY %8.sub0_sub1 - ; GFX11: %{{[0-9]+}}:vreg_64 = COPY killed %8.sub2_sub3 + ; GFX9-LABEL: name: gfx11_tbuffer_load_xy_xy_bothen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_xy_xy_bothen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_load_xy_xy_bothen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact]].sub0_sub1 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact]].sub2_sub3 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3617,11 +8049,35 @@ body: | name: gfx11_tbuffer_load_x_xyz_bothen_exact body: | bb.0.entry: - ; GFX11-LABEL: name: gfx11_tbuffer_load_x_xyz_bothen_exact - ; GFX11: %{{[0-9]+}}:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact %4, %5, 0, 0, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY %8.sub0 - ; GFX11: %{{[0-9]+}}:vreg_96 = COPY killed %8.sub1_sub2_sub3 + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_xyz_bothen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_xyz_bothen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_load_x_xyz_bothen_exact + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 0, 63, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vreg_96 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZW_BOTHEN_exact]].sub1_sub2_sub3 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3636,9 +8092,36 @@ body: | name: gfx11_tbuffer_load_xy_xy_bothen_exact_diff_vaddr body: | bb.0.entry: - ; GFX11-LABEL: name: gfx11_tbuffer_load_xy_xy_bothen_exact - ; GFX11: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact %4, %6, 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact %5, %6, 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-LABEL: name: gfx11_tbuffer_load_xy_xy_bothen_exact_diff_vaddr + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr1 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_xy_xy_bothen_exact_diff_vaddr + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr1 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_load_xy_xy_bothen_exact_diff_vaddr + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr1 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3655,9 +8138,39 @@ body: | name: gfx11_tbuffer_load_xy_xy_bothen_exact_diff_srsrc body: | bb.0.entry: - ; GFX11-LABEL: name: gfx11_tbuffer_load_xy_xy_bothen_exact - ; GFX11: {{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact %5, %6, 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - ; GFX11: {{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact %5, %7, 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-LABEL: name: gfx11_tbuffer_load_xy_xy_bothen_exact_diff_srsrc + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE1]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_xy_xy_bothen_exact_diff_srsrc + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE1]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_load_xy_xy_bothen_exact_diff_srsrc + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY5]], [[REG_SEQUENCE1]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3674,9 +8187,36 @@ body: | name: gfx11_tbuffer_load_xy_xy_idxen_exact_diff_vaddr body: | bb.0.entry: - ; GFX11-LABEL: name: gfx11_tbuffer_load_xy_xy_idxen_exact - ; GFX11: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact %4, %6, 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact %5, %6, 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-LABEL: name: gfx11_tbuffer_load_xy_xy_idxen_exact_diff_vaddr + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_xy_xy_idxen_exact_diff_vaddr + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_load_xy_xy_idxen_exact_diff_vaddr + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3692,9 +8232,39 @@ body: | name: gfx11_tbuffer_load_xy_xy_idxen_exact_diff_srsrc body: | bb.0.entry: - ; GFX11-LABEL: name: gfx11_tbuffer_load_xy_xy_idxen_exact - ; GFX11: {{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact %5, %6, 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - ; GFX11: {{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact %5, %7, 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-LABEL: name: gfx11_tbuffer_load_xy_xy_idxen_exact_diff_srsrc + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr4 + ; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE1]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_xy_xy_idxen_exact_diff_srsrc + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr4 + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE1]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-LABEL: name: gfx11_tbuffer_load_xy_xy_idxen_exact_diff_srsrc + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr4 + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE]], 0, 4, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_IDXEN_exact1:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_IDXEN_exact [[COPY5]], [[REG_SEQUENCE1]], 0, 12, 50, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3711,12 +8281,40 @@ body: | name: gfx11_tbuffer_load_x_x_x_bothen_exact body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_x_x_bothen_exact + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_x_x_bothen_exact + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 74, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact]].sub0_sub1 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact]].sub2 + ; GFX10-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY5]].sub0 + ; GFX10-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY killed [[COPY5]].sub1 ; GFX11-LABEL: name: gfx11_tbuffer_load_x_x_x_bothen_exact - ; GFX11: %{{[0-9]+}}:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact %4, %5, 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vreg_64 = COPY %10.sub0_sub1 - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %10.sub2 - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY %9.sub0 - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %9.sub1 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 60, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact]].sub0_sub1 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XYZ_BOTHEN_exact]].sub2 + ; GFX11-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY5]].sub0 + ; GFX11-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY killed [[COPY5]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 @@ -3732,11 +8330,38 @@ body: | name: gfx11_tbuffer_load_x_x_x_bothen_exact_swizzled_0 body: | bb.0.entry: + ; GFX9-LABEL: name: gfx11_tbuffer_load_x_x_x_bothen_exact_swizzled_0 + ; GFX9: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX9-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX9-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX9-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX9-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact1:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX9-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact2:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 12, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-LABEL: name: gfx11_tbuffer_load_x_x_x_bothen_exact_swizzled_0 + ; GFX10: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX10-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX10-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX10-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX10-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX10-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 64, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub0 + ; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub1 ; GFX11-LABEL: name: gfx11_tbuffer_load_x_x_x_bothen_exact_swizzled_0 - ; GFX11: %{{[0-9]+}}:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact %4, %5, 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact %4, %5, 0, 8, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY %9.sub0 - ; GFX11: %{{[0-9]+}}:vgpr_32 = COPY killed %9.sub1 + ; GFX11: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr0 + ; GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1 + ; GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2 + ; GFX11-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3 + ; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY $vgpr0 + ; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_X_BOTHEN_exact:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 4, 22, 0, 1, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 4) + ; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact [[COPY4]], [[REG_SEQUENCE]], 0, 8, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 4) + ; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub0 + ; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[TBUFFER_LOAD_FORMAT_XY_BOTHEN_exact]].sub1 %0:sgpr_32 = COPY $sgpr0 %1:sgpr_32 = COPY $sgpr1 %2:sgpr_32 = COPY $sgpr2 -- 2.7.4