From c8127fc67410f5d96fc480c50d10492f361e834b Mon Sep 17 00:00:00 2001 From: Stanislav Mekhanoshin Date: Fri, 9 Mar 2018 07:21:43 +0000 Subject: [PATCH] [AMDGPU] Fixed V_DIV_FIXUP_F16 selection on GFX9 GFX9 should select opsel version. Differential Revision: https://reviews.llvm.org/D44279 llvm-svn: 327106 --- llvm/lib/Target/AMDGPU/VOP3Instructions.td | 25 +++--- llvm/test/CodeGen/AMDGPU/fdiv.f16.ll | 117 +++++++++++++++-------------- 2 files changed, 71 insertions(+), 71 deletions(-) diff --git a/llvm/lib/Target/AMDGPU/VOP3Instructions.td b/llvm/lib/Target/AMDGPU/VOP3Instructions.td index aee6434..fd3cd24 100644 --- a/llvm/lib/Target/AMDGPU/VOP3Instructions.td +++ b/llvm/lib/Target/AMDGPU/VOP3Instructions.td @@ -406,16 +406,16 @@ def V_MAD_I64_I32 : VOP3Inst <"v_mad_i64_i32", VOP3b_I64_I1_I32_I32_I64>; } // End SubtargetPredicate = isCIVI -let SubtargetPredicate = Has16BitInsts in { - -let renamedInGFX9 = 1 in { -def V_DIV_FIXUP_F16 : VOP3Inst <"v_div_fixup_f16", VOP3_Profile, AMDGPUdiv_fixup>; +def V_DIV_FIXUP_F16 : VOP3Inst <"v_div_fixup_f16", VOP3_Profile, AMDGPUdiv_fixup> { + let Predicates = [Has16BitInsts, isVIOnly]; } -let SubtargetPredicate = isGFX9 in { -def V_DIV_FIXUP_F16_gfx9 : VOP3Inst <"v_div_fixup_f16_gfx9", VOP3_Profile>; +def V_DIV_FIXUP_F16_gfx9 : VOP3Inst <"v_div_fixup_f16_gfx9", + VOP3_Profile, AMDGPUdiv_fixup> { + let renamedInGFX9 = 1; + let Predicates = [Has16BitInsts, isGFX9]; } -let isCommutable = 1 in { +let SubtargetPredicate = Has16BitInsts, isCommutable = 1 in { let renamedInGFX9 = 1 in { def V_MAD_F16 : VOP3Inst <"v_mad_f16", VOP3_Profile, fmad>; @@ -436,8 +436,7 @@ def V_INTERP_P2_F16_gfx9 : VOP3Interp <"v_interp_p2_f16_gfx9", VOP3_INTERP16<[f1 def V_INTERP_P1LL_F16 : VOP3Interp <"v_interp_p1ll_f16", VOP3_INTERP16<[f32, f32, i32, untyped]>>; def V_INTERP_P1LV_F16 : VOP3Interp <"v_interp_p1lv_f16", VOP3_INTERP16<[f32, f32, i32, f16]>>; -} // End isCommutable = 1 -} // End SubtargetPredicate = Has16BitInsts +} // End SubtargetPredicate = Has16BitInsts, isCommutable = 1 let SubtargetPredicate = isVI in { def V_INTERP_P1_F32_e64 : VOP3Interp <"v_interp_p1_f32", VOP3_INTERP>; @@ -695,7 +694,7 @@ multiclass VOP3Interp_F16_Real_vi op> { let AssemblerPredicates = [isGFX9], DecoderNamespace = "GFX9" in { multiclass VOP3_F16_Real_gfx9 op, string OpName, string AsmName> { - def _vi : VOP3_Real(OpName), SIEncodingFamily.GFX9>, + def _gfx9 : VOP3_Real(OpName), SIEncodingFamily.GFX9>, VOP3e_vi (OpName).Pfl> { VOP3_Pseudo ps = !cast(OpName); let AsmString = AsmName # ps.AsmOperands; @@ -703,7 +702,7 @@ multiclass VOP3_F16_Real_gfx9 op, string OpName, string AsmName> { } multiclass VOP3OpSel_F16_Real_gfx9 op, string AsmName> { - def _vi : VOP3_Real(NAME), SIEncodingFamily.GFX9>, + def _gfx9 : VOP3_Real(NAME), SIEncodingFamily.GFX9>, VOP3OpSel_gfx9 (NAME).Pfl> { VOP3_Pseudo ps = !cast(NAME); let AsmString = AsmName # ps.AsmOperands; @@ -711,7 +710,7 @@ multiclass VOP3OpSel_F16_Real_gfx9 op, string AsmName> { } multiclass VOP3Interp_F16_Real_gfx9 op, string OpName, string AsmName> { - def _vi : VOP3_Real(OpName), SIEncodingFamily.GFX9>, + def _gfx9 : VOP3_Real(OpName), SIEncodingFamily.GFX9>, VOP3Interp_vi (OpName).Pfl> { VOP3_Pseudo ps = !cast(OpName); let AsmString = AsmName # ps.AsmOperands; @@ -719,7 +718,7 @@ multiclass VOP3Interp_F16_Real_gfx9 op, string OpName, string AsmName> } multiclass VOP3_Real_gfx9 op, string AsmName> { - def _vi : VOP3_Real(NAME), SIEncodingFamily.GFX9>, + def _gfx9 : VOP3_Real(NAME), SIEncodingFamily.GFX9>, VOP3e_vi (NAME).Pfl> { VOP3_Pseudo ps = !cast(NAME); let AsmString = AsmName # ps.AsmOperands; diff --git a/llvm/test/CodeGen/AMDGPU/fdiv.f16.ll b/llvm/test/CodeGen/AMDGPU/fdiv.f16.ll index 3331433..275a12d 100644 --- a/llvm/test/CodeGen/AMDGPU/fdiv.f16.ll +++ b/llvm/test/CodeGen/AMDGPU/fdiv.f16.ll @@ -1,6 +1,7 @@ ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s -; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-flat-for-global -mattr=+fp16-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s -; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-flat-for-global -mattr=-fp16-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s +; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-flat-for-global -mattr=+fp16-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX8_9 %s +; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-flat-for-global -mattr=-fp16-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX8_9 %s +; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX8_9 %s ; Make sure fdiv is promoted to f32. @@ -20,17 +21,17 @@ ; SI: v_div_fixup_f32 ; SI: v_cvt_f16_f32 -; VI: flat_load_ushort [[LHS:v[0-9]+]] -; VI: flat_load_ushort [[RHS:v[0-9]+]] +; GFX8_9: {{flat|global}}_load_ushort [[LHS:v[0-9]+]] +; GFX8_9: {{flat|global}}_load_ushort [[RHS:v[0-9]+]] -; VI-DAG: v_cvt_f32_f16_e32 [[CVT_LHS:v[0-9]+]], [[LHS]] -; VI-DAG: v_cvt_f32_f16_e32 [[CVT_RHS:v[0-9]+]], [[RHS]] +; GFX8_9-DAG: v_cvt_f32_f16_e32 [[CVT_LHS:v[0-9]+]], [[LHS]] +; GFX8_9-DAG: v_cvt_f32_f16_e32 [[CVT_RHS:v[0-9]+]], [[RHS]] -; VI-DAG: v_rcp_f32_e32 [[RCP_RHS:v[0-9]+]], [[CVT_RHS]] -; VI: v_mul_f32_e32 [[MUL:v[0-9]+]], [[CVT_LHS]], [[RCP_RHS]] -; VI: v_cvt_f16_f32_e32 [[CVT_BACK:v[0-9]+]], [[MUL]] -; VI: v_div_fixup_f16 [[RESULT:v[0-9]+]], [[CVT_BACK]], [[RHS]], [[LHS]] -; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]] +; GFX8_9-DAG: v_rcp_f32_e32 [[RCP_RHS:v[0-9]+]], [[CVT_RHS]] +; GFX8_9: v_mul_f32_e32 [[MUL:v[0-9]+]], [[CVT_LHS]], [[RCP_RHS]] +; GFX8_9: v_cvt_f16_f32_e32 [[CVT_BACK:v[0-9]+]], [[MUL]] +; GFX8_9: v_div_fixup_f16 [[RESULT:v[0-9]+]], [[CVT_BACK]], [[RHS]], [[LHS]] +; GFX8_9: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]] define amdgpu_kernel void @v_fdiv_f16( half addrspace(1)* %r, half addrspace(1)* %a, @@ -49,11 +50,11 @@ entry: } ; GCN-LABEL: {{^}}v_rcp_f16: -; VI: flat_load_ushort [[VAL:v[0-9]+]] -; VI-NOT: [[VAL]] -; VI: v_rcp_f16_e32 [[RESULT:v[0-9]+]], [[VAL]] -; VI-NOT: [[RESULT]] -; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]] +; GFX8_9: {{flat|global}}_load_ushort [[VAL:v[0-9]+]] +; GFX8_9-NOT: [[VAL]] +; GFX8_9: v_rcp_f16_e32 [[RESULT:v[0-9]+]], [[VAL]] +; GFX8_9-NOT: [[RESULT]] +; GFX8_9: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]] define amdgpu_kernel void @v_rcp_f16(half addrspace(1)* %r, half addrspace(1)* %b) #0 { entry: %tid = call i32 @llvm.amdgcn.workitem.id.x() @@ -67,11 +68,11 @@ entry: } ; GCN-LABEL: {{^}}v_rcp_f16_abs: -; VI: flat_load_ushort [[VAL:v[0-9]+]] -; VI-NOT: [[VAL]] -; VI: v_rcp_f16_e64 [[RESULT:v[0-9]+]], |[[VAL]]| -; VI-NOT: [RESULT]] -; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]] +; GFX8_9: {{flat|global}}_load_ushort [[VAL:v[0-9]+]] +; GFX8_9-NOT: [[VAL]] +; GFX8_9: v_rcp_f16_e64 [[RESULT:v[0-9]+]], |[[VAL]]| +; GFX8_9-NOT: [RESULT]] +; GFX8_9: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]] define amdgpu_kernel void @v_rcp_f16_abs(half addrspace(1)* %r, half addrspace(1)* %b) #0 { entry: %tid = call i32 @llvm.amdgcn.workitem.id.x() @@ -86,11 +87,11 @@ entry: } ; GCN-LABEL: {{^}}v_rcp_f16_arcp: -; VI: flat_load_ushort [[VAL:v[0-9]+]] -; VI-NOT: [[VAL]] -; VI: v_rcp_f16_e32 [[RESULT:v[0-9]+]], [[VAL]] -; VI-NOT: [[RESULT]] -; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]] +; GFX8_9: {{flat|global}}_load_ushort [[VAL:v[0-9]+]] +; GFX8_9-NOT: [[VAL]] +; GFX8_9: v_rcp_f16_e32 [[RESULT:v[0-9]+]], [[VAL]] +; GFX8_9-NOT: [[RESULT]] +; GFX8_9: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]] define amdgpu_kernel void @v_rcp_f16_arcp(half addrspace(1)* %r, half addrspace(1)* %b) #0 { entry: %tid = call i32 @llvm.amdgcn.workitem.id.x() @@ -104,11 +105,11 @@ entry: } ; GCN-LABEL: {{^}}v_rcp_f16_neg: -; VI: flat_load_ushort [[VAL:v[0-9]+]] -; VI-NOT: [[VAL]] -; VI: v_rcp_f16_e64 [[RESULT:v[0-9]+]], -[[VAL]] -; VI-NOT: [RESULT]] -; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]] +; GFX8_9: {{flat|global}}_load_ushort [[VAL:v[0-9]+]] +; GFX8_9-NOT: [[VAL]] +; GFX8_9: v_rcp_f16_e64 [[RESULT:v[0-9]+]], -[[VAL]] +; GFX8_9-NOT: [RESULT]] +; GFX8_9: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]] define amdgpu_kernel void @v_rcp_f16_neg(half addrspace(1)* %r, half addrspace(1)* %b) #0 { entry: %tid = call i32 @llvm.amdgcn.workitem.id.x() @@ -122,11 +123,11 @@ entry: } ; GCN-LABEL: {{^}}v_rsq_f16: -; VI: flat_load_ushort [[VAL:v[0-9]+]] -; VI-NOT: [[VAL]] -; VI: v_rsq_f16_e32 [[RESULT:v[0-9]+]], [[VAL]] -; VI-NOT: [RESULT]] -; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]] +; GFX8_9: {{flat|global}}_load_ushort [[VAL:v[0-9]+]] +; GFX8_9-NOT: [[VAL]] +; GFX8_9: v_rsq_f16_e32 [[RESULT:v[0-9]+]], [[VAL]] +; GFX8_9-NOT: [RESULT]] +; GFX8_9: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]] define amdgpu_kernel void @v_rsq_f16(half addrspace(1)* %r, half addrspace(1)* %b) #0 { entry: %tid = call i32 @llvm.amdgcn.workitem.id.x() @@ -141,12 +142,12 @@ entry: } ; GCN-LABEL: {{^}}v_rsq_f16_neg: -; VI: flat_load_ushort [[VAL:v[0-9]+]] -; VI-NOT: [[VAL]] -; VI: v_sqrt_f16_e32 [[SQRT:v[0-9]+]], [[VAL]] -; VI-NEXT: v_rcp_f16_e64 [[RESULT:v[0-9]+]], -[[SQRT]] -; VI-NOT: [RESULT]] -; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]] +; GFX8_9: {{flat|global}}_load_ushort [[VAL:v[0-9]+]] +; GFX8_9-NOT: [[VAL]] +; GFX8_9: v_sqrt_f16_e32 [[SQRT:v[0-9]+]], [[VAL]] +; GFX8_9-NEXT: v_rcp_f16_e64 [[RESULT:v[0-9]+]], -[[SQRT]] +; GFX8_9-NOT: [RESULT]] +; GFX8_9: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]] define amdgpu_kernel void @v_rsq_f16_neg(half addrspace(1)* %r, half addrspace(1)* %b) #0 { entry: %tid = call i32 @llvm.amdgcn.workitem.id.x() @@ -161,13 +162,13 @@ entry: } ; GCN-LABEL: {{^}}v_fdiv_f16_arcp: -; VI: flat_load_ushort [[LHS:v[0-9]+]] -; VI: flat_load_ushort [[RHS:v[0-9]+]] +; GFX8_9: {{flat|global}}_load_ushort [[LHS:v[0-9]+]] +; GFX8_9: {{flat|global}}_load_ushort [[RHS:v[0-9]+]] -; VI: v_rcp_f16_e32 [[RCP:v[0-9]+]], [[RHS]] -; VI: v_mul_f16_e32 [[RESULT:v[0-9]+]], [[LHS]], [[RCP]] +; GFX8_9: v_rcp_f16_e32 [[RCP:v[0-9]+]], [[RHS]] +; GFX8_9: v_mul_f16_e32 [[RESULT:v[0-9]+]], [[LHS]], [[RCP]] -; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]] +; GFX8_9: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]] define amdgpu_kernel void @v_fdiv_f16_arcp(half addrspace(1)* %r, half addrspace(1)* %a, half addrspace(1)* %b) #0 { entry: %tid = call i32 @llvm.amdgcn.workitem.id.x() @@ -183,13 +184,13 @@ entry: } ; GCN-LABEL: {{^}}v_fdiv_f16_unsafe: -; VI: flat_load_ushort [[LHS:v[0-9]+]] -; VI: flat_load_ushort [[RHS:v[0-9]+]] +; GFX8_9: {{flat|global}}_load_ushort [[LHS:v[0-9]+]] +; GFX8_9: {{flat|global}}_load_ushort [[RHS:v[0-9]+]] -; VI: v_rcp_f16_e32 [[RCP:v[0-9]+]], [[RHS]] -; VI: v_mul_f16_e32 [[RESULT:v[0-9]+]], [[LHS]], [[RCP]] +; GFX8_9: v_rcp_f16_e32 [[RCP:v[0-9]+]], [[RHS]] +; GFX8_9: v_mul_f16_e32 [[RESULT:v[0-9]+]], [[LHS]], [[RCP]] -; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]] +; GFX8_9: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]] define amdgpu_kernel void @v_fdiv_f16_unsafe(half addrspace(1)* %r, half addrspace(1)* %a, half addrspace(1)* %b) #2 { entry: %tid = call i32 @llvm.amdgcn.workitem.id.x() @@ -207,8 +208,8 @@ entry: ; FUNC-LABEL: {{^}}div_arcp_2_x_pat_f16: ; SI: v_mul_f32_e32 v{{[0-9]+}}, 0.5, v{{[0-9]+}} -; VI: v_mul_f16_e32 [[MUL:v[0-9]+]], 0.5, v{{[0-9]+}} -; VI: buffer_store_short [[MUL]] +; GFX8_9: v_mul_f16_e32 [[MUL:v[0-9]+]], 0.5, v{{[0-9]+}} +; GFX8_9: buffer_store_short [[MUL]] define amdgpu_kernel void @div_arcp_2_x_pat_f16(half addrspace(1)* %out) #0 { %x = load half, half addrspace(1)* undef %rcp = fdiv arcp half %x, 2.0 @@ -219,8 +220,8 @@ define amdgpu_kernel void @div_arcp_2_x_pat_f16(half addrspace(1)* %out) #0 { ; FUNC-LABEL: {{^}}div_arcp_k_x_pat_f16: ; SI: v_mul_f32_e32 v{{[0-9]+}}, 0x3dcccccd, v{{[0-9]+}} -; VI: v_mul_f16_e32 [[MUL:v[0-9]+]], 0x2e66, v{{[0-9]+}} -; VI: buffer_store_short [[MUL]] +; GFX8_9: v_mul_f16_e32 [[MUL:v[0-9]+]], 0x2e66, v{{[0-9]+}} +; GFX8_9: buffer_store_short [[MUL]] define amdgpu_kernel void @div_arcp_k_x_pat_f16(half addrspace(1)* %out) #0 { %x = load half, half addrspace(1)* undef %rcp = fdiv arcp half %x, 10.0 @@ -231,8 +232,8 @@ define amdgpu_kernel void @div_arcp_k_x_pat_f16(half addrspace(1)* %out) #0 { ; FUNC-LABEL: {{^}}div_arcp_neg_k_x_pat_f16: ; SI: v_mul_f32_e32 v{{[0-9]+}}, 0xbdcccccd, v{{[0-9]+}} -; VI: v_mul_f16_e32 [[MUL:v[0-9]+]], 0xae66, v{{[0-9]+}} -; VI: buffer_store_short [[MUL]] +; GFX8_9: v_mul_f16_e32 [[MUL:v[0-9]+]], 0xae66, v{{[0-9]+}} +; GFX8_9: buffer_store_short [[MUL]] define amdgpu_kernel void @div_arcp_neg_k_x_pat_f16(half addrspace(1)* %out) #0 { %x = load half, half addrspace(1)* undef %rcp = fdiv arcp half %x, -10.0 -- 2.7.4