From 38814fa2fd964623658a860914b6222e309a2bf9 Mon Sep 17 00:00:00 2001 From: Jan Vesely Date: Sat, 27 Aug 2016 19:09:43 +0000 Subject: [PATCH] AMDGPU/R600: Enable Load combine Fix and improve tests Differential Revision: https://reviews.llvm.org/D23899 llvm-svn: 279925 --- llvm/lib/Target/AMDGPU/R600ISelLowering.cpp | 1 + llvm/test/CodeGen/AMDGPU/load-constant-i16.ll | 209 ++++++++++-- llvm/test/CodeGen/AMDGPU/load-constant-i8.ll | 454 +++++++++++++++++++++++-- llvm/test/CodeGen/AMDGPU/load-global-i16.ll | 186 +++++++++-- llvm/test/CodeGen/AMDGPU/load-global-i8.ll | 457 ++++++++++++++++++++++++-- llvm/test/CodeGen/AMDGPU/load-local-i16.ll | 388 ++++++++++++++++++++-- llvm/test/CodeGen/AMDGPU/load-local-i8.ll | 392 +++++++++++++++++++++- 7 files changed, 1952 insertions(+), 135 deletions(-) diff --git a/llvm/lib/Target/AMDGPU/R600ISelLowering.cpp b/llvm/lib/Target/AMDGPU/R600ISelLowering.cpp index 481a6ac..d7d40a8 100644 --- a/llvm/lib/Target/AMDGPU/R600ISelLowering.cpp +++ b/llvm/lib/Target/AMDGPU/R600ISelLowering.cpp @@ -198,6 +198,7 @@ R600TargetLowering::R600TargetLowering(const TargetMachine &TM, setTargetDAGCombine(ISD::EXTRACT_VECTOR_ELT); setTargetDAGCombine(ISD::SELECT_CC); setTargetDAGCombine(ISD::INSERT_VECTOR_ELT); + setTargetDAGCombine(ISD::LOAD); } const R600Subtarget *R600TargetLowering::getSubtarget() const { diff --git a/llvm/test/CodeGen/AMDGPU/load-constant-i16.ll b/llvm/test/CodeGen/AMDGPU/load-constant-i16.ll index ef9791d..397e172 100644 --- a/llvm/test/CodeGen/AMDGPU/load-constant-i16.ll +++ b/llvm/test/CodeGen/AMDGPU/load-constant-i16.ll @@ -7,7 +7,7 @@ ; GCN-NOHSA: buffer_load_ushort v{{[0-9]+}} ; GCN-HSA: flat_load_ushort -; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0 +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @constant_load_i16(i16 addrspace(1)* %out, i16 addrspace(2)* %in) { entry: %ld = load i16, i16 addrspace(2)* %in @@ -18,7 +18,7 @@ entry: ; FUNC-LABEL: {{^}}constant_load_v2i16: ; GCN: s_load_dword s -; EG: VTX_READ_32 +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @constant_load_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(2)* %in) { entry: %ld = load <2 x i16>, <2 x i16> addrspace(2)* %in @@ -29,8 +29,8 @@ entry: ; FUNC-LABEL: {{^}}constant_load_v3i16: ; GCN: s_load_dwordx2 s -; EG-DAG: VTX_READ_32 -; EG-DAG: VTX_READ_16 +; EG-DAG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 4, #1 define void @constant_load_v3i16(<3 x i16> addrspace(1)* %out, <3 x i16> addrspace(2)* %in) { entry: %ld = load <3 x i16>, <3 x i16> addrspace(2)* %in @@ -41,7 +41,7 @@ entry: ; FUNC-LABEL: {{^}}constant_load_v4i16: ; GCN: s_load_dwordx2 -; EG: VTX_READ_64 +; EG: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0, #1 define void @constant_load_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> addrspace(2)* %in) { entry: %ld = load <4 x i16>, <4 x i16> addrspace(2)* %in @@ -52,7 +52,7 @@ entry: ; FUNC-LABEL: {{^}}constant_load_v8i16: ; GCN: s_load_dwordx4 -; EG: VTX_READ_128 +; EG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 define void @constant_load_v8i16(<8 x i16> addrspace(1)* %out, <8 x i16> addrspace(2)* %in) { entry: %ld = load <8 x i16>, <8 x i16> addrspace(2)* %in @@ -63,8 +63,8 @@ entry: ; FUNC-LABEL: {{^}}constant_load_v16i16: ; GCN: s_load_dwordx8 -; EG: VTX_READ_128 -; EG: VTX_READ_128 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 define void @constant_load_v16i16(<16 x i16> addrspace(1)* %out, <16 x i16> addrspace(2)* %in) { entry: %ld = load <16 x i16>, <16 x i16> addrspace(2)* %in @@ -79,7 +79,7 @@ entry: ; GCN-HSA: flat_load_ushort ; GCN-HSA: flat_store_dword -; EG: VTX_READ_16 T{{[0-9]+\.X, T[0-9]+\.X}} +; EG: VTX_READ_16 T{{[0-9]+\.X, T[0-9]+\.X}}, 0, #1 define void @constant_zextload_i16_to_i32(i32 addrspace(1)* %out, i16 addrspace(2)* %in) #0 { %a = load i16, i16 addrspace(2)* %in %ext = zext i16 %a to i32 @@ -94,7 +94,7 @@ define void @constant_zextload_i16_to_i32(i32 addrspace(1)* %out, i16 addrspace( ; GCN-HSA: flat_load_sshort ; GCN-HSA: flat_store_dword -; EG: VTX_READ_16 [[DST:T[0-9]\.[XYZW]]], [[DST]] +; EG: VTX_READ_16 [[DST:T[0-9]\.[XYZW]]], [[DST]], 0, #1 ; EG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST]], 0.0, literal ; EG: 16 define void @constant_sextload_i16_to_i32(i32 addrspace(1)* %out, i16 addrspace(2)* %in) #0 { @@ -107,6 +107,8 @@ define void @constant_sextload_i16_to_i32(i32 addrspace(1)* %out, i16 addrspace( ; FUNC-LABEL: {{^}}constant_zextload_v1i16_to_v1i32: ; GCN-NOHSA: buffer_load_ushort ; GCN-HSA: flat_load_ushort + +; EG: VTX_READ_16 T{{[0-9]+\.X, T[0-9]+\.X}}, 0, #1 define void @constant_zextload_v1i16_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x i16> addrspace(2)* %in) #0 { %load = load <1 x i16>, <1 x i16> addrspace(2)* %in %ext = zext <1 x i16> %load to <1 x i32> @@ -117,6 +119,10 @@ define void @constant_zextload_v1i16_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x ; FUNC-LABEL: {{^}}constant_sextload_v1i16_to_v1i32: ; GCN-NOHSA: buffer_load_sshort ; GCN-HSA: flat_load_sshort + +; EG: VTX_READ_16 [[DST:T[0-9]\.[XYZW]]], [[DST]], 0, #1 +; EG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST]], 0.0, literal +; EG: 16 define void @constant_sextload_v1i16_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x i16> addrspace(2)* %in) #0 { %load = load <1 x i16>, <1 x i16> addrspace(2)* %in %ext = sext <1 x i16> %load to <1 x i32> @@ -128,6 +134,12 @@ define void @constant_sextload_v1i16_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x ; GCN: s_load_dword s ; GCN-DAG: s_and_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0xffff{{$}} ; GCN-DAG: s_lshr_b32 s{{[0-9]+}}, s{{[0-9]+}}, 16 + +; v2i16 is naturally 4 byte aligned +; EG: VTX_READ_32 [[DST:T[0-9]\.[XYZW]]], [[DST]], 0, #1 +; TODO: This should use DST, but for some there are redundant MOVs +; EG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{PV.[XYZW]}}, literal +; EG: 16 define void @constant_zextload_v2i16_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i16> addrspace(2)* %in) #0 { %load = load <2 x i16>, <2 x i16> addrspace(2)* %in %ext = zext <2 x i16> %load to <2 x i32> @@ -140,10 +152,12 @@ define void @constant_zextload_v2i16_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x ; GCN-DAG: s_ashr_i32 ; GCN-DAG: s_sext_i32_i16 -; EG-DAG: VTX_READ_16 [[DST_X:T[0-9]\.[XYZW]]], [[DST_X]] -; EG-DAG: VTX_READ_16 [[DST_Y:T[0-9]\.[XYZW]]], [[DST_Y]] -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_X]], 0.0, literal -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_Y]], 0.0, literal +; v2i16 is naturally 4 byte aligned +; EG: VTX_READ_32 [[DST:T[0-9]\.[XYZW]]], [[DST]], 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; TODO: We should also use ASHR instead of LSHR + BFE +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{PV.[XYZW]}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{PV.[XYZW]}}, 0.0, literal ; EG-DAG: 16 ; EG-DAG: 16 define void @constant_sextload_v2i16_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i16> addrspace(2)* %in) #0 { @@ -155,6 +169,13 @@ define void @constant_sextload_v2i16_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x ; FUNC-LABEL: {{^}}constant_constant_zextload_v3i16_to_v3i32: ; GCN: s_load_dwordx2 + +; v3i16 is naturally 8 byte aligned +; EG-DAG: VTX_READ_32 [[DST_HI:T[0-9]\.[XYZW]]], [[DST_HI]], 0, #1 +; EG-DAG: VTX_READ_16 [[DST_LO:T[0-9]\.[XYZW]]], [[DST_LO]], 4, #1 +; TODO: This should use DST, but for some there are redundant MOVs +; EG: LSHR {{[* ]*}}{{T[0-9].[XYZW]}}, {{T[0-9].[XYZW]}}, literal +; EG: 16 define void @constant_constant_zextload_v3i16_to_v3i32(<3 x i32> addrspace(1)* %out, <3 x i16> addrspace(2)* %in) { entry: %ld = load <3 x i16>, <3 x i16> addrspace(2)* %in @@ -165,6 +186,16 @@ entry: ; FUNC-LABEL: {{^}}constant_constant_sextload_v3i16_to_v3i32: ; GCN: s_load_dwordx2 + +; v3i16 is naturally 8 byte aligned +; EG-DAG: VTX_READ_32 [[DST_HI:T[0-9]\.[XYZW]]], [[DST_HI]], 0, #1 +; EG-DAG: VTX_READ_16 [[DST_LO:T[0-9]\.[XYZW]]], [[DST_LO]], 4, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: ASHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{PV.[XYZW]}}, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{T[0-9].[XYZW]}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{T[0-9].[XYZW]}}, 0.0, literal +; EG-DAG: 16 +; EG-DAG: 16 define void @constant_constant_sextload_v3i16_to_v3i32(<3 x i32> addrspace(1)* %out, <3 x i16> addrspace(2)* %in) { entry: %ld = load <3 x i16>, <3 x i16> addrspace(2)* %in @@ -178,10 +209,13 @@ entry: ; GCN-DAG: s_and_b32 ; GCN-DAG: s_lshr_b32 -; EG: VTX_READ_16 -; EG: VTX_READ_16 -; EG: VTX_READ_16 -; EG: VTX_READ_16 +; v4i16 is naturally 8 byte aligned +; EG: VTX_READ_64 [[DST:T[0-9]\.XY]], {{T[0-9].[XYZW]}}, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{PV.[XYZW]}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{T[0-9].[XYZW]}}, literal +; EG-DAG: 16 +; EG-DAG: 16 define void @constant_constant_zextload_v4i16_to_v4i32(<4 x i32> addrspace(1)* %out, <4 x i16> addrspace(2)* %in) #0 { %load = load <4 x i16>, <4 x i16> addrspace(2)* %in %ext = zext <4 x i16> %load to <4 x i32> @@ -194,14 +228,14 @@ define void @constant_constant_zextload_v4i16_to_v4i32(<4 x i32> addrspace(1)* % ; GCN-DAG: s_ashr_i32 ; GCN-DAG: s_sext_i32_i16 -; EG-DAG: VTX_READ_16 [[DST_X:T[0-9]\.[XYZW]]], [[DST_X]] -; EG-DAG: VTX_READ_16 [[DST_Y:T[0-9]\.[XYZW]]], [[DST_Y]] -; EG-DAG: VTX_READ_16 [[DST_Z:T[0-9]\.[XYZW]]], [[DST_Z]] -; EG-DAG: VTX_READ_16 [[DST_W:T[0-9]\.[XYZW]]], [[DST_W]] -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_X]], 0.0, literal -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_Y]], 0.0, literal -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_Z]], 0.0, literal -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_W]], 0.0, literal +; v4i16 is naturally 8 byte aligned +; EG: VTX_READ_64 [[DST:T[0-9]\.XY]], {{T[0-9].[XYZW]}}, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; TODO: We should use ASHR instead of LSHR + BFE +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal ; EG-DAG: 16 ; EG-DAG: 16 ; EG-DAG: 16 @@ -217,6 +251,26 @@ define void @constant_sextload_v4i16_to_v4i32(<4 x i32> addrspace(1)* %out, <4 x ; GCN: s_load_dwordx4 ; GCN-DAG: s_and_b32 ; GCN-DAG: s_lshr_b32 + +; v8i16 is naturally 16 byte aligned +; EG: VTX_READ_128 [[DST:T[0-9]\.XYZW]], {{T[0-9].[XYZW]}}, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 define void @constant_zextload_v8i16_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x i16> addrspace(2)* %in) #0 { %load = load <8 x i16>, <8 x i16> addrspace(2)* %in %ext = zext <8 x i16> %load to <8 x i32> @@ -228,6 +282,27 @@ define void @constant_zextload_v8i16_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x ; GCN: s_load_dwordx4 ; GCN-DAG: s_ashr_i32 ; GCN-DAG: s_sext_i32_i16 + +; v8i16 is naturally 16 byte aligned +; EG: VTX_READ_128 [[DST:T[0-9]\.XYZW]], {{T[0-9].[XYZW]}}, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; TODO: We should use ASHR instead of LSHR + BFE +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 define void @constant_sextload_v8i16_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x i16> addrspace(2)* %in) #0 { %load = load <8 x i16>, <8 x i16> addrspace(2)* %in %ext = sext <8 x i16> %load to <8 x i32> @@ -239,6 +314,10 @@ define void @constant_sextload_v8i16_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x ; GCN: s_load_dwordx8 ; GCN-DAG: s_and_b32 ; GCN-DAG: s_lshr_b32 + +; v16i16 is naturally 32 byte aligned +; EG-DAG: VTX_READ_128 [[DST_HI:T[0-9]+\.XYZW]], {{T[0-9]+.[XYZW]}}, 0, #1 +; EG-DAG: VTX_READ_128 [[DST_LO:T[0-9]+\.XYZW]], {{T[0-9]+.[XYZW]}}, 16, #1 define void @constant_zextload_v16i16_to_v16i32(<16 x i32> addrspace(1)* %out, <16 x i16> addrspace(2)* %in) #0 { %load = load <16 x i16>, <16 x i16> addrspace(2)* %in %ext = zext <16 x i16> %load to <16 x i32> @@ -250,6 +329,10 @@ define void @constant_zextload_v16i16_to_v16i32(<16 x i32> addrspace(1)* %out, < ; GCN: s_load_dwordx8 ; GCN-DAG: s_ashr_i32 ; GCN-DAG: s_sext_i32_i16 + +; v16i16 is naturally 32 byte aligned +; EG-DAG: VTX_READ_128 [[DST_HI:T[0-9]+\.XYZW]], {{T[0-9]+\.[XYZW]}}, 0, #1 +; EG-DAG: VTX_READ_128 [[DST_LO:T[0-9]+\.XYZW]], {{T[0-9]+\.[XYZW]}}, 16, #1 define void @constant_sextload_v16i16_to_v16i32(<16 x i32> addrspace(1)* %out, <16 x i16> addrspace(2)* %in) #0 { %load = load <16 x i16>, <16 x i16> addrspace(2)* %in %ext = sext <16 x i16> %load to <16 x i32> @@ -262,6 +345,11 @@ define void @constant_sextload_v16i16_to_v16i32(<16 x i32> addrspace(1)* %out, < ; GCN-DAG: s_mov_b32 [[K:s[0-9]+]], 0xffff{{$}} ; GCN: s_and_b32 s{{[0-9]+}}, s{{[0-9]+}}, [[K]] ; GCN: s_lshr_b32 s{{[0-9]+}}, s{{[0-9]+}}, 16 + +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 0, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 16, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 32, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 48, #1 define void @constant_zextload_v32i16_to_v32i32(<32 x i32> addrspace(1)* %out, <32 x i16> addrspace(2)* %in) #0 { %load = load <32 x i16>, <32 x i16> addrspace(2)* %in %ext = zext <32 x i16> %load to <32 x i32> @@ -273,6 +361,11 @@ define void @constant_zextload_v32i16_to_v32i32(<32 x i32> addrspace(1)* %out, < ; GCN: s_load_dwordx16 ; GCN-DAG: s_ashr_i32 ; GCN-DAG: s_sext_i32_i16 + +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 0, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 16, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 32, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 48, #1 define void @constant_sextload_v32i16_to_v32i32(<32 x i32> addrspace(1)* %out, <32 x i16> addrspace(2)* %in) #0 { %load = load <32 x i16>, <32 x i16> addrspace(2)* %in %ext = sext <32 x i16> %load to <32 x i32> @@ -283,6 +376,15 @@ define void @constant_sextload_v32i16_to_v32i32(<32 x i32> addrspace(1)* %out, < ; FUNC-LABEL: {{^}}constant_zextload_v64i16_to_v64i32: ; GCN: s_load_dwordx16 ; GCN: s_load_dwordx16 + +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 0, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 16, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 32, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 48, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 64, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 80, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 96, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 112, #1 define void @constant_zextload_v64i16_to_v64i32(<64 x i32> addrspace(1)* %out, <64 x i16> addrspace(2)* %in) #0 { %load = load <64 x i16>, <64 x i16> addrspace(2)* %in %ext = zext <64 x i16> %load to <64 x i32> @@ -291,6 +393,15 @@ define void @constant_zextload_v64i16_to_v64i32(<64 x i32> addrspace(1)* %out, < } ; FUNC-LABEL: {{^}}constant_sextload_v64i16_to_v64i32: + +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 0, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 16, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 32, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 48, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 64, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 80, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 96, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+\.[XYZW]}}, 112, #1 define void @constant_sextload_v64i16_to_v64i32(<64 x i32> addrspace(1)* %out, <64 x i16> addrspace(2)* %in) #0 { %load = load <64 x i16>, <64 x i16> addrspace(2)* %in %ext = sext <64 x i16> %load to <64 x i32> @@ -305,6 +416,9 @@ define void @constant_sextload_v64i16_to_v64i32(<64 x i32> addrspace(1)* %out, < ; GCN-NOHSA: buffer_store_dwordx2 v{{\[}}[[LO]]:[[HI]]] ; GCN-HSA: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[LO]]:[[HI]]{{\]}} + +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG: MOV {{.*}}, 0.0 define void @constant_zextload_i16_to_i64(i64 addrspace(1)* %out, i16 addrspace(2)* %in) #0 { %a = load i16, i16 addrspace(2)* %in %ext = zext i16 %a to i64 @@ -319,6 +433,11 @@ define void @constant_zextload_i16_to_i64(i64 addrspace(1)* %out, i16 addrspace( ; GCN-NOHSA: buffer_store_dwordx2 v{{\[}}[[LO]]:[[HI]]] ; GCN-HSA: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[LO]]:[[HI]]{{\]}} + +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG: ASHR {{\**}} {{T[0-9]\.[XYZW]}}, {{.*}}, literal +; TODO: Why not 15 ? +; EG: 31 define void @constant_sextload_i16_to_i64(i64 addrspace(1)* %out, i16 addrspace(2)* %in) #0 { %a = load i16, i16 addrspace(2)* %in %ext = sext i16 %a to i64 @@ -327,6 +446,9 @@ define void @constant_sextload_i16_to_i64(i64 addrspace(1)* %out, i16 addrspace( } ; FUNC-LABEL: {{^}}constant_zextload_v1i16_to_v1i64: + +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG: MOV {{.*}}, 0.0 define void @constant_zextload_v1i16_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i16> addrspace(2)* %in) #0 { %load = load <1 x i16>, <1 x i16> addrspace(2)* %in %ext = zext <1 x i16> %load to <1 x i64> @@ -335,6 +457,11 @@ define void @constant_zextload_v1i16_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x } ; FUNC-LABEL: {{^}}constant_sextload_v1i16_to_v1i64: + +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG: ASHR {{\**}} {{T[0-9]\.[XYZW]}}, {{.*}}, literal +; TODO: Why not 15 ? +; EG: 31 define void @constant_sextload_v1i16_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i16> addrspace(2)* %in) #0 { %load = load <1 x i16>, <1 x i16> addrspace(2)* %in %ext = sext <1 x i16> %load to <1 x i64> @@ -343,6 +470,8 @@ define void @constant_sextload_v1i16_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x } ; FUNC-LABEL: {{^}}constant_zextload_v2i16_to_v2i64: + +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @constant_zextload_v2i16_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i16> addrspace(2)* %in) #0 { %load = load <2 x i16>, <2 x i16> addrspace(2)* %in %ext = zext <2 x i16> %load to <2 x i64> @@ -351,6 +480,8 @@ define void @constant_zextload_v2i16_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x } ; FUNC-LABEL: {{^}}constant_sextload_v2i16_to_v2i64: + +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @constant_sextload_v2i16_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i16> addrspace(2)* %in) #0 { %load = load <2 x i16>, <2 x i16> addrspace(2)* %in %ext = sext <2 x i16> %load to <2 x i64> @@ -359,6 +490,8 @@ define void @constant_sextload_v2i16_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x } ; FUNC-LABEL: {{^}}constant_zextload_v4i16_to_v4i64: + +; EG: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0, #1 define void @constant_zextload_v4i16_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i16> addrspace(2)* %in) #0 { %load = load <4 x i16>, <4 x i16> addrspace(2)* %in %ext = zext <4 x i16> %load to <4 x i64> @@ -367,6 +500,8 @@ define void @constant_zextload_v4i16_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x } ; FUNC-LABEL: {{^}}constant_sextload_v4i16_to_v4i64: + +; EG: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0, #1 define void @constant_sextload_v4i16_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i16> addrspace(2)* %in) #0 { %load = load <4 x i16>, <4 x i16> addrspace(2)* %in %ext = sext <4 x i16> %load to <4 x i64> @@ -375,6 +510,8 @@ define void @constant_sextload_v4i16_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x } ; FUNC-LABEL: {{^}}constant_zextload_v8i16_to_v8i64: + +; EG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 define void @constant_zextload_v8i16_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i16> addrspace(2)* %in) #0 { %load = load <8 x i16>, <8 x i16> addrspace(2)* %in %ext = zext <8 x i16> %load to <8 x i64> @@ -383,6 +520,8 @@ define void @constant_zextload_v8i16_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x } ; FUNC-LABEL: {{^}}constant_sextload_v8i16_to_v8i64: + +; EG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 define void @constant_sextload_v8i16_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i16> addrspace(2)* %in) #0 { %load = load <8 x i16>, <8 x i16> addrspace(2)* %in %ext = sext <8 x i16> %load to <8 x i64> @@ -391,6 +530,9 @@ define void @constant_sextload_v8i16_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x } ; FUNC-LABEL: {{^}}constant_zextload_v16i16_to_v16i64: + +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 define void @constant_zextload_v16i16_to_v16i64(<16 x i64> addrspace(1)* %out, <16 x i16> addrspace(2)* %in) #0 { %load = load <16 x i16>, <16 x i16> addrspace(2)* %in %ext = zext <16 x i16> %load to <16 x i64> @@ -399,6 +541,9 @@ define void @constant_zextload_v16i16_to_v16i64(<16 x i64> addrspace(1)* %out, < } ; FUNC-LABEL: {{^}}constant_sextload_v16i16_to_v16i64: + +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 define void @constant_sextload_v16i16_to_v16i64(<16 x i64> addrspace(1)* %out, <16 x i16> addrspace(2)* %in) #0 { %load = load <16 x i16>, <16 x i16> addrspace(2)* %in %ext = sext <16 x i16> %load to <16 x i64> @@ -407,6 +552,11 @@ define void @constant_sextload_v16i16_to_v16i64(<16 x i64> addrspace(1)* %out, < } ; FUNC-LABEL: {{^}}constant_zextload_v32i16_to_v32i64: + +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 32, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 48, #1 define void @constant_zextload_v32i16_to_v32i64(<32 x i64> addrspace(1)* %out, <32 x i16> addrspace(2)* %in) #0 { %load = load <32 x i16>, <32 x i16> addrspace(2)* %in %ext = zext <32 x i16> %load to <32 x i64> @@ -415,6 +565,11 @@ define void @constant_zextload_v32i16_to_v32i64(<32 x i64> addrspace(1)* %out, < } ; FUNC-LABEL: {{^}}constant_sextload_v32i16_to_v32i64: + +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 32, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 48, #1 define void @constant_sextload_v32i16_to_v32i64(<32 x i64> addrspace(1)* %out, <32 x i16> addrspace(2)* %in) #0 { %load = load <32 x i16>, <32 x i16> addrspace(2)* %in %ext = sext <32 x i16> %load to <32 x i64> @@ -422,6 +577,8 @@ define void @constant_sextload_v32i16_to_v32i64(<32 x i64> addrspace(1)* %out, < ret void } +; These trigger undefined register machine verifier errors + ; ; XFUNC-LABEL: {{^}}constant_zextload_v64i16_to_v64i64: ; define void @constant_zextload_v64i16_to_v64i64(<64 x i64> addrspace(1)* %out, <64 x i16> addrspace(2)* %in) #0 { ; %load = load <64 x i16>, <64 x i16> addrspace(2)* %in diff --git a/llvm/test/CodeGen/AMDGPU/load-constant-i8.ll b/llvm/test/CodeGen/AMDGPU/load-constant-i8.ll index 8782898..39a299f 100644 --- a/llvm/test/CodeGen/AMDGPU/load-constant-i8.ll +++ b/llvm/test/CodeGen/AMDGPU/load-constant-i8.ll @@ -8,7 +8,8 @@ ; GCN-NOHSA: buffer_load_ubyte v{{[0-9]+}} ; GCN-HSA: flat_load_ubyte -; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0 +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; TODO: NOT AND define void @constant_load_i8(i8 addrspace(1)* %out, i8 addrspace(2)* %in) #0 { entry: %ld = load i8, i8 addrspace(2)* %in @@ -20,7 +21,7 @@ entry: ; GCN-NOHSA: buffer_load_ushort v ; GCN-HSA: flat_load_ushort v -; EG: VTX_READ_16 +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @constant_load_v2i8(<2 x i8> addrspace(1)* %out, <2 x i8> addrspace(2)* %in) #0 { entry: %ld = load <2 x i8>, <2 x i8> addrspace(2)* %in @@ -31,7 +32,7 @@ entry: ; FUNC-LABEL: {{^}}constant_load_v3i8: ; GCN: s_load_dword s -; EG-DAG: VTX_READ_32 +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @constant_load_v3i8(<3 x i8> addrspace(1)* %out, <3 x i8> addrspace(2)* %in) #0 { entry: %ld = load <3 x i8>, <3 x i8> addrspace(2)* %in @@ -42,7 +43,7 @@ entry: ; FUNC-LABEL: {{^}}constant_load_v4i8: ; GCN: s_load_dword s -; EG: VTX_READ_32 +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @constant_load_v4i8(<4 x i8> addrspace(1)* %out, <4 x i8> addrspace(2)* %in) #0 { entry: %ld = load <4 x i8>, <4 x i8> addrspace(2)* %in @@ -53,7 +54,7 @@ entry: ; FUNC-LABEL: {{^}}constant_load_v8i8: ; GCN: s_load_dwordx2 -; EG: VTX_READ_64 +; EG: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0, #1 define void @constant_load_v8i8(<8 x i8> addrspace(1)* %out, <8 x i8> addrspace(2)* %in) #0 { entry: %ld = load <8 x i8>, <8 x i8> addrspace(2)* %in @@ -64,7 +65,7 @@ entry: ; FUNC-LABEL: {{^}}constant_load_v16i8: ; GCN: s_load_dwordx4 -; EG: VTX_READ_128 +; EG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 define void @constant_load_v16i8(<16 x i8> addrspace(1)* %out, <16 x i8> addrspace(2)* %in) #0 { entry: %ld = load <16 x i8>, <16 x i8> addrspace(2)* %in @@ -76,7 +77,7 @@ entry: ; GCN-NOHSA: buffer_load_ubyte v{{[0-9]+}}, ; GCN-HSA: flat_load_ubyte -; EG: VTX_READ_8 T{{[0-9]+\.X, T[0-9]+\.X}} +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @constant_zextload_i8_to_i32(i32 addrspace(1)* %out, i8 addrspace(2)* %in) #0 { %a = load i8, i8 addrspace(2)* %in %ext = zext i8 %a to i32 @@ -88,7 +89,7 @@ define void @constant_zextload_i8_to_i32(i32 addrspace(1)* %out, i8 addrspace(2) ; GCN-NOHSA: buffer_load_sbyte ; GCN-HSA: flat_load_sbyte -; EG: VTX_READ_8 [[DST:T[0-9]\.[XYZW]]], [[DST]] +; EG: VTX_READ_8 [[DST:T[0-9]+\.X]], T{{[0-9]+}}.X, 0, #1 ; EG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST]], 0.0, literal ; EG: 8 define void @constant_sextload_i8_to_i32(i32 addrspace(1)* %out, i8 addrspace(2)* %in) #0 { @@ -99,6 +100,8 @@ define void @constant_sextload_i8_to_i32(i32 addrspace(1)* %out, i8 addrspace(2) } ; FUNC-LABEL: {{^}}constant_zextload_v1i8_to_v1i32: + +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @constant_zextload_v1i8_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x i8> addrspace(2)* %in) #0 { %load = load <1 x i8>, <1 x i8> addrspace(2)* %in %ext = zext <1 x i8> %load to <1 x i32> @@ -107,6 +110,10 @@ define void @constant_zextload_v1i8_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x } ; FUNC-LABEL: {{^}}constant_sextload_v1i8_to_v1i32: + +; EG: VTX_READ_8 [[DST:T[0-9]+\.X]], T{{[0-9]+}}.X, 0, #1 +; EG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST]], 0.0, literal +; EG: 8 define void @constant_sextload_v1i8_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x i8> addrspace(2)* %in) #0 { %load = load <1 x i8>, <1 x i8> addrspace(2)* %in %ext = sext <1 x i8> %load to <1 x i32> @@ -117,8 +124,11 @@ define void @constant_sextload_v1i8_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x ; FUNC-LABEL: {{^}}constant_zextload_v2i8_to_v2i32: ; GCN-NOHSA: buffer_load_ushort ; GCN-HSA: flat_load_ushort -; EG: VTX_READ_8 -; EG: VTX_READ_8 + +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; TODO: This should use DST, but for some there are redundant MOVs +; EG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG: 8 define void @constant_zextload_v2i8_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i8> addrspace(2)* %in) #0 { %load = load <2 x i8>, <2 x i8> addrspace(2)* %in %ext = zext <2 x i8> %load to <2 x i32> @@ -134,10 +144,10 @@ define void @constant_zextload_v2i8_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x ; GCN: v_bfe_i32 ; GCN: v_bfe_i32 -; EG-DAG: VTX_READ_8 [[DST_X:T[0-9]\.[XYZW]]], [[DST_X]] -; EG-DAG: VTX_READ_8 [[DST_Y:T[0-9]\.[XYZW]]], [[DST_Y]] -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_X]], 0.0, literal -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_Y]], 0.0, literal +; EG: VTX_READ_16 [[DST:T[0-9]+\.X]], T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal ; EG-DAG: 8 ; EG-DAG: 8 define void @constant_sextload_v2i8_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i8> addrspace(2)* %in) #0 { @@ -153,6 +163,13 @@ define void @constant_sextload_v2i8_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x ; GCN-DAG: s_bfe_u32 ; GCN-DAG: s_bfe_u32 ; GCN-DAG: s_and_b32 + +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: 8 +; EG-DAG: 8 define void @constant_zextload_v3i8_to_v3i32(<3 x i32> addrspace(1)* %out, <3 x i8> addrspace(2)* %in) #0 { entry: %ld = load <3 x i8>, <3 x i8> addrspace(2)* %in @@ -167,6 +184,15 @@ entry: ; GCN-DAG: s_bfe_i32 ; GCN-DAG: s_bfe_i32 ; GCN-DAG: s_bfe_i32 + +; EG: VTX_READ_32 [[DST:T[0-9]+\.X]], T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 define void @constant_sextload_v3i8_to_v3i32(<3 x i32> addrspace(1)* %out, <3 x i8> addrspace(2)* %in) #0 { entry: %ld = load <3 x i8>, <3 x i8> addrspace(2)* %in @@ -180,10 +206,14 @@ entry: ; GCN-DAG: s_and_b32 ; GCN-DAG: s_lshr_b32 -; EG: VTX_READ_8 -; EG: VTX_READ_8 -; EG: VTX_READ_8 -; EG: VTX_READ_8 +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 define void @constant_zextload_v4i8_to_v4i32(<4 x i32> addrspace(1)* %out, <4 x i8> addrspace(2)* %in) #0 { %load = load <4 x i8>, <4 x i8> addrspace(2)* %in %ext = zext <4 x i8> %load to <4 x i32> @@ -196,14 +226,12 @@ define void @constant_zextload_v4i8_to_v4i32(<4 x i32> addrspace(1)* %out, <4 x ; GCN-DAG: s_sext_i32_i8 ; GCN-DAG: s_ashr_i32 -; EG-DAG: VTX_READ_8 [[DST_X:T[0-9]\.[XYZW]]], [[DST_X]] -; EG-DAG: VTX_READ_8 [[DST_Y:T[0-9]\.[XYZW]]], [[DST_Y]] -; EG-DAG: VTX_READ_8 [[DST_Z:T[0-9]\.[XYZW]]], [[DST_Z]] -; EG-DAG: VTX_READ_8 [[DST_W:T[0-9]\.[XYZW]]], [[DST_W]] -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_X]], 0.0, literal -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_Y]], 0.0, literal -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_Z]], 0.0, literal -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_W]], 0.0, literal +; EG: VTX_READ_32 [[DST:T[0-9]+\.X]], T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal ; EG-DAG: 8 ; EG-DAG: 8 ; EG-DAG: 8 @@ -219,6 +247,23 @@ define void @constant_sextload_v4i8_to_v4i32(<4 x i32> addrspace(1)* %out, <4 x ; GCN: s_load_dwordx2 ; GCN-DAG: s_and_b32 ; GCN-DAG: s_lshr_b32 + +; EG: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 define void @constant_zextload_v8i8_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x i8> addrspace(2)* %in) #0 { %load = load <8 x i8>, <8 x i8> addrspace(2)* %in %ext = zext <8 x i8> %load to <8 x i32> @@ -230,6 +275,25 @@ define void @constant_zextload_v8i8_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x ; GCN: s_load_dwordx2 ; GCN-DAG: s_ashr_i32 ; GCN-DAG: s_sext_i32_i8 + +; EG: VTX_READ_64 [[DST:T[0-9]+\.XY]], T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 define void @constant_sextload_v8i8_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x i8> addrspace(2)* %in) #0 { %load = load <8 x i8>, <8 x i8> addrspace(2)* %in %ext = sext <8 x i8> %load to <8 x i32> @@ -238,6 +302,39 @@ define void @constant_sextload_v8i8_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x } ; FUNC-LABEL: {{^}}constant_zextload_v16i8_to_v16i32: + +; EG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 define void @constant_zextload_v16i8_to_v16i32(<16 x i32> addrspace(1)* %out, <16 x i8> addrspace(2)* %in) #0 { %load = load <16 x i8>, <16 x i8> addrspace(2)* %in %ext = zext <16 x i8> %load to <16 x i32> @@ -246,6 +343,41 @@ define void @constant_zextload_v16i8_to_v16i32(<16 x i32> addrspace(1)* %out, <1 } ; FUNC-LABEL: {{^}}constant_sextload_v16i8_to_v16i32: + +; EG: VTX_READ_128 [[DST:T[0-9]+\.XYZW]], T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 define void @constant_sextload_v16i8_to_v16i32(<16 x i32> addrspace(1)* %out, <16 x i8> addrspace(2)* %in) #0 { %load = load <16 x i8>, <16 x i8> addrspace(2)* %in %ext = sext <16 x i8> %load to <16 x i32> @@ -254,6 +386,70 @@ define void @constant_sextload_v16i8_to_v16i32(<16 x i32> addrspace(1)* %out, <1 } ; FUNC-LABEL: {{^}}constant_zextload_v32i8_to_v32i32: + +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 define void @constant_zextload_v32i8_to_v32i32(<32 x i32> addrspace(1)* %out, <32 x i8> addrspace(2)* %in) #0 { %load = load <32 x i8>, <32 x i8> addrspace(2)* %in %ext = zext <32 x i8> %load to <32 x i32> @@ -262,6 +458,74 @@ define void @constant_zextload_v32i8_to_v32i32(<32 x i32> addrspace(1)* %out, <3 } ; FUNC-LABEL: {{^}}constant_sextload_v32i8_to_v32i32: + +; EG-DAG: VTX_READ_128 [[DST_LO:T[0-9]+\.XYZW]], T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 [[DST_HI:T[0-9]+\.XYZW]], T{{[0-9]+}}.X, 16, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 define void @constant_sextload_v32i8_to_v32i32(<32 x i32> addrspace(1)* %out, <32 x i8> addrspace(2)* %in) #0 { %load = load <32 x i8>, <32 x i8> addrspace(2)* %in %ext = sext <32 x i8> %load to <32 x i32> @@ -270,6 +534,11 @@ define void @constant_sextload_v32i8_to_v32i32(<32 x i32> addrspace(1)* %out, <3 } ; FUNC-LABEL: {{^}}constant_zextload_v64i8_to_v64i32: + +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, T{{[0-9]+}}.X, 16, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, T{{[0-9]+}}.X, 32, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, T{{[0-9]+}}.X, 48, #1 define void @constant_zextload_v64i8_to_v64i32(<64 x i32> addrspace(1)* %out, <64 x i8> addrspace(2)* %in) #0 { %load = load <64 x i8>, <64 x i8> addrspace(2)* %in %ext = zext <64 x i8> %load to <64 x i32> @@ -278,6 +547,11 @@ define void @constant_zextload_v64i8_to_v64i32(<64 x i32> addrspace(1)* %out, <6 } ; FUNC-LABEL: {{^}}constant_sextload_v64i8_to_v64i32: + +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, T{{[0-9]+}}.X, 16, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, T{{[0-9]+}}.X, 32, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, T{{[0-9]+}}.X, 48, #1 define void @constant_sextload_v64i8_to_v64i32(<64 x i32> addrspace(1)* %out, <64 x i8> addrspace(2)* %in) #0 { %load = load <64 x i8>, <64 x i8> addrspace(2)* %in %ext = sext <64 x i8> %load to <64 x i32> @@ -293,6 +567,9 @@ define void @constant_sextload_v64i8_to_v64i32(<64 x i32> addrspace(1)* %out, <6 ; GCN-HSA-DAG: flat_load_ubyte v[[LO:[0-9]+]], ; GCN-HSA: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[LO]]:[[HI]]] + +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG: MOV {{.*}}, 0.0 define void @constant_zextload_i8_to_i64(i64 addrspace(1)* %out, i8 addrspace(2)* %in) #0 { %a = load i8, i8 addrspace(2)* %in %ext = zext i8 %a to i64 @@ -307,6 +584,11 @@ define void @constant_zextload_i8_to_i64(i64 addrspace(1)* %out, i8 addrspace(2) ; GCN-NOHSA: buffer_store_dwordx2 v{{\[}}[[LO]]:[[HI]]{{\]}} ; GCN-HSA: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[LO]]:[[HI]]{{\]}} + +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG: ASHR {{\**}} {{T[0-9]\.[XYZW]}}, {{.*}}, literal +; TODO: Why not 7 ? +; EG: 31 define void @constant_sextload_i8_to_i64(i64 addrspace(1)* %out, i8 addrspace(2)* %in) #0 { %a = load i8, i8 addrspace(2)* %in %ext = sext i8 %a to i64 @@ -315,6 +597,9 @@ define void @constant_sextload_i8_to_i64(i64 addrspace(1)* %out, i8 addrspace(2) } ; FUNC-LABEL: {{^}}constant_zextload_v1i8_to_v1i64: + +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG: MOV {{.*}}, 0.0 define void @constant_zextload_v1i8_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i8> addrspace(2)* %in) #0 { %load = load <1 x i8>, <1 x i8> addrspace(2)* %in %ext = zext <1 x i8> %load to <1 x i64> @@ -323,6 +608,11 @@ define void @constant_zextload_v1i8_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x } ; FUNC-LABEL: {{^}}constant_sextload_v1i8_to_v1i64: + +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG: ASHR {{\**}} {{T[0-9]\.[XYZW]}}, {{.*}}, literal +; TODO: Why not 7 ? +; EG: 31 define void @constant_sextload_v1i8_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i8> addrspace(2)* %in) #0 { %load = load <1 x i8>, <1 x i8> addrspace(2)* %in %ext = sext <1 x i8> %load to <1 x i64> @@ -331,6 +621,8 @@ define void @constant_sextload_v1i8_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x } ; FUNC-LABEL: {{^}}constant_zextload_v2i8_to_v2i64: + +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @constant_zextload_v2i8_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i8> addrspace(2)* %in) #0 { %load = load <2 x i8>, <2 x i8> addrspace(2)* %in %ext = zext <2 x i8> %load to <2 x i64> @@ -339,6 +631,8 @@ define void @constant_zextload_v2i8_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x } ; FUNC-LABEL: {{^}}constant_sextload_v2i8_to_v2i64: + +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @constant_sextload_v2i8_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i8> addrspace(2)* %in) #0 { %load = load <2 x i8>, <2 x i8> addrspace(2)* %in %ext = sext <2 x i8> %load to <2 x i64> @@ -347,6 +641,8 @@ define void @constant_sextload_v2i8_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x } ; FUNC-LABEL: {{^}}constant_zextload_v4i8_to_v4i64: + +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @constant_zextload_v4i8_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i8> addrspace(2)* %in) #0 { %load = load <4 x i8>, <4 x i8> addrspace(2)* %in %ext = zext <4 x i8> %load to <4 x i64> @@ -355,6 +651,8 @@ define void @constant_zextload_v4i8_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x } ; FUNC-LABEL: {{^}}constant_sextload_v4i8_to_v4i64: + +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @constant_sextload_v4i8_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i8> addrspace(2)* %in) #0 { %load = load <4 x i8>, <4 x i8> addrspace(2)* %in %ext = sext <4 x i8> %load to <4 x i64> @@ -363,6 +661,8 @@ define void @constant_sextload_v4i8_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x } ; FUNC-LABEL: {{^}}constant_zextload_v8i8_to_v8i64: + +; EG: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0, #1 define void @constant_zextload_v8i8_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i8> addrspace(2)* %in) #0 { %load = load <8 x i8>, <8 x i8> addrspace(2)* %in %ext = zext <8 x i8> %load to <8 x i64> @@ -371,6 +671,8 @@ define void @constant_zextload_v8i8_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x } ; FUNC-LABEL: {{^}}constant_sextload_v8i8_to_v8i64: + +; EG: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0, #1 define void @constant_sextload_v8i8_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i8> addrspace(2)* %in) #0 { %load = load <8 x i8>, <8 x i8> addrspace(2)* %in %ext = sext <8 x i8> %load to <8 x i64> @@ -379,6 +681,8 @@ define void @constant_sextload_v8i8_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x } ; FUNC-LABEL: {{^}}constant_zextload_v16i8_to_v16i64: + +; EG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 define void @constant_zextload_v16i8_to_v16i64(<16 x i64> addrspace(1)* %out, <16 x i8> addrspace(2)* %in) #0 { %load = load <16 x i8>, <16 x i8> addrspace(2)* %in %ext = zext <16 x i8> %load to <16 x i64> @@ -387,6 +691,8 @@ define void @constant_zextload_v16i8_to_v16i64(<16 x i64> addrspace(1)* %out, <1 } ; FUNC-LABEL: {{^}}constant_sextload_v16i8_to_v16i64: + +; EG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 define void @constant_sextload_v16i8_to_v16i64(<16 x i64> addrspace(1)* %out, <16 x i8> addrspace(2)* %in) #0 { %load = load <16 x i8>, <16 x i8> addrspace(2)* %in %ext = sext <16 x i8> %load to <16 x i64> @@ -395,6 +701,9 @@ define void @constant_sextload_v16i8_to_v16i64(<16 x i64> addrspace(1)* %out, <1 } ; FUNC-LABEL: {{^}}constant_zextload_v32i8_to_v32i64: + +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 define void @constant_zextload_v32i8_to_v32i64(<32 x i64> addrspace(1)* %out, <32 x i8> addrspace(2)* %in) #0 { %load = load <32 x i8>, <32 x i8> addrspace(2)* %in %ext = zext <32 x i8> %load to <32 x i64> @@ -403,6 +712,9 @@ define void @constant_zextload_v32i8_to_v32i64(<32 x i64> addrspace(1)* %out, <3 } ; FUNC-LABEL: {{^}}constant_sextload_v32i8_to_v32i64: + +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 define void @constant_sextload_v32i8_to_v32i64(<32 x i64> addrspace(1)* %out, <32 x i8> addrspace(2)* %in) #0 { %load = load <32 x i8>, <32 x i8> addrspace(2)* %in %ext = sext <32 x i8> %load to <32 x i64> @@ -445,6 +757,8 @@ define void @constant_zextload_i8_to_i16(i16 addrspace(1)* %out, i8 addrspace(2) ; GCN-NOHSA: buffer_store_short v[[VAL]] ; GCN-HSA: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, v[[VAL]] + +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @constant_sextload_i8_to_i16(i16 addrspace(1)* %out, i8 addrspace(2)* %in) #0 { %a = load i8, i8 addrspace(2)* %in %ext = sext i8 %a to i16 @@ -461,6 +775,9 @@ define void @constant_zextload_v1i8_to_v1i16(<1 x i16> addrspace(1)* %out, <1 x } ; FUNC-LABEL: {{^}}constant_sextload_v1i8_to_v1i16: + +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal define void @constant_sextload_v1i8_to_v1i16(<1 x i16> addrspace(1)* %out, <1 x i8> addrspace(2)* %in) #0 { %load = load <1 x i8>, <1 x i8> addrspace(2)* %in %ext = sext <1 x i8> %load to <1 x i16> @@ -469,6 +786,8 @@ define void @constant_sextload_v1i8_to_v1i16(<1 x i16> addrspace(1)* %out, <1 x } ; FUNC-LABEL: {{^}}constant_zextload_v2i8_to_v2i16: + +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @constant_zextload_v2i8_to_v2i16(<2 x i16> addrspace(1)* %out, <2 x i8> addrspace(2)* %in) #0 { %load = load <2 x i8>, <2 x i8> addrspace(2)* %in %ext = zext <2 x i8> %load to <2 x i16> @@ -477,6 +796,10 @@ define void @constant_zextload_v2i8_to_v2i16(<2 x i16> addrspace(1)* %out, <2 x } ; FUNC-LABEL: {{^}}constant_sextload_v2i8_to_v2i16: + +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal define void @constant_sextload_v2i8_to_v2i16(<2 x i16> addrspace(1)* %out, <2 x i8> addrspace(2)* %in) #0 { %load = load <2 x i8>, <2 x i8> addrspace(2)* %in %ext = sext <2 x i8> %load to <2 x i16> @@ -485,6 +808,8 @@ define void @constant_sextload_v2i8_to_v2i16(<2 x i16> addrspace(1)* %out, <2 x } ; FUNC-LABEL: {{^}}constant_zextload_v4i8_to_v4i16: + +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @constant_zextload_v4i8_to_v4i16(<4 x i16> addrspace(1)* %out, <4 x i8> addrspace(2)* %in) #0 { %load = load <4 x i8>, <4 x i8> addrspace(2)* %in %ext = zext <4 x i8> %load to <4 x i16> @@ -493,6 +818,12 @@ define void @constant_zextload_v4i8_to_v4i16(<4 x i16> addrspace(1)* %out, <4 x } ; FUNC-LABEL: {{^}}constant_sextload_v4i8_to_v4i16: + +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal define void @constant_sextload_v4i8_to_v4i16(<4 x i16> addrspace(1)* %out, <4 x i8> addrspace(2)* %in) #0 { %load = load <4 x i8>, <4 x i8> addrspace(2)* %in %ext = sext <4 x i8> %load to <4 x i16> @@ -501,6 +832,8 @@ define void @constant_sextload_v4i8_to_v4i16(<4 x i16> addrspace(1)* %out, <4 x } ; FUNC-LABEL: {{^}}constant_zextload_v8i8_to_v8i16: + +; EG: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0, #1 define void @constant_zextload_v8i8_to_v8i16(<8 x i16> addrspace(1)* %out, <8 x i8> addrspace(2)* %in) #0 { %load = load <8 x i8>, <8 x i8> addrspace(2)* %in %ext = zext <8 x i8> %load to <8 x i16> @@ -509,6 +842,17 @@ define void @constant_zextload_v8i8_to_v8i16(<8 x i16> addrspace(1)* %out, <8 x } ; FUNC-LABEL: {{^}}constant_sextload_v8i8_to_v8i16: + +; EG: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal + define void @constant_sextload_v8i8_to_v8i16(<8 x i16> addrspace(1)* %out, <8 x i8> addrspace(2)* %in) #0 { %load = load <8 x i8>, <8 x i8> addrspace(2)* %in %ext = sext <8 x i8> %load to <8 x i16> @@ -517,6 +861,8 @@ define void @constant_sextload_v8i8_to_v8i16(<8 x i16> addrspace(1)* %out, <8 x } ; FUNC-LABEL: {{^}}constant_zextload_v16i8_to_v16i16: + +; EG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 define void @constant_zextload_v16i8_to_v16i16(<16 x i16> addrspace(1)* %out, <16 x i8> addrspace(2)* %in) #0 { %load = load <16 x i8>, <16 x i8> addrspace(2)* %in %ext = zext <16 x i8> %load to <16 x i16> @@ -525,6 +871,24 @@ define void @constant_zextload_v16i8_to_v16i16(<16 x i16> addrspace(1)* %out, <1 } ; FUNC-LABEL: {{^}}constant_sextload_v16i8_to_v16i16: + +; EG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal define void @constant_sextload_v16i8_to_v16i16(<16 x i16> addrspace(1)* %out, <16 x i8> addrspace(2)* %in) #0 { %load = load <16 x i8>, <16 x i8> addrspace(2)* %in %ext = sext <16 x i8> %load to <16 x i16> @@ -533,6 +897,9 @@ define void @constant_sextload_v16i8_to_v16i16(<16 x i16> addrspace(1)* %out, <1 } ; FUNC-LABEL: {{^}}constant_zextload_v32i8_to_v32i16: + +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 define void @constant_zextload_v32i8_to_v32i16(<32 x i16> addrspace(1)* %out, <32 x i8> addrspace(2)* %in) #0 { %load = load <32 x i8>, <32 x i8> addrspace(2)* %in %ext = zext <32 x i8> %load to <32 x i16> @@ -541,6 +908,41 @@ define void @constant_zextload_v32i8_to_v32i16(<32 x i16> addrspace(1)* %out, <3 } ; FUNC-LABEL: {{^}}constant_sextload_v32i8_to_v32i16: + +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal define void @constant_sextload_v32i8_to_v32i16(<32 x i16> addrspace(1)* %out, <32 x i8> addrspace(2)* %in) #0 { %load = load <32 x i8>, <32 x i8> addrspace(2)* %in %ext = sext <32 x i8> %load to <32 x i16> diff --git a/llvm/test/CodeGen/AMDGPU/load-global-i16.ll b/llvm/test/CodeGen/AMDGPU/load-global-i16.ll index 11e6b10..86739ff 100644 --- a/llvm/test/CodeGen/AMDGPU/load-global-i16.ll +++ b/llvm/test/CodeGen/AMDGPU/load-global-i16.ll @@ -10,7 +10,7 @@ ; GCN-NOHSA: buffer_load_ushort v{{[0-9]+}} ; GCN-HSA: flat_load_ushort -; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0 +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @global_load_i16(i16 addrspace(1)* %out, i16 addrspace(1)* %in) { entry: %ld = load i16, i16 addrspace(1)* %in @@ -22,7 +22,7 @@ entry: ; GCN-NOHSA: buffer_load_dword v ; GCN-HSA: flat_load_dword v -; EG: VTX_READ_32 +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @global_load_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) { entry: %ld = load <2 x i16>, <2 x i16> addrspace(1)* %in @@ -34,8 +34,8 @@ entry: ; GCN-NOHSA: buffer_load_dwordx2 v ; GCN-HSA: flat_load_dwordx2 v -; EG-DAG: VTX_READ_32 -; EG-DAG: VTX_READ_16 +; EG-DAG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 4, #1 define void @global_load_v3i16(<3 x i16> addrspace(1)* %out, <3 x i16> addrspace(1)* %in) { entry: %ld = load <3 x i16>, <3 x i16> addrspace(1)* %in @@ -47,7 +47,7 @@ entry: ; GCN-NOHSA: buffer_load_dwordx2 ; GCN-HSA: flat_load_dwordx2 -; EG: VTX_READ_64 +; EG: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0, #1 define void @global_load_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> addrspace(1)* %in) { entry: %ld = load <4 x i16>, <4 x i16> addrspace(1)* %in @@ -59,7 +59,7 @@ entry: ; GCN-NOHSA: buffer_load_dwordx4 ; GCN-HSA: flat_load_dwordx4 -; EG: VTX_READ_128 +; EG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 define void @global_load_v8i16(<8 x i16> addrspace(1)* %out, <8 x i16> addrspace(1)* %in) { entry: %ld = load <8 x i16>, <8 x i16> addrspace(1)* %in @@ -74,8 +74,8 @@ entry: ; GCN-HSA: flat_load_dwordx4 ; GCN-HSA: flat_load_dwordx4 -; EG: VTX_READ_128 -; EG: VTX_READ_128 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 define void @global_load_v16i16(<16 x i16> addrspace(1)* %out, <16 x i16> addrspace(1)* %in) { entry: %ld = load <16 x i16>, <16 x i16> addrspace(1)* %in @@ -90,7 +90,7 @@ entry: ; GCN-HSA: flat_load_ushort ; GCN-HSA: flat_store_dword -; EG: VTX_READ_16 T{{[0-9]+\.X, T[0-9]+\.X}} +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @global_zextload_i16_to_i32(i32 addrspace(1)* %out, i16 addrspace(1)* %in) #0 { %a = load i16, i16 addrspace(1)* %in %ext = zext i16 %a to i32 @@ -105,7 +105,7 @@ define void @global_zextload_i16_to_i32(i32 addrspace(1)* %out, i16 addrspace(1) ; GCN-HSA: flat_load_sshort ; GCN-HSA: flat_store_dword -; EG: VTX_READ_16 [[DST:T[0-9]\.[XYZW]]], [[DST]] +; EG: VTX_READ_16 [[DST:T[0-9]\.[XYZW]]], T{{[0-9]+}}.X, 0, #1 ; EG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST]], 0.0, literal ; EG: 16 define void @global_sextload_i16_to_i32(i32 addrspace(1)* %out, i16 addrspace(1)* %in) #0 { @@ -118,6 +118,8 @@ define void @global_sextload_i16_to_i32(i32 addrspace(1)* %out, i16 addrspace(1) ; FUNC-LABEL: {{^}}global_zextload_v1i16_to_v1i32: ; GCN-NOHSA: buffer_load_ushort ; GCN-HSA: flat_load_ushort + +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @global_zextload_v1i16_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x i16> addrspace(1)* %in) #0 { %load = load <1 x i16>, <1 x i16> addrspace(1)* %in %ext = zext <1 x i16> %load to <1 x i32> @@ -128,6 +130,10 @@ define void @global_zextload_v1i16_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x i ; FUNC-LABEL: {{^}}global_sextload_v1i16_to_v1i32: ; GCN-NOHSA: buffer_load_sshort ; GCN-HSA: flat_load_sshort + +; EG: VTX_READ_16 [[DST:T[0-9]\.[XYZW]]], T{{[0-9]+}}.X, 0, #1 +; EG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST]], 0.0, literal +; EG: 16 define void @global_sextload_v1i16_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x i16> addrspace(1)* %in) #0 { %load = load <1 x i16>, <1 x i16> addrspace(1)* %in %ext = sext <1 x i16> %load to <1 x i32> @@ -138,6 +144,11 @@ define void @global_sextload_v1i16_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x i ; FUNC-LABEL: {{^}}global_zextload_v2i16_to_v2i32: ; GCN-NOHSA: buffer_load_dword ; GCN-HSA: flat_load_dword + +; EG: VTX_READ_32 [[DST:T[0-9]\.[XYZW]]], [[DST]], 0, #1 +; TODO: This should use DST, but for some there are redundant MOVs +; EG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{PV.[XYZW]}}, literal +; EG: 16 define void @global_zextload_v2i16_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 { %load = load <2 x i16>, <2 x i16> addrspace(1)* %in %ext = zext <2 x i16> %load to <2 x i32> @@ -150,10 +161,11 @@ define void @global_zextload_v2i16_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i ; GCN-HSA: flat_load_dword -; EG-DAG: VTX_READ_16 [[DST_X:T[0-9]\.[XYZW]]], [[DST_X]] -; EG-DAG: VTX_READ_16 [[DST_Y:T[0-9]\.[XYZW]]], [[DST_Y]] -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_X]], 0.0, literal -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_Y]], 0.0, literal +; EG: VTX_READ_32 [[DST:T[0-9]\.[XYZW]]], [[DST]], 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; TODO: We should also use ASHR instead of LSHR + BFE +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{PV.[XYZW]}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{PV.[XYZW]}}, 0.0, literal ; EG-DAG: 16 ; EG-DAG: 16 define void @global_sextload_v2i16_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 { @@ -166,6 +178,12 @@ define void @global_sextload_v2i16_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i ; FUNC-LABEL: {{^}}global_global_zextload_v3i16_to_v3i32: ; GCN-NOHSA: buffer_load_dwordx2 ; GCN-HSA: flat_load_dwordx2 + +; EG-DAG: VTX_READ_32 [[DST_HI:T[0-9]\.[XYZW]]], [[DST_HI]], 0, #1 +; EG-DAG: VTX_READ_16 [[DST_LO:T[0-9]\.[XYZW]]], [[DST_LO]], 4, #1 +; TODO: This should use DST, but for some there are redundant MOVs +; EG: LSHR {{[* ]*}}{{T[0-9].[XYZW]}}, {{T[0-9].[XYZW]}}, literal +; EG: 16 define void @global_global_zextload_v3i16_to_v3i32(<3 x i32> addrspace(1)* %out, <3 x i16> addrspace(1)* %in) { entry: %ld = load <3 x i16>, <3 x i16> addrspace(1)* %in @@ -177,6 +195,15 @@ entry: ; FUNC-LABEL: {{^}}global_global_sextload_v3i16_to_v3i32: ; GCN-NOHSA: buffer_load_dwordx2 ; GCN-HSA: flat_load_dwordx2 + +; EG-DAG: VTX_READ_32 [[DST_HI:T[0-9]\.[XYZW]]], [[DST_HI]], 0, #1 +; EG-DAG: VTX_READ_16 [[DST_LO:T[0-9]\.[XYZW]]], [[DST_LO]], 4, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: ASHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{PV.[XYZW]}}, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{T[0-9].[XYZW]}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{T[0-9].[XYZW]}}, 0.0, literal +; EG-DAG: 16 +; EG-DAG: 16 define void @global_global_sextload_v3i16_to_v3i32(<3 x i32> addrspace(1)* %out, <3 x i16> addrspace(1)* %in) { entry: %ld = load <3 x i16>, <3 x i16> addrspace(1)* %in @@ -190,10 +217,12 @@ entry: ; GCN-HSA: flat_load_dwordx2 -; EG: VTX_READ_16 -; EG: VTX_READ_16 -; EG: VTX_READ_16 -; EG: VTX_READ_16 +; EG: VTX_READ_64 [[DST:T[0-9]\.XY]], {{T[0-9].[XYZW]}}, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{PV.[XYZW]}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{T[0-9].[XYZW]}}, literal +; EG-DAG: 16 +; EG-DAG: 16 define void @global_global_zextload_v4i16_to_v4i32(<4 x i32> addrspace(1)* %out, <4 x i16> addrspace(1)* %in) #0 { %load = load <4 x i16>, <4 x i16> addrspace(1)* %in %ext = zext <4 x i16> %load to <4 x i32> @@ -206,14 +235,13 @@ define void @global_global_zextload_v4i16_to_v4i32(<4 x i32> addrspace(1)* %out, ; GCN-HSA: flat_load_dwordx2 -; EG-DAG: VTX_READ_16 [[DST_X:T[0-9]\.[XYZW]]], [[DST_X]] -; EG-DAG: VTX_READ_16 [[DST_Y:T[0-9]\.[XYZW]]], [[DST_Y]] -; EG-DAG: VTX_READ_16 [[DST_Z:T[0-9]\.[XYZW]]], [[DST_Z]] -; EG-DAG: VTX_READ_16 [[DST_W:T[0-9]\.[XYZW]]], [[DST_W]] -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_X]], 0.0, literal -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_Y]], 0.0, literal -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_Z]], 0.0, literal -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_W]], 0.0, literal +; EG: VTX_READ_64 [[DST:T[0-9]\.XY]], {{T[0-9].[XYZW]}}, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; TODO: We should use ASHR instead of LSHR + BFE +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal ; EG-DAG: 16 ; EG-DAG: 16 ; EG-DAG: 16 @@ -228,6 +256,17 @@ define void @global_sextload_v4i16_to_v4i32(<4 x i32> addrspace(1)* %out, <4 x i ; FUNC-LABEL: {{^}}global_zextload_v8i16_to_v8i32: ; GCN-NOHSA: buffer_load_dwordx4 ; GCN-HSA: flat_load_dwordx4 + +; EG: VTX_READ_128 [[DST:T[0-9]\.XYZW]], {{T[0-9].[XYZW]}}, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 define void @global_zextload_v8i16_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x i16> addrspace(1)* %in) #0 { %load = load <8 x i16>, <8 x i16> addrspace(1)* %in %ext = zext <8 x i16> %load to <8 x i32> @@ -238,6 +277,25 @@ define void @global_zextload_v8i16_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x i ; FUNC-LABEL: {{^}}global_sextload_v8i16_to_v8i32: ; GCN-NOHSA: buffer_load_dwordx4 ; GCN-HSA: flat_load_dwordx4 + +; EG: VTX_READ_128 [[DST:T[0-9]\.XYZW]], {{T[0-9].[XYZW]}}, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: LSHR {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 +; EG-DAG: 16 define void @global_sextload_v8i16_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x i16> addrspace(1)* %in) #0 { %load = load <8 x i16>, <8 x i16> addrspace(1)* %in %ext = sext <8 x i16> %load to <8 x i32> @@ -251,6 +309,9 @@ define void @global_sextload_v8i16_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x i ; GCN-HSA: flat_load_dwordx4 ; GCN-HSA: flat_load_dwordx4 + +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 0, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 16, #1 define void @global_zextload_v16i16_to_v16i32(<16 x i32> addrspace(1)* %out, <16 x i16> addrspace(1)* %in) #0 { %load = load <16 x i16>, <16 x i16> addrspace(1)* %in %ext = zext <16 x i16> %load to <16 x i32> @@ -259,6 +320,9 @@ define void @global_zextload_v16i16_to_v16i32(<16 x i32> addrspace(1)* %out, <16 } ; FUNC-LABEL: {{^}}global_sextload_v16i16_to_v16i32: + +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 0, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 16, #1 define void @global_sextload_v16i16_to_v16i32(<16 x i32> addrspace(1)* %out, <16 x i16> addrspace(1)* %in) #0 { %load = load <16 x i16>, <16 x i16> addrspace(1)* %in %ext = sext <16 x i16> %load to <16 x i32> @@ -276,6 +340,11 @@ define void @global_sextload_v16i16_to_v16i32(<16 x i32> addrspace(1)* %out, <16 ; GCN-HSA: flat_load_dwordx4 ; GCN-HSA: flat_load_dwordx4 ; GCN-HSA: flat_load_dwordx4 + +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 0, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 16, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 32, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 48, #1 define void @global_zextload_v32i16_to_v32i32(<32 x i32> addrspace(1)* %out, <32 x i16> addrspace(1)* %in) #0 { %load = load <32 x i16>, <32 x i16> addrspace(1)* %in %ext = zext <32 x i16> %load to <32 x i32> @@ -293,6 +362,11 @@ define void @global_zextload_v32i16_to_v32i32(<32 x i32> addrspace(1)* %out, <32 ; GCN-HSA: flat_load_dwordx4 ; GCN-HSA: flat_load_dwordx4 ; GCN-HSA: flat_load_dwordx4 + +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 0, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 16, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 32, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 48, #1 define void @global_sextload_v32i16_to_v32i32(<32 x i32> addrspace(1)* %out, <32 x i16> addrspace(1)* %in) #0 { %load = load <32 x i16>, <32 x i16> addrspace(1)* %in %ext = sext <32 x i16> %load to <32 x i32> @@ -318,6 +392,15 @@ define void @global_sextload_v32i16_to_v32i32(<32 x i32> addrspace(1)* %out, <32 ; GCN-HSA: flat_load_dwordx4 ; GCN-HSA: flat_load_dwordx4 ; GCN-HSA: flat_load_dwordx4 + +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 0, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 16, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 32, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 48, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 64, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 80, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 96, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 112, #1 define void @global_zextload_v64i16_to_v64i32(<64 x i32> addrspace(1)* %out, <64 x i16> addrspace(1)* %in) #0 { %load = load <64 x i16>, <64 x i16> addrspace(1)* %in %ext = zext <64 x i16> %load to <64 x i32> @@ -326,6 +409,15 @@ define void @global_zextload_v64i16_to_v64i32(<64 x i32> addrspace(1)* %out, <64 } ; FUNC-LABEL: {{^}}global_sextload_v64i16_to_v64i32: + +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 0, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 16, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 32, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 48, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 64, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 80, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 96, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, {{T[0-9]+.[XYZW]}}, 112, #1 define void @global_sextload_v64i16_to_v64i32(<64 x i32> addrspace(1)* %out, <64 x i16> addrspace(1)* %in) #0 { %load = load <64 x i16>, <64 x i16> addrspace(1)* %in %ext = sext <64 x i16> %load to <64 x i32> @@ -340,6 +432,9 @@ define void @global_sextload_v64i16_to_v64i32(<64 x i32> addrspace(1)* %out, <64 ; GCN-NOHSA: buffer_store_dwordx2 v{{\[}}[[LO]]:[[HI]]] ; GCN-HSA: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[LO]]:[[HI]]{{\]}} + +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG: MOV {{.*}}, 0.0 define void @global_zextload_i16_to_i64(i64 addrspace(1)* %out, i16 addrspace(1)* %in) #0 { %a = load i16, i16 addrspace(1)* %in %ext = zext i16 %a to i64 @@ -354,6 +449,11 @@ define void @global_zextload_i16_to_i64(i64 addrspace(1)* %out, i16 addrspace(1) ; GCN-NOHSA: buffer_store_dwordx2 v{{\[}}[[LO]]:[[HI]]] ; GCN-HSA: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[LO]]:[[HI]]{{\]}} + +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG: ASHR {{\**}} {{T[0-9]\.[XYZW]}}, {{.*}}, literal +; TODO: Why not 15 ? +; EG: 31 define void @global_sextload_i16_to_i64(i64 addrspace(1)* %out, i16 addrspace(1)* %in) #0 { %a = load i16, i16 addrspace(1)* %in %ext = sext i16 %a to i64 @@ -362,6 +462,9 @@ define void @global_sextload_i16_to_i64(i64 addrspace(1)* %out, i16 addrspace(1) } ; FUNC-LABEL: {{^}}global_zextload_v1i16_to_v1i64: + +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG: MOV {{.*}}, 0.0 define void @global_zextload_v1i16_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i16> addrspace(1)* %in) #0 { %load = load <1 x i16>, <1 x i16> addrspace(1)* %in %ext = zext <1 x i16> %load to <1 x i64> @@ -370,6 +473,11 @@ define void @global_zextload_v1i16_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i } ; FUNC-LABEL: {{^}}global_sextload_v1i16_to_v1i64: + +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG: ASHR {{\**}} {{T[0-9]\.[XYZW]}}, {{.*}}, literal +; TODO: Why not 15 ? +; EG: 31 define void @global_sextload_v1i16_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i16> addrspace(1)* %in) #0 { %load = load <1 x i16>, <1 x i16> addrspace(1)* %in %ext = sext <1 x i16> %load to <1 x i64> @@ -386,6 +494,8 @@ define void @global_zextload_v2i16_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i } ; FUNC-LABEL: {{^}}global_sextload_v2i16_to_v2i64: + +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @global_sextload_v2i16_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 { %load = load <2 x i16>, <2 x i16> addrspace(1)* %in %ext = sext <2 x i16> %load to <2 x i64> @@ -394,6 +504,8 @@ define void @global_sextload_v2i16_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i } ; FUNC-LABEL: {{^}}global_zextload_v4i16_to_v4i64: + +; EG: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0, #1 define void @global_zextload_v4i16_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i16> addrspace(1)* %in) #0 { %load = load <4 x i16>, <4 x i16> addrspace(1)* %in %ext = zext <4 x i16> %load to <4 x i64> @@ -402,6 +514,8 @@ define void @global_zextload_v4i16_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i } ; FUNC-LABEL: {{^}}global_sextload_v4i16_to_v4i64: + +; EG: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0, #1 define void @global_sextload_v4i16_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i16> addrspace(1)* %in) #0 { %load = load <4 x i16>, <4 x i16> addrspace(1)* %in %ext = sext <4 x i16> %load to <4 x i64> @@ -410,6 +524,8 @@ define void @global_sextload_v4i16_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i } ; FUNC-LABEL: {{^}}global_zextload_v8i16_to_v8i64: + +; EG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 define void @global_zextload_v8i16_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i16> addrspace(1)* %in) #0 { %load = load <8 x i16>, <8 x i16> addrspace(1)* %in %ext = zext <8 x i16> %load to <8 x i64> @@ -418,6 +534,8 @@ define void @global_zextload_v8i16_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i } ; FUNC-LABEL: {{^}}global_sextload_v8i16_to_v8i64: + +; EG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 define void @global_sextload_v8i16_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i16> addrspace(1)* %in) #0 { %load = load <8 x i16>, <8 x i16> addrspace(1)* %in %ext = sext <8 x i16> %load to <8 x i64> @@ -426,6 +544,9 @@ define void @global_sextload_v8i16_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i } ; FUNC-LABEL: {{^}}global_zextload_v16i16_to_v16i64: + +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 define void @global_zextload_v16i16_to_v16i64(<16 x i64> addrspace(1)* %out, <16 x i16> addrspace(1)* %in) #0 { %load = load <16 x i16>, <16 x i16> addrspace(1)* %in %ext = zext <16 x i16> %load to <16 x i64> @@ -434,6 +555,9 @@ define void @global_zextload_v16i16_to_v16i64(<16 x i64> addrspace(1)* %out, <16 } ; FUNC-LABEL: {{^}}global_sextload_v16i16_to_v16i64: + +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 define void @global_sextload_v16i16_to_v16i64(<16 x i64> addrspace(1)* %out, <16 x i16> addrspace(1)* %in) #0 { %load = load <16 x i16>, <16 x i16> addrspace(1)* %in %ext = sext <16 x i16> %load to <16 x i64> @@ -442,6 +566,11 @@ define void @global_sextload_v16i16_to_v16i64(<16 x i64> addrspace(1)* %out, <16 } ; FUNC-LABEL: {{^}}global_zextload_v32i16_to_v32i64: + +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 32, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 48, #1 define void @global_zextload_v32i16_to_v32i64(<32 x i64> addrspace(1)* %out, <32 x i16> addrspace(1)* %in) #0 { %load = load <32 x i16>, <32 x i16> addrspace(1)* %in %ext = zext <32 x i16> %load to <32 x i64> @@ -450,6 +579,11 @@ define void @global_zextload_v32i16_to_v32i64(<32 x i64> addrspace(1)* %out, <32 } ; FUNC-LABEL: {{^}}global_sextload_v32i16_to_v32i64: + +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 32, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 48, #1 define void @global_sextload_v32i16_to_v32i64(<32 x i64> addrspace(1)* %out, <32 x i16> addrspace(1)* %in) #0 { %load = load <32 x i16>, <32 x i16> addrspace(1)* %in %ext = sext <32 x i16> %load to <32 x i64> diff --git a/llvm/test/CodeGen/AMDGPU/load-global-i8.ll b/llvm/test/CodeGen/AMDGPU/load-global-i8.ll index b697967..ae032a9 100644 --- a/llvm/test/CodeGen/AMDGPU/load-global-i8.ll +++ b/llvm/test/CodeGen/AMDGPU/load-global-i8.ll @@ -9,7 +9,8 @@ ; GCN-NOHSA: buffer_load_ubyte v{{[0-9]+}} ; GCN-HSA: flat_load_ubyte -; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0 +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; TODO: NOT AND define void @global_load_i8(i8 addrspace(1)* %out, i8 addrspace(1)* %in) #0 { entry: %ld = load i8, i8 addrspace(1)* %in @@ -21,7 +22,7 @@ entry: ; GCN-NOHSA: buffer_load_ushort v ; GCN-HSA: flat_load_ushort v -; EG: VTX_READ_16 +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @global_load_v2i8(<2 x i8> addrspace(1)* %out, <2 x i8> addrspace(1)* %in) #0 { entry: %ld = load <2 x i8>, <2 x i8> addrspace(1)* %in @@ -33,7 +34,7 @@ entry: ; GCN-NOHSA: buffer_load_dword v ; GCN-HSA: flat_load_dword v -; EG-DAG: VTX_READ_32 +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @global_load_v3i8(<3 x i8> addrspace(1)* %out, <3 x i8> addrspace(1)* %in) #0 { entry: %ld = load <3 x i8>, <3 x i8> addrspace(1)* %in @@ -45,7 +46,7 @@ entry: ; GCN-NOHSA: buffer_load_dword v ; GCN-HSA: flat_load_dword v -; EG: VTX_READ_32 +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @global_load_v4i8(<4 x i8> addrspace(1)* %out, <4 x i8> addrspace(1)* %in) #0 { entry: %ld = load <4 x i8>, <4 x i8> addrspace(1)* %in @@ -57,7 +58,7 @@ entry: ; GCN-NOHSA: buffer_load_dwordx2 ; GCN-HSA: flat_load_dwordx2 -; EG: VTX_READ_64 +; EG: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0, #1 define void @global_load_v8i8(<8 x i8> addrspace(1)* %out, <8 x i8> addrspace(1)* %in) #0 { entry: %ld = load <8 x i8>, <8 x i8> addrspace(1)* %in @@ -70,7 +71,7 @@ entry: ; GCN-HSA: flat_load_dwordx4 -; EG: VTX_READ_128 +; EG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 define void @global_load_v16i8(<16 x i8> addrspace(1)* %out, <16 x i8> addrspace(1)* %in) #0 { entry: %ld = load <16 x i8>, <16 x i8> addrspace(1)* %in @@ -82,7 +83,7 @@ entry: ; GCN-NOHSA: buffer_load_ubyte v{{[0-9]+}}, ; GCN-HSA: flat_load_ubyte -; EG: VTX_READ_8 T{{[0-9]+\.X, T[0-9]+\.X}} +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @global_zextload_i8_to_i32(i32 addrspace(1)* %out, i8 addrspace(1)* %in) #0 { %a = load i8, i8 addrspace(1)* %in %ext = zext i8 %a to i32 @@ -94,7 +95,7 @@ define void @global_zextload_i8_to_i32(i32 addrspace(1)* %out, i8 addrspace(1)* ; GCN-NOHSA: buffer_load_sbyte ; GCN-HSA: flat_load_sbyte -; EG: VTX_READ_8 [[DST:T[0-9]\.[XYZW]]], [[DST]] +; EG: VTX_READ_8 [[DST:T[0-9]\.[XYZW]]], [[DST]], 0, #1 ; EG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST]], 0.0, literal ; EG: 8 define void @global_sextload_i8_to_i32(i32 addrspace(1)* %out, i8 addrspace(1)* %in) #0 { @@ -105,6 +106,8 @@ define void @global_sextload_i8_to_i32(i32 addrspace(1)* %out, i8 addrspace(1)* } ; FUNC-LABEL: {{^}}global_zextload_v1i8_to_v1i32: + +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @global_zextload_v1i8_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x i8> addrspace(1)* %in) #0 { %load = load <1 x i8>, <1 x i8> addrspace(1)* %in %ext = zext <1 x i8> %load to <1 x i32> @@ -113,6 +116,10 @@ define void @global_zextload_v1i8_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x i8 } ; FUNC-LABEL: {{^}}global_sextload_v1i8_to_v1i32: + +; EG: VTX_READ_8 [[DST:T[0-9]\.[XYZW]]], [[DST]], 0, #1 +; EG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST]], 0.0, literal +; EG: 8 define void @global_sextload_v1i8_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x i8> addrspace(1)* %in) #0 { %load = load <1 x i8>, <1 x i8> addrspace(1)* %in %ext = sext <1 x i8> %load to <1 x i32> @@ -124,8 +131,10 @@ define void @global_sextload_v1i8_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x i8 ; GCN-NOHSA: buffer_load_ushort ; GCN-HSA: flat_load_ushort -; EG: VTX_READ_8 -; EG: VTX_READ_8 +; EG: VTX_READ_16 [[DST:T[0-9]+\.X]], T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, literal +; EG-DAG: 8 define void @global_zextload_v2i8_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i8> addrspace(1)* %in) #0 { %load = load <2 x i8>, <2 x i8> addrspace(1)* %in %ext = zext <2 x i8> %load to <2 x i32> @@ -137,10 +146,10 @@ define void @global_zextload_v2i8_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i8 ; GCN-NOHSA: buffer_load_ushort ; GCN-HSA: flat_load_ushort -; EG-DAG: VTX_READ_8 [[DST_X:T[0-9]\.[XYZW]]], [[DST_X]] -; EG-DAG: VTX_READ_8 [[DST_Y:T[0-9]\.[XYZW]]], [[DST_Y]] -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_X]], 0.0, literal -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_Y]], 0.0, literal +; EG: VTX_READ_16 [[DST:T[0-9]+\.X]], T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal ; EG-DAG: 8 ; EG-DAG: 8 define void @global_sextload_v2i8_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i8> addrspace(1)* %in) #0 { @@ -157,6 +166,13 @@ define void @global_sextload_v2i8_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i8 ; GCN-DAG: v_bfe_u32 v{{[0-9]+}}, v{{[0-9]+}}, 8, 8 ; GCN-DAG: v_bfe_u32 v{{[0-9]+}}, v{{[0-9]+}}, 16, 8 ; GCN-DAG: v_and_b32_e32 v{{[0-9]+}}, 0xff, + +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: 8 +; EG-DAG: 8 define void @global_zextload_v3i8_to_v3i32(<3 x i32> addrspace(1)* %out, <3 x i8> addrspace(1)* %in) #0 { entry: %ld = load <3 x i8>, <3 x i8> addrspace(1)* %in @@ -172,6 +188,15 @@ entry: ; GCN-DAG: v_bfe_i32 v{{[0-9]+}}, v{{[0-9]+}}, 8, 8 ; GCN-DAG: v_bfe_i32 v{{[0-9]+}}, v{{[0-9]+}}, 0, 8 ; GCN-DAG: v_bfe_i32 v{{[0-9]+}}, v{{[0-9]+}}, 16, 8 + +; EG: VTX_READ_32 [[DST:T[0-9]+\.X]], T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 define void @global_sextload_v3i8_to_v3i32(<3 x i32> addrspace(1)* %out, <3 x i8> addrspace(1)* %in) #0 { entry: %ld = load <3 x i8>, <3 x i8> addrspace(1)* %in @@ -184,10 +209,14 @@ entry: ; GCN-NOHSA: buffer_load_dword ; GCN-HSA: flat_load_dword -; EG: VTX_READ_8 -; EG: VTX_READ_8 -; EG: VTX_READ_8 -; EG: VTX_READ_8 +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 define void @global_zextload_v4i8_to_v4i32(<4 x i32> addrspace(1)* %out, <4 x i8> addrspace(1)* %in) #0 { %load = load <4 x i8>, <4 x i8> addrspace(1)* %in %ext = zext <4 x i8> %load to <4 x i32> @@ -199,14 +228,12 @@ define void @global_zextload_v4i8_to_v4i32(<4 x i32> addrspace(1)* %out, <4 x i8 ; GCN-NOHSA: buffer_load_dword ; GCN-HSA: flat_load_dword -; EG-DAG: VTX_READ_8 [[DST_X:T[0-9]\.[XYZW]]], [[DST_X]] -; EG-DAG: VTX_READ_8 [[DST_Y:T[0-9]\.[XYZW]]], [[DST_Y]] -; EG-DAG: VTX_READ_8 [[DST_Z:T[0-9]\.[XYZW]]], [[DST_Z]] -; EG-DAG: VTX_READ_8 [[DST_W:T[0-9]\.[XYZW]]], [[DST_W]] -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_X]], 0.0, literal -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_Y]], 0.0, literal -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_Z]], 0.0, literal -; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, [[DST_W]], 0.0, literal +; EG: VTX_READ_32 [[DST:T[0-9]+\.X]], T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal ; EG-DAG: 8 ; EG-DAG: 8 ; EG-DAG: 8 @@ -219,6 +246,23 @@ define void @global_sextload_v4i8_to_v4i32(<4 x i32> addrspace(1)* %out, <4 x i8 } ; FUNC-LABEL: {{^}}global_zextload_v8i8_to_v8i32: + +; EG: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 define void @global_zextload_v8i8_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x i8> addrspace(1)* %in) #0 { %load = load <8 x i8>, <8 x i8> addrspace(1)* %in %ext = zext <8 x i8> %load to <8 x i32> @@ -227,6 +271,25 @@ define void @global_zextload_v8i8_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x i8 } ; FUNC-LABEL: {{^}}global_sextload_v8i8_to_v8i32: + +; EG: VTX_READ_64 [[DST:T[0-9]+\.XY]], T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 define void @global_sextload_v8i8_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x i8> addrspace(1)* %in) #0 { %load = load <8 x i8>, <8 x i8> addrspace(1)* %in %ext = sext <8 x i8> %load to <8 x i32> @@ -235,6 +298,39 @@ define void @global_sextload_v8i8_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x i8 } ; FUNC-LABEL: {{^}}global_zextload_v16i8_to_v16i32: + +; EG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 define void @global_zextload_v16i8_to_v16i32(<16 x i32> addrspace(1)* %out, <16 x i8> addrspace(1)* %in) #0 { %load = load <16 x i8>, <16 x i8> addrspace(1)* %in %ext = zext <16 x i8> %load to <16 x i32> @@ -243,6 +339,41 @@ define void @global_zextload_v16i8_to_v16i32(<16 x i32> addrspace(1)* %out, <16 } ; FUNC-LABEL: {{^}}global_sextload_v16i8_to_v16i32: + +; EG: VTX_READ_128 [[DST:T[0-9]+\.XYZW]], T{{[0-9]+}}.X, 0, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 define void @global_sextload_v16i8_to_v16i32(<16 x i32> addrspace(1)* %out, <16 x i8> addrspace(1)* %in) #0 { %load = load <16 x i8>, <16 x i8> addrspace(1)* %in %ext = sext <16 x i8> %load to <16 x i32> @@ -251,6 +382,70 @@ define void @global_sextload_v16i8_to_v16i32(<16 x i32> addrspace(1)* %out, <16 } ; FUNC-LABEL: {{^}}global_zextload_v32i8_to_v32i32: + +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: BFE_UINT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, {{.*}}literal +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 define void @global_zextload_v32i8_to_v32i32(<32 x i32> addrspace(1)* %out, <32 x i8> addrspace(1)* %in) #0 { %load = load <32 x i8>, <32 x i8> addrspace(1)* %in %ext = zext <32 x i8> %load to <32 x i32> @@ -259,6 +454,74 @@ define void @global_zextload_v32i8_to_v32i32(<32 x i32> addrspace(1)* %out, <32 } ; FUNC-LABEL: {{^}}global_sextload_v32i8_to_v32i32: + +; EG-DAG: VTX_READ_128 [[DST_LO:T[0-9]+\.XYZW]], T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 [[DST_HI:T[0-9]+\.XYZW]], T{{[0-9]+}}.X, 16, #1 +; TODO: These should use DST, but for some there are redundant MOVs +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9]+.[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 +; EG-DAG: 8 define void @global_sextload_v32i8_to_v32i32(<32 x i32> addrspace(1)* %out, <32 x i8> addrspace(1)* %in) #0 { %load = load <32 x i8>, <32 x i8> addrspace(1)* %in %ext = sext <32 x i8> %load to <32 x i32> @@ -267,6 +530,11 @@ define void @global_sextload_v32i8_to_v32i32(<32 x i32> addrspace(1)* %out, <32 } ; FUNC-LABEL: {{^}}global_zextload_v64i8_to_v64i32: + +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, T{{[0-9]+}}.X, 16, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, T{{[0-9]+}}.X, 32, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, T{{[0-9]+}}.X, 48, #1 define void @global_zextload_v64i8_to_v64i32(<64 x i32> addrspace(1)* %out, <64 x i8> addrspace(1)* %in) #0 { %load = load <64 x i8>, <64 x i8> addrspace(1)* %in %ext = zext <64 x i8> %load to <64 x i32> @@ -275,6 +543,11 @@ define void @global_zextload_v64i8_to_v64i32(<64 x i32> addrspace(1)* %out, <64 } ; FUNC-LABEL: {{^}}global_sextload_v64i8_to_v64i32: + +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, T{{[0-9]+}}.X, 16, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, T{{[0-9]+}}.X, 32, #1 +; EG-DAG: VTX_READ_128 {{T[0-9]+\.XYZW}}, T{{[0-9]+}}.X, 48, #1 define void @global_sextload_v64i8_to_v64i32(<64 x i32> addrspace(1)* %out, <64 x i8> addrspace(1)* %in) #0 { %load = load <64 x i8>, <64 x i8> addrspace(1)* %in %ext = sext <64 x i8> %load to <64 x i32> @@ -290,6 +563,9 @@ define void @global_sextload_v64i8_to_v64i32(<64 x i32> addrspace(1)* %out, <64 ; GCN-HSA-DAG: flat_load_ubyte v[[LO:[0-9]+]], ; GCN-HSA: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[LO]]:[[HI]]] + +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG: MOV {{.*}}, 0.0 define void @global_zextload_i8_to_i64(i64 addrspace(1)* %out, i8 addrspace(1)* %in) #0 { %a = load i8, i8 addrspace(1)* %in %ext = zext i8 %a to i64 @@ -304,6 +580,11 @@ define void @global_zextload_i8_to_i64(i64 addrspace(1)* %out, i8 addrspace(1)* ; GCN-NOHSA: buffer_store_dwordx2 v{{\[}}[[LO]]:[[HI]]{{\]}} ; GCN-HSA: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[LO]]:[[HI]]{{\]}} + +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG: ASHR {{\**}} {{T[0-9]\.[XYZW]}}, {{.*}}, literal +; TODO: Why not 7 ? +; EG: 31 define void @global_sextload_i8_to_i64(i64 addrspace(1)* %out, i8 addrspace(1)* %in) #0 { %a = load i8, i8 addrspace(1)* %in %ext = sext i8 %a to i64 @@ -312,6 +593,9 @@ define void @global_sextload_i8_to_i64(i64 addrspace(1)* %out, i8 addrspace(1)* } ; FUNC-LABEL: {{^}}global_zextload_v1i8_to_v1i64: + +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG: MOV {{.*}}, 0.0 define void @global_zextload_v1i8_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i8> addrspace(1)* %in) #0 { %load = load <1 x i8>, <1 x i8> addrspace(1)* %in %ext = zext <1 x i8> %load to <1 x i64> @@ -320,6 +604,11 @@ define void @global_zextload_v1i8_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i8 } ; FUNC-LABEL: {{^}}global_sextload_v1i8_to_v1i64: + +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG: ASHR {{\**}} {{T[0-9]\.[XYZW]}}, {{.*}}, literal +; TODO: Why not 7 ? +; EG: 31 define void @global_sextload_v1i8_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i8> addrspace(1)* %in) #0 { %load = load <1 x i8>, <1 x i8> addrspace(1)* %in %ext = sext <1 x i8> %load to <1 x i64> @@ -328,6 +617,8 @@ define void @global_sextload_v1i8_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i8 } ; FUNC-LABEL: {{^}}global_zextload_v2i8_to_v2i64: + +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @global_zextload_v2i8_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i8> addrspace(1)* %in) #0 { %load = load <2 x i8>, <2 x i8> addrspace(1)* %in %ext = zext <2 x i8> %load to <2 x i64> @@ -336,6 +627,8 @@ define void @global_zextload_v2i8_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i8 } ; FUNC-LABEL: {{^}}global_sextload_v2i8_to_v2i64: + +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @global_sextload_v2i8_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i8> addrspace(1)* %in) #0 { %load = load <2 x i8>, <2 x i8> addrspace(1)* %in %ext = sext <2 x i8> %load to <2 x i64> @@ -344,6 +637,8 @@ define void @global_sextload_v2i8_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i8 } ; FUNC-LABEL: {{^}}global_zextload_v4i8_to_v4i64: + +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @global_zextload_v4i8_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i8> addrspace(1)* %in) #0 { %load = load <4 x i8>, <4 x i8> addrspace(1)* %in %ext = zext <4 x i8> %load to <4 x i64> @@ -352,6 +647,8 @@ define void @global_zextload_v4i8_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i8 } ; FUNC-LABEL: {{^}}global_sextload_v4i8_to_v4i64: + +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @global_sextload_v4i8_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i8> addrspace(1)* %in) #0 { %load = load <4 x i8>, <4 x i8> addrspace(1)* %in %ext = sext <4 x i8> %load to <4 x i64> @@ -360,6 +657,8 @@ define void @global_sextload_v4i8_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i8 } ; FUNC-LABEL: {{^}}global_zextload_v8i8_to_v8i64: + +; EG: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0, #1 define void @global_zextload_v8i8_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i8> addrspace(1)* %in) #0 { %load = load <8 x i8>, <8 x i8> addrspace(1)* %in %ext = zext <8 x i8> %load to <8 x i64> @@ -368,6 +667,8 @@ define void @global_zextload_v8i8_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i8 } ; FUNC-LABEL: {{^}}global_sextload_v8i8_to_v8i64: + +; EG: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0, #1 define void @global_sextload_v8i8_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i8> addrspace(1)* %in) #0 { %load = load <8 x i8>, <8 x i8> addrspace(1)* %in %ext = sext <8 x i8> %load to <8 x i64> @@ -376,6 +677,8 @@ define void @global_sextload_v8i8_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i8 } ; FUNC-LABEL: {{^}}global_zextload_v16i8_to_v16i64: + +; EG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 define void @global_zextload_v16i8_to_v16i64(<16 x i64> addrspace(1)* %out, <16 x i8> addrspace(1)* %in) #0 { %load = load <16 x i8>, <16 x i8> addrspace(1)* %in %ext = zext <16 x i8> %load to <16 x i64> @@ -384,6 +687,8 @@ define void @global_zextload_v16i8_to_v16i64(<16 x i64> addrspace(1)* %out, <16 } ; FUNC-LABEL: {{^}}global_sextload_v16i8_to_v16i64: + +; EG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 define void @global_sextload_v16i8_to_v16i64(<16 x i64> addrspace(1)* %out, <16 x i8> addrspace(1)* %in) #0 { %load = load <16 x i8>, <16 x i8> addrspace(1)* %in %ext = sext <16 x i8> %load to <16 x i64> @@ -392,6 +697,9 @@ define void @global_sextload_v16i8_to_v16i64(<16 x i64> addrspace(1)* %out, <16 } ; FUNC-LABEL: {{^}}global_zextload_v32i8_to_v32i64: + +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 define void @global_zextload_v32i8_to_v32i64(<32 x i64> addrspace(1)* %out, <32 x i8> addrspace(1)* %in) #0 { %load = load <32 x i8>, <32 x i8> addrspace(1)* %in %ext = zext <32 x i8> %load to <32 x i64> @@ -400,6 +708,9 @@ define void @global_zextload_v32i8_to_v32i64(<32 x i64> addrspace(1)* %out, <32 } ; FUNC-LABEL: {{^}}global_sextload_v32i8_to_v32i64: + +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 define void @global_sextload_v32i8_to_v32i64(<32 x i64> addrspace(1)* %out, <32 x i8> addrspace(1)* %in) #0 { %load = load <32 x i8>, <32 x i8> addrspace(1)* %in %ext = sext <32 x i8> %load to <32 x i64> @@ -429,6 +740,8 @@ define void @global_sextload_v32i8_to_v32i64(<32 x i64> addrspace(1)* %out, <32 ; GCN-HSA: flat_load_ubyte v[[VAL:[0-9]+]], ; GCN-HSA: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, v[[VAL]] + +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @global_zextload_i8_to_i16(i16 addrspace(1)* %out, i8 addrspace(1)* %in) #0 { %a = load i8, i8 addrspace(1)* %in %ext = zext i8 %a to i16 @@ -442,6 +755,9 @@ define void @global_zextload_i8_to_i16(i16 addrspace(1)* %out, i8 addrspace(1)* ; GCN-NOHSA: buffer_store_short v[[VAL]] ; GCN-HSA: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, v[[VAL]] + +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal define void @global_sextload_i8_to_i16(i16 addrspace(1)* %out, i8 addrspace(1)* %in) #0 { %a = load i8, i8 addrspace(1)* %in %ext = sext i8 %a to i16 @@ -450,6 +766,8 @@ define void @global_sextload_i8_to_i16(i16 addrspace(1)* %out, i8 addrspace(1)* } ; FUNC-LABEL: {{^}}global_zextload_v1i8_to_v1i16: + +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @global_zextload_v1i8_to_v1i16(<1 x i16> addrspace(1)* %out, <1 x i8> addrspace(1)* %in) #0 { %load = load <1 x i8>, <1 x i8> addrspace(1)* %in %ext = zext <1 x i8> %load to <1 x i16> @@ -458,6 +776,9 @@ define void @global_zextload_v1i8_to_v1i16(<1 x i16> addrspace(1)* %out, <1 x i8 } ; FUNC-LABEL: {{^}}global_sextload_v1i8_to_v1i16: + +; EG: VTX_READ_8 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal define void @global_sextload_v1i8_to_v1i16(<1 x i16> addrspace(1)* %out, <1 x i8> addrspace(1)* %in) #0 { %load = load <1 x i8>, <1 x i8> addrspace(1)* %in %ext = sext <1 x i8> %load to <1 x i16> @@ -466,6 +787,8 @@ define void @global_sextload_v1i8_to_v1i16(<1 x i16> addrspace(1)* %out, <1 x i8 } ; FUNC-LABEL: {{^}}global_zextload_v2i8_to_v2i16: + +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @global_zextload_v2i8_to_v2i16(<2 x i16> addrspace(1)* %out, <2 x i8> addrspace(1)* %in) #0 { %load = load <2 x i8>, <2 x i8> addrspace(1)* %in %ext = zext <2 x i8> %load to <2 x i16> @@ -474,6 +797,10 @@ define void @global_zextload_v2i8_to_v2i16(<2 x i16> addrspace(1)* %out, <2 x i8 } ; FUNC-LABEL: {{^}}global_sextload_v2i8_to_v2i16: + +; EG: VTX_READ_16 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal define void @global_sextload_v2i8_to_v2i16(<2 x i16> addrspace(1)* %out, <2 x i8> addrspace(1)* %in) #0 { %load = load <2 x i8>, <2 x i8> addrspace(1)* %in %ext = sext <2 x i8> %load to <2 x i16> @@ -482,6 +809,8 @@ define void @global_sextload_v2i8_to_v2i16(<2 x i16> addrspace(1)* %out, <2 x i8 } ; FUNC-LABEL: {{^}}global_zextload_v4i8_to_v4i16: + +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 define void @global_zextload_v4i8_to_v4i16(<4 x i16> addrspace(1)* %out, <4 x i8> addrspace(1)* %in) #0 { %load = load <4 x i8>, <4 x i8> addrspace(1)* %in %ext = zext <4 x i8> %load to <4 x i16> @@ -490,6 +819,12 @@ define void @global_zextload_v4i8_to_v4i16(<4 x i16> addrspace(1)* %out, <4 x i8 } ; FUNC-LABEL: {{^}}global_sextload_v4i8_to_v4i16: + +; EG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal define void @global_sextload_v4i8_to_v4i16(<4 x i16> addrspace(1)* %out, <4 x i8> addrspace(1)* %in) #0 { %load = load <4 x i8>, <4 x i8> addrspace(1)* %in %ext = sext <4 x i8> %load to <4 x i16> @@ -498,6 +833,8 @@ define void @global_sextload_v4i8_to_v4i16(<4 x i16> addrspace(1)* %out, <4 x i8 } ; FUNC-LABEL: {{^}}global_zextload_v8i8_to_v8i16: + +; EG: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0, #1 define void @global_zextload_v8i8_to_v8i16(<8 x i16> addrspace(1)* %out, <8 x i8> addrspace(1)* %in) #0 { %load = load <8 x i8>, <8 x i8> addrspace(1)* %in %ext = zext <8 x i8> %load to <8 x i16> @@ -506,6 +843,16 @@ define void @global_zextload_v8i8_to_v8i16(<8 x i16> addrspace(1)* %out, <8 x i8 } ; FUNC-LABEL: {{^}}global_sextload_v8i8_to_v8i16: + +; EG: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal define void @global_sextload_v8i8_to_v8i16(<8 x i16> addrspace(1)* %out, <8 x i8> addrspace(1)* %in) #0 { %load = load <8 x i8>, <8 x i8> addrspace(1)* %in %ext = sext <8 x i8> %load to <8 x i16> @@ -514,6 +861,8 @@ define void @global_sextload_v8i8_to_v8i16(<8 x i16> addrspace(1)* %out, <8 x i8 } ; FUNC-LABEL: {{^}}global_zextload_v16i8_to_v16i16: + +; EG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 define void @global_zextload_v16i8_to_v16i16(<16 x i16> addrspace(1)* %out, <16 x i8> addrspace(1)* %in) #0 { %load = load <16 x i8>, <16 x i8> addrspace(1)* %in %ext = zext <16 x i8> %load to <16 x i16> @@ -522,6 +871,24 @@ define void @global_zextload_v16i8_to_v16i16(<16 x i16> addrspace(1)* %out, <16 } ; FUNC-LABEL: {{^}}global_sextload_v16i8_to_v16i16: + +; EG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal define void @global_sextload_v16i8_to_v16i16(<16 x i16> addrspace(1)* %out, <16 x i8> addrspace(1)* %in) #0 { %load = load <16 x i8>, <16 x i8> addrspace(1)* %in %ext = sext <16 x i8> %load to <16 x i16> @@ -530,6 +897,9 @@ define void @global_sextload_v16i8_to_v16i16(<16 x i16> addrspace(1)* %out, <16 } ; FUNC-LABEL: {{^}}global_zextload_v32i8_to_v32i16: + +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 define void @global_zextload_v32i8_to_v32i16(<32 x i16> addrspace(1)* %out, <32 x i8> addrspace(1)* %in) #0 { %load = load <32 x i8>, <32 x i8> addrspace(1)* %in %ext = zext <32 x i8> %load to <32 x i16> @@ -538,6 +908,41 @@ define void @global_zextload_v32i8_to_v32i16(<32 x i16> addrspace(1)* %out, <32 } ; FUNC-LABEL: {{^}}global_sextload_v32i8_to_v32i16: + +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0, #1 +; EG-DAG: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 16, #1 +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal +; EG-DAG: BFE_INT {{[* ]*}}T{{[0-9].[XYZW]}}, {{.*}}, 0.0, literal define void @global_sextload_v32i8_to_v32i16(<32 x i16> addrspace(1)* %out, <32 x i8> addrspace(1)* %in) #0 { %load = load <32 x i8>, <32 x i8> addrspace(1)* %in %ext = sext <32 x i8> %load to <32 x i16> diff --git a/llvm/test/CodeGen/AMDGPU/load-local-i16.ll b/llvm/test/CodeGen/AMDGPU/load-local-i16.ll index 0af378d..ce311bf 100644 --- a/llvm/test/CodeGen/AMDGPU/load-local-i16.ll +++ b/llvm/test/CodeGen/AMDGPU/load-local-i16.ll @@ -5,7 +5,11 @@ ; FUNC-LABEL: {{^}}local_load_i16: ; GCN: ds_read_u16 v{{[0-9]+}} -; EG: LDS_USHORT_READ_RET +; EG: MOV {{[* ]*}}[[FROM:T[0-9]+\.[XYZW]]], KC0[2].Z +; EG: LDS_USHORT_READ_RET {{.*}} [[FROM]] +; EG-DAG: MOV {{[* ]*}}[[DATA:T[0-9]+\.[XYZW]]], OQAP +; EG-DAG: MOV {{[* ]*}}[[TO:T[0-9]+\.[XYZW]]], KC0[2].Y +; EG: LDS_SHORT_WRITE {{\*?}} [[TO]], [[DATA]] define void @local_load_i16(i16 addrspace(3)* %out, i16 addrspace(3)* %in) { entry: %ld = load i16, i16 addrspace(3)* %in @@ -16,7 +20,11 @@ entry: ; FUNC-LABEL: {{^}}local_load_v2i16: ; GCN: ds_read_b32 -; EG: LDS_READ_RET +; EG: MOV {{[* ]*}}[[FROM:T[0-9]+\.[XYZW]]], KC0[2].Z +; EG: LDS_READ_RET {{.*}} [[FROM]] +; EG-DAG: MOV {{[* ]*}}[[DATA:T[0-9]+\.[XYZW]]], OQAP +; EG-DAG: MOV {{[* ]*}}[[TO:T[0-9]+\.[XYZW]]], KC0[2].Y +; EG: LDS_WRITE {{\*?}} [[TO]], [[DATA]] define void @local_load_v2i16(<2 x i16> addrspace(3)* %out, <2 x i16> addrspace(3)* %in) { entry: %ld = load <2 x i16>, <2 x i16> addrspace(3)* %in @@ -89,7 +97,11 @@ entry: ; GCN: ds_read_u16 ; GCN: ds_write_b32 -; EG: LDS_USHORT_READ_RET +; EG: MOV {{[* ]*}}[[FROM:T[0-9]+\.[XYZW]]], KC0[2].Z +; EG: LDS_USHORT_READ_RET {{.*}} [[FROM]] +; EG-DAG: MOV {{[* ]*}}[[DATA:T[0-9]+\.[XYZW]]], OQAP +; EG-DAG: MOV {{[* ]*}}[[TO:T[0-9]+\.[XYZW]]], KC0[2].Y +; EG: LDS_WRITE {{\*?}} [[TO]], [[DATA]] define void @local_zextload_i16_to_i32(i32 addrspace(3)* %out, i16 addrspace(3)* %in) #0 { %a = load i16, i16 addrspace(3)* %in %ext = zext i16 %a to i32 @@ -102,8 +114,13 @@ define void @local_zextload_i16_to_i32(i32 addrspace(3)* %out, i16 addrspace(3)* ; GCN: s_mov_b32 m0 ; GCN: ds_read_i16 -; EG: LDS_USHORT_READ_RET -; EG: BFE_INT +; EG: MOV {{[* ]*}}[[FROM:T[0-9]+\.[XYZW]]], KC0[2].Z +; EG: LDS_USHORT_READ_RET {{.*}} [[FROM]] +; EG-DAG: MOV {{[* ]*}}[[TMP:T[0-9]+\.[XYZW]]], OQAP +; EG-DAG: MOV {{[* ]*}}[[TO:T[0-9]+\.[XYZW]]], KC0[2].Y +; EG-DAG: BFE_INT {{[* ]*}}[[DATA:T[0-9]+\.[XYZW]]], {{.*}}, 0.0, literal +; EG: 16 +; EG: LDS_WRITE {{\*?}} [[TO]], [[DATA]] define void @local_sextload_i16_to_i32(i32 addrspace(3)* %out, i16 addrspace(3)* %in) #0 { %a = load i16, i16 addrspace(3)* %in %ext = sext i16 %a to i32 @@ -113,6 +130,12 @@ define void @local_sextload_i16_to_i32(i32 addrspace(3)* %out, i16 addrspace(3)* ; FUNC-LABEL: {{^}}local_zextload_v1i16_to_v1i32: ; GCN: ds_read_u16 + +; EG: MOV {{[* ]*}}[[FROM:T[0-9]+\.[XYZW]]], KC0[2].Z +; EG: LDS_USHORT_READ_RET {{.*}} [[FROM]] +; EG-DAG: MOV {{[* ]*}}[[DATA:T[0-9]+\.[XYZW]]], OQAP +; EG-DAG: MOV {{[* ]*}}[[TO:T[0-9]+\.[XYZW]]], KC0[2].Y +; EG: LDS_WRITE {{\*?}} [[TO]], [[DATA]] define void @local_zextload_v1i16_to_v1i32(<1 x i32> addrspace(3)* %out, <1 x i16> addrspace(3)* %in) #0 { %load = load <1 x i16>, <1 x i16> addrspace(3)* %in %ext = zext <1 x i16> %load to <1 x i32> @@ -122,6 +145,14 @@ define void @local_zextload_v1i16_to_v1i32(<1 x i32> addrspace(3)* %out, <1 x i1 ; FUNC-LABEL: {{^}}local_sextload_v1i16_to_v1i32: ; GCN: ds_read_i16 + +; EG: MOV {{[* ]*}}[[FROM:T[0-9]+\.[XYZW]]], KC0[2].Z +; EG: LDS_USHORT_READ_RET {{.*}} [[FROM]] +; EG-DAG: MOV {{[* ]*}}[[TMP:T[0-9]+\.[XYZW]]], OQAP +; EG-DAG: MOV {{[* ]*}}[[TO:T[0-9]+\.[XYZW]]], KC0[2].Y +; EG-DAG: BFE_INT {{[* ]*}}[[DATA:T[0-9]+\.[XYZW]]], {{.*}}, 0.0, literal +; EG: 16 +; EG: LDS_WRITE {{\*?}} [[TO]], [[DATA]] define void @local_sextload_v1i16_to_v1i32(<1 x i32> addrspace(3)* %out, <1 x i16> addrspace(3)* %in) #0 { %load = load <1 x i16>, <1 x i16> addrspace(3)* %in %ext = sext <1 x i16> %load to <1 x i32> @@ -134,8 +165,7 @@ define void @local_sextload_v1i16_to_v1i32(<1 x i32> addrspace(3)* %out, <1 x i1 ; GCN: s_mov_b32 m0 ; GCN: ds_read_b32 -; EG: LDS_USHORT_READ_RET -; EG: LDS_USHORT_READ_RET +; EG: LDS_READ_RET define void @local_zextload_v2i16_to_v2i32(<2 x i32> addrspace(3)* %out, <2 x i16> addrspace(3)* %in) #0 { %load = load <2 x i16>, <2 x i16> addrspace(3)* %in %ext = zext <2 x i16> %load to <2 x i32> @@ -148,10 +178,9 @@ define void @local_zextload_v2i16_to_v2i32(<2 x i32> addrspace(3)* %out, <2 x i1 ; GCN: s_mov_b32 m0 ; GCN: ds_read_b32 -; EG-DAG: LDS_USHORT_READ_RET -; EG-DAG: LDS_USHORT_READ_RET -; EG-DAG: BFE_INT -; EG-DAG: BFE_INT +; EG: LDS_READ_RET +; EG: BFE_INT +; EG: BFE_INT define void @local_sextload_v2i16_to_v2i32(<2 x i32> addrspace(3)* %out, <2 x i16> addrspace(3)* %in) #0 { %load = load <2 x i16>, <2 x i16> addrspace(3)* %in %ext = sext <2 x i16> %load to <2 x i32> @@ -163,6 +192,8 @@ define void @local_sextload_v2i16_to_v2i32(<2 x i32> addrspace(3)* %out, <2 x i1 ; GCN: ds_read_b64 ; GCN-DAG: ds_write_b32 ; GCN-DAG: ds_write_b64 + +; EG: LDS_READ_RET define void @local_local_zextload_v3i16_to_v3i32(<3 x i32> addrspace(3)* %out, <3 x i16> addrspace(3)* %in) { entry: %ld = load <3 x i16>, <3 x i16> addrspace(3)* %in @@ -175,6 +206,11 @@ entry: ; GCN: ds_read_b64 ; GCN-DAG: ds_write_b32 ; GCN-DAG: ds_write_b64 + +; EG: LDS_READ_RET +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT define void @local_local_sextload_v3i16_to_v3i32(<3 x i32> addrspace(3)* %out, <3 x i16> addrspace(3)* %in) { entry: %ld = load <3 x i16>, <3 x i16> addrspace(3)* %in @@ -188,10 +224,8 @@ entry: ; GCN: s_mov_b32 m0 ; GCN: ds_read_b64 -; EG: LDS_USHORT_READ_RET -; EG: LDS_USHORT_READ_RET -; EG: LDS_USHORT_READ_RET -; EG: LDS_USHORT_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET define void @local_local_zextload_v4i16_to_v4i32(<4 x i32> addrspace(3)* %out, <4 x i16> addrspace(3)* %in) #0 { %load = load <4 x i16>, <4 x i16> addrspace(3)* %in %ext = zext <4 x i16> %load to <4 x i32> @@ -204,10 +238,8 @@ define void @local_local_zextload_v4i16_to_v4i32(<4 x i32> addrspace(3)* %out, < ; GCN: s_mov_b32 m0 ; GCN: ds_read_b64 -; EG-DAG: LDS_USHORT_READ_RET -; EG-DAG: LDS_USHORT_READ_RET -; EG-DAG: LDS_USHORT_READ_RET -; EG-DAG: LDS_USHORT_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET ; EG-DAG: BFE_INT ; EG-DAG: BFE_INT ; EG-DAG: BFE_INT @@ -221,6 +253,11 @@ define void @local_sextload_v4i16_to_v4i32(<4 x i32> addrspace(3)* %out, <4 x i1 ; FUNC-LABEL: {{^}}local_zextload_v8i16_to_v8i32: ; GCN: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}} + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET define void @local_zextload_v8i16_to_v8i32(<8 x i32> addrspace(3)* %out, <8 x i16> addrspace(3)* %in) #0 { %load = load <8 x i16>, <8 x i16> addrspace(3)* %in %ext = zext <8 x i16> %load to <8 x i32> @@ -230,6 +267,19 @@ define void @local_zextload_v8i16_to_v8i32(<8 x i32> addrspace(3)* %out, <8 x i1 ; FUNC-LABEL: {{^}}local_sextload_v8i16_to_v8i32: ; GCN: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}} + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT define void @local_sextload_v8i16_to_v8i32(<8 x i32> addrspace(3)* %out, <8 x i16> addrspace(3)* %in) #0 { %load = load <8 x i16>, <8 x i16> addrspace(3)* %in %ext = sext <8 x i16> %load to <8 x i32> @@ -247,6 +297,15 @@ define void @local_sextload_v8i16_to_v8i32(<8 x i32> addrspace(3)* %out, <8 x i1 ; GCN: ds_write2_b64 ; GCN: ds_write2_b64 ; GCN: ds_write2_b64 + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET define void @local_zextload_v16i16_to_v16i32(<16 x i32> addrspace(3)* %out, <16 x i16> addrspace(3)* %in) #0 { %load = load <16 x i16>, <16 x i16> addrspace(3)* %in %ext = zext <16 x i16> %load to <16 x i32> @@ -258,6 +317,31 @@ define void @local_zextload_v16i16_to_v16i32(<16 x i32> addrspace(3)* %out, <16 ; GCN-DAG: ds_read_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+$}} ; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:1 offset1:3{{$}} ; GCN-DAG: ds_read_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset:16{{$}} + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT define void @local_sextload_v16i16_to_v16i32(<16 x i32> addrspace(3)* %out, <16 x i16> addrspace(3)* %in) #0 { %load = load <16 x i16>, <16 x i16> addrspace(3)* %in %ext = sext <16 x i16> %load to <16 x i32> @@ -270,6 +354,23 @@ define void @local_sextload_v16i16_to_v16i32(<16 x i32> addrspace(3)* %out, <16 ; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:2 offset1:3 ; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:4 offset1:5 ; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:6 offset1:7 + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET define void @local_zextload_v32i16_to_v32i32(<32 x i32> addrspace(3)* %out, <32 x i16> addrspace(3)* %in) #0 { %load = load <32 x i16>, <32 x i16> addrspace(3)* %in %ext = zext <32 x i16> %load to <32 x i32> @@ -282,6 +383,23 @@ define void @local_zextload_v32i16_to_v32i32(<32 x i32> addrspace(3)* %out, <32 ; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:3 offset1:4 ; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:5{{$}} ; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:6 offset1:7 + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET define void @local_sextload_v32i16_to_v32i32(<32 x i32> addrspace(3)* %out, <32 x i16> addrspace(3)* %in) #0 { %load = load <32 x i16>, <32 x i16> addrspace(3)* %in %ext = sext <32 x i16> %load to <32 x i32> @@ -300,6 +418,39 @@ define void @local_sextload_v32i16_to_v32i32(<32 x i32> addrspace(3)* %out, <32 ; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:9 offset1:10 ; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:12 offset1:13 ; GCN-DAG: ds_read_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset:112 + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET define void @local_zextload_v64i16_to_v64i32(<64 x i32> addrspace(3)* %out, <64 x i16> addrspace(3)* %in) #0 { %load = load <64 x i16>, <64 x i16> addrspace(3)* %in %ext = zext <64 x i16> %load to <64 x i32> @@ -308,6 +459,39 @@ define void @local_zextload_v64i16_to_v64i32(<64 x i32> addrspace(3)* %out, <64 } ; FUNC-LABEL: {{^}}local_sextload_v64i16_to_v64i32: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET define void @local_sextload_v64i16_to_v64i32(<64 x i32> addrspace(3)* %out, <64 x i16> addrspace(3)* %in) #0 { %load = load <64 x i16>, <64 x i16> addrspace(3)* %in %ext = sext <64 x i16> %load to <64 x i32> @@ -320,6 +504,13 @@ define void @local_sextload_v64i16_to_v64i32(<64 x i32> addrspace(3)* %out, <64 ; GCN-DAG: v_mov_b32_e32 v[[HI:[0-9]+]], 0{{$}} ; GCN: ds_write_b64 v{{[0-9]+}}, v{{\[}}[[LO]]:[[HI]]] + +; EG: MOV {{[* ]*}}[[FROM:T[0-9]+\.[XYZW]]], KC0[2].Z +; EG: LDS_USHORT_READ_RET {{.*}} [[FROM]] +; EG-DAG: MOV {{[* ]*}}[[DATA:T[0-9]+\.[XYZW]]], OQAP +; EG-DAG: MOV {{[* ]*}}[[TO:T[0-9]+\.[XYZW]]], KC0[2].Y +; EG-DAG: LDS_WRITE +; EG: LDS_WRITE {{\*?}} [[TO]], [[DATA]] define void @local_zextload_i16_to_i64(i64 addrspace(3)* %out, i16 addrspace(3)* %in) #0 { %a = load i16, i16 addrspace(3)* %in %ext = zext i16 %a to i64 @@ -332,6 +523,15 @@ define void @local_zextload_i16_to_i64(i64 addrspace(3)* %out, i16 addrspace(3)* ; GCN-DAG: v_ashrrev_i32_e32 v[[HI:[0-9]+]], 31, v[[LO]] ; GCN: ds_write_b64 v{{[0-9]+}}, v{{\[}}[[LO]]:[[HI]]] + +; EG: MOV {{[* ]*}}[[FROM:T[0-9]+\.[XYZW]]], KC0[2].Z +; EG: LDS_USHORT_READ_RET {{.*}} [[FROM]] +; EG-DAG: MOV {{[* ]*}}[[TMP:T[0-9]+\.[XYZW]]], OQAP +; EG-DAG: MOV {{[* ]*}}[[TO:T[0-9]+\.[XYZW]]], KC0[2].Y +; EG-DAG: BFE_INT {{[* ]*}}[[DATA:T[0-9]+\.[XYZW]]], {{.*}}, 0.0, literal +; EG-DAG: LDS_WRITE +; EG-DAG: 16 +; EG: LDS_WRITE {{\*?}} [[TO]], [[DATA]] define void @local_sextload_i16_to_i64(i64 addrspace(3)* %out, i16 addrspace(3)* %in) #0 { %a = load i16, i16 addrspace(3)* %in %ext = sext i16 %a to i64 @@ -340,6 +540,13 @@ define void @local_sextload_i16_to_i64(i64 addrspace(3)* %out, i16 addrspace(3)* } ; FUNC-LABEL: {{^}}local_zextload_v1i16_to_v1i64: + +; EG: MOV {{[* ]*}}[[FROM:T[0-9]+\.[XYZW]]], KC0[2].Z +; EG: LDS_USHORT_READ_RET {{.*}} [[FROM]] +; EG-DAG: MOV {{[* ]*}}[[DATA:T[0-9]+\.[XYZW]]], OQAP +; EG-DAG: MOV {{[* ]*}}[[TO:T[0-9]+\.[XYZW]]], KC0[2].Y +; EG-DAG: LDS_WRITE +; EG: LDS_WRITE {{\*?}} [[TO]], [[DATA]] define void @local_zextload_v1i16_to_v1i64(<1 x i64> addrspace(3)* %out, <1 x i16> addrspace(3)* %in) #0 { %load = load <1 x i16>, <1 x i16> addrspace(3)* %in %ext = zext <1 x i16> %load to <1 x i64> @@ -348,6 +555,15 @@ define void @local_zextload_v1i16_to_v1i64(<1 x i64> addrspace(3)* %out, <1 x i1 } ; FUNC-LABEL: {{^}}local_sextload_v1i16_to_v1i64: + +; EG: MOV {{[* ]*}}[[FROM:T[0-9]+\.[XYZW]]], KC0[2].Z +; EG: LDS_USHORT_READ_RET {{.*}} [[FROM]] +; EG-DAG: MOV {{[* ]*}}[[TMP:T[0-9]+\.[XYZW]]], OQAP +; EG-DAG: MOV {{[* ]*}}[[TO:T[0-9]+\.[XYZW]]], KC0[2].Y +; EG-DAG: BFE_INT {{[* ]*}}[[DATA:T[0-9]+\.[XYZW]]], {{.*}}, 0.0, literal +; EG-DAG: LDS_WRITE +; EG-DAG: 16 +; EG: LDS_WRITE {{\*?}} [[TO]], [[DATA]] define void @local_sextload_v1i16_to_v1i64(<1 x i64> addrspace(3)* %out, <1 x i16> addrspace(3)* %in) #0 { %load = load <1 x i16>, <1 x i16> addrspace(3)* %in %ext = sext <1 x i16> %load to <1 x i64> @@ -356,6 +572,8 @@ define void @local_sextload_v1i16_to_v1i64(<1 x i64> addrspace(3)* %out, <1 x i1 } ; FUNC-LABEL: {{^}}local_zextload_v2i16_to_v2i64: + +; EG: LDS_READ_RET define void @local_zextload_v2i16_to_v2i64(<2 x i64> addrspace(3)* %out, <2 x i16> addrspace(3)* %in) #0 { %load = load <2 x i16>, <2 x i16> addrspace(3)* %in %ext = zext <2 x i16> %load to <2 x i64> @@ -364,6 +582,10 @@ define void @local_zextload_v2i16_to_v2i64(<2 x i64> addrspace(3)* %out, <2 x i1 } ; FUNC-LABEL: {{^}}local_sextload_v2i16_to_v2i64: + +; EG: LDS_READ_RET +; EG-DAG: BFE_INT +; EG-DAG: ASHR define void @local_sextload_v2i16_to_v2i64(<2 x i64> addrspace(3)* %out, <2 x i16> addrspace(3)* %in) #0 { %load = load <2 x i16>, <2 x i16> addrspace(3)* %in %ext = sext <2 x i16> %load to <2 x i64> @@ -372,6 +594,9 @@ define void @local_sextload_v2i16_to_v2i64(<2 x i64> addrspace(3)* %out, <2 x i1 } ; FUNC-LABEL: {{^}}local_zextload_v4i16_to_v4i64: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET define void @local_zextload_v4i16_to_v4i64(<4 x i64> addrspace(3)* %out, <4 x i16> addrspace(3)* %in) #0 { %load = load <4 x i16>, <4 x i16> addrspace(3)* %in %ext = zext <4 x i16> %load to <4 x i64> @@ -380,6 +605,13 @@ define void @local_zextload_v4i16_to_v4i64(<4 x i64> addrspace(3)* %out, <4 x i1 } ; FUNC-LABEL: {{^}}local_sextload_v4i16_to_v4i64: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG-DAG: ASHR define void @local_sextload_v4i16_to_v4i64(<4 x i64> addrspace(3)* %out, <4 x i16> addrspace(3)* %in) #0 { %load = load <4 x i16>, <4 x i16> addrspace(3)* %in %ext = sext <4 x i16> %load to <4 x i64> @@ -388,6 +620,11 @@ define void @local_sextload_v4i16_to_v4i64(<4 x i64> addrspace(3)* %out, <4 x i1 } ; FUNC-LABEL: {{^}}local_zextload_v8i16_to_v8i64: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET define void @local_zextload_v8i16_to_v8i64(<8 x i64> addrspace(3)* %out, <8 x i16> addrspace(3)* %in) #0 { %load = load <8 x i16>, <8 x i16> addrspace(3)* %in %ext = zext <8 x i16> %load to <8 x i64> @@ -396,6 +633,19 @@ define void @local_zextload_v8i16_to_v8i64(<8 x i64> addrspace(3)* %out, <8 x i1 } ; FUNC-LABEL: {{^}}local_sextload_v8i16_to_v8i64: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG-DAG: ASHR define void @local_sextload_v8i16_to_v8i64(<8 x i64> addrspace(3)* %out, <8 x i16> addrspace(3)* %in) #0 { %load = load <8 x i16>, <8 x i16> addrspace(3)* %in %ext = sext <8 x i16> %load to <8 x i64> @@ -404,6 +654,15 @@ define void @local_sextload_v8i16_to_v8i64(<8 x i64> addrspace(3)* %out, <8 x i1 } ; FUNC-LABEL: {{^}}local_zextload_v16i16_to_v16i64: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET define void @local_zextload_v16i16_to_v16i64(<16 x i64> addrspace(3)* %out, <16 x i16> addrspace(3)* %in) #0 { %load = load <16 x i16>, <16 x i16> addrspace(3)* %in %ext = zext <16 x i16> %load to <16 x i64> @@ -412,6 +671,31 @@ define void @local_zextload_v16i16_to_v16i64(<16 x i64> addrspace(3)* %out, <16 } ; FUNC-LABEL: {{^}}local_sextload_v16i16_to_v16i64: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG-DAG: ASHR define void @local_sextload_v16i16_to_v16i64(<16 x i64> addrspace(3)* %out, <16 x i16> addrspace(3)* %in) #0 { %load = load <16 x i16>, <16 x i16> addrspace(3)* %in %ext = sext <16 x i16> %load to <16 x i64> @@ -420,6 +704,23 @@ define void @local_sextload_v16i16_to_v16i64(<16 x i64> addrspace(3)* %out, <16 } ; FUNC-LABEL: {{^}}local_zextload_v32i16_to_v32i64: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET define void @local_zextload_v32i16_to_v32i64(<32 x i64> addrspace(3)* %out, <32 x i16> addrspace(3)* %in) #0 { %load = load <32 x i16>, <32 x i16> addrspace(3)* %in %ext = zext <32 x i16> %load to <32 x i64> @@ -428,6 +729,55 @@ define void @local_zextload_v32i16_to_v32i64(<32 x i64> addrspace(3)* %out, <32 } ; FUNC-LABEL: {{^}}local_sextload_v32i16_to_v32i64: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG-DAG: ASHR define void @local_sextload_v32i16_to_v32i64(<32 x i64> addrspace(3)* %out, <32 x i16> addrspace(3)* %in) #0 { %load = load <32 x i16>, <32 x i16> addrspace(3)* %in %ext = sext <32 x i16> %load to <32 x i64> diff --git a/llvm/test/CodeGen/AMDGPU/load-local-i8.ll b/llvm/test/CodeGen/AMDGPU/load-local-i8.ll index 3c618e9..6b5b1cf 100644 --- a/llvm/test/CodeGen/AMDGPU/load-local-i8.ll +++ b/llvm/test/CodeGen/AMDGPU/load-local-i8.ll @@ -106,6 +106,8 @@ define void @local_sextload_i8_to_i32(i32 addrspace(3)* %out, i8 addrspace(3)* % } ; FUNC-LABEL: {{^}}local_zextload_v1i8_to_v1i32: + +; EG: LDS_UBYTE_READ_RET define void @local_zextload_v1i8_to_v1i32(<1 x i32> addrspace(3)* %out, <1 x i8> addrspace(3)* %in) #0 { %load = load <1 x i8>, <1 x i8> addrspace(3)* %in %ext = zext <1 x i8> %load to <1 x i32> @@ -114,6 +116,9 @@ define void @local_zextload_v1i8_to_v1i32(<1 x i32> addrspace(3)* %out, <1 x i8> } ; FUNC-LABEL: {{^}}local_sextload_v1i8_to_v1i32: + +; EG: LDS_UBYTE_READ_RET +; EG: BFE_INT define void @local_sextload_v1i8_to_v1i32(<1 x i32> addrspace(3)* %out, <1 x i8> addrspace(3)* %in) #0 { %load = load <1 x i8>, <1 x i8> addrspace(3)* %in %ext = sext <1 x i8> %load to <1 x i32> @@ -124,8 +129,7 @@ define void @local_sextload_v1i8_to_v1i32(<1 x i32> addrspace(3)* %out, <1 x i8> ; FUNC-LABEL: {{^}}local_zextload_v2i8_to_v2i32: ; GCN: ds_read_u16 -; EG: LDS_UBYTE_READ_RET -; EG: LDS_UBYTE_READ_RET +; EG: LDS_USHORT_READ_RET define void @local_zextload_v2i8_to_v2i32(<2 x i32> addrspace(3)* %out, <2 x i8> addrspace(3)* %in) #0 { %load = load <2 x i8>, <2 x i8> addrspace(3)* %in %ext = zext <2 x i8> %load to <2 x i32> @@ -140,8 +144,7 @@ define void @local_zextload_v2i8_to_v2i32(<2 x i32> addrspace(3)* %out, <2 x i8> ; GCN-DAG: v_bfe_i32 v{{[0-9]+}}, v{{[0-9]+}}, 8, 8 ; GCN-DAG: v_bfe_i32 v{{[0-9]+}}, v{{[0-9]+}}, 0, 8 -; EG-DAG: LDS_UBYTE_READ_RET -; EG-DAG: LDS_UBYTE_READ_RET +; EG: LDS_USHORT_READ_RET ; EG-DAG: BFE_INT ; EG-DAG: BFE_INT define void @local_sextload_v2i8_to_v2i32(<2 x i32> addrspace(3)* %out, <2 x i8> addrspace(3)* %in) #0 { @@ -157,6 +160,8 @@ define void @local_sextload_v2i8_to_v2i32(<2 x i32> addrspace(3)* %out, <2 x i8> ; GCN-DAG: v_bfe_u32 v{{[0-9]+}}, v{{[0-9]+}}, 8, 8 ; GCN-DAG: v_bfe_u32 v{{[0-9]+}}, v{{[0-9]+}}, 16, 8 ; GCN-DAG: v_and_b32_e32 v{{[0-9]+}}, 0xff, + +; EG: LDS_READ_RET define void @local_zextload_v3i8_to_v3i32(<3 x i32> addrspace(3)* %out, <3 x i8> addrspace(3)* %in) #0 { entry: %ld = load <3 x i8>, <3 x i8> addrspace(3)* %in @@ -178,6 +183,10 @@ entry: ; GCN-DAG: ds_write_b64 ; GCN-DAG: ds_write_b32 +; EG: LDS_READ_RET +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT define void @local_sextload_v3i8_to_v3i32(<3 x i32> addrspace(3)* %out, <3 x i8> addrspace(3)* %in) #0 { entry: %ld = load <3 x i8>, <3 x i8> addrspace(3)* %in @@ -191,10 +200,10 @@ entry: ; GCN: s_mov_b32 m0 ; GCN: ds_read_b32 -; EG: LDS_UBYTE_READ_RET -; EG: LDS_UBYTE_READ_RET -; EG: LDS_UBYTE_READ_RET -; EG: LDS_UBYTE_READ_RET +; EG: LDS_READ_RET +; EG-DAG: BFE_UINT +; EG-DAG: BFE_UINT +; EG-DAG: BFE_UINT define void @local_zextload_v4i8_to_v4i32(<4 x i32> addrspace(3)* %out, <4 x i8> addrspace(3)* %in) #0 { %load = load <4 x i8>, <4 x i8> addrspace(3)* %in %ext = zext <4 x i8> %load to <4 x i32> @@ -207,10 +216,7 @@ define void @local_zextload_v4i8_to_v4i32(<4 x i32> addrspace(3)* %out, <4 x i8> ; GCN: s_mov_b32 m0 ; GCN: ds_read_b32 -; EG-DAG: LDS_UBYTE_READ_RET -; EG-DAG: LDS_UBYTE_READ_RET -; EG-DAG: LDS_UBYTE_READ_RET -; EG-DAG: LDS_UBYTE_READ_RET +; EG-DAG: LDS_READ_RET ; EG-DAG: BFE_INT ; EG-DAG: BFE_INT ; EG-DAG: BFE_INT @@ -223,6 +229,15 @@ define void @local_sextload_v4i8_to_v4i32(<4 x i32> addrspace(3)* %out, <4 x i8> } ; FUNC-LABEL: {{^}}local_zextload_v8i8_to_v8i32: + +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: BFE_UINT +; EG-DAG: BFE_UINT +; EG-DAG: BFE_UINT +; EG-DAG: BFE_UINT +; EG-DAG: BFE_UINT +; EG-DAG: BFE_UINT define void @local_zextload_v8i8_to_v8i32(<8 x i32> addrspace(3)* %out, <8 x i8> addrspace(3)* %in) #0 { %load = load <8 x i8>, <8 x i8> addrspace(3)* %in %ext = zext <8 x i8> %load to <8 x i32> @@ -231,6 +246,17 @@ define void @local_zextload_v8i8_to_v8i32(<8 x i32> addrspace(3)* %out, <8 x i8> } ; FUNC-LABEL: {{^}}local_sextload_v8i8_to_v8i32: + +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT define void @local_sextload_v8i8_to_v8i32(<8 x i32> addrspace(3)* %out, <8 x i8> addrspace(3)* %in) #0 { %load = load <8 x i8>, <8 x i8> addrspace(3)* %in %ext = sext <8 x i8> %load to <8 x i32> @@ -239,6 +265,23 @@ define void @local_sextload_v8i8_to_v8i32(<8 x i32> addrspace(3)* %out, <8 x i8> } ; FUNC-LABEL: {{^}}local_zextload_v16i8_to_v16i32: + +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: BFE_UINT +; EG-DAG: BFE_UINT +; EG-DAG: BFE_UINT +; EG-DAG: BFE_UINT +; EG-DAG: BFE_UINT +; EG-DAG: BFE_UINT +; EG-DAG: BFE_UINT +; EG-DAG: BFE_UINT +; EG-DAG: BFE_UINT +; EG-DAG: BFE_UINT +; EG-DAG: BFE_UINT +; EG-DAG: BFE_UINT define void @local_zextload_v16i8_to_v16i32(<16 x i32> addrspace(3)* %out, <16 x i8> addrspace(3)* %in) #0 { %load = load <16 x i8>, <16 x i8> addrspace(3)* %in %ext = zext <16 x i8> %load to <16 x i32> @@ -247,6 +290,27 @@ define void @local_zextload_v16i8_to_v16i32(<16 x i32> addrspace(3)* %out, <16 x } ; FUNC-LABEL: {{^}}local_sextload_v16i8_to_v16i32: + +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT define void @local_sextload_v16i8_to_v16i32(<16 x i32> addrspace(3)* %out, <16 x i8> addrspace(3)* %in) #0 { %load = load <16 x i8>, <16 x i8> addrspace(3)* %in %ext = sext <16 x i8> %load to <16 x i32> @@ -255,6 +319,15 @@ define void @local_sextload_v16i8_to_v16i32(<16 x i32> addrspace(3)* %out, <16 x } ; FUNC-LABEL: {{^}}local_zextload_v32i8_to_v32i32: + +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET define void @local_zextload_v32i8_to_v32i32(<32 x i32> addrspace(3)* %out, <32 x i8> addrspace(3)* %in) #0 { %load = load <32 x i8>, <32 x i8> addrspace(3)* %in %ext = zext <32 x i8> %load to <32 x i32> @@ -263,6 +336,15 @@ define void @local_zextload_v32i8_to_v32i32(<32 x i32> addrspace(3)* %out, <32 x } ; FUNC-LABEL: {{^}}local_sextload_v32i8_to_v32i32: + +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET define void @local_sextload_v32i8_to_v32i32(<32 x i32> addrspace(3)* %out, <32 x i8> addrspace(3)* %in) #0 { %load = load <32 x i8>, <32 x i8> addrspace(3)* %in %ext = sext <32 x i8> %load to <32 x i32> @@ -271,6 +353,23 @@ define void @local_sextload_v32i8_to_v32i32(<32 x i32> addrspace(3)* %out, <32 x } ; FUNC-LABEL: {{^}}local_zextload_v64i8_to_v64i32: + +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET define void @local_zextload_v64i8_to_v64i32(<64 x i32> addrspace(3)* %out, <64 x i8> addrspace(3)* %in) #0 { %load = load <64 x i8>, <64 x i8> addrspace(3)* %in %ext = zext <64 x i8> %load to <64 x i32> @@ -279,6 +378,23 @@ define void @local_zextload_v64i8_to_v64i32(<64 x i32> addrspace(3)* %out, <64 x } ; FUNC-LABEL: {{^}}local_sextload_v64i8_to_v64i32: + +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET +; EG-DAG: LDS_READ_RET define void @local_sextload_v64i8_to_v64i32(<64 x i32> addrspace(3)* %out, <64 x i8> addrspace(3)* %in) #0 { %load = load <64 x i8>, <64 x i8> addrspace(3)* %in %ext = sext <64 x i8> %load to <64 x i32> @@ -290,6 +406,10 @@ define void @local_sextload_v64i8_to_v64i32(<64 x i32> addrspace(3)* %out, <64 x ; GCN-DAG: v_mov_b32_e32 v[[HI:[0-9]+]], 0{{$}} ; GCN-DAG: ds_read_u8 v[[LO:[0-9]+]], ; GCN: ds_write_b64 v{{[0-9]+}}, v{{\[}}[[LO]]:[[HI]]] + +; EG: LDS_UBYTE_READ_RET +; EG: MOV {{.*}}, literal +; EG: 0.0 define void @local_zextload_i8_to_i64(i64 addrspace(3)* %out, i8 addrspace(3)* %in) #0 { %a = load i8, i8 addrspace(3)* %in %ext = zext i8 %a to i64 @@ -302,6 +422,11 @@ define void @local_zextload_i8_to_i64(i64 addrspace(3)* %out, i8 addrspace(3)* % ; GCN: v_ashrrev_i32_e32 v[[HI:[0-9]+]], 31, v[[LO]] ; GCN: ds_write_b64 v{{[0-9]+}}, v{{\[}}[[LO]]:[[HI]]{{\]}} + +; EG: LDS_UBYTE_READ_RET +; EG: ASHR +; TODO: why not 7? +; EG: 31 define void @local_sextload_i8_to_i64(i64 addrspace(3)* %out, i8 addrspace(3)* %in) #0 { %a = load i8, i8 addrspace(3)* %in %ext = sext i8 %a to i64 @@ -310,6 +435,11 @@ define void @local_sextload_i8_to_i64(i64 addrspace(3)* %out, i8 addrspace(3)* % } ; FUNC-LABEL: {{^}}local_zextload_v1i8_to_v1i64: + +; EG: LDS_UBYTE_READ_RET +; EG: MOV {{.*}}, literal +; TODO: merge? +; EG: 0.0 define void @local_zextload_v1i8_to_v1i64(<1 x i64> addrspace(3)* %out, <1 x i8> addrspace(3)* %in) #0 { %load = load <1 x i8>, <1 x i8> addrspace(3)* %in %ext = zext <1 x i8> %load to <1 x i64> @@ -318,6 +448,11 @@ define void @local_zextload_v1i8_to_v1i64(<1 x i64> addrspace(3)* %out, <1 x i8> } ; FUNC-LABEL: {{^}}local_sextload_v1i8_to_v1i64: + +; EG: LDS_UBYTE_READ_RET +; EG: ASHR +; TODO: why not 7? +; EG: 31 define void @local_sextload_v1i8_to_v1i64(<1 x i64> addrspace(3)* %out, <1 x i8> addrspace(3)* %in) #0 { %load = load <1 x i8>, <1 x i8> addrspace(3)* %in %ext = sext <1 x i8> %load to <1 x i64> @@ -326,6 +461,8 @@ define void @local_sextload_v1i8_to_v1i64(<1 x i64> addrspace(3)* %out, <1 x i8> } ; FUNC-LABEL: {{^}}local_zextload_v2i8_to_v2i64: + +; EG: LDS_USHORT_READ_RET define void @local_zextload_v2i8_to_v2i64(<2 x i64> addrspace(3)* %out, <2 x i8> addrspace(3)* %in) #0 { %load = load <2 x i8>, <2 x i8> addrspace(3)* %in %ext = zext <2 x i8> %load to <2 x i64> @@ -334,6 +471,10 @@ define void @local_zextload_v2i8_to_v2i64(<2 x i64> addrspace(3)* %out, <2 x i8> } ; FUNC-LABEL: {{^}}local_sextload_v2i8_to_v2i64: + +; EG: LDS_USHORT_READ_RET +; EG: BFE_INT +; EG: BFE_INT define void @local_sextload_v2i8_to_v2i64(<2 x i64> addrspace(3)* %out, <2 x i8> addrspace(3)* %in) #0 { %load = load <2 x i8>, <2 x i8> addrspace(3)* %in %ext = sext <2 x i8> %load to <2 x i64> @@ -342,6 +483,8 @@ define void @local_sextload_v2i8_to_v2i64(<2 x i64> addrspace(3)* %out, <2 x i8> } ; FUNC-LABEL: {{^}}local_zextload_v4i8_to_v4i64: + +; EG: LDS_READ_RET define void @local_zextload_v4i8_to_v4i64(<4 x i64> addrspace(3)* %out, <4 x i8> addrspace(3)* %in) #0 { %load = load <4 x i8>, <4 x i8> addrspace(3)* %in %ext = zext <4 x i8> %load to <4 x i64> @@ -350,6 +493,8 @@ define void @local_zextload_v4i8_to_v4i64(<4 x i64> addrspace(3)* %out, <4 x i8> } ; FUNC-LABEL: {{^}}local_sextload_v4i8_to_v4i64: + +; EG: LDS_READ_RET define void @local_sextload_v4i8_to_v4i64(<4 x i64> addrspace(3)* %out, <4 x i8> addrspace(3)* %in) #0 { %load = load <4 x i8>, <4 x i8> addrspace(3)* %in %ext = sext <4 x i8> %load to <4 x i64> @@ -358,6 +503,9 @@ define void @local_sextload_v4i8_to_v4i64(<4 x i64> addrspace(3)* %out, <4 x i8> } ; FUNC-LABEL: {{^}}local_zextload_v8i8_to_v8i64: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET define void @local_zextload_v8i8_to_v8i64(<8 x i64> addrspace(3)* %out, <8 x i8> addrspace(3)* %in) #0 { %load = load <8 x i8>, <8 x i8> addrspace(3)* %in %ext = zext <8 x i8> %load to <8 x i64> @@ -366,6 +514,18 @@ define void @local_zextload_v8i8_to_v8i64(<8 x i64> addrspace(3)* %out, <8 x i8> } ; FUNC-LABEL: {{^}}local_sextload_v8i8_to_v8i64: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT define void @local_sextload_v8i8_to_v8i64(<8 x i64> addrspace(3)* %out, <8 x i8> addrspace(3)* %in) #0 { %load = load <8 x i8>, <8 x i8> addrspace(3)* %in %ext = sext <8 x i8> %load to <8 x i64> @@ -374,6 +534,11 @@ define void @local_sextload_v8i8_to_v8i64(<8 x i64> addrspace(3)* %out, <8 x i8> } ; FUNC-LABEL: {{^}}local_zextload_v16i8_to_v16i64: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET define void @local_zextload_v16i8_to_v16i64(<16 x i64> addrspace(3)* %out, <16 x i8> addrspace(3)* %in) #0 { %load = load <16 x i8>, <16 x i8> addrspace(3)* %in %ext = zext <16 x i8> %load to <16 x i64> @@ -382,6 +547,11 @@ define void @local_zextload_v16i8_to_v16i64(<16 x i64> addrspace(3)* %out, <16 x } ; FUNC-LABEL: {{^}}local_sextload_v16i8_to_v16i64: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET define void @local_sextload_v16i8_to_v16i64(<16 x i64> addrspace(3)* %out, <16 x i8> addrspace(3)* %in) #0 { %load = load <16 x i8>, <16 x i8> addrspace(3)* %in %ext = sext <16 x i8> %load to <16 x i64> @@ -390,6 +560,15 @@ define void @local_sextload_v16i8_to_v16i64(<16 x i64> addrspace(3)* %out, <16 x } ; FUNC-LABEL: {{^}}local_zextload_v32i8_to_v32i64: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET define void @local_zextload_v32i8_to_v32i64(<32 x i64> addrspace(3)* %out, <32 x i8> addrspace(3)* %in) #0 { %load = load <32 x i8>, <32 x i8> addrspace(3)* %in %ext = zext <32 x i8> %load to <32 x i64> @@ -398,6 +577,15 @@ define void @local_zextload_v32i8_to_v32i64(<32 x i64> addrspace(3)* %out, <32 x } ; FUNC-LABEL: {{^}}local_sextload_v32i8_to_v32i64: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET define void @local_sextload_v32i8_to_v32i64(<32 x i64> addrspace(3)* %out, <32 x i8> addrspace(3)* %in) #0 { %load = load <32 x i8>, <32 x i8> addrspace(3)* %in %ext = sext <32 x i8> %load to <32 x i64> @@ -424,6 +612,9 @@ define void @local_sextload_v32i8_to_v32i64(<32 x i64> addrspace(3)* %out, <32 x ; FUNC-LABEL: {{^}}local_zextload_i8_to_i16: ; GCN: ds_read_u8 v[[VAL:[0-9]+]], ; GCN: ds_write_b16 v[[VAL:[0-9]+]] + +; EG: LDS_UBYTE_READ_RET +; EG: LDS_SHORT_WRITE define void @local_zextload_i8_to_i16(i16 addrspace(3)* %out, i8 addrspace(3)* %in) #0 { %a = load i8, i8 addrspace(3)* %in %ext = zext i8 %a to i16 @@ -434,6 +625,10 @@ define void @local_zextload_i8_to_i16(i16 addrspace(3)* %out, i8 addrspace(3)* % ; FUNC-LABEL: {{^}}local_sextload_i8_to_i16: ; GCN: ds_read_i8 v[[VAL:[0-9]+]], ; GCN: ds_write_b16 v{{[0-9]+}}, v[[VAL]] + +; EG: LDS_UBYTE_READ_RET +; EG: BFE_INT +; EG: LDS_SHORT_WRITE define void @local_sextload_i8_to_i16(i16 addrspace(3)* %out, i8 addrspace(3)* %in) #0 { %a = load i8, i8 addrspace(3)* %in %ext = sext i8 %a to i16 @@ -442,6 +637,9 @@ define void @local_sextload_i8_to_i16(i16 addrspace(3)* %out, i8 addrspace(3)* % } ; FUNC-LABEL: {{^}}local_zextload_v1i8_to_v1i16: + +; EG: LDS_UBYTE_READ_RET +; EG: LDS_SHORT_WRITE define void @local_zextload_v1i8_to_v1i16(<1 x i16> addrspace(3)* %out, <1 x i8> addrspace(3)* %in) #0 { %load = load <1 x i8>, <1 x i8> addrspace(3)* %in %ext = zext <1 x i8> %load to <1 x i16> @@ -450,6 +648,10 @@ define void @local_zextload_v1i8_to_v1i16(<1 x i16> addrspace(3)* %out, <1 x i8> } ; FUNC-LABEL: {{^}}local_sextload_v1i8_to_v1i16: + +; EG: LDS_UBYTE_READ_RET +; EG: BFE_INT +; EG: LDS_SHORT_WRITE define void @local_sextload_v1i8_to_v1i16(<1 x i16> addrspace(3)* %out, <1 x i8> addrspace(3)* %in) #0 { %load = load <1 x i8>, <1 x i8> addrspace(3)* %in %ext = sext <1 x i8> %load to <1 x i16> @@ -458,6 +660,9 @@ define void @local_sextload_v1i8_to_v1i16(<1 x i16> addrspace(3)* %out, <1 x i8> } ; FUNC-LABEL: {{^}}local_zextload_v2i8_to_v2i16: + +; EG: LDS_USHORT_READ_RET +; EG: LDS_WRITE define void @local_zextload_v2i8_to_v2i16(<2 x i16> addrspace(3)* %out, <2 x i8> addrspace(3)* %in) #0 { %load = load <2 x i8>, <2 x i8> addrspace(3)* %in %ext = zext <2 x i8> %load to <2 x i16> @@ -466,6 +671,11 @@ define void @local_zextload_v2i8_to_v2i16(<2 x i16> addrspace(3)* %out, <2 x i8> } ; FUNC-LABEL: {{^}}local_sextload_v2i8_to_v2i16: + +; EG: LDS_USHORT_READ_RET +; EG: BFE_INT +; EG: BFE_INT +; EG: LDS_WRITE define void @local_sextload_v2i8_to_v2i16(<2 x i16> addrspace(3)* %out, <2 x i8> addrspace(3)* %in) #0 { %load = load <2 x i8>, <2 x i8> addrspace(3)* %in %ext = sext <2 x i8> %load to <2 x i16> @@ -474,6 +684,10 @@ define void @local_sextload_v2i8_to_v2i16(<2 x i16> addrspace(3)* %out, <2 x i8> } ; FUNC-LABEL: {{^}}local_zextload_v4i8_to_v4i16: + +; EG: LDS_READ_RET +; EG: LDS_WRITE +; EG: LDS_WRITE define void @local_zextload_v4i8_to_v4i16(<4 x i16> addrspace(3)* %out, <4 x i8> addrspace(3)* %in) #0 { %load = load <4 x i8>, <4 x i8> addrspace(3)* %in %ext = zext <4 x i8> %load to <4 x i16> @@ -482,6 +696,14 @@ define void @local_zextload_v4i8_to_v4i16(<4 x i16> addrspace(3)* %out, <4 x i8> } ; FUNC-LABEL: {{^}}local_sextload_v4i8_to_v4i16: + +; EG: LDS_READ_RET +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG: LDS_WRITE +; EG: LDS_WRITE define void @local_sextload_v4i8_to_v4i16(<4 x i16> addrspace(3)* %out, <4 x i8> addrspace(3)* %in) #0 { %load = load <4 x i8>, <4 x i8> addrspace(3)* %in %ext = sext <4 x i8> %load to <4 x i16> @@ -490,6 +712,13 @@ define void @local_sextload_v4i8_to_v4i16(<4 x i16> addrspace(3)* %out, <4 x i8> } ; FUNC-LABEL: {{^}}local_zextload_v8i8_to_v8i16: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE define void @local_zextload_v8i8_to_v8i16(<8 x i16> addrspace(3)* %out, <8 x i8> addrspace(3)* %in) #0 { %load = load <8 x i8>, <8 x i8> addrspace(3)* %in %ext = zext <8 x i8> %load to <8 x i16> @@ -498,6 +727,21 @@ define void @local_zextload_v8i8_to_v8i16(<8 x i16> addrspace(3)* %out, <8 x i8> } ; FUNC-LABEL: {{^}}local_sextload_v8i8_to_v8i16: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE define void @local_sextload_v8i8_to_v8i16(<8 x i16> addrspace(3)* %out, <8 x i8> addrspace(3)* %in) #0 { %load = load <8 x i8>, <8 x i8> addrspace(3)* %in %ext = sext <8 x i8> %load to <8 x i16> @@ -506,6 +750,19 @@ define void @local_sextload_v8i8_to_v8i16(<8 x i16> addrspace(3)* %out, <8 x i8> } ; FUNC-LABEL: {{^}}local_zextload_v16i8_to_v16i16: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE define void @local_zextload_v16i8_to_v16i16(<16 x i16> addrspace(3)* %out, <16 x i8> addrspace(3)* %in) #0 { %load = load <16 x i8>, <16 x i8> addrspace(3)* %in %ext = zext <16 x i8> %load to <16 x i16> @@ -514,6 +771,35 @@ define void @local_zextload_v16i8_to_v16i16(<16 x i16> addrspace(3)* %out, <16 x } ; FUNC-LABEL: {{^}}local_sextload_v16i8_to_v16i16: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE define void @local_sextload_v16i8_to_v16i16(<16 x i16> addrspace(3)* %out, <16 x i8> addrspace(3)* %in) #0 { %load = load <16 x i8>, <16 x i8> addrspace(3)* %in %ext = sext <16 x i8> %load to <16 x i16> @@ -522,6 +808,31 @@ define void @local_sextload_v16i8_to_v16i16(<16 x i16> addrspace(3)* %out, <16 x } ; FUNC-LABEL: {{^}}local_zextload_v32i8_to_v32i16: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE define void @local_zextload_v32i8_to_v32i16(<32 x i16> addrspace(3)* %out, <32 x i8> addrspace(3)* %in) #0 { %load = load <32 x i8>, <32 x i8> addrspace(3)* %in %ext = zext <32 x i8> %load to <32 x i16> @@ -530,6 +841,63 @@ define void @local_zextload_v32i8_to_v32i16(<32 x i16> addrspace(3)* %out, <32 x } ; FUNC-LABEL: {{^}}local_sextload_v32i8_to_v32i16: + +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG: LDS_READ_RET +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: BFE_INT +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG-DAG: ASHR +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE +; EG: LDS_WRITE define void @local_sextload_v32i8_to_v32i16(<32 x i16> addrspace(3)* %out, <32 x i8> addrspace(3)* %in) #0 { %load = load <32 x i8>, <32 x i8> addrspace(3)* %in %ext = sext <32 x i8> %load to <32 x i16> -- 2.7.4