diff options
author | Jan Vesely <jan.vesely@rutgers.edu> | 2016-08-27 19:09:43 +0000 |
---|---|---|
committer | Jan Vesely <jan.vesely@rutgers.edu> | 2016-08-27 19:09:43 +0000 |
commit | 38814fa2fd964623658a860914b6222e309a2bf9 (patch) | |
tree | a9b708e919d37b27e2f143b1efdddf89465cff69 /llvm/test/CodeGen/AMDGPU/load-local-i16.ll | |
parent | 6943aa306ee5f27ed3bdd95daa863bed99f2a51b (diff) | |
download | bcm5719-llvm-38814fa2fd964623658a860914b6222e309a2bf9.tar.gz bcm5719-llvm-38814fa2fd964623658a860914b6222e309a2bf9.zip |
AMDGPU/R600: Enable Load combine
Fix and improve tests
Differential Revision: https://reviews.llvm.org/D23899
llvm-svn: 279925
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/load-local-i16.ll')
-rw-r--r-- | llvm/test/CodeGen/AMDGPU/load-local-i16.ll | 388 |
1 files changed, 369 insertions, 19 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/load-local-i16.ll b/llvm/test/CodeGen/AMDGPU/load-local-i16.ll index 0af378dcd25..ce311bfc1e6 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> |