diff options
author | Matt Arsenault <Matthew.Arsenault@amd.com> | 2016-07-01 22:47:50 +0000 |
---|---|---|
committer | Matt Arsenault <Matthew.Arsenault@amd.com> | 2016-07-01 22:47:50 +0000 |
commit | 327bb5ad82328bb92907bb2acf87e2282593e1e4 (patch) | |
tree | 96f20ec1942050bdab64dc965a2f9f5e292994ce /llvm/test/CodeGen/AMDGPU/load-local-i16.ll | |
parent | 591ff8376bbf3ee0cc6f6006300878ae96475a49 (diff) | |
download | bcm5719-llvm-327bb5ad82328bb92907bb2acf87e2282593e1e4.tar.gz bcm5719-llvm-327bb5ad82328bb92907bb2acf87e2282593e1e4.zip |
AMDGPU: Improve load/store of illegal types.
There was a combine before to handle the simple copy case.
Split this into handling loads and stores separately.
We might want to change how this handles some of the vector
extloads, since this can result in large code size increases.
llvm-svn: 274394
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/load-local-i16.ll')
-rw-r--r-- | llvm/test/CodeGen/AMDGPU/load-local-i16.ll | 226 |
1 files changed, 36 insertions, 190 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/load-local-i16.ll b/llvm/test/CodeGen/AMDGPU/load-local-i16.ll index 6db67d77425..d3c0af469dd 100644 --- a/llvm/test/CodeGen/AMDGPU/load-local-i16.ll +++ b/llvm/test/CodeGen/AMDGPU/load-local-i16.ll @@ -132,8 +132,7 @@ define void @local_sextload_v1i16_to_v1i32(<1 x i32> addrspace(3)* %out, <1 x i1 ; FUNC-LABEL: {{^}}local_zextload_v2i16_to_v2i32: ; GCN-NOT: s_wqm_b64 ; GCN: s_mov_b32 m0 -; GCN: ds_read_u16 -; GCN: ds_read_u16 +; GCN: ds_read_b32 ; EG: LDS_USHORT_READ_RET ; EG: LDS_USHORT_READ_RET @@ -147,8 +146,7 @@ define void @local_zextload_v2i16_to_v2i32(<2 x i32> addrspace(3)* %out, <2 x i1 ; FUNC-LABEL: {{^}}local_sextload_v2i16_to_v2i32: ; GCN-NOT: s_wqm_b64 ; GCN: s_mov_b32 m0 -; GCN: ds_read_i16 -; GCN: ds_read_i16 +; GCN: ds_read_b32 ; EG-DAG: LDS_USHORT_READ_RET ; EG-DAG: LDS_USHORT_READ_RET @@ -188,10 +186,7 @@ entry: ; FUNC-LABEL: {{^}}local_local_zextload_v4i16_to_v4i32: ; GCN-NOT: s_wqm_b64 ; GCN: s_mov_b32 m0 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 +; GCN: ds_read_b64 ; EG: LDS_USHORT_READ_RET ; EG: LDS_USHORT_READ_RET @@ -207,10 +202,7 @@ define void @local_local_zextload_v4i16_to_v4i32(<4 x i32> addrspace(3)* %out, < ; FUNC-LABEL: {{^}}local_sextload_v4i16_to_v4i32: ; GCN-NOT: s_wqm_b64 ; GCN: s_mov_b32 m0 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 +; GCN: ds_read_b64 ; EG-DAG: LDS_USHORT_READ_RET ; EG-DAG: LDS_USHORT_READ_RET @@ -228,14 +220,7 @@ 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_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 +; GCN: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}} 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> @@ -244,14 +229,7 @@ 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_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 +; GCN: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}} 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> @@ -259,23 +237,16 @@ define void @local_sextload_v8i16_to_v8i32(<8 x i32> addrspace(3)* %out, <8 x i1 ret void } +; FIXME: Should have 2 ds_read_b64 ; FUNC-LABEL: {{^}}local_zextload_v16i16_to_v16i32: -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 +; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:1 offset1:2{{$}} +; GCN-DAG: ds_read_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+$}} +; GCN-DAG: ds_read_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset:24 + +; GCN: ds_write2_b64 +; GCN: ds_write2_b64 +; GCN: ds_write2_b64 +; GCN: ds_write2_b64 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> @@ -284,22 +255,9 @@ define void @local_zextload_v16i16_to_v16i32(<16 x i32> addrspace(3)* %out, <16 } ; FUNC-LABEL: {{^}}local_sextload_v16i16_to_v16i32: -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 +; 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]+}} offset0:3 offset1:1{{$}} +; GCN-DAG: ds_read_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset:16{{$}} 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> @@ -308,39 +266,10 @@ define void @local_sextload_v16i16_to_v16i32(<16 x i32> addrspace(3)* %out, <16 } ; FUNC-LABEL: {{^}}local_zextload_v32i16_to_v32i32: -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 - +; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}} +; 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 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> @@ -349,38 +278,10 @@ define void @local_zextload_v32i16_to_v32i32(<32 x i32> addrspace(3)* %out, <32 } ; FUNC-LABEL: {{^}}local_sextload_v32i16_to_v32i32: -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 -; GCN: ds_read_i16 +; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:1 offset1:2{{$}} +; 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]+}} offset0:5{{$}} +; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:6 offset1:7 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> @@ -388,72 +289,17 @@ define void @local_sextload_v32i16_to_v32i32(<32 x i32> addrspace(3)* %out, <32 ret void } +; FIXME: Missed read2 ; FUNC-LABEL: {{^}}local_zextload_v64i16_to_v64i32: -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 -; GCN: ds_read_u16 - +; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:11 offset1:15 +; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}} +; 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 +; GCN-DAG: ds_read_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset:64 +; 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 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> |