diff options
| author | Aaron Watry <awatry@gmail.com> | 2013-08-12 14:42:50 +0000 |
|---|---|---|
| committer | Aaron Watry <awatry@gmail.com> | 2013-08-12 14:42:50 +0000 |
| commit | c0aa6e02915887d0ba4ac6c2b70c5a283ffdfd91 (patch) | |
| tree | 8c4472edc837bade2e87dcbba1f90d0743a70186 /libclc/r600 | |
| parent | 7d525653213542746e4e18589d5c4bf046f614d6 (diff) | |
| download | llvm-c0aa6e02915887d0ba4ac6c2b70c5a283ffdfd91.tar.gz | |
Enable assembly vload3 int/uint constant/global for R600
It's supported by the R600 LLVM back-end now, at least for evergreen.
Signed-off-by: Aaron Watry <awatry@gmail.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 188180
Diffstat (limited to 'libclc/r600')
| -rw-r--r-- | libclc/r600/lib/shared/vload.cl | 18 |
1 files changed, 2 insertions, 16 deletions
diff --git a/libclc/r600/lib/shared/vload.cl b/libclc/r600/lib/shared/vload.cl index 60502d1ca0ad..49309c32e01b 100644 --- a/libclc/r600/lib/shared/vload.cl +++ b/libclc/r600/lib/shared/vload.cl @@ -51,22 +51,10 @@ VLOAD_VECTORIZE(int, __local) VLOAD_VECTORIZE(uint, __private) VLOAD_VECTORIZE(uint, __local) -_CLC_OVERLOAD _CLC_DEF int3 vload3(size_t offset, const global int *x) { - return (int3)(vload2(0, &x[3*offset]), x[3*offset+2]); -} -_CLC_OVERLOAD _CLC_DEF uint3 vload3(size_t offset, const global uint *x) { - return (uint3)(vload2(0, &x[3*offset]), x[3*offset+2]); -} -_CLC_OVERLOAD _CLC_DEF int3 vload3(size_t offset, const constant int *x) { - return (int3)(vload2(0, &x[3*offset]), x[3*offset+2]); -} -_CLC_OVERLOAD _CLC_DEF uint3 vload3(size_t offset, const constant uint *x) { - return (uint3)(vload2(0, &x[3*offset]), x[3*offset+2]); -} - //We only define functions for typeN vloadN(), and then just bitcast the result for unsigned types #define _CLC_VLOAD_ASM_DECL(PRIM_TYPE,LLVM_SCALAR_TYPE,ADDR_SPACE,ADDR_SPACE_ID) \ _CLC_DECL PRIM_TYPE##2 __clc_vload2_##LLVM_SCALAR_TYPE##__addr##ADDR_SPACE_ID (const ADDR_SPACE PRIM_TYPE *); \ +_CLC_DECL PRIM_TYPE##3 __clc_vload3_##LLVM_SCALAR_TYPE##__addr##ADDR_SPACE_ID (const ADDR_SPACE PRIM_TYPE *); \ _CLC_DECL PRIM_TYPE##4 __clc_vload4_##LLVM_SCALAR_TYPE##__addr##ADDR_SPACE_ID (const ADDR_SPACE PRIM_TYPE *); \ _CLC_DECL PRIM_TYPE##8 __clc_vload8_##LLVM_SCALAR_TYPE##__addr##ADDR_SPACE_ID (const ADDR_SPACE PRIM_TYPE *); \ _CLC_DECL PRIM_TYPE##16 __clc_vload16_##LLVM_SCALAR_TYPE##__addr##ADDR_SPACE_ID (const ADDR_SPACE PRIM_TYPE *); \ @@ -76,11 +64,9 @@ _CLC_DECL PRIM_TYPE##16 __clc_vload16_##LLVM_SCALAR_TYPE##__addr##ADDR_SPACE_ID return __builtin_astype(__clc_vload##VEC_WIDTH##_##LLVM_SCALAR_TYPE##__addr##ADDR_SPACE_ID ((const ADDR_SPACE S_PRIM_TYPE *)&x[VEC_WIDTH * offset]), PRIM_TYPE##VEC_WIDTH); \ } \ -/*Note: R600 back-end doesn't support load <3 x ?>... so - * those functions aren't actually overridden here - */ #define _CLC_VLOAD_ASM_OVERLOAD_SIZES(PRIM_TYPE,S_PRIM_TYPE,LLVM_TYPE,ADDR_SPACE,ADDR_SPACE_ID) \ _CLC_VLOAD_ASM_DEFINE(PRIM_TYPE, S_PRIM_TYPE, LLVM_TYPE, 2, ADDR_SPACE, ADDR_SPACE_ID) \ + _CLC_VLOAD_ASM_DEFINE(PRIM_TYPE, S_PRIM_TYPE, LLVM_TYPE, 3, ADDR_SPACE, ADDR_SPACE_ID) \ _CLC_VLOAD_ASM_DEFINE(PRIM_TYPE, S_PRIM_TYPE, LLVM_TYPE, 4, ADDR_SPACE, ADDR_SPACE_ID) \ _CLC_VLOAD_ASM_DEFINE(PRIM_TYPE, S_PRIM_TYPE, LLVM_TYPE, 8, ADDR_SPACE, ADDR_SPACE_ID) \ _CLC_VLOAD_ASM_DEFINE(PRIM_TYPE, S_PRIM_TYPE, LLVM_TYPE, 16, ADDR_SPACE, ADDR_SPACE_ID) \ |