maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

shared: Implement aligned vector stores (vstorea_half) 

Float version passes newly posted piglit tests on turks, float and double pass on carrizo.
v2: scalar vstorea_half
v3: fix typo

Reviewer: Aaron Watry
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 316291
This commit is contained in:
Jan Vesely 2017-10-22 14:21:59 +00:00
parent 12061c7125
commit 7ab2d0bdcd
3 changed files with 59 additions and 33 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

41
libclc/generic/include/clc/shared/vstore.h
View File

@ -16,37 +16,52 @@
#define _CLC_VECTOR_VSTORE_PRIM1(PRIM_TYPE) \
_CLC_VECTOR_VSTORE_PRIM3(,PRIM_TYPE, PRIM_TYPE) \
#define _CLC_VECTOR_VSTORE_PRIM() \
_CLC_VECTOR_VSTORE_PRIM1(char) \
_CLC_VECTOR_VSTORE_PRIM1(uchar) \
_CLC_VECTOR_VSTORE_PRIM1(short) \
_CLC_VECTOR_VSTORE_PRIM1(ushort) \
_CLC_VECTOR_VSTORE_PRIM1(int) \
_CLC_VECTOR_VSTORE_PRIM1(uint) \
_CLC_VECTOR_VSTORE_PRIM1(long) \
_CLC_VECTOR_VSTORE_PRIM1(ulong) \
_CLC_VECTOR_VSTORE_PRIM1(float) \
_CLC_VECTOR_VSTORE_PRIM3(_half, half, float)
_CLC_VECTOR_VSTORE_PRIM1(char)
_CLC_VECTOR_VSTORE_PRIM1(uchar)
_CLC_VECTOR_VSTORE_PRIM1(short)
_CLC_VECTOR_VSTORE_PRIM1(ushort)
_CLC_VECTOR_VSTORE_PRIM1(int)
_CLC_VECTOR_VSTORE_PRIM1(uint)
_CLC_VECTOR_VSTORE_PRIM1(long)
_CLC_VECTOR_VSTORE_PRIM1(ulong)
_CLC_VECTOR_VSTORE_PRIM1(float)
_CLC_VECTOR_VSTORE_PRIM3(_half, half, float)
// Use suffix to declare aligned vstorea_halfN
_CLC_VECTOR_VSTORE_PRIM3(a_half, half, float)
#ifdef cl_khr_fp64
_CLC_VECTOR_VSTORE_PRIM1(double)
_CLC_VECTOR_VSTORE_PRIM3(_half, half, double)
// Use suffix to declare aligned vstorea_halfN
_CLC_VECTOR_VSTORE_PRIM3(a_half, half, double)
// Scalar vstore_half also needs to be declared
_CLC_VSTORE_DECL(_half, half, double, , __private)
_CLC_VSTORE_DECL(_half, half, double, , __local)
_CLC_VSTORE_DECL(_half, half, double, , __global)
// Scalar vstorea_half is not part of the specs but CTS expects it
_CLC_VSTORE_DECL(a_half, half, double, , __private)
_CLC_VSTORE_DECL(a_half, half, double, , __local)
_CLC_VSTORE_DECL(a_half, half, double, , __global)
#endif
#ifdef cl_khr_fp16
_CLC_VECTOR_VSTORE_PRIM1(half)
#endif
_CLC_VECTOR_VSTORE_PRIM()
// Scalar vstore_half also needs to be declared
_CLC_VSTORE_DECL(_half, half, float, , __private)
_CLC_VSTORE_DECL(_half, half, float, , __local)
_CLC_VSTORE_DECL(_half, half, float, , __global)
// Scalar vstorea_half is not part of the specs but CTS expects it
_CLC_VSTORE_DECL(a_half, half, float, , __private)
_CLC_VSTORE_DECL(a_half, half, float, , __local)
_CLC_VSTORE_DECL(a_half, half, float, , __global)
#undef _CLC_VSTORE_DECL
#undef _CLC_VECTOR_VSTORE_DECL
#undef _CLC_VECTOR_VSTORE_PRIM3
#undef _CLC_VECTOR_VSTORE_PRIM1
#undef _CLC_VECTOR_VSTORE_PRIM

30
libclc/generic/lib/shared/vstore.cl
View File

@ -33,23 +33,22 @@
VSTORE_VECTORIZE(__CLC_SCALAR___CLC_GENTYPE, __local) \
VSTORE_VECTORIZE(__CLC_SCALAR___CLC_GENTYPE, __global) \
#define VSTORE_TYPES() \
VSTORE_ADDR_SPACES(char) \
VSTORE_ADDR_SPACES(uchar) \
VSTORE_ADDR_SPACES(short) \
VSTORE_ADDR_SPACES(ushort) \
VSTORE_ADDR_SPACES(int) \
VSTORE_ADDR_SPACES(uint) \
VSTORE_ADDR_SPACES(long) \
VSTORE_ADDR_SPACES(ulong) \
VSTORE_ADDR_SPACES(float) \
VSTORE_ADDR_SPACES(char)
VSTORE_ADDR_SPACES(uchar)
VSTORE_ADDR_SPACES(short)
VSTORE_ADDR_SPACES(ushort)
VSTORE_ADDR_SPACES(int)
VSTORE_ADDR_SPACES(uint)
VSTORE_ADDR_SPACES(long)
VSTORE_ADDR_SPACES(ulong)
VSTORE_ADDR_SPACES(float)
VSTORE_TYPES()
#ifdef cl_khr_fp64
#pragma OPENCL EXTENSION cl_khr_fp64 : enable
VSTORE_ADDR_SPACES(double)
#endif
#ifdef cl_khr_fp16
#pragma OPENCL EXTENSION cl_khr_fp16 : enable
VSTORE_ADDR_SPACES(half)
@ -95,13 +94,17 @@ DECLARE_HELPER(double, __local, __builtin_store_half);
VEC_STORE8(STYPE, AS, val.lo) \
VEC_STORE8(STYPE, AS, val.hi)
#define __FUNC(SUFFIX, VEC_SIZE, TYPE, STYPE, AS) \
#define __FUNC(SUFFIX, VEC_SIZE, OFFSET, TYPE, STYPE, AS) \
_CLC_OVERLOAD _CLC_DEF void vstore_half##SUFFIX(TYPE vec, size_t offset, AS half *mem) { \
offset *= VEC_SIZE; \
VEC_STORE##VEC_SIZE(STYPE, AS, vec) \
} \
_CLC_OVERLOAD _CLC_DEF void vstorea_half##SUFFIX(TYPE vec, size_t offset, AS half *mem) { \
offset *= OFFSET; \
VEC_STORE##VEC_SIZE(STYPE, AS, vec) \
}
#define FUNC(SUFFIX, VEC_SIZE, TYPE, STYPE, AS) __FUNC(SUFFIX, VEC_SIZE, TYPE, STYPE, AS)
#define FUNC(SUFFIX, VEC_SIZE, OFFSET, TYPE, STYPE, AS) __FUNC(SUFFIX, VEC_SIZE, OFFSET, TYPE, STYPE, AS)
#define __CLC_BODY "vstore_half.inc"
#include <clc/math/gentype.inc>
@ -115,6 +118,5 @@ DECLARE_HELPER(double, __local, __builtin_store_half);
#undef VEC_LOAD2
#undef VEC_LOAD1
#undef DECLARE_HELPER
#undef VSTORE_TYPES
#undef VSTORE_ADDR_SPACES
#undef VSTORE_VECTORIZE

21
libclc/generic/lib/shared/vstore_half.inc
View File

@ -1,10 +1,19 @@
#ifdef __CLC_VECSIZE
FUNC(__CLC_VECSIZE, __CLC_VECSIZE, __CLC_GENTYPE, __CLC_SCALAR_GENTYPE, __private);
FUNC(__CLC_VECSIZE, __CLC_VECSIZE, __CLC_GENTYPE, __CLC_SCALAR_GENTYPE, __local);
FUNC(__CLC_VECSIZE, __CLC_VECSIZE, __CLC_GENTYPE, __CLC_SCALAR_GENTYPE, __global);
#if __CLC_VECSIZE == 3
# define __CLC_OFFSET 4
#else
FUNC(, 1, __CLC_GENTYPE, __CLC_SCALAR_GENTYPE, __private);
FUNC(, 1, __CLC_GENTYPE, __CLC_SCALAR_GENTYPE, __local);
FUNC(, 1, __CLC_GENTYPE, __CLC_SCALAR_GENTYPE, __global);
# define __CLC_OFFSET __CLC_VECSIZE
#endif
FUNC(__CLC_VECSIZE, __CLC_VECSIZE, __CLC_OFFSET, __CLC_GENTYPE, __CLC_SCALAR_GENTYPE, __private);
FUNC(__CLC_VECSIZE, __CLC_VECSIZE, __CLC_OFFSET, __CLC_GENTYPE, __CLC_SCALAR_GENTYPE, __local);
FUNC(__CLC_VECSIZE, __CLC_VECSIZE, __CLC_OFFSET, __CLC_GENTYPE, __CLC_SCALAR_GENTYPE, __global);
#undef __CLC_OFFSET
#else
FUNC(, 1, 1, __CLC_GENTYPE, __CLC_SCALAR_GENTYPE, __private);
FUNC(, 1, 1, __CLC_GENTYPE, __CLC_SCALAR_GENTYPE, __local);
FUNC(, 1, 1, __CLC_GENTYPE, __CLC_SCALAR_GENTYPE, __global);
#endif