summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorRuiling Song <ruiling.song@intel.com>2016-01-19 11:29:09 +0800
committerYang Rong <rong.r.yang@intel.com>2016-02-01 16:06:18 +0800
commit51b987e90f37898144cfdfe65e933cca537d6c3a (patch)
treed2986ae9e54e64452d2caa219500db113b3521e4
parent6b294a22113b14864084ebda662d7730b479d1c4 (diff)
libocl: change prototype of vload/vstore to match ocl2.0 spec.OCL20
ocl2.0 spec only declare generic and constant address space vload, and only generic space vstore. private/global/local version are all removed. Signed-off-by: Ruiling Song <ruiling.song@intel.com> Reviewed-by: Yang Rong <rong.r.yang@intel.com>
-rw-r--r--backend/src/libocl/include/ocl_vload.h22
-rw-r--r--backend/src/libocl/src/ocl_vload.cl20
2 files changed, 11 insertions, 31 deletions
diff --git a/backend/src/libocl/include/ocl_vload.h b/backend/src/libocl/include/ocl_vload.h
index b1b1a32f..9e2706e3 100644
--- a/backend/src/libocl/include/ocl_vload.h
+++ b/backend/src/libocl/include/ocl_vload.h
@@ -56,10 +56,8 @@ OVERLOADABLE TYPE##3 vload3(size_t offset, const SPACE TYPE *p);
DECL_UNTYPED_RD_SPACE_N(TYPE, 16, SPACE)
#define DECL_UNTYPED_RW_ALL(TYPE) \
- DECL_UNTYPED_RW_ALL_SPACE(TYPE, __global) \
- DECL_UNTYPED_RW_ALL_SPACE(TYPE, __local) \
DECL_UNTYPED_RD_ALL_SPACE(TYPE, __constant) \
- DECL_UNTYPED_RW_ALL_SPACE(TYPE, __private)
+ DECL_UNTYPED_RW_ALL_SPACE(TYPE, __generic)
#define DECL_BYTE_RD_SPACE(TYPE, SPACE) \
OVERLOADABLE TYPE##2 vload2(size_t offset, const SPACE TYPE *p); \
@@ -76,13 +74,9 @@ OVERLOADABLE void vstore8(TYPE##8 v, size_t offset, SPACE TYPE *p); \
OVERLOADABLE void vstore16(TYPE##16 v, size_t offset, SPACE TYPE *p);
#define DECL_BYTE_RW_ALL(TYPE) \
- DECL_BYTE_RD_SPACE(TYPE, __global) \
- DECL_BYTE_RD_SPACE(TYPE, __local) \
- DECL_BYTE_RD_SPACE(TYPE, __private) \
- DECL_BYTE_RD_SPACE(TYPE, __constant) \
- DECL_BYTE_WR_SPACE(TYPE, __global) \
- DECL_BYTE_WR_SPACE(TYPE, __local) \
- DECL_BYTE_WR_SPACE(TYPE, __private)
+ DECL_BYTE_RD_SPACE(TYPE, __generic) \
+ DECL_BYTE_WR_SPACE(TYPE, __generic) \
+ DECL_BYTE_RD_SPACE(TYPE, __constant)
DECL_BYTE_RW_ALL(char)
DECL_BYTE_RW_ALL(uchar)
@@ -137,14 +131,10 @@ OVERLOADABLE void vstorea_half16##ROUND(float16 data, size_t offset, SPACE half
DECL_HALF_ST_SPACE_ROUND(SPACE, _rtp, dummy) \
DECL_HALF_ST_SPACE_ROUND(SPACE, _rtn, dummy) \
-DECL_HALF_LD_SPACE(__global)
-DECL_HALF_LD_SPACE(__local)
DECL_HALF_LD_SPACE(__constant)
-DECL_HALF_LD_SPACE(__private)
+DECL_HALF_LD_SPACE(__generic)
-DECL_HALF_ST_SPACE(__global)
-DECL_HALF_ST_SPACE(__local)
-DECL_HALF_ST_SPACE(__private)
+DECL_HALF_ST_SPACE(__generic)
//#undef DECL_UNTYPED_RW_ALL_SPACE
#undef DECL_HALF_LD_SPACE
diff --git a/backend/src/libocl/src/ocl_vload.cl b/backend/src/libocl/src/ocl_vload.cl
index fa5e04fd..2d0befb6 100644
--- a/backend/src/libocl/src/ocl_vload.cl
+++ b/backend/src/libocl/src/ocl_vload.cl
@@ -65,10 +65,8 @@ OVERLOADABLE TYPE##3 vload3(size_t offset, const SPACE TYPE *p) { \
DECL_UNTYPED_RD_SPACE_N(TYPE, 16, SPACE)
#define DECL_UNTYPED_RW_ALL(TYPE) \
- DECL_UNTYPED_RW_ALL_SPACE(TYPE, __global) \
- DECL_UNTYPED_RW_ALL_SPACE(TYPE, __local) \
DECL_UNTYPED_RD_ALL_SPACE(TYPE, __constant) \
- DECL_UNTYPED_RW_ALL_SPACE(TYPE, __private)
+ DECL_UNTYPED_RW_ALL_SPACE(TYPE, __generic)
#define DECL_BYTE_RD_SPACE(TYPE, SPACE) \
OVERLOADABLE TYPE##2 vload2(size_t offset, const SPACE TYPE *p) { \
@@ -111,13 +109,9 @@ OVERLOADABLE void vstore16(TYPE##16 v, size_t offset, SPACE TYPE *p) { \
}
#define DECL_BYTE_RW_ALL(TYPE) \
- DECL_BYTE_RD_SPACE(TYPE, __global) \
- DECL_BYTE_RD_SPACE(TYPE, __local) \
- DECL_BYTE_RD_SPACE(TYPE, __private) \
+ DECL_BYTE_RD_SPACE(TYPE, __generic) \
DECL_BYTE_RD_SPACE(TYPE, __constant) \
- DECL_BYTE_WR_SPACE(TYPE, __global) \
- DECL_BYTE_WR_SPACE(TYPE, __local) \
- DECL_BYTE_WR_SPACE(TYPE, __private)
+ DECL_BYTE_WR_SPACE(TYPE, __generic)
DECL_BYTE_RW_ALL(char)
DECL_BYTE_RW_ALL(uchar)
@@ -259,14 +253,10 @@ OVERLOADABLE void vstorea_half16##ROUND(float16 data, size_t offset, SPACE half
DECL_HALF_ST_SPACE_ROUND(SPACE, _rtp, f32to16_rtp) \
DECL_HALF_ST_SPACE_ROUND(SPACE, _rtn, f32to16_rtn) \
-DECL_HALF_LD_SPACE(__global)
-DECL_HALF_LD_SPACE(__local)
DECL_HALF_LD_SPACE(__constant)
-DECL_HALF_LD_SPACE(__private)
+DECL_HALF_LD_SPACE(__generic)
-DECL_HALF_ST_SPACE(__global)
-DECL_HALF_ST_SPACE(__local)
-DECL_HALF_ST_SPACE(__private)
+DECL_HALF_ST_SPACE(__generic)
//#undef DECL_UNTYPED_RW_ALL_SPACE
#undef DECL_HALF_LD_SPACE