[Libclc-dev] [PATCH 3/7] vstore_half: Consolidate declarations

[Jan Vesely via Libclc-dev]

Add support for rounding suffix

Signed-off-by: Jan Vesely <jan.vesely at rutgers.edu>
---
 generic/include/clc/shared/vstore.h | 65 +++++++++++++++----------------------
 1 file changed, 26 insertions(+), 39 deletions(-)

diff --git a/generic/include/clc/shared/vstore.h b/generic/include/clc/shared/vstore.h
index a246d52..e479e33 100644
--- a/generic/include/clc/shared/vstore.h
+++ b/generic/include/clc/shared/vstore.h
@@ -1,20 +1,30 @@
-#define _CLC_VSTORE_DECL(SUFFIX, PRIM_TYPE, VEC_TYPE, WIDTH, ADDR_SPACE) \
-  _CLC_OVERLOAD _CLC_DECL void vstore##SUFFIX##WIDTH(VEC_TYPE vec, size_t offset, ADDR_SPACE PRIM_TYPE *out);
+#define _CLC_VSTORE_DECL(SUFFIX, PRIM_TYPE, VEC_TYPE, WIDTH, ADDR_SPACE, RND) \
+  _CLC_OVERLOAD _CLC_DECL void vstore##SUFFIX##WIDTH##RND(VEC_TYPE vec, size_t offset, ADDR_SPACE PRIM_TYPE *out);
 
-#define _CLC_VECTOR_VSTORE_DECL(SUFFIX, MEM_TYPE, PRIM_TYPE, ADDR_SPACE) \
-  _CLC_VSTORE_DECL(SUFFIX, MEM_TYPE, PRIM_TYPE##2, 2, ADDR_SPACE) \
-  _CLC_VSTORE_DECL(SUFFIX, MEM_TYPE, PRIM_TYPE##3, 3, ADDR_SPACE) \
-  _CLC_VSTORE_DECL(SUFFIX, MEM_TYPE, PRIM_TYPE##4, 4, ADDR_SPACE) \
-  _CLC_VSTORE_DECL(SUFFIX, MEM_TYPE, PRIM_TYPE##8, 8, ADDR_SPACE) \
-  _CLC_VSTORE_DECL(SUFFIX, MEM_TYPE, PRIM_TYPE##16, 16, ADDR_SPACE)
+#define _CLC_VECTOR_VSTORE_DECL(SUFFIX, MEM_TYPE, PRIM_TYPE, ADDR_SPACE, RND) \
+  _CLC_VSTORE_DECL(SUFFIX, MEM_TYPE, PRIM_TYPE##2, 2, ADDR_SPACE, RND) \
+  _CLC_VSTORE_DECL(SUFFIX, MEM_TYPE, PRIM_TYPE##3, 3, ADDR_SPACE, RND) \
+  _CLC_VSTORE_DECL(SUFFIX, MEM_TYPE, PRIM_TYPE##4, 4, ADDR_SPACE, RND) \
+  _CLC_VSTORE_DECL(SUFFIX, MEM_TYPE, PRIM_TYPE##8, 8, ADDR_SPACE, RND) \
+  _CLC_VSTORE_DECL(SUFFIX, MEM_TYPE, PRIM_TYPE##16, 16, ADDR_SPACE, RND)
 
-#define _CLC_VECTOR_VSTORE_PRIM3(SUFFIX, MEM_TYPE, PRIM_TYPE) \
-  _CLC_VECTOR_VSTORE_DECL(SUFFIX, MEM_TYPE, PRIM_TYPE, __private) \
-  _CLC_VECTOR_VSTORE_DECL(SUFFIX, MEM_TYPE, PRIM_TYPE, __local) \
-  _CLC_VECTOR_VSTORE_DECL(SUFFIX, MEM_TYPE, PRIM_TYPE, __global) \
+#define _CLC_VECTOR_VSTORE_PRIM3(SUFFIX, MEM_TYPE, PRIM_TYPE, RND) \
+  _CLC_VECTOR_VSTORE_DECL(SUFFIX, MEM_TYPE, PRIM_TYPE, __private, RND) \
+  _CLC_VECTOR_VSTORE_DECL(SUFFIX, MEM_TYPE, PRIM_TYPE, __local, RND) \
+  _CLC_VECTOR_VSTORE_DECL(SUFFIX, MEM_TYPE, PRIM_TYPE, __global, RND)
 
 #define _CLC_VECTOR_VSTORE_PRIM1(PRIM_TYPE) \
-  _CLC_VECTOR_VSTORE_PRIM3(,PRIM_TYPE, PRIM_TYPE) \
+  _CLC_VECTOR_VSTORE_PRIM3(,PRIM_TYPE, PRIM_TYPE, )
+
+#define _CLC_VECTOR_VSTORE_HALF_PRIM1(PRIM_TYPE, RND) \
+	_CLC_VSTORE_DECL(_half, half, PRIM_TYPE, , __private, RND) \
+	_CLC_VSTORE_DECL(_half, half, PRIM_TYPE, , __local, RND) \
+	_CLC_VSTORE_DECL(_half, half, PRIM_TYPE, , __global, RND) \
+	_CLC_VECTOR_VSTORE_PRIM3(_half, half, PRIM_TYPE, RND) \
+	_CLC_VSTORE_DECL(a_half, half, PRIM_TYPE, , __private, RND) \
+	_CLC_VSTORE_DECL(a_half, half, PRIM_TYPE, , __local, RND) \
+	_CLC_VSTORE_DECL(a_half, half, PRIM_TYPE, , __global, RND) \
+	_CLC_VECTOR_VSTORE_PRIM3(a_half, half, PRIM_TYPE, RND)
 
 _CLC_VECTOR_VSTORE_PRIM1(char)
 _CLC_VECTOR_VSTORE_PRIM1(uchar)
@@ -25,41 +35,18 @@ _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)
+
+_CLC_VECTOR_VSTORE_HALF_PRIM1(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)
+  _CLC_VECTOR_VSTORE_HALF_PRIM1(double,)
 #endif
 
 #ifdef cl_khr_fp16
   _CLC_VECTOR_VSTORE_PRIM1(half)
 #endif
 
-// 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
-- 
2.14.3