[SYCL][libclc] Fix identity for multiplication (intel#4337)

This resolves intel#4336 issue which is a bug related to 0 being used as the identity in the CUDA back-end.

Signed-off-by: Michel Migdal [email protected]

Author: Michoumichmich

libclc/ptx-nvidiacl/libspirv/group/collectives.cl

@@ -211,17 +211,17 @@ __CLC_SUBGROUP_COLLECTIVE(FAdd, __CLC_ADD, half, 0)
 __CLC_SUBGROUP_COLLECTIVE(FAdd, __CLC_ADD, float, 0)
 __CLC_SUBGROUP_COLLECTIVE(FAdd, __CLC_ADD, double, 0)
 
-__CLC_SUBGROUP_COLLECTIVE(IMul, __CLC_MUL, char, 0)
-__CLC_SUBGROUP_COLLECTIVE(IMul, __CLC_MUL, uchar, 0)
-__CLC_SUBGROUP_COLLECTIVE(IMul, __CLC_MUL, short, 0)
-__CLC_SUBGROUP_COLLECTIVE(IMul, __CLC_MUL, ushort, 0)
-__CLC_SUBGROUP_COLLECTIVE(IMul, __CLC_MUL, int, 0)
-__CLC_SUBGROUP_COLLECTIVE(IMul, __CLC_MUL, uint, 0)
-__CLC_SUBGROUP_COLLECTIVE(IMul, __CLC_MUL, long, 0)
-__CLC_SUBGROUP_COLLECTIVE(IMul, __CLC_MUL, ulong, 0)
-__CLC_SUBGROUP_COLLECTIVE(FMul, __CLC_MUL, half, 0)
-__CLC_SUBGROUP_COLLECTIVE(FMul, __CLC_MUL, float, 0)
-__CLC_SUBGROUP_COLLECTIVE(FMul, __CLC_MUL, double, 0)
+__CLC_SUBGROUP_COLLECTIVE(IMul, __CLC_MUL, char, 1)
+__CLC_SUBGROUP_COLLECTIVE(IMul, __CLC_MUL, uchar, 1)
+__CLC_SUBGROUP_COLLECTIVE(IMul, __CLC_MUL, short, 1)
+__CLC_SUBGROUP_COLLECTIVE(IMul, __CLC_MUL, ushort, 1)
+__CLC_SUBGROUP_COLLECTIVE(IMul, __CLC_MUL, int, 1)
+__CLC_SUBGROUP_COLLECTIVE(IMul, __CLC_MUL, uint, 1)
+__CLC_SUBGROUP_COLLECTIVE(IMul, __CLC_MUL, long, 1)
+__CLC_SUBGROUP_COLLECTIVE(IMul, __CLC_MUL, ulong, 1)
+__CLC_SUBGROUP_COLLECTIVE(FMul, __CLC_MUL, half, 1)
+__CLC_SUBGROUP_COLLECTIVE(FMul, __CLC_MUL, float, 1)
+__CLC_SUBGROUP_COLLECTIVE(FMul, __CLC_MUL, double, 1)
 
 __CLC_SUBGROUP_COLLECTIVE(SMin, __CLC_MIN, char, CHAR_MAX)
 __CLC_SUBGROUP_COLLECTIVE(UMin, __CLC_MIN, uchar, UCHAR_MAX)
@@ -332,17 +332,17 @@ __CLC_GROUP_COLLECTIVE(FAdd, __CLC_ADD, float, 0)
 __CLC_GROUP_COLLECTIVE(FAdd, __CLC_ADD, double, 0)
 
 // There is no Mul group op in SPIR-V, use non-uniform variant instead.
-__CLC_GROUP_COLLECTIVE(NonUniformIMul, IMul, __CLC_MUL, char, 0)
-__CLC_GROUP_COLLECTIVE(NonUniformIMul, IMul, __CLC_MUL, uchar, 0)
-__CLC_GROUP_COLLECTIVE(NonUniformIMul, IMul, __CLC_MUL, short, 0)
-__CLC_GROUP_COLLECTIVE(NonUniformIMul, IMul, __CLC_MUL, ushort, 0)
-__CLC_GROUP_COLLECTIVE(NonUniformIMul, IMul, __CLC_MUL, int, 0)
-__CLC_GROUP_COLLECTIVE(NonUniformIMul, IMul, __CLC_MUL, uint, 0)
-__CLC_GROUP_COLLECTIVE(NonUniformIMul, IMul, __CLC_MUL, long, 0)
-__CLC_GROUP_COLLECTIVE(NonUniformIMul, IMul, __CLC_MUL, ulong, 0)
-__CLC_GROUP_COLLECTIVE(NonUniformFMul, FMul, __CLC_MUL, half, 0)
-__CLC_GROUP_COLLECTIVE(NonUniformFMul, FMul, __CLC_MUL, float, 0)
-__CLC_GROUP_COLLECTIVE(NonUniformFMul, FMul, __CLC_MUL, double, 0)
+__CLC_GROUP_COLLECTIVE(NonUniformIMul, IMul, __CLC_MUL, char, 1)
+__CLC_GROUP_COLLECTIVE(NonUniformIMul, IMul, __CLC_MUL, uchar, 1)
+__CLC_GROUP_COLLECTIVE(NonUniformIMul, IMul, __CLC_MUL, short, 1)
+__CLC_GROUP_COLLECTIVE(NonUniformIMul, IMul, __CLC_MUL, ushort, 1)
+__CLC_GROUP_COLLECTIVE(NonUniformIMul, IMul, __CLC_MUL, int, 1)
+__CLC_GROUP_COLLECTIVE(NonUniformIMul, IMul, __CLC_MUL, uint, 1)
+__CLC_GROUP_COLLECTIVE(NonUniformIMul, IMul, __CLC_MUL, long, 1)
+__CLC_GROUP_COLLECTIVE(NonUniformIMul, IMul, __CLC_MUL, ulong, 1)
+__CLC_GROUP_COLLECTIVE(NonUniformFMul, FMul, __CLC_MUL, half, 1)
+__CLC_GROUP_COLLECTIVE(NonUniformFMul, FMul, __CLC_MUL, float, 1)
+__CLC_GROUP_COLLECTIVE(NonUniformFMul, FMul, __CLC_MUL, double, 1)
 
 __CLC_GROUP_COLLECTIVE(SMin, __CLC_MIN, char, CHAR_MAX)
 __CLC_GROUP_COLLECTIVE(UMin, __CLC_MIN, uchar, UCHAR_MAX)