[SYCL][CUDA][HIP][libclc] Rename libclc NonUniform binops to KHR extensions

#5705 changed the use of NonUniform
group binary operators to the KHR extension operators. These changes
rename the implementation of these in libclc for both amdgcn and PTX to
the new naming.

Signed-off-by: Larsen, Steffen <[email protected]>

Author: steffenlarsen

libclc/amdgcn-amdhsa/libspirv/group/collectives.cl

@@ -235,16 +235,16 @@ __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, 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, float, 1)
-__CLC_GROUP_COLLECTIVE(NonUniformFMul, FMul, __CLC_MUL, double, 1)
+__CLC_GROUP_COLLECTIVE(IMulKHR, IMul, __CLC_MUL, char, 1)
+__CLC_GROUP_COLLECTIVE(IMulKHR, IMul, __CLC_MUL, uchar, 1)
+__CLC_GROUP_COLLECTIVE(IMulKHR, IMul, __CLC_MUL, short, 1)
+__CLC_GROUP_COLLECTIVE(IMulKHR, IMul, __CLC_MUL, ushort, 1)
+__CLC_GROUP_COLLECTIVE(IMulKHR, IMul, __CLC_MUL, int, 1)
+__CLC_GROUP_COLLECTIVE(IMulKHR, IMul, __CLC_MUL, uint, 1)
+__CLC_GROUP_COLLECTIVE(IMulKHR, IMul, __CLC_MUL, long, 1)
+__CLC_GROUP_COLLECTIVE(IMulKHR, IMul, __CLC_MUL, ulong, 1)
+__CLC_GROUP_COLLECTIVE(FMulKHR, FMul, __CLC_MUL, float, 1)
+__CLC_GROUP_COLLECTIVE(FMulKHR, FMul, __CLC_MUL, double, 1)
 
 __CLC_GROUP_COLLECTIVE(SMin, __CLC_MIN, char, CHAR_MAX)
 __CLC_GROUP_COLLECTIVE(UMin, __CLC_MIN, uchar, UCHAR_MAX)

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

@@ -249,12 +249,12 @@ __CLC_SUBGROUP_COLLECTIVE(FMax, __CLC_MAX, half, -HALF_MAX)
 __CLC_SUBGROUP_COLLECTIVE(FMax, __CLC_MAX, float, -FLT_MAX)
 __CLC_SUBGROUP_COLLECTIVE(FMax, __CLC_MAX, double, -DBL_MAX)
 
-__CLC_SUBGROUP_COLLECTIVE_REDUX(NonUniformBitwiseAnd, __CLC_AND, and, uint, ~0)
-__CLC_SUBGROUP_COLLECTIVE_REDUX(NonUniformBitwiseOr, __CLC_OR, or, uint, 0)
-__CLC_SUBGROUP_COLLECTIVE_REDUX(NonUniformBitwiseXor, __CLC_XOR, xor, uint, 0)
-__CLC_SUBGROUP_COLLECTIVE_REDUX(NonUniformBitwiseAnd, __CLC_AND, and, int, ~0)
-__CLC_SUBGROUP_COLLECTIVE_REDUX(NonUniformBitwiseOr, __CLC_OR, or, int, 0)
-__CLC_SUBGROUP_COLLECTIVE_REDUX(NonUniformBitwiseXor, __CLC_XOR, xor, int, 0)
+__CLC_SUBGROUP_COLLECTIVE_REDUX(BitwiseAndKHR, __CLC_AND, and, uint, ~0)
+__CLC_SUBGROUP_COLLECTIVE_REDUX(BitwiseOrKHR, __CLC_OR, or, uint, 0)
+__CLC_SUBGROUP_COLLECTIVE_REDUX(BitwiseXorKHR, __CLC_XOR, xor, uint, 0)
+__CLC_SUBGROUP_COLLECTIVE_REDUX(BitwiseAndKHR, __CLC_AND, and, int, ~0)
+__CLC_SUBGROUP_COLLECTIVE_REDUX(BitwiseOrKHR, __CLC_OR, or, int, 0)
+__CLC_SUBGROUP_COLLECTIVE_REDUX(BitwiseXorKHR, __CLC_XOR, xor, int, 0)
 
 #undef __CLC_SUBGROUP_COLLECTIVE_BODY
 #undef __CLC_SUBGROUP_COLLECTIVE
@@ -340,18 +340,17 @@ __CLC_GROUP_COLLECTIVE(FAdd, __CLC_ADD, half, 0)
 __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, 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(IMulKHR, IMul, __CLC_MUL, char, 1)
+__CLC_GROUP_COLLECTIVE(IMulKHR, IMul, __CLC_MUL, uchar, 1)
+__CLC_GROUP_COLLECTIVE(IMulKHR, IMul, __CLC_MUL, short, 1)
+__CLC_GROUP_COLLECTIVE(IMulKHR, IMul, __CLC_MUL, ushort, 1)
+__CLC_GROUP_COLLECTIVE(IMulKHR, IMul, __CLC_MUL, int, 1)
+__CLC_GROUP_COLLECTIVE(IMulKHR, IMul, __CLC_MUL, uint, 1)
+__CLC_GROUP_COLLECTIVE(IMulKHR, IMul, __CLC_MUL, long, 1)
+__CLC_GROUP_COLLECTIVE(IMulKHR, IMul, __CLC_MUL, ulong, 1)
+__CLC_GROUP_COLLECTIVE(FMulKHR, FMul, __CLC_MUL, half, 1)
+__CLC_GROUP_COLLECTIVE(FMulKHR, FMul, __CLC_MUL, float, 1)
+__CLC_GROUP_COLLECTIVE(FMulKHR, FMul, __CLC_MUL, double, 1)
 
 __CLC_GROUP_COLLECTIVE(SMin, __CLC_MIN, char, CHAR_MAX)
 __CLC_GROUP_COLLECTIVE(UMin, __CLC_MIN, uchar, UCHAR_MAX)
@@ -377,12 +376,12 @@ __CLC_GROUP_COLLECTIVE(FMax, __CLC_MAX, half, -HALF_MAX)
 __CLC_GROUP_COLLECTIVE(FMax, __CLC_MAX, float, -FLT_MAX)
 __CLC_GROUP_COLLECTIVE(FMax, __CLC_MAX, double, -DBL_MAX)
 
-__CLC_GROUP_COLLECTIVE(NonUniformBitwiseAnd, __CLC_AND, uint, ~0)
-__CLC_GROUP_COLLECTIVE(NonUniformBitwiseOr, __CLC_OR, uint, 0)
-__CLC_GROUP_COLLECTIVE(NonUniformBitwiseXor, __CLC_XOR, uint, 0)
-__CLC_GROUP_COLLECTIVE(NonUniformBitwiseAnd, __CLC_AND, int, ~0)
-__CLC_GROUP_COLLECTIVE(NonUniformBitwiseOr, __CLC_OR, int, 0)
-__CLC_GROUP_COLLECTIVE(NonUniformBitwiseXor, __CLC_XOR, int, 0)
+__CLC_GROUP_COLLECTIVE(BitwiseAndKHR, __CLC_AND, uint, ~0)
+__CLC_GROUP_COLLECTIVE(BitwiseOrKHR, __CLC_OR, uint, 0)
+__CLC_GROUP_COLLECTIVE(BitwiseXorKHR, __CLC_XOR, uint, 0)
+__CLC_GROUP_COLLECTIVE(BitwiseAndKHR, __CLC_AND, int, ~0)
+__CLC_GROUP_COLLECTIVE(BitwiseOrKHR, __CLC_OR, int, 0)
+__CLC_GROUP_COLLECTIVE(BitwiseXorKHR, __CLC_XOR, int, 0)
 
 // half requires additional mangled entry points
 _CLC_DEF _CLC_CONVERGENT half _Z17__spirv_GroupFAddjjDF16_(uint scope, uint op,