[SYCL][CUDA][libclc] Add bitwise group collective functions for remaining integral types (#6405)

Adds bitwise group collective functions for signed and unsigned versions of types char, short and long.

Changes to test suite in intel/llvm-test-suite#1077.

Author: t4c1

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

@@ -249,13 +249,34 @@ __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(BitwiseAndKHR, __CLC_AND, and, uchar, ~0)
+__CLC_SUBGROUP_COLLECTIVE_REDUX(BitwiseOrKHR, __CLC_OR, or, uchar, 0)
+__CLC_SUBGROUP_COLLECTIVE_REDUX(BitwiseXorKHR, __CLC_XOR, xor, uchar, 0)
+__CLC_SUBGROUP_COLLECTIVE_REDUX(BitwiseAndKHR, __CLC_AND, and, char, ~0)
+__CLC_SUBGROUP_COLLECTIVE_REDUX(BitwiseOrKHR, __CLC_OR, or, char, 0)
+__CLC_SUBGROUP_COLLECTIVE_REDUX(BitwiseXorKHR, __CLC_XOR, xor, char, 0)
+
+__CLC_SUBGROUP_COLLECTIVE_REDUX(BitwiseAndKHR, __CLC_AND, and, ushort, ~0)
+__CLC_SUBGROUP_COLLECTIVE_REDUX(BitwiseOrKHR, __CLC_OR, or, ushort, 0)
+__CLC_SUBGROUP_COLLECTIVE_REDUX(BitwiseXorKHR, __CLC_XOR, xor, ushort, 0)
+__CLC_SUBGROUP_COLLECTIVE_REDUX(BitwiseAndKHR, __CLC_AND, and, short, ~0)
+__CLC_SUBGROUP_COLLECTIVE_REDUX(BitwiseOrKHR, __CLC_OR, or, short, 0)
+__CLC_SUBGROUP_COLLECTIVE_REDUX(BitwiseXorKHR, __CLC_XOR, xor, short, 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)
 
+__CLC_SUBGROUP_COLLECTIVE(BitwiseAndKHR, __CLC_AND, ulong, ~0l)
+__CLC_SUBGROUP_COLLECTIVE(BitwiseOrKHR, __CLC_OR, ulong, 0l)
+__CLC_SUBGROUP_COLLECTIVE(BitwiseXorKHR, __CLC_XOR, ulong, 0l)
+__CLC_SUBGROUP_COLLECTIVE(BitwiseAndKHR, __CLC_AND, long, ~0l)
+__CLC_SUBGROUP_COLLECTIVE(BitwiseOrKHR, __CLC_OR, long, 0l)
+__CLC_SUBGROUP_COLLECTIVE(BitwiseXorKHR, __CLC_XOR, long, 0l)
+
 #undef __CLC_SUBGROUP_COLLECTIVE_BODY
 #undef __CLC_SUBGROUP_COLLECTIVE
 #undef __CLC_SUBGROUP_COLLECTIVE_REDUX
@@ -376,13 +397,34 @@ __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(BitwiseAndKHR, __CLC_AND, uchar, ~0)
+__CLC_GROUP_COLLECTIVE(BitwiseOrKHR, __CLC_OR, uchar, 0)
+__CLC_GROUP_COLLECTIVE(BitwiseXorKHR, __CLC_XOR, uchar, 0)
+__CLC_GROUP_COLLECTIVE(BitwiseAndKHR, __CLC_AND, char, ~0)
+__CLC_GROUP_COLLECTIVE(BitwiseOrKHR, __CLC_OR, char, 0)
+__CLC_GROUP_COLLECTIVE(BitwiseXorKHR, __CLC_XOR, char, 0)
+
+__CLC_GROUP_COLLECTIVE(BitwiseAndKHR, __CLC_AND, ushort, ~0)
+__CLC_GROUP_COLLECTIVE(BitwiseOrKHR, __CLC_OR, ushort, 0)
+__CLC_GROUP_COLLECTIVE(BitwiseXorKHR, __CLC_XOR, ushort, 0)
+__CLC_GROUP_COLLECTIVE(BitwiseAndKHR, __CLC_AND, short, ~0)
+__CLC_GROUP_COLLECTIVE(BitwiseOrKHR, __CLC_OR, short, 0)
+__CLC_GROUP_COLLECTIVE(BitwiseXorKHR, __CLC_XOR, short, 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)
 
+__CLC_GROUP_COLLECTIVE(BitwiseAndKHR, __CLC_AND, ulong, ~0l)
+__CLC_GROUP_COLLECTIVE(BitwiseOrKHR, __CLC_OR, ulong, 0l)
+__CLC_GROUP_COLLECTIVE(BitwiseXorKHR, __CLC_XOR, ulong, 0l)
+__CLC_GROUP_COLLECTIVE(BitwiseAndKHR, __CLC_AND, long, ~0l)
+__CLC_GROUP_COLLECTIVE(BitwiseOrKHR, __CLC_OR, long, 0l)
+__CLC_GROUP_COLLECTIVE(BitwiseXorKHR, __CLC_XOR, long, 0l)
+
 // half requires additional mangled entry points
 _CLC_DEF _CLC_CONVERGENT half _Z17__spirv_GroupFAddjjDF16_(uint scope, uint op,
                                                            half x) {