[SYCL] Use SPIR-V built-in function call for all targets and add BuiltIn to the name (#19359)

Before this PR, SPIR-V built-in is represented as global variable for
SPIR/SPIR-V targets and as function call for other targets in
include/sycl/__spirv/spirv_vars.hpp.

According to
https://github.com/llvm/llvm-project/blob/main/llvm/docs/SPIRVUsage.rst,
SPIR-V built-in variable can be mapped to either function call or global
variable. So function call representation should work for SPIR-V target
as well.
The benefit of choosing function call over global variable is that
function call works better for AOT targets that bypass SPIR-V. It is
easier for these targets to implements SPIR-V built-in function via
built-in modules like libclc rather than special handling for lowering
global variable in a custom pass.

Add `BuiltIn` to the name to align with SPIR-V friendly IR and global
variable name.

Author: wenju-he

libclc/libspirv/include/libspirv/async/common.h

@@ -10,12 +10,13 @@
 #define CLC_ASYNC_COMMON
 
 #define SET_GROUP_SIZE_AND_ID(SIZE, ID)                                        \
-  SIZE = __spirv_WorkgroupSize_x() * __spirv_WorkgroupSize_y() *               \
-         __spirv_WorkgroupSize_z();                                            \
-  ID = (__spirv_WorkgroupSize_y() * __spirv_WorkgroupSize_x() *                \
-        __spirv_LocalInvocationId_z()) +                                       \
-       (__spirv_WorkgroupSize_x() * __spirv_LocalInvocationId_y()) +           \
-       __spirv_LocalInvocationId_x();
+  SIZE = __spirv_BuiltInWorkgroupSize(0) * __spirv_BuiltInWorkgroupSize(1) *   \
+         __spirv_BuiltInWorkgroupSize(2);                                      \
+  ID = (__spirv_BuiltInWorkgroupSize(1) * __spirv_BuiltInWorkgroupSize(0) *    \
+        __spirv_BuiltInLocalInvocationId(2)) +                                 \
+       (__spirv_BuiltInWorkgroupSize(0) *                                      \
+        __spirv_BuiltInLocalInvocationId(1)) +                                 \
+       __spirv_BuiltInLocalInvocationId(0);
 
 // Macro used by all data types, for generic and nvidia, for async copy when
 // arch < sm80

libclc/libspirv/include/libspirv/workitem/get_global_offset.h

@@ -6,6 +6,4 @@
 //
 //===----------------------------------------------------------------------===//
 
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_GlobalOffset_x();
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_GlobalOffset_y();
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_GlobalOffset_z();
+_CLC_DECL _CLC_OVERLOAD size_t __spirv_BuiltInGlobalOffset(int);

libclc/libspirv/include/libspirv/workitem/get_global_size.h

@@ -6,6 +6,4 @@
 //
 //===----------------------------------------------------------------------===//
 
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_GlobalSize_x();
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_GlobalSize_y();
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_GlobalSize_z();
+_CLC_DECL _CLC_OVERLOAD size_t __spirv_BuiltInGlobalSize(int);

libclc/libspirv/include/libspirv/workitem/get_group_id.h

@@ -6,6 +6,4 @@
 //
 //===----------------------------------------------------------------------===//
 
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_WorkgroupId_x();
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_WorkgroupId_y();
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_WorkgroupId_z();
+_CLC_DECL _CLC_OVERLOAD size_t __spirv_BuiltInWorkgroupId(int);

libclc/libspirv/include/libspirv/workitem/get_local_id.h

@@ -6,6 +6,4 @@
 //
 //===----------------------------------------------------------------------===//
 
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_LocalInvocationId_x();
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_LocalInvocationId_y();
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_LocalInvocationId_z();
+_CLC_DECL _CLC_OVERLOAD size_t __spirv_BuiltInLocalInvocationId(int);

libclc/libspirv/include/libspirv/workitem/get_local_linear_id.h

@@ -6,4 +6,4 @@
 //
 //===----------------------------------------------------------------------===//
 
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_LocalInvocationIndex();
+_CLC_DECL _CLC_OVERLOAD size_t __spirv_BuiltInLocalInvocationIndex();

libclc/libspirv/include/libspirv/workitem/get_local_size.h

@@ -6,6 +6,4 @@
 //
 //===----------------------------------------------------------------------===//
 
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_WorkgroupSize_x();
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_WorkgroupSize_y();
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_WorkgroupSize_z();
+_CLC_DECL _CLC_OVERLOAD size_t __spirv_BuiltInWorkgroupSize(int);

libclc/libspirv/include/libspirv/workitem/get_max_sub_group_size.h

@@ -6,4 +6,4 @@
 //
 //===----------------------------------------------------------------------===//
 
-_CLC_DEF _CLC_OVERLOAD uint __spirv_SubgroupMaxSize();
+_CLC_DEF _CLC_OVERLOAD uint __spirv_BuiltInSubgroupMaxSize();

libclc/libspirv/include/libspirv/workitem/get_num_groups.h

@@ -6,6 +6,4 @@
 //
 //===----------------------------------------------------------------------===//
 
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_NumWorkgroups_x();
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_NumWorkgroups_y();
-_CLC_DECL _CLC_OVERLOAD size_t __spirv_NumWorkgroups_z();
+_CLC_DECL _CLC_OVERLOAD size_t __spirv_BuiltInNumWorkgroups(int);

libclc/libspirv/include/libspirv/workitem/get_num_sub_groups.h

@@ -6,4 +6,4 @@
 //
 //===----------------------------------------------------------------------===//
 
-_CLC_DEF _CLC_OVERLOAD uint __spirv_NumSubgroups();
+_CLC_DEF _CLC_OVERLOAD uint __spirv_BuiltInNumSubgroups();