libclc: Stop using asm declarations for r600 on amdgcn for get_global_size (#128692)

Comparing the case where each dimension is used alone, the only codegen
difference is a missed addressing mode fold for the constant offset in the old
version due to an ancient bug.

Author: arsenm

libclc/amdgcn/lib/workitem/get_global_size.cl

@@ -1,17 +1,13 @@
 #include <clc/clc.h>
 
-uint __clc_amdgcn_get_global_size_x(void) __asm("llvm.r600.read.global.size.x");
-uint __clc_amdgcn_get_global_size_y(void) __asm("llvm.r600.read.global.size.y");
-uint __clc_amdgcn_get_global_size_z(void) __asm("llvm.r600.read.global.size.z");
-
 _CLC_DEF _CLC_OVERLOAD size_t get_global_size(uint dim) {
   switch (dim) {
   case 0:
-    return __clc_amdgcn_get_global_size_x();
+    return __builtin_amdgcn_grid_size_x();
   case 1:
-    return __clc_amdgcn_get_global_size_y();
+    return __builtin_amdgcn_grid_size_y();
   case 2:
-    return __clc_amdgcn_get_global_size_z();
+    return __builtin_amdgcn_grid_size_z();
   default:
     return 1;
   }