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

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();

libclc/libspirv/include/libspirv/workitem/get_sub_group_id.h

@@ -6,4 +6,4 @@
 //
 //===----------------------------------------------------------------------===//
 
-_CLC_DEF _CLC_OVERLOAD uint __spirv_SubgroupId();
+_CLC_DEF _CLC_OVERLOAD uint __spirv_BuiltInSubgroupId();

libclc/libspirv/include/libspirv/workitem/get_sub_group_local_id.h

@@ -6,4 +6,4 @@
 //
 //===----------------------------------------------------------------------===//
 
-_CLC_DEF _CLC_OVERLOAD uint __spirv_SubgroupLocalInvocationId();
+_CLC_DEF _CLC_OVERLOAD uint __spirv_BuiltInSubgroupLocalInvocationId();

libclc/libspirv/include/libspirv/workitem/get_sub_group_size.h

@@ -6,4 +6,4 @@
 //
 //===----------------------------------------------------------------------===//
 
-_CLC_DEF _CLC_OVERLOAD uint __spirv_SubgroupSize();
+_CLC_DEF _CLC_OVERLOAD uint __spirv_BuiltInSubgroupSize();

libclc/libspirv/lib/amdgcn-amdhsa/assert/__assert_fail.ll

@@ -14,13 +14,9 @@ declare i64 @__ockl_fprintf_stderr_begin() local_unnamed_addr
 declare i64 @__ockl_fprintf_append_string_n(i64, i8* readonly, i64, i32) local_unnamed_addr
 declare i64 @__ockl_fprintf_append_args(i64, i32, i64, i64, i64, i64, i64, i64, i64, i32) local_unnamed_addr
 
-declare dso_local i64 @_Z28__spirv_GlobalInvocationId_xv() local_unnamed_addr
-declare dso_local i64 @_Z28__spirv_GlobalInvocationId_yv() local_unnamed_addr
-declare dso_local i64 @_Z28__spirv_GlobalInvocationId_zv() local_unnamed_addr
+declare dso_local i64 @_Z33__spirv_BuiltInGlobalInvocationIdi(i32) local_unnamed_addr
 
-declare dso_local i64 @_Z27__spirv_LocalInvocationId_xv() local_unnamed_addr
-declare dso_local i64 @_Z27__spirv_LocalInvocationId_yv() local_unnamed_addr
-declare dso_local i64 @_Z27__spirv_LocalInvocationId_zv() local_unnamed_addr
+declare dso_local i64 @_Z32__spirv_BuiltInLocalInvocationIdi(i32) local_unnamed_addr
 
 define dso_local hidden noundef i64 @__strlen_assert(i8* noundef %str) local_unnamed_addr {
 entry:
@@ -50,12 +46,12 @@ entry:
   %msg.3 = call i64 @__ockl_fprintf_append_args(i64 %msg.2, i32 1, i64 %line.i64, i64 0, i64 0, i64 0, i64 0, i64 0, i64 0, i32 0)
   %len.func = call i64 @__strlen_assert(i8* %function)
   %msg.4 = call i64 @__ockl_fprintf_append_string_n(i64 %msg.3, i8* readonly %function, i64 %len.func, i32 0)
-  %gidx = tail call i64 @_Z28__spirv_GlobalInvocationId_xv()
-  %gidy = tail call i64 @_Z28__spirv_GlobalInvocationId_yv()
-  %gidz = tail call i64 @_Z28__spirv_GlobalInvocationId_zv()
-  %lidx = tail call i64 @_Z27__spirv_LocalInvocationId_xv()
-  %lidy = tail call i64 @_Z27__spirv_LocalInvocationId_yv()
-  %lidz = tail call i64 @_Z27__spirv_LocalInvocationId_zv()
+  %gidx = tail call i64 @_Z33__spirv_BuiltInGlobalInvocationIdi(i32 0)
+  %gidy = tail call i64 @_Z33__spirv_BuiltInGlobalInvocationIdi(i32 1)
+  %gidz = tail call i64 @_Z33__spirv_BuiltInGlobalInvocationIdi(i32 2)
+  %lidx = tail call i64 @_Z32__spirv_BuiltInLocalInvocationIdi(i32 0)
+  %lidy = tail call i64 @_Z32__spirv_BuiltInLocalInvocationIdi(i32 1)
+  %lidz = tail call i64 @_Z32__spirv_BuiltInLocalInvocationIdi(i32 2)
   %msg.5 = call i64 @__ockl_fprintf_append_args(i64 %msg.4, i32 6, i64 %gidx, i64 %gidy, i64 %gidz, i64 %lidx, i64 %lidy, i64 %lidz, i64 0, i32 0)
   %len.assertion = call i64 @__strlen_assert(i8* %assertion)
   %msg.6 = call i64 @__ockl_fprintf_append_string_n(i64 %msg.4, i8* readonly %assertion, i64 %len.assertion, i32 1)

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

@@ -51,9 +51,9 @@ __clc__get_group_scratch_double() __asm("__clc__get_group_scratch_double");
 #define __CLC_LOGICAL_AND(x, y) (x && y)
 
 #define __CLC_SUBGROUP_COLLECTIVE_BODY(OP, TYPE, IDENTITY)                     \
-  uint sg_lid = __spirv_SubgroupLocalInvocationId();                           \
+  uint sg_lid = __spirv_BuiltInSubgroupLocalInvocationId();                    \
   /* Can't use XOR/butterfly shuffles; some lanes may be inactive */           \
-  for (int o = 1; o < __spirv_SubgroupMaxSize(); o *= 2) {                     \
+  for (int o = 1; o < __spirv_BuiltInSubgroupMaxSize(); o *= 2) {              \
     TYPE contribution = __spirv_SubgroupShuffleUpINTEL(x, x, o);               \
     bool inactive = (sg_lid < o);                                              \
     contribution = (inactive) ? IDENTITY : contribution;                       \
@@ -62,7 +62,8 @@ __clc__get_group_scratch_double() __asm("__clc__get_group_scratch_double");
   /* For Reduce, broadcast result from highest active lane */                  \
   TYPE result;                                                                 \
   if (op == Reduce) {                                                          \
-    result = __spirv_SubgroupShuffleINTEL(x, __spirv_SubgroupSize() - 1);      \
+    result =                                                                   \
+        __spirv_SubgroupShuffleINTEL(x, __spirv_BuiltInSubgroupSize() - 1);    \
     *carry = result;                                                           \
   } /* For InclusiveScan, use results as computed */                           \
   else if (op == InclusiveScan) {                                              \
@@ -171,18 +172,18 @@ __CLC_SUBGROUP_COLLECTIVE(LogicalAndKHR, __CLC_LOGICAL_AND, bool, true)
 
 #define __CLC_GROUP_COLLECTIVE_INNER(SPIRV_NAME, CLC_NAME, OP, TYPE, IDENTITY) \
   _CLC_DEF _CLC_OVERLOAD _CLC_CONVERGENT TYPE __CLC_APPEND(                    \
-      __spirv_Group, SPIRV_NAME)(int scope, int op, TYPE x) {                \
+      __spirv_Group, SPIRV_NAME)(int scope, int op, TYPE x) {                  \
     TYPE carry = IDENTITY;                                                     \
     /* Perform GroupOperation within sub-group */                              \
     TYPE sg_x = __CLC_APPEND(__clc__Subgroup, CLC_NAME)(op, x, &carry);        \
     if (scope == Subgroup) {                                                   \
       return sg_x;                                                             \
     }                                                                          \
     __local TYPE *scratch = __CLC_APPEND(__clc__get_group_scratch_, TYPE)();   \
-    uint sg_id = __spirv_SubgroupId();                                         \
-    uint num_sg = __spirv_NumSubgroups();                                      \
-    uint sg_lid = __spirv_SubgroupLocalInvocationId();                         \
-    uint sg_size = __spirv_SubgroupSize();                                     \
+    uint sg_id = __spirv_BuiltInSubgroupId();                                  \
+    uint num_sg = __spirv_BuiltInNumSubgroups();                               \
+    uint sg_lid = __spirv_BuiltInSubgroupLocalInvocationId();                  \
+    uint sg_size = __spirv_BuiltInSubgroupSize();                              \
     /* Share carry values across sub-groups */                                 \
     if (sg_lid == sg_size - 1) {                                               \
       scratch[sg_id] = carry;                                                  \
@@ -329,15 +330,15 @@ __CLC_GROUP_COLLECTIVE(LogicalAndKHR, __CLC_LOGICAL_AND, bool, true)
 #undef __CLC_MUL
 
 long __clc__2d_to_linear_local_id(ulong2 id) {
-  size_t size_x = __spirv_WorkgroupSize_x();
-  size_t size_y = __spirv_WorkgroupSize_y();
+  size_t size_x = __spirv_BuiltInWorkgroupSize(0);
+  size_t size_y = __spirv_BuiltInWorkgroupSize(1);
   return (id.y * size_x + id.x);
 }
 
 long __clc__3d_to_linear_local_id(ulong3 id) {
-  size_t size_x = __spirv_WorkgroupSize_x();
-  size_t size_y = __spirv_WorkgroupSize_y();
-  size_t size_z = __spirv_WorkgroupSize_z();
+  size_t size_x = __spirv_BuiltInWorkgroupSize(0);
+  size_t size_y = __spirv_BuiltInWorkgroupSize(1);
+  size_t size_z = __spirv_BuiltInWorkgroupSize(2);
   return (id.z * size_y * size_x + id.y * size_x + id.x);
 }
 
@@ -347,7 +348,7 @@ long __clc__3d_to_linear_local_id(ulong3 id) {
     if (scope == Subgroup) {                                                   \
       return __spirv_SubgroupShuffleINTEL(x, local_id);                        \
     }                                                                          \
-    bool source = (__spirv_LocalInvocationIndex() == local_id);                \
+    bool source = (__spirv_BuiltInLocalInvocationIndex() == local_id);         \
     __local TYPE *scratch = __CLC_APPEND(__clc__get_group_scratch_, TYPE)();   \
     if (source) {                                                              \
       *scratch = x;                                                            \

libclc/libspirv/lib/amdgcn-amdhsa/misc/sub_group_shuffle.cl

@@ -8,8 +8,8 @@
 
 #include <libspirv/spirv.h>
 
-#define SELF __spirv_SubgroupLocalInvocationId();
-#define SUBGROUP_SIZE __spirv_SubgroupMaxSize()
+#define SELF __spirv_BuiltInSubgroupLocalInvocationId();
+#define SUBGROUP_SIZE __spirv_BuiltInSubgroupMaxSize()
 
 // Shuffle
 _CLC_OVERLOAD _CLC_DEF int

libclc/libspirv/lib/amdgcn-amdhsa/workitem/get_global_offset.ll

@@ -9,28 +9,35 @@
 ; Function Attrs: nounwind readnone speculatable
 declare i32 addrspace(5)* @llvm.amdgcn.implicit.offset()
 
-define hidden i64 @_Z22__spirv_GlobalOffset_xv() nounwind alwaysinline {
+define hidden i64 @_Z27__spirv_BuiltInGlobalOffseti(i32 %dim) nounwind alwaysinline {
 entry:
+  switch i32 %dim, label %return [
+    i32 0, label %sw.bb
+    i32 1, label %sw.bb1
+    i32 2, label %sw.bb3
+  ]
+
+sw.bb:                                            ; preds = %entry
   %0 = tail call i32 addrspace(5)* @llvm.amdgcn.implicit.offset()
   %1 = load i32, i32 addrspace(5)* %0, align 4
   %zext = zext i32 %1 to i64
-  ret i64 %zext
-}
+  br label %return
 
-define hidden i64 @_Z22__spirv_GlobalOffset_yv() nounwind alwaysinline {
-entry:
-  %0 = tail call i32 addrspace(5)* @llvm.amdgcn.implicit.offset()
-  %arrayidx = getelementptr inbounds i32, i32 addrspace(5)* %0, i64 1
-  %1 = load i32, i32 addrspace(5)* %arrayidx, align 4
-  %zext = zext i32 %1 to i64
-  ret i64 %zext
-}
+sw.bb1:                                           ; preds = %entry
+  %2 = tail call i32 addrspace(5)* @llvm.amdgcn.implicit.offset()
+  %arrayidx = getelementptr inbounds i32, i32 addrspace(5)* %2, i64 1
+  %3 = load i32, i32 addrspace(5)* %arrayidx, align 4
+  %zext2 = zext i32 %3 to i64
+  br label %return
 
-define hidden i64 @_Z22__spirv_GlobalOffset_zv() nounwind alwaysinline {
-entry:
-  %0 = tail call i32 addrspace(5)* @llvm.amdgcn.implicit.offset()
-  %arrayidx = getelementptr inbounds i32, i32 addrspace(5)* %0, i64 2
-  %1 = load i32, i32 addrspace(5)* %arrayidx, align 4
-  %zext = zext i32 %1 to i64
-  ret i64 %zext
+sw.bb3:                                           ; preds = %entry
+  %4 = tail call i32 addrspace(5)* @llvm.amdgcn.implicit.offset()
+  %arrayidx2 = getelementptr inbounds i32, i32 addrspace(5)* %4, i64 2
+  %5 = load i32, i32 addrspace(5)* %arrayidx2, align 4
+  %zext3 = zext i32 %5 to i64
+  br label %return
+
+return:                                           ; preds = %entry, %sw.bb3, %sw.bb1, %sw.bb
+  %retval = phi i64 [ %zext, %sw.bb ], [ %zext2, %sw.bb1 ], [ %zext3, %sw.bb3 ], [ 0, %entry ]
+  ret i64 %retval
 }

libclc/libspirv/lib/amdgcn-amdhsa/workitem/get_global_size.cl

@@ -24,17 +24,16 @@ CONST_AS uchar *
 __clc_amdgcn_dispatch_ptr(void) __asm("llvm.amdgcn.dispatch.ptr");
 #endif
 
-_CLC_DEF _CLC_OVERLOAD size_t __spirv_GlobalSize_x() {
-    CONST_AS uint * ptr = (CONST_AS uint *) __dispatch_ptr();
+_CLC_DEF _CLC_OVERLOAD size_t __spirv_BuiltInGlobalSize(int dim) {
+  CONST_AS uint *ptr = (CONST_AS uint *)__dispatch_ptr();
+  switch (dim) {
+  case 0:
     return ptr[3];
-}
-
-_CLC_DEF _CLC_OVERLOAD size_t __spirv_GlobalSize_y() {
-    CONST_AS uint * ptr = (CONST_AS uint *) __dispatch_ptr();
+  case 1:
     return ptr[4];
-}
-
-_CLC_DEF _CLC_OVERLOAD size_t __spirv_GlobalSize_z() {
-    CONST_AS uint * ptr = (CONST_AS uint *) __dispatch_ptr();
+  case 2:
     return ptr[5];
+  default:
+    return 1;
+  }
 }

libclc/libspirv/lib/amdgcn-amdhsa/workitem/get_local_size.cl

@@ -26,17 +26,16 @@ __clc_amdgcn_dispatch_ptr(void) __asm("llvm.amdgcn.dispatch.ptr");
 
 // Mimic `EmitAMDGPUWorkGroupSize` in `clang/lib/CodeGen/CGBuiltin.cpp`.
 
-_CLC_DEF _CLC_OVERLOAD size_t __spirv_WorkgroupSize_x() {
-    CONST_AS ushort * ptr = (CONST_AS ushort *) __dispatch_ptr();
+_CLC_DEF _CLC_OVERLOAD size_t __spirv_BuiltInWorkgroupSize(int dim) {
+  CONST_AS ushort *ptr = (CONST_AS ushort *)__dispatch_ptr();
+  switch (dim) {
+  case 0:
     return ptr[2];
-}
-
-_CLC_DEF _CLC_OVERLOAD size_t __spirv_WorkgroupSize_y() {
-    CONST_AS ushort * ptr = (CONST_AS ushort *) __dispatch_ptr();
+  case 1:
     return ptr[3];
-}
-
-_CLC_DEF _CLC_OVERLOAD size_t __spirv_WorkgroupSize_z() {
-    CONST_AS ushort * ptr = (CONST_AS ushort *) __dispatch_ptr();
+  case 2:
     return ptr[4];
+  default:
+    return 1;
+  }
 }

libclc/libspirv/lib/amdgcn-amdhsa/workitem/get_max_sub_group_size.cl

@@ -14,7 +14,7 @@
 // 32.
 extern constant unsigned char __oclc_wavefrontsize64;
 
-_CLC_DEF _CLC_OVERLOAD uint __spirv_SubgroupMaxSize() {
+_CLC_DEF _CLC_OVERLOAD uint __spirv_BuiltInSubgroupMaxSize() {
   if (__oclc_wavefrontsize64 == 1) {
     return 64;
   }

libclc/libspirv/lib/amdgcn-amdhsa/workitem/get_num_groups.cl

@@ -8,32 +8,11 @@
 
 #include <libspirv/spirv.h>
 
-_CLC_DEF _CLC_OVERLOAD size_t __spirv_NumWorkgroups_x() {
-  size_t global_size = __spirv_GlobalSize_x();
-  size_t local_size = __spirv_WorkgroupSize_x();
+_CLC_DEF _CLC_OVERLOAD size_t __spirv_BuiltInNumWorkgroups(int dim) {
+  size_t global_size = __spirv_BuiltInGlobalSize(dim);
+  size_t local_size = __spirv_BuiltInWorkgroupSize(dim);
   size_t num_groups = global_size / local_size;
-  if (global_size % local_size != 0) {
+  if (global_size % local_size != 0)
     num_groups++;
-  }
-  return num_groups;
-}
-
-_CLC_DEF _CLC_OVERLOAD size_t __spirv_NumWorkgroups_y() {
-  size_t global_size = __spirv_GlobalSize_y();
-  size_t local_size = __spirv_WorkgroupSize_y();
-  size_t num_groups = global_size / local_size;
-  if (global_size % local_size != 0) {
-    num_groups++;
-  }
-  return num_groups;
-}
-
-_CLC_DEF _CLC_OVERLOAD size_t __spirv_NumWorkgroups_z() {
-  size_t global_size = __spirv_GlobalSize_z();
-  size_t local_size = __spirv_WorkgroupSize_z();
-  size_t num_groups = global_size / local_size;
-  if (global_size % local_size != 0) {
-    num_groups++;
-  }
   return num_groups;
 }

libclc/libspirv/lib/amdgcn-amdhsa/workitem/get_sub_group_id.cl

@@ -8,12 +8,12 @@
 
 #include <libspirv/spirv.h>
 
-_CLC_DEF _CLC_OVERLOAD uint __spirv_SubgroupId() {
-  size_t id_x = __spirv_LocalInvocationId_x();
-  size_t id_y = __spirv_LocalInvocationId_y();
-  size_t id_z = __spirv_LocalInvocationId_z();
-  size_t size_x = __spirv_WorkgroupSize_x();
-  size_t size_y = __spirv_WorkgroupSize_y();
-  uint sg_size = __spirv_SubgroupMaxSize();
+_CLC_DEF _CLC_OVERLOAD uint __spirv_BuiltInSubgroupId() {
+  size_t id_x = __spirv_BuiltInLocalInvocationId(0);
+  size_t id_y = __spirv_BuiltInLocalInvocationId(1);
+  size_t id_z = __spirv_BuiltInLocalInvocationId(2);
+  size_t size_x = __spirv_BuiltInWorkgroupSize(0);
+  size_t size_y = __spirv_BuiltInWorkgroupSize(1);
+  uint sg_size = __spirv_BuiltInSubgroupMaxSize();
   return (id_z * size_y * size_x + id_y * size_x + id_x) / sg_size;
 }