[OpenCL] Adjust diagnostic for subgroup support.

[azabaznov]

Summary

OpenCL C 3.0 __opencl_c_subgroups feature is slightly different
then other equivalent features and extensions (fp64 and 3d image writes):
OpenCL C 3.0 device can support the extension but not the feature.
cl_khr_subgroups requires subgroup independent forward progress.

This patch adjusts the check which is used when translating language
builtins to check either the extension or feature is supported.

[Anastasia]

Btw can you point us to the spec reference please, I can't seem to find anything related to this: https://www.khronos.org/registry/OpenCL/specs/3.0-unified/html/OpenCL_C.html#subgroup-functions.

It would be nice to add the reference in the code comment too.

clang/lib/Sema/SemaChecking.cpp

1055

Ideally we should reduce the number of separate diagnostics, so could we just print something like:
"cl_khr_subgroups extension or __opencl_c_subgroups OpenCL C 3.0 feature"
when using diag::err_opencl_requires_extension?

[azabaznov]

It seems hard to find a direct mention in the spec, but in the API spec:

CL_DEVICE_SUB_GROUP_INDEPENDENT_FORWARD_PROGRESS: Is CL_TRUE if this device supports independent forward progress of sub-groups, CL_FALSE otherwise. This query must return CL_TRUE for devices that support the cl_khr_subgroups extension, and must return CL_FALSE for devices that do not support subgroups.

[Anastasia]

In D118999#3296899, @azabaznov wrote:

It seems hard to find a direct mention in the spec, but in the API spec:

CL_DEVICE_SUB_GROUP_INDEPENDENT_FORWARD_PROGRESS: Is CL_TRUE if this device supports independent forward progress of sub-groups, CL_FALSE otherwise. This query must return CL_TRUE for devices that support the cl_khr_subgroups extension, and must return CL_FALSE for devices that do not support subgroups.

Ok, I would agree that the wording doesn't seem identical since the extension explicitly details forward progress behavior but the feature doesn't:
https://www.khronos.org/registry/OpenCL/specs/3.0-unified/html/OpenCL_Ext.html#cl_khr_subgroups

But I think it would make sense to add wording of it in the feature spec too otherwise it appears unspecified?

Anyway, at present, since they are not identical having separate handling of these teo seems reasonable.

[azabaznov]

Use existing diagnostics; add the comment that device can support extension but not the feature. Will follow up if we need explicit mention in the language spec.

[Anastasia]

LGTM! Thanks

clang/lib/Sema/SemaChecking.cpp

if (!(*I)->isPointerType() || !(*I)->getPointeeType()->isVoidType() ||
       IllegalParams = true;
     }
   }
 
   return IllegalParams;
 }
 
 static bool checkOpenCLSubgroupExt(Sema &S, CallExpr *Call) {
-  if (!S.getOpenCLOptions().isSupported("cl_khr_subgroups", S.getLangOpts())) {
+  // OpenCL device can support extension but not the feature as extension
+  // requires subgroup independent forward progress, but subgroup independent
+  // forward progress is optional in OpenCL C 3.0 __opencl_c_subgroups feature.
+  if (!S.getOpenCLOptions().isSupported("cl_khr_subgroups", S.getLangOpts()) &&
+      !S.getOpenCLOptions().isSupported("__opencl_c_subgroups",
+                                        S.getLangOpts())) {
     S.Diag(Call->getBeginLoc(), diag::err_opencl_requires_extension)
-        << 1 << Call->getDirectCallee() << "cl_khr_subgroups";
+        << 1 << Call->getDirectCallee()
+        << "cl_khr_subgroups or __opencl_c_subgroups";
     return true;
   }
   return false;
 }
 
 static bool SemaOpenCLBuiltinNDRangeAndBlock(Sema &S, CallExpr *TheCall) {
   if (checkArgCount(S, TheCall, 2))
     return true;
 
   if (checkOpenCLSubgroupExt(S, TheCall))

clang/test/SemaOpenCL/cl20-device-side-enqueue.cl

 // RUN: %clang_cc1 %s -cl-std=CL2.0 -triple "spir-unknown-unknown" -verify -pedantic -fsyntax-only -DB32 -DQUALS=
 // RUN: %clang_cc1 %s -cl-std=CL2.0 -triple "spir-unknown-unknown" -verify -pedantic -fsyntax-only -DB32 -DQUALS= -cl-ext=-cl_khr_subgroups
 // RUN: %clang_cc1 %s -cl-std=CL2.0 -triple "spir-unknown-unknown" -verify -pedantic -fsyntax-only -DB32 -DQUALS="const volatile"
 // RUN: %clang_cc1 %s -cl-std=CL2.0 -triple "spir64-unknown-unknown" -verify -pedantic -fsyntax-only -Wconversion -DWCONV -DQUALS=
 // RUN: %clang_cc1 %s -cl-std=CL2.0 -triple "spir64-unknown-unknown" -verify -pedantic -fsyntax-only -Wconversion -DWCONV -DQUALS="const volatile"
 // RUN: %clang_cc1 %s -cl-std=CL3.0 -triple "spir-unknown-unknown" -verify -pedantic -fsyntax-only -DB32 -DQUALS=
-// RUN: %clang_cc1 %s -cl-std=CL3.0 -triple "spir-unknown-unknown" -verify -pedantic -fsyntax-only -DB32 -DQUALS= -cl-ext=-cl_khr_subgroups
+// RUN: %clang_cc1 %s -cl-std=CL3.0 -triple "spir-unknown-unknown" -verify -pedantic -fsyntax-only -DB32 -DQUALS= -cl-ext=-cl_khr_subgroups,-__opencl_c_subgroups
 // RUN: %clang_cc1 %s -cl-std=CL3.0 -triple "spir-unknown-unknown" -verify -pedantic -fsyntax-only -DB32 -DQUALS="const volatile"
 // RUN: %clang_cc1 %s -cl-std=CL3.0 -triple "spir64-unknown-unknown" -verify -pedantic -fsyntax-only -Wconversion -DWCONV -DQUALS=
 // RUN: %clang_cc1 %s -cl-std=CL3.0 -triple "spir64-unknown-unknown" -verify -pedantic -fsyntax-only -Wconversion -DWCONV -DQUALS="const volatile"
 
 typedef struct {int a;} ndrange_t;
 // Diagnostic tests for different overloads of enqueue_kernel from Table 6.13.17.1 of OpenCL 2.0 Spec.
 kernel void enqueue_kernel_tests() {
   queue_t default_queue;
kernel void bar(global unsigned int *buf)
   buf[0] = get_kernel_sub_group_count_for_ndrange(n, 1); // expected-error{{illegal call to 'get_kernel_sub_group_count_for_ndrange', expected block argument type}}
 }
 
 #pragma OPENCL EXTENSION cl_khr_subgroups : disable
 #else
 kernel void foo1(global unsigned int *buf)
 {
   ndrange_t n;
-  buf[0] = get_kernel_max_sub_group_size_for_ndrange(n, ^(){}); // expected-error {{use of declaration 'get_kernel_max_sub_group_size_for_ndrange' requires cl_khr_subgroups support}}
+  buf[0] = get_kernel_max_sub_group_size_for_ndrange(n, ^(){}); // expected-error {{use of declaration 'get_kernel_max_sub_group_size_for_ndrange' requires cl_khr_subgroups or __opencl_c_subgroups support}}
 }
 
 kernel void bar1(global unsigned int *buf)
 {
   ndrange_t n;
-  buf[0] = get_kernel_sub_group_count_for_ndrange(n, ^(){}); // expected-error {{use of declaration 'get_kernel_sub_group_count_for_ndrange' requires cl_khr_subgroups support}}
+  buf[0] = get_kernel_sub_group_count_for_ndrange(n, ^(){}); // expected-error {{use of declaration 'get_kernel_sub_group_count_for_ndrange' requires cl_khr_subgroups or __opencl_c_subgroups support}}
 }
 #endif // ifdef cl_khr_subgroups