[OpenCL] Add support for missing sub_group functions.

[joey]

Summary

This adds get_kernel_max_sub_group_size_for_ndrange and get_kernel_sub_group_count_for_ndrange.

Note this also changes err_opencl_requires_extension to print the name of the function that the diagnostic is warning about.

[bader]

Please, split this patch into two parts:

1. Improve diagnostics on extension enabling.
2. Add missing sub_group_* built-in functions.

Sema/SemaChecking.cpp

1091–1092

I suggest leaving SemaBuiltinReserveRWPipe as is (i.e. two parameter) and modify the check for sub_group_* functions only:

if (checkOpenCLSubgroupExt(S, TheCall) || SemaBuiltinReserveRWPipe(*this, TheCall))
  return ExprError();

It think it's more readable than looking at SemaBuiltinReserveRWPipe declaration or leaving the comment like this:

if (SemaBuiltinReserveRWPipe(*this, TheCall, false /*isSubgroup*/))

Please, apply the same approach to SemaBuiltinCommitRWPipe.

In addition to that, it makes sense to set the TheCall type inside the SemaBuiltinReserveRWPipe to avoid code duplication.

SemaOpenCL/cl20-device-side-enqueue.cl

219

Please, add a test case(s) on invalid block parameters to cover the checks you added for new sub_group_ built-ins..

SemaOpenCL/sub-group-bifs.cl

1 ↗

(On Diff #101576)

I don't think it makes sense to add this test. This test look like a duplicate of test cases added to SemaOpenCL/cl20-device-side-enqueue.cl.

clang/Basic/Builtins.def

1402–1403

This built-in functions should return unsigned integers: "i." -> "Ui.".
Please, fix get_kernel_work_group_size and get_kernel_preferred_work_group_size_multiple too.

clang/Basic/DiagnosticSemaKinds.td

8423–8424 ↗

(On Diff #101576)

Since, this message is not specific to enqueue_kernel anymore, I suggest either rename it or better re-use existing diagnostics if possible. I think there are already message to report argument mismatch + probably additional note diagnostics that hints correct argument type.

[joey]

Updated all the comments you made and rebased.

Sorry for the long delay.

[bader]

Thanks!
Overall the patch looks good, but I would suggest splitting it into three commits (as they seems to be independent):

1. [OpenCL] Check that cl_khr_subgroups pragma is enabled if respective extension is used.
2. [OpenCL] Add support for missing sub_group functions.
3. [OpenCL] Fix return type for reserve pipe built-ins.

Please, add a regression test for the part #3.

You might also review this patch with @Anastasia (OpenCL code owner).

Sema/SemaChecking.cpp

685–689

This change is not covered with regression tests.

[Anastasia]

LGTM! Thanks!

[joey]

I committed all the parts separately: r309567 (with r309571 to fix a test), r309678 and r310477.

CodeGen/CGBuiltin.cpp

 
   case Builtin::BI__builtin_coro_size: {
     auto & Context = getContext();
     auto SizeTy = Context.getSizeType();
     auto T = Builder.getIntNTy(Context.getTypeSize(SizeTy));
     Value *F = CGM.getIntrinsic(Intrinsic::coro_size, T);
     return RValue::get(Builder.CreateCall(F));
   }
-
   case Builtin::BI__builtin_coro_id:
     return EmitCoroutineIntrinsic(E, Intrinsic::coro_id);
   case Builtin::BI__builtin_coro_promise:
     return EmitCoroutineIntrinsic(E, Intrinsic::coro_promise);
   case Builtin::BI__builtin_coro_resume:
     return EmitCoroutineIntrinsic(E, Intrinsic::coro_resume);
   case Builtin::BI__builtin_coro_frame:
     return EmitCoroutineIntrinsic(E, Intrinsic::coro_frame);
case Builtin::BIget_kernel_preferred_work_group_size_multiple: {
     Value *Arg = EmitScalarExpr(E->getArg(0));
     Arg = Builder.CreatePointerCast(Arg, GenericVoidPtrTy);
     return RValue::get(Builder.CreateCall(
         CGM.CreateRuntimeFunction(
             llvm::FunctionType::get(IntTy, GenericVoidPtrTy, false),
             "__get_kernel_preferred_work_group_multiple_impl"),
         Arg));
   }
+  case Builtin::BIget_kernel_max_sub_group_size_for_ndrange:
+  case Builtin::BIget_kernel_sub_group_count_for_ndrange: {
+    llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy(
+        getContext().getTargetAddressSpace(LangAS::opencl_generic));
+    LValue NDRangeL = EmitAggExprToLValue(E->getArg(0));
+    llvm::Value *NDRange = NDRangeL.getAddress().getPointer();
+    Value *Block = EmitScalarExpr(E->getArg(1));
+    Block = Builder.CreatePointerCast(Block, GenericVoidPtrTy);
+    const char *Name =
+        BuiltinID == Builtin::BIget_kernel_max_sub_group_size_for_ndrange
+            ? "__get_kernel_max_sub_group_size_for_ndrange_impl"
+            : "__get_kernel_sub_group_count_for_ndrange_impl";
+    return RValue::get(Builder.CreateCall(
+        CGM.CreateRuntimeFunction(
+            llvm::FunctionType::get(
+                IntTy, {NDRange->getType(), GenericVoidPtrTy}, false),
+            Name),
+        {NDRange, Block}));
+  }
   case Builtin::BIprintf:
     if (getTarget().getTriple().isNVPTX())
       return EmitNVPTXDevicePrintfCallExpr(E, ReturnValue);
     break;
   case Builtin::BI__builtin_canonicalize:
   case Builtin::BI__builtin_canonicalizef:
   case Builtin::BI__builtin_canonicalizel:
     return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::canonicalize));

CodeGenOpenCL/cl20-device-side-enqueue.cl

 // RUN: %clang_cc1 %s -cl-std=CL2.0 -ffake-address-space-map -O0 -emit-llvm -o - -triple "spir-unknown-unknown" | FileCheck %s --check-prefix=COMMON --check-prefix=B32
 // RUN: %clang_cc1 %s -cl-std=CL2.0 -ffake-address-space-map -O0 -emit-llvm -o - -triple "spir64-unknown-unknown" | FileCheck %s --check-prefix=COMMON --check-prefix=B64
 
+#pragma OPENCL EXTENSION cl_khr_subgroups : enable
+
 typedef void (^bl_t)(local void *);
 typedef struct {int a;} ndrange_t;
 
 // N.B. The check here only exists to set BL_GLOBAL
 // COMMON: @block_G =  addrspace(1) constant void (i8 addrspace(3)*) addrspace(4)* addrspacecast (void (i8 addrspace(3)*) addrspace(1)* bitcast ({ i8**, i32, i32, i8*, %struct.__block_descriptor addrspace(2)* } addrspace(1)* [[BL_GLOBAL:@__block_literal_global(\.[0-9]+)?]] to void (i8 addrspace(3)*) addrspace(1)*) to void (i8 addrspace(3)*) addrspace(4)*)
 const bl_t block_G = (bl_t) ^ (local void *a) {};
 
 kernel void device_side_enqueue(global int *a, global int *b, int i) {
kernel void device_side_enqueue(global int *a, global int *b, int i) {
   // COMMON: call i32 @__get_kernel_work_group_size_impl(i8 addrspace(4)* addrspacecast (i8 addrspace(1)* bitcast ({ i8**, i32, i32, i8*, %struct.__block_descriptor addrspace(2)* } addrspace(1)* [[BL_A]] to i8 addrspace(1)*) to i8 addrspace(4)*))
   unsigned size = get_kernel_work_group_size(block_A);
   // COMMON: call i32 @__get_kernel_work_group_size_impl(i8 addrspace(4)* addrspacecast (i8 addrspace(1)* bitcast ({ i8**, i32, i32, i8*, %struct.__block_descriptor addrspace(2)* } addrspace(1)* [[BL_B]] to i8 addrspace(1)*) to i8 addrspace(4)*))
   size = get_kernel_work_group_size(block_B);
   // COMMON: call i32 @__get_kernel_preferred_work_group_multiple_impl(i8 addrspace(4)* addrspacecast (i8 addrspace(1)* bitcast ({ i8**, i32, i32, i8*, %struct.__block_descriptor addrspace(2)* } addrspace(1)* [[BL_A]] to i8 addrspace(1)*) to i8 addrspace(4)*))
   size = get_kernel_preferred_work_group_size_multiple(block_A);
   // COMMON: call i32 @__get_kernel_preferred_work_group_multiple_impl(i8 addrspace(4)* addrspacecast (i8 addrspace(1)* bitcast ({ i8**, i32, i32, i8*, %struct.__block_descriptor addrspace(2)* } addrspace(1)* [[BL_GLOBAL]] to i8 addrspace(1)*) to i8 addrspace(4)*))
   size = get_kernel_preferred_work_group_size_multiple(block_G);
+
+  // COMMON: call i32 @__get_kernel_max_sub_group_size_for_ndrange_impl(%struct.ndrange_t* {{.*}}, i8 addrspace(4)* addrspacecast (i8 addrspace(1)* bitcast ({ i8**, i32, i32, i8*, %struct.__block_descriptor addrspace(2)* } addrspace(1)* {{.*}} to i8 addrspace(1)*) to i8 addrspace(4)*))
+  size = get_kernel_max_sub_group_size_for_ndrange(ndrange, ^(){});
+  // COMMON: call i32 @__get_kernel_sub_group_count_for_ndrange_impl(%struct.ndrange_t* {{.*}}, i8 addrspace(4)* addrspacecast (i8 addrspace(1)* bitcast ({ i8**, i32, i32, i8*, %struct.__block_descriptor addrspace(2)* } addrspace(1)* {{.*}} to i8 addrspace(1)*) to i8 addrspace(4)*))
+  size = get_kernel_sub_group_count_for_ndrange(ndrange, ^(){});
 }

CodeGenOpenCL/pipe_builtin.cl

 // RUN: %clang_cc1 -emit-llvm -O0 -cl-std=CL2.0 -o - %s | FileCheck %s
 
 // CHECK: %opencl.pipe_t = type opaque
 // CHECK: %opencl.reserve_id_t = type opaque
 
+#pragma OPENCL EXTENSION cl_khr_subgroups : enable
+
 void test1(read_only pipe int p, global int *ptr) {
   // CHECK: call i32 @__read_pipe_2(%opencl.pipe_t* %{{.*}}, i8* %{{.*}}, i32 4, i32 4)
   read_pipe(p, ptr);
   // CHECK: call %opencl.reserve_id_t* @__reserve_read_pipe(%opencl.pipe_t* %{{.*}}, i32 {{.*}}, i32 4, i32 4)
   reserve_id_t rid = reserve_read_pipe(p, 2);
   // CHECK: call i32 @__read_pipe_4(%opencl.pipe_t* %{{.*}}, %opencl.reserve_id_t* %{{.*}}, i32 {{.*}}, i8* %{{.*}}, i32 4, i32 4)
   read_pipe(p, rid, 2, ptr);
   // CHECK: call void @__commit_read_pipe(%opencl.pipe_t* %{{.*}}, %opencl.reserve_id_t* %{{.*}}, i32 4, i32 4)

Sema/SemaChecking.cpp

if (!(*I)->isPointerType() || !(*I)->getPointeeType()->isVoidType() ||
              diag::err_opencl_enqueue_kernel_blocks_non_local_void_args);
       IllegalParams = true;
     }
   }
 
   return IllegalParams;
 }
 
+static bool checkOpenCLSubgroupExt(Sema &S, CallExpr *Call) {
+  if (!S.getOpenCLOptions().isEnabled("cl_khr_subgroups")) {
+    S.Diag(Call->getLocStart(), diag::err_opencl_requires_extension)
+          << 1 << Call->getDirectCallee() << "cl_khr_subgroups";
+    return true;
+  }
+  return false;
+}
+
+static bool SemaOpenCLBuiltinNDRangeAndBlock(Sema &S, CallExpr *TheCall) {
+  if (checkArgCount(S, TheCall, 2))
+    return true;
+
+  if (checkOpenCLSubgroupExt(S, TheCall))
+    return true;
+
+  // First argument is an ndrange_t type.
+  Expr *NDRangeArg = TheCall->getArg(0);
+  if (NDRangeArg->getType().getAsString() != "ndrange_t") {
+    S.Diag(NDRangeArg->getLocStart(),
+           diag::err_opencl_builtin_expected_type)
+        << TheCall->getDirectCallee() << "'ndrange_t'";
+    return true;
+  }
+
+  Expr *BlockArg = TheCall->getArg(1);
+  if (!isBlockPointer(BlockArg)) {
+    S.Diag(BlockArg->getLocStart(),
+           diag::err_opencl_builtin_expected_type)
+        << TheCall->getDirectCallee() << "block";
+    return true;
+  }
+  return checkOpenCLBlockArgs(S, BlockArg);
+}
+
 /// OpenCL C v2.0, s6.13.17.6 - Check the argument to the
 /// get_kernel_work_group_size
 /// and get_kernel_preferred_work_group_size_multiple builtin functions.
 static bool SemaOpenCLBuiltinKernelWorkGroupSize(Sema &S, CallExpr *TheCall) {
   if (checkArgCount(S, TheCall, 1))
     return true;
 
   Expr *BlockArg = TheCall->getArg(0);
static bool SemaBuiltinReserveRWPipe(Sema &S, CallExpr *Call) {
   if (!Call->getArg(1)->getType()->isIntegerType() &&
       !Call->getArg(1)->getType()->isUnsignedIntegerType()) {
     S.Diag(Call->getLocStart(), diag::err_opencl_builtin_pipe_invalid_arg)
         << Call->getDirectCallee() << S.Context.UnsignedIntTy
         << Call->getArg(1)->getType() << Call->getArg(1)->getSourceRange();
     return true;
   }
 
+  // Since return type of reserve_read/write_pipe built-in function is
+  // reserve_id_t, which is not defined in the builtin def file , we used int
+  // as return type and need to override the return type of these functions.
+  Call->setType(S.Context.OCLReserveIDTy);
+
   return false;
 }
 
 // \brief Performs a semantic analysis on {work_group_/sub_group_
 //        /_}commit_{read/write}_pipe
 // \param S Reference to the semantic analyzer.
 // \param Call The call to the builtin function to be analyzed.
 // \return True if a semantic error was found, false otherwise.
case Builtin::BIwrite_pipe:
     if (SemaBuiltinRWPipe(*this, TheCall))
       return ExprError();
     TheCall->setType(Context.IntTy);
     break;
   case Builtin::BIreserve_read_pipe:
   case Builtin::BIreserve_write_pipe:
   case Builtin::BIwork_group_reserve_read_pipe:
   case Builtin::BIwork_group_reserve_write_pipe:
+    if (SemaBuiltinReserveRWPipe(*this, TheCall))
+      return ExprError();
+    break;
   case Builtin::BIsub_group_reserve_read_pipe:
   case Builtin::BIsub_group_reserve_write_pipe:
-    if (SemaBuiltinReserveRWPipe(*this, TheCall))
+    if (checkOpenCLSubgroupExt(*this, TheCall) ||
+        SemaBuiltinReserveRWPipe(*this, TheCall))
       return ExprError();
-    // Since return type of reserve_read/write_pipe built-in function is
-    // reserve_id_t, which is not defined in the builtin def file , we used int
-    // as return type and need to override the return type of these functions.
-    TheCall->setType(Context.OCLReserveIDTy);
     break;
   case Builtin::BIcommit_read_pipe:
   case Builtin::BIcommit_write_pipe:
   case Builtin::BIwork_group_commit_read_pipe:
   case Builtin::BIwork_group_commit_write_pipe:
+    if (SemaBuiltinCommitRWPipe(*this, TheCall))
+      return ExprError();
+    break;
   case Builtin::BIsub_group_commit_read_pipe:
   case Builtin::BIsub_group_commit_write_pipe:
-    if (SemaBuiltinCommitRWPipe(*this, TheCall))
+    if (checkOpenCLSubgroupExt(*this, TheCall) ||
+        SemaBuiltinCommitRWPipe(*this, TheCall))
       return ExprError();
     break;
   case Builtin::BIget_pipe_num_packets:
   case Builtin::BIget_pipe_max_packets:
     if (SemaBuiltinPipePackets(*this, TheCall))
       return ExprError();
     TheCall->setType(Context.UnsignedIntTy);
     break;
   case Builtin::BIto_global:
   case Builtin::BIto_local:
   case Builtin::BIto_private:
     if (SemaOpenCLBuiltinToAddr(*this, BuiltinID, TheCall))
       return ExprError();
     break;
   // OpenCL v2.0, s6.13.17 - Enqueue kernel functions.
   case Builtin::BIenqueue_kernel:
     if (SemaOpenCLBuiltinEnqueueKernel(*this, TheCall))
       return ExprError();
     break;
   case Builtin::BIget_kernel_work_group_size:
   case Builtin::BIget_kernel_preferred_work_group_size_multiple:
     if (SemaOpenCLBuiltinKernelWorkGroupSize(*this, TheCall))
       return ExprError();
     break;
+    break;
+  case Builtin::BIget_kernel_max_sub_group_size_for_ndrange:
+  case Builtin::BIget_kernel_sub_group_count_for_ndrange:
+    if (SemaOpenCLBuiltinNDRangeAndBlock(*this, TheCall))
+      return ExprError();
+    break;
   case Builtin::BI__builtin_os_log_format:
   case Builtin::BI__builtin_os_log_format_buffer_size:
     if (SemaBuiltinOSLogFormat(TheCall)) {
       return ExprError();
     }
     break;
   }
 

SemaOpenCL/cl20-device-side-enqueue.cl

size = get_kernel_preferred_work_group_size_multiple(^(int a) {  // expected-error {{blocks used in enqueue_kernel call are expected to have parameters of type 'local void*'}}
     return;
   });
   size = get_kernel_preferred_work_group_size_multiple(block_B);   // expected-error {{blocks used in enqueue_kernel call are expected to have parameters of type 'local void*'}}
   size = get_kernel_preferred_work_group_size_multiple(block_D);   // expected-error {{blocks used in enqueue_kernel call are expected to have parameters of type 'local void*'}}
   size = get_kernel_preferred_work_group_size_multiple();          // expected-error {{too few arguments to function call, expected 1, have 0}}
   size = get_kernel_preferred_work_group_size_multiple(1);         // expected-error{{expected block argument}}
   size = get_kernel_preferred_work_group_size_multiple(block_A, 1); // expected-error{{too many arguments to function call, expected 1, have 2}}
 }
+
+#pragma OPENCL EXTENSION cl_khr_subgroups : enable
+
+kernel void foo(global int *buf)
+{
+  ndrange_t n;
+  buf[0] = get_kernel_max_sub_group_size_for_ndrange(n, ^(){});
+  buf[0] = get_kernel_max_sub_group_size_for_ndrange(0, ^(){}); // expected-error{{illegal call to 'get_kernel_max_sub_group_size_for_ndrange', expected 'ndrange_t' argument type}}
+  buf[0] = get_kernel_max_sub_group_size_for_ndrange(n, 1); // expected-error{{illegal call to 'get_kernel_max_sub_group_size_for_ndrange', expected block argument type}}
+}
+
+kernel void bar(global int *buf)
+{
+  ndrange_t n;
+  buf[0] = get_kernel_sub_group_count_for_ndrange(n, ^(){});
+  buf[0] = get_kernel_sub_group_count_for_ndrange(0, ^(){}); // expected-error{{illegal call to 'get_kernel_sub_group_count_for_ndrange', expected 'ndrange_t' argument type}}
+  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
+
+kernel void foo1(global 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 extension to be enabled}}
+}
+
+kernel void bar1(global 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 extension to be enabled}}
+}

SemaOpenCL/invalid-pipe-builtin-cl2.0.cl

 // RUN: %clang_cc1 %s -verify -pedantic -fsyntax-only -cl-std=CL2.0
 
+#pragma OPENCL EXTENSION cl_khr_subgroups : enable
+
 void test1(read_only pipe int p, global int* ptr){
   int tmp;
   reserve_id_t rid;
 
   // read/write_pipe
   read_pipe(p, &tmp);
   read_pipe(p, ptr);
   read_pipe(tmp, p);    // expected-error {{first argument to 'read_pipe' must be a pipe type}}

clang/Basic/Builtins.def

 LANGBUILTIN(work_group_commit_write_pipe, "v.", "tn", OCLC20_LANG)
 
 LANGBUILTIN(get_pipe_num_packets, "Ui.", "tn", OCLC20_LANG)
 LANGBUILTIN(get_pipe_max_packets, "Ui.", "tn", OCLC20_LANG)
 
 // OpenCL v2.0 s6.13.17 - Enqueue kernel functions.
 // Custom builtin check allows to perform special check of passed block arguments.
 LANGBUILTIN(enqueue_kernel, "i.", "tn", OCLC20_LANG)
-LANGBUILTIN(get_kernel_work_group_size, "i.", "tn", OCLC20_LANG)
-LANGBUILTIN(get_kernel_preferred_work_group_size_multiple, "i.", "tn", OCLC20_LANG)
+LANGBUILTIN(get_kernel_work_group_size, "Ui.", "tn", OCLC20_LANG)
+LANGBUILTIN(get_kernel_preferred_work_group_size_multiple, "Ui.", "tn", OCLC20_LANG)
+LANGBUILTIN(get_kernel_max_sub_group_size_for_ndrange, "Ui.", "tn", OCLC20_LANG)
+LANGBUILTIN(get_kernel_sub_group_count_for_ndrange, "Ui.", "tn", OCLC20_LANG)
 
 // OpenCL v2.0 s6.13.9 - Address space qualifier functions.
 LANGBUILTIN(to_global, "v*v*", "tn", OCLC20_LANG)
 LANGBUILTIN(to_local, "v*v*", "tn", OCLC20_LANG)
 LANGBUILTIN(to_private, "v*v*", "tn", OCLC20_LANG)
 
 // Builtins for os_log/os_trace
 BUILTIN(__builtin_os_log_format_buffer_size, "zcC*.", "p:0:nut")