[SYCL] Add template parameter support for max_global_work_dim attribute

clang/include/clang/Basic/Attr.td

@@ -1293,7 +1293,7 @@ def SYCLIntelMaxWorkGroupSize : InheritableAttr {
 def SYCLIntelMaxGlobalWorkDim : InheritableAttr {
   let Spellings = [CXX11<"intelfpga","max_global_work_dim">,
                    CXX11<"intel","max_global_work_dim">];
-  let Args = [UnsignedArgument<"Number">];
+  let Args = [ExprArgument<"Value">];
   let LangOpts = [SYCLIsDevice, SYCLIsHost];
   let Subjects = SubjectList<[Function], ErrorDiag>;
   let Documentation = [SYCLIntelMaxGlobalWorkDimAttrDocs];

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -11134,8 +11134,6 @@ def err_sycl_function_attribute_mismatch : Error<
   "SYCL kernel without %0 attribute can't call a function with this attribute">;
 def err_sycl_x_y_z_arguments_must_be_one : Error<
   "%0 X-, Y- and Z- sizes must be 1 when %1 attribute is used with value 0">;
-def err_intel_attribute_argument_is_not_in_range: Error<
-   "The value of %0 attribute must be in range from 0 to 3">;
 def warn_boolean_attribute_argument_is_not_valid: Warning<
    "The value of %0 attribute should be 0 or 1. Adjusted to 1">,
    InGroup<AdjustedAttributes>;

clang/include/clang/Sema/Sema.h

@@ -12963,10 +12963,25 @@ void Sema::addIntelSYCLSingleArgFunctionAttr(Decl *D,
       return;
     }
     int32_t ArgInt = ArgVal->getSExtValue();
-    if (ArgInt <= 0) {
-      Diag(E->getExprLoc(), diag::err_attribute_requires_positive_integer)
-          << CI.getAttrName() << /*positive*/ 0;
-      return;
+    if (CI.getParsedKind() == ParsedAttr::AT_SYCLIntelNumSimdWorkItems ||
+        CI.getParsedKind() == ParsedAttr::AT_IntelReqdSubGroupSize) {
+      if (ArgInt <= 0) {
+        Diag(E->getExprLoc(), diag::err_attribute_requires_positive_integer)
+            << CI.getAttrName() << /*positive*/ 0;
+        return;
+      }
+    }
+    if (CI.getParsedKind() == ParsedAttr::AT_SYCLIntelMaxGlobalWorkDim) {
+      if (ArgInt < 0) {
+        Diag(E->getExprLoc(), diag::err_attribute_requires_positive_integer)
+            << CI.getAttrName() << /*non-negative*/ 1;
+        return;
+      }
+      if (ArgInt > 3) {
+        Diag(E->getBeginLoc(), diag::err_attribute_argument_out_of_range)
+            << CI.getAttrName() << 0 << 3 << E->getSourceRange();
+        return;
+      }
     }
   }
 

clang/lib/CodeGen/CodeGenFunction.cpp

@@ -682,8 +682,12 @@ void CodeGenFunction::EmitOpenCLKernelMetadata(const FunctionDecl *FD,
 
   if (const SYCLIntelMaxGlobalWorkDimAttr *A =
       FD->getAttr<SYCLIntelMaxGlobalWorkDimAttr>()) {
-    llvm::Metadata *AttrMDArgs[] = {
-        llvm::ConstantAsMetadata::get(Builder.getInt32(A->getNumber()))};
+    llvm::LLVMContext &Context = getLLVMContext();
+    Optional<llvm::APSInt> ArgVal =
+        A->getValue()->getIntegerConstantExpr(FD->getASTContext());
+    assert(ArgVal.hasValue() && "Not an integer constant expression");
+    llvm::Metadata *AttrMDArgs[] = {llvm::ConstantAsMetadata::get(
+        Builder.getInt32(ArgVal->getSExtValue()))};
     Fn->setMetadata("max_global_work_dim",
                     llvm::MDNode::get(Context, AttrMDArgs));
   }

clang/lib/Sema/SemaDeclAttr.cpp

@@ -2925,10 +2925,13 @@ static bool checkWorkGroupSizeValues(Sema &S, Decl *D, const ParsedAttr &Attr,
     return Result;
   }
 
-  if (const auto *A = D->getAttr<SYCLIntelMaxGlobalWorkDimAttr>())
-    if (A->getNumber() == 0)
+  if (const auto *A = D->getAttr<SYCLIntelMaxGlobalWorkDimAttr>()) {
+    int64_t AttrValue =
+        A->getValue()->getIntegerConstantExpr(S.Context)->getSExtValue();
+    if (AttrValue == 0)
       Result &= checkZeroDim(A, WGSize[0], WGSize[1], WGSize[2],
                              /*ReverseAttrs=*/true);
+  }
 
   if (const auto *A = D->getAttr<SYCLIntelMaxWorkGroupSizeAttr>()) {
     if (!(WGSize[0] <= A->getXDim() && WGSize[1] <= A->getYDim() &&
@@ -3087,34 +3090,23 @@ static void handleMaxGlobalWorkDimAttr(Sema &S, Decl *D,
   if (D->isInvalidDecl())
     return;
 
-  uint32_t MaxGlobalWorkDim;
-  const Expr *E = Attr.getArgAsExpr(0);
-  if (!checkUInt32Argument(S, Attr, E, MaxGlobalWorkDim, 0,
-                           /*StrictlyUnsigned=*/true))
-    return;
+  Expr *E = Attr.getArgAsExpr(0);
 
-  if (MaxGlobalWorkDim > 3) {
-    S.Diag(Attr.getLoc(), diag::err_intel_attribute_argument_is_not_in_range)
-      << Attr;
+  uint32_t WGSize[3] = {1, 1, 1};
+  if (!checkWorkGroupSizeValues(S, D, Attr, WGSize)) {
+    D->setInvalidDecl();
     return;
   }
 
-  if (MaxGlobalWorkDim == 0) {
-    uint32_t WGSize[3] = {1, 1, 1};
-    if (!checkWorkGroupSizeValues(S, D, Attr, WGSize)) {
-      D->setInvalidDecl();
-      return;
-    }
-  }
   if (D->getAttr<SYCLIntelMaxGlobalWorkDimAttr>())
     S.Diag(Attr.getLoc(), diag::warn_duplicate_attribute) << Attr;
 
   if (checkDeprecatedSYCLAttributeSpelling(S, Attr))
     S.Diag(Attr.getLoc(), diag::note_spelling_suggestion)
         << "'intel::max_global_work_dim'";
 
-  D->addAttr(::new (S.Context) SYCLIntelMaxGlobalWorkDimAttr(
-        S.Context, Attr, MaxGlobalWorkDim));
+  S.addIntelSYCLSingleArgFunctionAttr<SYCLIntelMaxGlobalWorkDimAttr>(D, Attr,
+                                                                     E);
 }
 
 static void handleVecTypeHint(Sema &S, Decl *D, const ParsedAttr &AL) {

clang/lib/Sema/SemaTemplateInstantiateDecl.cpp

@@ -769,6 +769,12 @@ void Sema::InstantiateAttrs(const MultiLevelTemplateArgumentList &TemplateArgs,
           *this, TemplateArgs, SYCLIntelSchedulerTargetFmaxMhz, New);
       continue;
     }
+    if (const auto *SYCLIntelMaxGlobalWorkDim =
+            dyn_cast<SYCLIntelMaxGlobalWorkDimAttr>(TmplAttr)) {
+      instantiateIntelSYCLFunctionAttr<SYCLIntelMaxGlobalWorkDimAttr>(
+          *this, TemplateArgs, SYCLIntelMaxGlobalWorkDim, New);
+      continue;
+    }
     // Existing DLL attribute on the instantiation takes precedence.
     if (TmplAttr->getKind() == attr::DLLExport ||
         TmplAttr->getKind() == attr::DLLImport) {

clang/test/CodeGenSYCL/intel-max-global-work-dim.cpp

@@ -1,24 +1,37 @@
-// RUN: %clang_cc1 -fsycl -fsycl-is-device -triple spir64-unknown-unknown-sycldevice -disable-llvm-passes -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -fsycl -fsycl-is-device -internal-isystem %S/Inputs -triple spir64-unknown-unknown-sycldevice -disable-llvm-passes -emit-llvm -o - %s | FileCheck %s
+
+#include "sycl.hpp"
+
+using namespace cl::sycl;
+queue q;
 
 class Foo {
 public:
   [[intel::max_global_work_dim(1)]] void operator()() const {}
 };
 
-template <typename name, typename Func>
-__attribute__((sycl_kernel)) void kernel(const Func &kernelFunc) {
-  kernelFunc();
-}
+template <int SIZE>
+class Functor {
+public:
+  [[intel::max_global_work_dim(SIZE)]] void operator()() const {}
+};
+
+int main() {
+  q.submit([&](handler &h) {
+    Foo boo;
+    h.single_task<class kernel_name1>(boo);
 
-void bar() {
-  Foo boo;
-  kernel<class kernel_name1>(boo);
+    h.single_task<class kernel_name2>(
+        []() [[intel::max_global_work_dim(2)]]{});
 
-  kernel<class kernel_name2>(
-      []() [[intel::max_global_work_dim(2)]]{});
+    Functor<2> f;
+    h.single_task<class kernel_name3>(f);
+  });
+  return 0;
 }
 
-// CHECK: define spir_kernel void @{{.*}}kernel_name1() {{.*}} !max_global_work_dim ![[NUM1:[0-9]+]]
-// CHECK: define spir_kernel void @{{.*}}kernel_name2() {{.*}} !max_global_work_dim ![[NUM8:[0-9]+]]
+// CHECK: define spir_kernel void @{{.*}}kernel_name1"() #0 {{.*}} !max_global_work_dim ![[NUM1:[0-9]+]]
+// CHECK: define spir_kernel void @{{.*}}kernel_name2"() #0 {{.*}} !max_global_work_dim ![[NUM2:[0-9]+]]
+// CHECK: define spir_kernel void @{{.*}}kernel_name3"() #0 {{.*}} !max_global_work_dim ![[NUM2]]
 // CHECK: ![[NUM1]] = !{i32 1}
-// CHECK: ![[NUM8]] = !{i32 2}
+// CHECK: ![[NUM2]] = !{i32 2}

clang/test/CodeGenSYCL/num-simd-work-items.cpp

@@ -1,4 +1,9 @@
-// RUN: %clang_cc1 -fsycl -fsycl-is-device -triple spir64-unknown-unknown-sycldevice -disable-llvm-passes -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -fsycl -fsycl-is-device -internal-isystem %S/Inputs -triple spir64-unknown-unknown-sycldevice -disable-llvm-passes -emit-llvm -o - %s | FileCheck %s
+
+#include "sycl.hpp"
+
+using namespace cl::sycl;
+queue q;
 
 class Foo {
 public:
@@ -11,25 +16,23 @@ class Functor {
   [[intel::num_simd_work_items(SIZE)]] void operator()() const {}
 };
 
-template <typename name, typename Func>
-__attribute__((sycl_kernel)) void kernel(const Func &kernelFunc) {
-  kernelFunc();
-}
-
-void bar() {
-  Foo boo;
-  kernel<class kernel_name1>(boo);
+int main() {
+  q.submit([&](handler &h) {
+    Foo boo;
+    h.single_task<class kernel_name1>(boo);
 
-  kernel<class kernel_name2>(
-      []() [[intel::num_simd_work_items(42)]]{});
+    h.single_task<class kernel_name2>(
+        []() [[intel::num_simd_work_items(42)]]{});
 
-  Functor<2> f;
-  kernel<class kernel_name3>(f);
+    Functor<2> f;
+    h.single_task<class kernel_name3>(f);
+  });
+  return 0;
 }
 
-// CHECK: define spir_kernel void @{{.*}}kernel_name1() {{.*}} !num_simd_work_items ![[NUM1:[0-9]+]]
-// CHECK: define spir_kernel void @{{.*}}kernel_name2() {{.*}} !num_simd_work_items ![[NUM42:[0-9]+]]
-// CHECK: define spir_kernel void @{{.*}}kernel_name3() {{.*}} !num_simd_work_items ![[NUM2:[0-9]+]]
+// CHECK: define spir_kernel void @{{.*}}kernel_name1"() #0 {{.*}} !num_simd_work_items ![[NUM1:[0-9]+]]
+// CHECK: define spir_kernel void @{{.*}}kernel_name2"() #0 {{.*}} !num_simd_work_items ![[NUM42:[0-9]+]]
+// CHECK: define spir_kernel void @{{.*}}kernel_name3"() #0 {{.*}} !num_simd_work_items ![[NUM2:[0-9]+]]
 // CHECK: ![[NUM1]] = !{i32 1}
 // CHECK: ![[NUM42]] = !{i32 42}
 // CHECK: ![[NUM2]] = !{i32 2}

clang/test/CodeGenSYCL/reqd-sub-group-size.cpp

@@ -1,4 +1,9 @@
-// RUN: %clang_cc1 -fsycl -fsycl-is-device -disable-llvm-passes -triple spir64-unknown-unknown-sycldevice -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -fsycl -fsycl-is-device -internal-isystem %S/Inputs -disable-llvm-passes -triple spir64-unknown-unknown-sycldevice -emit-llvm -o - %s | FileCheck %s
+
+#include "sycl.hpp"
+
+using namespace cl::sycl;
+queue q;
 
 class Functor16 {
 public:
@@ -7,42 +12,40 @@ class Functor16 {
 
 [[intel::reqd_sub_group_size(8)]] void foo() {}
 
-class Functor {
+class Functor8 {
 public:
   void operator()() const {
     foo();
   }
 };
 
 template <int SIZE>
-class Functor5 {
+class Functor2 {
 public:
   [[intel::reqd_sub_group_size(SIZE)]] void operator()() const {}
 };
 
-template <typename name, typename Func>
-__attribute__((sycl_kernel)) void kernel(const Func &kernelFunc) {
-  kernelFunc();
-}
-
-void bar() {
-  Functor16 f16;
-  kernel<class kernel_name1>(f16);
+int main() {
+  q.submit([&](handler &h) {
+    Functor16 f16;
+    h.single_task<class kernel_name1>(f16);
 
-  Functor f;
-  kernel<class kernel_name2>(f);
+    Functor8 f8;
+    h.single_task<class kernel_name2>(f8);
 
-  kernel<class kernel_name3>(
-      []() [[intel::reqd_sub_group_size(4)]]{});
+    h.single_task<class kernel_name3>(
+        []() [[intel::reqd_sub_group_size(4)]]{});
 
-  Functor5<2> f5;
-  kernel<class kernel_name4>(f5);
+    Functor2<2> f2;
+    h.single_task<class kernel_name4>(f2);
+  });
+  return 0;
 }
 
-// CHECK: define spir_kernel void @{{.*}}kernel_name1() {{.*}} !intel_reqd_sub_group_size ![[SGSIZE16:[0-9]+]]
-// CHECK: define spir_kernel void @{{.*}}kernel_name2() {{.*}} !intel_reqd_sub_group_size ![[SGSIZE8:[0-9]+]]
-// CHECK: define spir_kernel void @{{.*}}kernel_name3() {{.*}} !intel_reqd_sub_group_size ![[SGSIZE4:[0-9]+]]
-// CHECK: define spir_kernel void @{{.*}}kernel_name4() {{.*}} !intel_reqd_sub_group_size ![[SGSIZE2:[0-9]+]]
+// CHECK: define spir_kernel void @{{.*}}kernel_name1"() #0 {{.*}} !intel_reqd_sub_group_size ![[SGSIZE16:[0-9]+]]
+// CHECK: define spir_kernel void @{{.*}}kernel_name2"() #0 {{.*}} !intel_reqd_sub_group_size ![[SGSIZE8:[0-9]+]]
+// CHECK: define spir_kernel void @{{.*}}kernel_name3"() #0 {{.*}} !intel_reqd_sub_group_size ![[SGSIZE4:[0-9]+]]
+// CHECK: define spir_kernel void @{{.*}}kernel_name4"() #0 {{.*}} !intel_reqd_sub_group_size ![[SGSIZE2:[0-9]+]]
 // CHECK: ![[SGSIZE16]] = !{i32 16}
 // CHECK: ![[SGSIZE8]] = !{i32 8}
 // CHECK: ![[SGSIZE4]] = !{i32 4}