[SYCL][FPGA] Implement num_simd_work_items kernel attribute

Applies to a device function/lambda function. Indicates the number
of X work items that should be processed in parallel. Valid values
are positive integers.

Signed-off-by: Dmitry Sidorov <[email protected]>

Author: MrSidims

clang/include/clang/Basic/Attr.td

@@ -1114,6 +1114,15 @@ def SYCLIntelKernelArgsRestrict : InheritableAttr {
   let Documentation = [ SYCLIntelKernelArgsRestrictDocs ];
 }
 
+def SYCLIntelNumSimdWorkItems : InheritableAttr {
+  let Spellings = [CXX11<"intelfpga","num_simd_work_items">];
+  let Args = [UnsignedArgument<"Number">];
+  let LangOpts = [SYCLIsDevice, SYCLIsHost];
+  let Subjects = SubjectList<[Function], ErrorDiag>;
+  let Documentation = [SYCLIntelNumSimdWorkItemsAttrDocs];
+  let PragmaAttributeSupport = 0;
+}
+
 def C11NoReturn : InheritableAttr {
   let Spellings = [Keyword<"_Noreturn">];
   let Subjects = SubjectList<[Function], ErrorDiag>;

clang/include/clang/Basic/AttrDocs.td

@@ -1958,6 +1958,17 @@ function, and no effect otherwise.
   }];
 }
 
+def SYCLIntelNumSimdWorkItemsAttrDocs : Documentation {
+  let Category = DocCatFunction;
+  let Heading = "num_simd_work_items (IntelFPGA)";
+  let Content = [{
+Applies to a device function/lambda function. Indicates the number of work
+items that should be processed in parallel. Valid values are positive integers.
+If ``intelfpga::num_simd_work_items`` is applied to a function called from a
+device kernel, the attribute is ignored and it is not propagated to a kernel.
+  }];
+}
+
 def SYCLFPGAPipeDocs : Documentation {
   let Category = DocCatStmt;
   let Heading = "pipe (read_only, write_only)";

clang/include/clang/Basic/AttributeCommonInfo.h

@@ -154,7 +154,8 @@ class AttributeCommonInfo {
     // to the call operator declaration.
     auto ParsedAttr = getParsedKind();
     if (ParsedAttr == AT_SYCLIntelKernelArgsRestrict ||
-        (ParsedAttr == AT_ReqdWorkGroupSize && isCXX11Attribute()))
+        (ParsedAttr == AT_ReqdWorkGroupSize && isCXX11Attribute()) ||
+        ParsedAttr == AT_SYCLIntelNumSimdWorkItems)
       return true;
 
     return false;

clang/lib/CodeGen/CodeGenFunction.cpp

@@ -587,6 +587,14 @@ void CodeGenFunction::EmitOpenCLKernelMetadata(const FunctionDecl *FD,
     Fn->setMetadata("intel_reqd_sub_group_size",
                     llvm::MDNode::get(Context, AttrMDArgs));
   }
+
+  if (const SYCLIntelNumSimdWorkItemsAttr *A =
+      FD->getAttr<SYCLIntelNumSimdWorkItemsAttr>()) {
+    llvm::Metadata *AttrMDArgs[] = {
+        llvm::ConstantAsMetadata::get(Builder.getInt32(A->getNumber()))};
+    Fn->setMetadata("num_simd_work_items",
+                    llvm::MDNode::get(Context, AttrMDArgs));
+  }
 }
 
 /// Determine whether the function F ends with a return stmt.

clang/lib/Sema/SemaDeclAttr.cpp

@@ -2903,6 +2903,31 @@ static void handleSubGroupSize(Sema &S, Decl *D, const ParsedAttr &AL) {
                  IntelReqdSubGroupSizeAttr(S.Context, AL, SGSize));
 }
 
+// Handles num_simd_work_items.
+static void handleNumSimdWorkItemsAttr(Sema &S, Decl *D,
+                                       const ParsedAttr &Attr) {
+  if (D->isInvalidDecl())
+    return;
+
+  uint32_t NumSimdWorkItems = 0;
+  const Expr *E = Attr.getArgAsExpr(0);
+  if (!checkUInt32Argument(S, Attr, E, NumSimdWorkItems, 0,
+                           /*StrictlyUnsigned=*/true))
+    return;
+
+  if (NumSimdWorkItems == 0) {
+    S.Diag(Attr.getLoc(), diag::err_attribute_argument_is_zero)
+      << Attr << E->getSourceRange();
+    return;
+  }
+
+  if (D->getAttr<SYCLIntelNumSimdWorkItemsAttr>())
+    S.Diag(Attr.getLoc(), diag::warn_duplicate_attribute) << Attr;
+
+  D->addAttr(::new (S.Context) SYCLIntelNumSimdWorkItemsAttr(
+        S.Context, Attr, NumSimdWorkItems));
+}
+
 static void handleVecTypeHint(Sema &S, Decl *D, const ParsedAttr &AL) {
   if (!AL.hasParsedType()) {
     S.Diag(AL.getLoc(), diag::err_attribute_wrong_number_arguments) << AL << 1;
@@ -7323,6 +7348,9 @@ static void ProcessDeclAttribute(Sema &S, Scope *scope, Decl *D,
   case ParsedAttr::AT_IntelReqdSubGroupSize:
     handleSubGroupSize(S, D, AL);
     break;
+  case ParsedAttr::AT_SYCLIntelNumSimdWorkItems:
+    handleNumSimdWorkItemsAttr(S, D, AL);
+    break;
   case ParsedAttr::AT_VecTypeHint:
     handleVecTypeHint(S, D, AL);
     break;

clang/lib/Sema/SemaSYCL.cpp

@@ -428,18 +428,27 @@ class MarkDeviceFunction : public RecursiveASTVisitor<MarkDeviceFunction> {
         Attrs.insert(A);
       if (auto *A = FD->getAttr<ReqdWorkGroupSizeAttr>())
         Attrs.insert(A);
+
+      // Allow the following kernel attributes only on lambda functions and
+      // function objects that are called directly from a kernel (i.e. the one
+      // passed to the parallel_for function). For all other cases,
+      // emit a warning and ignore.
       if (auto *A = FD->getAttr<SYCLIntelKernelArgsRestrictAttr>()) {
-        // Allow the intel::kernel_args_restrict only on the lambda (function
-        // object) function, that is called directly from a kernel (i.e. the one
-        // passed to the parallel_for function). Emit a warning and ignore all
-        // other cases.
         if (ParentFD == SYCLKernel) {
           Attrs.insert(A);
         } else {
           SemaRef.Diag(A->getLocation(), diag::warn_attribute_ignored) << A;
           FD->dropAttr<SYCLIntelKernelArgsRestrictAttr>();
         }
       }
+      if (auto *A = FD->getAttr<SYCLIntelNumSimdWorkItemsAttr>()) {
+        if (ParentFD == SYCLKernel) {
+          Attrs.insert(A);
+        } else {
+          SemaRef.Diag(A->getLocation(), diag::warn_attribute_ignored) << A;
+          FD->dropAttr<SYCLIntelNumSimdWorkItemsAttr>();
+        }
+      }
 
       // TODO: vec_len_hint should be handled here
 
@@ -1338,7 +1347,8 @@ void Sema::MarkDevice(void) {
           }
           break;
         }
-        case attr::Kind::SYCLIntelKernelArgsRestrict: {
+        case attr::Kind::SYCLIntelKernelArgsRestrict:
+        case attr::Kind::SYCLIntelNumSimdWorkItems: {
           SYCLKernel->addAttr(A);
           break;
         }

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

@@ -0,0 +1,26 @@
+// RUN: %clang_cc1 -std=c++11 -triple spir64-unknown-unknown-sycldevice -disable-llvm-passes -fsycl-is-device -emit-llvm -o - %s | FileCheck %s
+
+class Foo {
+public:
+  [[intelfpga::num_simd_work_items(1)]] void operator()() {}
+};
+
+template <typename name, typename Func>
+__attribute__((sycl_kernel)) void kernel(Func kernelFunc) {
+  kernelFunc();
+}
+
+void bar() {
+  Foo boo;
+  kernel<class kernel_name1>(boo);
+
+  kernel<class kernel_name2>(
+  []() [[intelfpga::num_simd_work_items(42)]] {});
+
+}
+
+// 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: ![[NUM1]] = !{i32 1}
+// CHECK: ![[NUM42]] = !{i32 42}
+

clang/test/SemaSYCL/num_simd_work_items.cpp

@@ -0,0 +1,65 @@
+// RUN: %clang %s -fsyntax-only -fsycl-device-only -DTRIGGER_ERROR -Xclang -verify
+// RUN: %clang %s -fsyntax-only -Xclang -ast-dump -fsycl-device-only | FileCheck %s
+// RUN: %clang_cc1 -fsycl-is-host -fsyntax-only -verify %s
+
+#ifndef __SYCL_DEVICE_ONLY__
+struct FuncObj {
+  [[intelfpga::num_simd_work_items(42)]] // expected-no-diagnostics
+  void operator()() {}
+};
+
+template <typename name, typename Func>
+void kernel(Func kernelFunc) {
+  kernelFunc();
+}
+
+void foo() {
+  kernel<class test_kernel1>(
+      FuncObj());
+}
+
+#else // __SYCL_DEVICE_ONLY__
+
+[[intelfpga::num_simd_work_items(2)]] // expected-warning{{'num_simd_work_items' attribute ignored}}
+void func_ignore() {}
+
+struct FuncObj {
+  [[intelfpga::num_simd_work_items(42)]]
+  void operator()() {}
+};
+
+template <typename name, typename Func>
+__attribute__((sycl_kernel)) void kernel(Func kernelFunc) {
+  kernelFunc();
+}
+
+int main() {
+  // CHECK-LABEL: FunctionDecl {{.*}} _ZTSZ4mainE12test_kernel1
+  // CHECK:       SYCLIntelNumSimdWorkItemsAttr {{.*}} 42
+  kernel<class test_kernel1>(
+      FuncObj());
+
+  // CHECK-LABEL: FunctionDecl {{.*}} _ZTSZ4mainE12test_kernel2
+  // CHECK:       SYCLIntelNumSimdWorkItemsAttr {{.*}} 8
+  kernel<class test_kernel2>(
+      []() [[intelfpga::num_simd_work_items(8)]] {});
+
+  // CHECK-LABEL: FunctionDecl {{.*}} _ZTSZ4mainE12test_kernel3
+  // CHECK-NOT:   SYCLIntelNumSimdWorkItemsAttr {{.*}} 2
+  kernel<class test_kernel3>(
+      []() {func_ignore();});
+
+#ifdef TRIGGER_ERROR
+  [[intelfpga::num_simd_work_items(0)]] int Var = 0; // expected-error{{'num_simd_work_items' attribute only applies to functions}}
+
+  kernel<class test_kernel4>(
+      []() [[intelfpga::num_simd_work_items(0)]] {}); // expected-error{{'num_simd_work_items' attribute must be greater than 0}}
+
+  kernel<class test_kernel5>(
+      []() [[intelfpga::num_simd_work_items(-42)]] {}); // expected-error{{'num_simd_work_items' attribute requires a non-negative integral compile time constant expression}}
+
+  kernel<class test_kernel6>(
+      []() [[intelfpga::num_simd_work_items(1), intelfpga::num_simd_work_items(2)]] {}); // expected-warning{{attribute 'num_simd_work_items' is already applied with different parameters}}
+#endif // TRIGGER_ERROR
+}
+#endif // __SYCL_DEVICE_ONLY__