Merge branch 'sycl' into enable_spirv_test

Author: YixingZhang007

.github/workflows/sycl-linux-precommit.yml

@@ -62,6 +62,42 @@ jobs:
       e2e_binaries_artifact: e2e_bin
       e2e_binaries_preview_artifact: e2e_bin_preview
 
+  # Build and run native cpu e2e tests separately as cannot currently
+  # build all the e2e tests
+  build_run_native_cpu_e2e_tests:
+    if: ${{ always() && !cancelled() && needs.build.outputs.build_conclusion == 'success' }}  
+    runs-on: [Linux, build]
+    needs: [build]
+    container:
+      image: ghcr.io/intel/llvm/sycl_ubuntu2404_nightly:latest
+      options: -u 1001:1001
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          sparse-checkout: |
+            devops/
+
+      # download build artefact
+      - name: Download toolchain
+        uses: actions/download-artifact@v4
+        with:
+          name: sycl_linux_default
+      - name: Extract SYCL toolchain
+        shell: bash
+        run: |
+          mkdir toolchain
+          tar -xf llvm_sycl.tar.zst -C toolchain
+          rm llvm_sycl.tar.zst
+      - name: Build and run E2E tests
+        uses: ./devops/actions/run-tests/e2e
+        with:
+          ref: ${{ inputs.ref || github.sha }}
+          testing_mode: full
+          target_devices: native_cpu:cpu
+          sycl_compiler: $GITHUB_WORKSPACE/toolchain/bin/clang++
+          extra_lit_opts: --param sycl_build_targets="native_cpu"
+          extra_cmake_args: -DSYCL_TEST_E2E_TARGETS="native_cpu:cpu" -DSYCL_TEST_E2E_STANDALONE=ON
+
   # If a PR changes CUDA adapter, run the build on Ubuntu 22.04 as well.
   # Ubuntu 22.04 container has CUDA 12.1 installed while Ubuntu 24.0 image
   # has CUDA 12.6.1 installed.

clang/lib/Driver/ToolChains/Clang.cpp

@@ -10097,8 +10097,13 @@ void OffloadWrapper::ConstructJob(Compilation &C, const JobAction &JA,
     // clang-offload-wrapper
     //   -o=<outputfile>.bc
     //   -host=x86_64-pc-linux-gnu -kind=sycl
+#ifndef __INTEL_PREVIEW_BREAKING_CHANGES
     //   -format=spirv <inputfile1>.spv <manifest1>(optional)
     //   -format=spirv <inputfile2>.spv <manifest2>(optional)
+#else
+    //   -format=spirv <inputfile1>.spv
+    //   -format=spirv <inputfile2>.spv
+#endif
     //  ...
     ArgStringList WrapperArgs;
 
@@ -10163,6 +10168,12 @@ void OffloadWrapper::ConstructJob(Compilation &C, const JobAction &JA,
     WrapperArgs.push_back(
         C.getArgs().MakeArgString(Twine("-kind=") + Twine(Kind)));
 
+    // Enable preview breaking changes in clang-offload-wrapper,
+    // in case it needs to introduce any ABI breaking changes.
+    // For example, changes in offload binary descriptor format.
+    if (C.getArgs().hasArg(options::OPT_fpreview_breaking_changes))
+      WrapperArgs.push_back("-fpreview-breaking-changes");
+
     assert((Inputs.size() > 0) && "no inputs for clang-offload-wrapper");
     assert(((Inputs[0].getType() != types::TY_Tempfiletable) ||
             (Inputs.size() == 1)) &&
@@ -11580,6 +11591,12 @@ void LinkerWrapper::ConstructJob(Compilation &C, const JobAction &JA,
           Args.MakeArgString("--wrapper-jobs=" + Twine(NumThreads)));
   }
 
+  // Enable preview breaking changes in clang-linker-wrapper,
+  // in case it needs to introduce any ABI breaking changes.
+  // For example, changes in offload binary descriptor format.
+  if (Args.hasArg(options::OPT_fpreview_breaking_changes))
+    CmdArgs.push_back("-fpreview-breaking-changes");
+
   const char *Exec =
       Args.MakeArgString(getToolChain().GetProgramPath("clang-linker-wrapper"));
 

clang/lib/Sema/SemaSYCL.cpp

@@ -6626,10 +6626,12 @@ class FreeFunctionPrinter {
   raw_ostream &O;
   PrintingPolicy &Policy;
   bool NSInserted = false;
+  ASTContext &Context;
 
 public:
-  FreeFunctionPrinter(raw_ostream &O, PrintingPolicy &PrintPolicy)
-      : O(O), Policy(PrintPolicy) {}
+  FreeFunctionPrinter(raw_ostream &O, PrintingPolicy &PrintPolicy,
+                      ASTContext &Context)
+      : O(O), Policy(PrintPolicy), Context(Context) {}
 
   /// Emits the function declaration of template free function.
   /// \param FTD The function declaration to print.
@@ -6822,22 +6824,46 @@ class FreeFunctionPrinter {
         continue;
       }
 
-      TemplateName TN = TST->getTemplateName();
-      auto SpecArgs = TST->template_arguments();
-      auto DeclArgs = CTST->template_arguments();
-
-      TN.getAsTemplateDecl()->printQualifiedName(ParmListOstream);
+      TemplateName CTN = CTST->getTemplateName();
+      CTN.getAsTemplateDecl()->printQualifiedName(ParmListOstream);
       ParmListOstream << "<";
 
+      ArrayRef<TemplateArgument> SpecArgs = TST->template_arguments();
+      ArrayRef<TemplateArgument> DeclArgs = CTST->template_arguments();
+
+      auto TemplateArgPrinter = [&](const TemplateArgument &Arg) {
+        if (Arg.getKind() != TemplateArgument::ArgKind::Expression ||
+            Arg.isInstantiationDependent()) {
+          Arg.print(Policy, ParmListOstream, /* IncludeType = */ false);
+          return;
+        }
+
+        Expr *E = Arg.getAsExpr();
+        assert(E && "Failed to get an Expr for an Expression template arg?");
+        if (E->getType().getTypePtr()->isScopedEnumeralType()) {
+          // Scoped enumerations can't be implicitly cast from integers, so
+          // we don't need to evaluate them.
+          Arg.print(Policy, ParmListOstream, /* IncludeType = */ false);
+          return;
+        }
+
+        Expr::EvalResult Res;
+        [[maybe_unused]] bool Success =
+            Arg.getAsExpr()->EvaluateAsConstantExpr(Res, Context);
+        assert(Success && "invalid non-type template argument?");
+        assert(!Res.Val.isAbsent() && "couldn't read the evaulation result?");
+        Res.Val.printPretty(ParmListOstream, Policy, Arg.getAsExpr()->getType(),
+                            &Context);
+      };
+
       for (size_t I = 0, E = std::max(DeclArgs.size(), SpecArgs.size()),
                   SE = SpecArgs.size();
            I < E; ++I) {
         if (I != 0)
           ParmListOstream << ", ";
-        if (I < SE) // A specialized argument exists, use it
-          SpecArgs[I].print(Policy, ParmListOstream, false /* IncludeType */);
-        else // Print a canonical form of a default argument
-          DeclArgs[I].print(Policy, ParmListOstream, false /* IncludeType */);
+        // If we have a specialized argument, use it. Otherwise fallback to a
+        // default argument.
+        TemplateArgPrinter(I < SE ? SpecArgs[I] : DeclArgs[I]);
       }
 
       ParmListOstream << ">";
@@ -7236,7 +7262,7 @@ void SYCLIntegrationHeader::emit(raw_ostream &O) {
     // template arguments that match default template arguments while printing
     // template-ids, even if the source code doesn't reference them.
     Policy.EnforceDefaultTemplateArgs = true;
-    FreeFunctionPrinter FFPrinter(O, Policy);
+    FreeFunctionPrinter FFPrinter(O, Policy, S.getASTContext());
     if (FTD) {
       FFPrinter.printFreeFunctionDeclaration(FTD);
       if (const auto kind = K.SyclKernel->getTemplateSpecializationKind();

clang/test/CodeGenSYCL/free-function-kernel-expr-as-template-arg.cpp

@@ -0,0 +1,110 @@
+// RUN: %clang_cc1 -fsycl-is-device -internal-isystem %S/Inputs -triple spir64-unknown-unknown -sycl-std=2020 -fsycl-int-header=%t.h %s
+// RUN: FileCheck -input-file=%t.h %s
+//
+// The purpose of this test is to ensure that forward declarations of free
+// function kernels are emitted properly.
+// However, this test checks a specific scenario:
+// - free function argument is a template which accepts constant expressions as
+//   arguments
+
+constexpr int A = 2;
+constexpr int B = 3;
+
+namespace ns {
+
+constexpr int C = 4;
+
+struct Foo {
+  static constexpr int D = 5;
+};
+
+enum non_class_enum {
+  VAL_A,
+  VAL_B
+};
+
+enum class class_enum {
+  VAL_A,
+  VAL_B
+};
+
+enum non_class_enum_typed : int {
+  VAL_C,
+  VAL_D
+};
+
+enum class class_enum_typed : int {
+  VAL_C,
+  VAL_D
+};
+
+constexpr int bar(int arg) {
+  return arg + 42;
+}
+
+} // namespace ns
+
+template<int V>
+struct Arg {};
+
+template<ns::non_class_enum V>
+struct Arg2 {};
+
+template<ns::non_class_enum_typed V>
+struct Arg3 {};
+
+template<ns::class_enum V>
+struct Arg4 {};
+
+template<ns::class_enum_typed V>
+struct Arg5 {};
+
+[[__sycl_detail__::add_ir_attributes_function("sycl-nd-range-kernel", 2)]]
+void constant(Arg<1>) {}
+
+// CHECK: void constant(Arg<1> );
+
+[[__sycl_detail__::add_ir_attributes_function("sycl-nd-range-kernel", 2)]]
+void constexpr_v(Arg<A>) {}
+
+// CHECK: void constexpr_v(Arg<2> );
+
+[[__sycl_detail__::add_ir_attributes_function("sycl-nd-range-kernel", 2)]]
+void constexpr_expr(Arg<A * B>) {}
+
+// CHECK: void constexpr_expr(Arg<6> );
+
+[[__sycl_detail__::add_ir_attributes_function("sycl-nd-range-kernel", 2)]]
+void constexpr_ns(Arg<ns::C>) {}
+
+// CHECK: void constexpr_ns(Arg<4> );
+
+[[__sycl_detail__::add_ir_attributes_function("sycl-nd-range-kernel", 2)]]
+void constexpr_ns2(Arg<ns::Foo::D>) {}
+
+// CHECK: void constexpr_ns2(Arg<5> );
+
+[[__sycl_detail__::add_ir_attributes_function("sycl-nd-range-kernel", 2)]]
+void constexpr_ns2(Arg2<ns::non_class_enum::VAL_A>) {}
+
+// CHECK: void constexpr_ns2(Arg2<ns::VAL_A> );
+
+[[__sycl_detail__::add_ir_attributes_function("sycl-nd-range-kernel", 2)]]
+void constexpr_ns2(Arg3<ns::non_class_enum_typed::VAL_C>) {}
+
+// CHECK: void constexpr_ns2(Arg3<ns::VAL_C> );
+
+[[__sycl_detail__::add_ir_attributes_function("sycl-nd-range-kernel", 2)]]
+void constexpr_ns2(Arg4<ns::class_enum::VAL_A>) {}
+
+// CHECK: void constexpr_ns2(Arg4<ns::class_enum::VAL_A> );
+
+[[__sycl_detail__::add_ir_attributes_function("sycl-nd-range-kernel", 2)]]
+void constexpr_ns2(Arg5<ns::class_enum_typed::VAL_C>) {}
+
+// CHECK: void constexpr_ns2(Arg5<ns::class_enum_typed::VAL_C> );
+
+[[__sycl_detail__::add_ir_attributes_function("sycl-nd-range-kernel", 2)]]
+void constexpr_call(Arg<ns::bar(B)>) {}
+
+// CHECK: void constexpr_call(Arg<45> );

clang/test/CodeGenSYCL/free-function-kernel-type-alias-arg.cpp

@@ -0,0 +1,63 @@
+// RUN: %clang_cc1 -fsycl-is-device -internal-isystem %S/Inputs -triple spir64-unknown-unknown -sycl-std=2020 -fsycl-int-header=%t.h %s
+// RUN: FileCheck -input-file=%t.h %s
+//
+// The purpose of this test is to ensure that forward declarations of free
+// function kernels are emitted properly.
+// However, this test checks a specific scenario:
+// - free function arguments are type aliases (through using or typedef)
+
+namespace ns {
+
+using IntUsing = int;
+typedef int IntTypedef;
+
+template <typename T>
+struct Foo {};
+
+using FooIntUsing = Foo<int>;
+typedef Foo<int> FooIntTypedef;
+
+template <typename T1, typename T2>
+struct Bar {};
+
+template<typename T1>
+using BarUsing = Bar<T1, float>;
+
+class Baz {
+public:
+  using type = BarUsing<double>;
+};
+
+} // namespace ns
+
+[[__sycl_detail__::add_ir_attributes_function("sycl-nd-range-kernel", 2)]]
+void int_using(ns::IntUsing Arg) {}
+
+// CHECK: void int_using(int Arg);
+
+[[__sycl_detail__::add_ir_attributes_function("sycl-nd-range-kernel", 2)]]
+void int_typedef(ns::IntTypedef Arg) {}
+
+// CHECK: void int_typedef(int Arg);
+
+[[__sycl_detail__::add_ir_attributes_function("sycl-nd-range-kernel", 2)]]
+void foo_using(ns::FooIntUsing Arg) {}
+
+// CHECK: void foo_using(ns::Foo<int> Arg);
+
+[[__sycl_detail__::add_ir_attributes_function("sycl-nd-range-kernel", 2)]]
+void foo_typedef(ns::FooIntTypedef Arg) {}
+
+// CHECK: void foo_typedef(ns::Foo<int> Arg);
+
+template<typename T>
+[[__sycl_detail__::add_ir_attributes_function("sycl-nd-range-kernel", 2)]]
+void bar_using(ns::BarUsing<T> Arg) {}
+template void bar_using(ns::BarUsing<int>);
+
+// CHECK: template <typename T> void bar_using(ns::Bar<T, float>);
+
+[[__sycl_detail__::add_ir_attributes_function("sycl-nd-range-kernel", 2)]]
+void baz_type(ns::Baz::type Arg) {}
+
+// CHECK: void baz_type(ns::Bar<double, float> Arg);