Merge remote-tracking branch 'upstream/sycl' into sycl-mlir

Author: whitneywhtsang

.github/workflows/sycl_post_commit.yml

@@ -19,12 +19,14 @@ jobs:
   # This job generates matrix of tests for LLVM Test Suite
   test_matrix:
     name: Generate Test Matrix
+    if: github.repository == 'intel/llvm'
     uses: ./.github/workflows/sycl_gen_test_matrix.yml
     with:
       lts_config: "l0_gen9"
       cts_config: "cuda"
   linux_default:
     name: Linux Default
+    if: github.repository == 'intel/llvm'
     needs: test_matrix
     uses: ./.github/workflows/sycl_linux_build_and_test.yml
     secrets: inherit
@@ -36,6 +38,7 @@ jobs:
       lts_aws_matrix: ${{ needs.test_matrix.outputs.lts_aws_matrix }}
   linux_no_assert:
     name: Linux (no assert)
+    if: github.repository == 'intel/llvm'
     uses: ./.github/workflows/sycl_linux_build_and_test.yml
     secrets: inherit
     with:

.github/workflows/windows_test_comment_trigger.yml

@@ -56,9 +56,10 @@ jobs:
   windows_test_completion:
     runs-on: ubuntu-latest
     needs: [windows_test_preparation, windows_default]
+    if: always()
     steps:
     - name: update_pr_status_success
-      if: always() && success()
+      if: needs.windows_default.result == 'success'
       env:
         GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         OWNER: ${{ github.repository_owner }}
@@ -67,7 +68,7 @@ jobs:
       run: |
           gh api --method POST -H "Accept: application/vnd.github+json" /repos/$OWNER/$REPO/statuses/$SHA -f state='success' -f target_url='${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}' -f description='The operation completed successfully' -f context='Windows pre-commit'
     - name: update_pr_status_failure
-      if: failure()
+      if: needs.windows_default.result == 'failure'
       env:
         GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         OWNER: ${{ github.repository_owner }}

CONTRIBUTING.md

@@ -90,6 +90,8 @@ the PR with:
   - link to existing open issue
   - if author cannot identify any of these, the minimal action expected is to
     open a new [issue](/../../issues)
+  - list of unrelated tests failing in pre-commit CI. This would enable easy
+    access to them via github search functionality.
 
 ### Merge
 

buildbot/configure.py

@@ -23,6 +23,11 @@ def do_configure(args):
     libclc_amd_target_names = ';amdgcn--;amdgcn--amdhsa'
     libclc_nvidia_target_names = ';nvptx64--;nvptx64--nvidiacl'
 
+    sycl_enable_fusion = "OFF"
+    if not args.disable_fusion:
+        llvm_external_projects += ";sycl-fusion"
+        sycl_enable_fusion = "ON"
+
     if args.llvm_external_projects:
         llvm_external_projects += ";" + args.llvm_external_projects.replace(",", ";")
 
@@ -35,6 +40,7 @@ def do_configure(args):
     mlir_dir = os.path.join(abs_src_dir, "mlir")
     mlir_sycl_dir = os.path.join(abs_src_dir, "mlir-sycl")
     polygeist_dir = os.path.join(abs_src_dir, "polygeist")
+    fusion_dir = os.path.join(abs_src_dir, "sycl-fusion")
     llvm_targets_to_build = args.host_target
     llvm_enable_projects = 'clang;' + llvm_external_projects
     libclc_targets_to_build = ''
@@ -164,6 +170,7 @@ def do_configure(args):
         "-DLLVM_EXTERNAL_MLIR_SOURCE_DIR={}".format(mlir_dir),
         "-DLLVM_EXTERNAL_MLIR_SYCL_SOURCE_DIR={}".format(mlir_sycl_dir),
         "-DLLVM_EXTERNAL_POLYGEIST_SOURCE_DIR={}".format(polygeist_dir),
+        "-DLLVM_EXTERNAL_SYCL_FUSION_SOURCE_DIR={}".format(fusion_dir),
         "-DLLVM_ENABLE_PROJECTS={}".format(llvm_enable_projects),
         "-DLIBCLC_TARGETS_TO_BUILD={}".format(libclc_targets_to_build),
         "-DLIBCLC_GENERATE_REMANGLED_VARIANTS={}".format(libclc_gen_remangled_variants),
@@ -182,7 +189,8 @@ def do_configure(args):
         "-DSYCL_ENABLE_PLUGINS={}".format(';'.join(set(sycl_enabled_plugins))),
         "-DCMAKE_C_COMPILER={}".format(build_compiler_c),
         "-DCMAKE_CXX_COMPILER={}".format(build_compiler_cpp),
-        "-DCMAKE_VERBOSE_MAKEFILE={}".format(verbose)
+        "-DCMAKE_VERBOSE_MAKEFILE={}".format(verbose),
+        "-DSYCL_ENABLE_KERNEL_FUSION={}".format(sycl_enable_fusion)
     ]
 
     if args.l0_headers and args.l0_loader:
@@ -262,8 +270,9 @@ def main():
     parser.add_argument("--ci-defaults", action="store_true", help="Enable default CI parameters")
     parser.add_argument("--enable-plugin", action='append', help="Enable SYCL plugin")
     parser.add_argument("--build-compiler-c", metavar="BUILD_COMPILER_C", help="C compiler to use to build the project")
-    parser.add_argument("--build-compiler-cpp", metavar="BUILD_COMPILER_CPP", help="C++ compiler to use to build the project"),
-    parser.add_argument("--verbose", default='OFF', help="Verbose build"),
+    parser.add_argument("--build-compiler-cpp", metavar="BUILD_COMPILER_CPP", help="C++ compiler to use to build the project")
+    parser.add_argument("--verbose", default='OFF', help="Verbose build")
+    parser.add_argument("--disable-fusion", action="store_true", help="Disable the kernel fusion JIT compiler")
 
     args = parser.parse_args()
 

clang/include/clang/Basic/Attr.td

@@ -3497,35 +3497,55 @@ def SYCLReqdWorkGroupSize : InheritableAttr, LanguageOptionsSpecificAttr {
   let HasCustomParsing = 1;
 }
 
-def WorkGroupSizeHint :  InheritableAttr {
+def WorkGroupSizeHint :  InheritableAttr, LanguageOptionsSpecificAttr {
+  let Spellings = [GNU<"work_group_size_hint">,
+                   CXX11<"sycl", "work_group_size_hint">];
+  let Args = [UnsignedArgument<"XDim">,
+              UnsignedArgument<"YDim">,
+              UnsignedArgument<"ZDim">];
+  let Subjects = SubjectList<[Function], ErrorDiag>;
+  let Documentation = [WorkGroupSizeHintAttrDocs];
+  let LangOpts = [NotSYCL];
+  let SupportsNonconformingLambdaSyntax = 1;
+  let ParseKind = "WorkGroupSizeHint";
+  let HasCustomParsing = 1;
+}
+
+def SYCLWorkGroupSizeHint :  InheritableAttr, LanguageOptionsSpecificAttr {
   let Spellings = [GNU<"work_group_size_hint">,
                    CXX11<"sycl", "work_group_size_hint">];
   let Args = [ExprArgument<"XDim">,
               ExprArgument<"YDim", /*optional*/1>,
               ExprArgument<"ZDim", /*optional*/1>];
   let Subjects = SubjectList<[Function], ErrorDiag>;
   // In OpenCL C this attribute takes no default values whereas in SYCL it does. 
-  // To avoid confusing diagnostics, the checks are deferred to "handleWorkGroupSizeHint". 
+  // To avoid confusing diagnostics, the checks are deferred to
+  // "handleWorkGroupSizeHint". 
   let HasCustomParsing = 1;
   let AdditionalMembers = [{
     Optional<llvm::APSInt> getXDimVal() const {
+      // X-dimension is not optional.
       if (const auto *CE = dyn_cast<ConstantExpr>(getXDim()))
         return CE->getResultAsAPSInt();
       return std::nullopt;
     }
     Optional<llvm::APSInt> getYDimVal() const {
-      if (const auto *CE = dyn_cast<ConstantExpr>(getYDim()))
+      // Y-dimension is optional so a nullptr value is allowed.
+      if (const auto *CE = dyn_cast_or_null<ConstantExpr>(getYDim()))
         return CE->getResultAsAPSInt();
       return std::nullopt;
     }
     Optional<llvm::APSInt> getZDimVal() const {
-      if (const auto *CE = dyn_cast<ConstantExpr>(getZDim()))
+      // Z-dimension is optional so a nullptr value is allowed.
+      if (const auto *CE = dyn_cast_or_null<ConstantExpr>(getZDim()))
         return CE->getResultAsAPSInt();
       return std::nullopt;
     }
   }];
   let Documentation = [WorkGroupSizeHintAttrDocs];
+  let LangOpts = [SYCLIsDevice, SilentlyIgnoreSYCLIsHost];
   let SupportsNonconformingLambdaSyntax = 1;
+  let ParseKind = "WorkGroupSizeHint";
 }
 
 def InitPriority : InheritableAttr, TargetSpecificAttr<TargetSupportsInitPriority> {

clang/include/clang/Sema/Sema.h

@@ -10997,10 +10997,10 @@ class Sema final {
   bool AllWorkGroupSizesSame(const Expr *LHSXDim, const Expr *LHSYDim,
                              const Expr *LHSZDim, const Expr *RHSXDim,
                              const Expr *RHSYDim, const Expr *RHSZDim);
-  void AddWorkGroupSizeHintAttr(Decl *D, const AttributeCommonInfo &CI,
-                                Expr *XDim, Expr *YDim, Expr *ZDim);
-  WorkGroupSizeHintAttr *
-  MergeWorkGroupSizeHintAttr(Decl *D, const WorkGroupSizeHintAttr &A);
+  void AddSYCLWorkGroupSizeHintAttr(Decl *D, const AttributeCommonInfo &CI,
+                                    Expr *XDim, Expr *YDim, Expr *ZDim);
+  SYCLWorkGroupSizeHintAttr *
+  MergeSYCLWorkGroupSizeHintAttr(Decl *D, const SYCLWorkGroupSizeHintAttr &A);
   void AddIntelReqdSubGroupSize(Decl *D, const AttributeCommonInfo &CI,
                                 Expr *E);
   IntelReqdSubGroupSizeAttr *

clang/lib/Basic/Targets/AMDGPU.cpp

@@ -415,6 +415,7 @@ void AMDGPUTargetInfo::getTargetDefines(const LangOptions &Opts,
       assert(CanonName.startswith("gfx") && "Invalid amdgcn canonical name");
       Builder.defineMacro(Twine("__") + Twine(CanonName.drop_back(2).upper()) +
                           Twine("__"));
+      Builder.defineMacro("__CUDA_ARCH__", "0");
     }
     if (isAMDGCN(getTriple())) {
       Builder.defineMacro("__amdgcn_processor__",

clang/lib/CodeGen/CodeGenFunction.cpp

@@ -613,12 +613,33 @@ void CodeGenFunction::EmitKernelMetadata(const FunctionDecl *FD,
 
   if (const WorkGroupSizeHintAttr *A = FD->getAttr<WorkGroupSizeHintAttr>()) {
     llvm::Metadata *AttrMDArgs[] = {
-        llvm::ConstantAsMetadata::get(Builder.getInt(*A->getXDimVal())),
-        llvm::ConstantAsMetadata::get(Builder.getInt(*A->getYDimVal())),
-        llvm::ConstantAsMetadata::get(Builder.getInt(*A->getZDimVal()))};
+        llvm::ConstantAsMetadata::get(Builder.getInt32(A->getXDim())),
+        llvm::ConstantAsMetadata::get(Builder.getInt32(A->getYDim())),
+        llvm::ConstantAsMetadata::get(Builder.getInt32(A->getZDim()))};
     Fn->setMetadata("work_group_size_hint", llvm::MDNode::get(Context, AttrMDArgs));
   }
 
+  if (const SYCLWorkGroupSizeHintAttr *A =
+          FD->getAttr<SYCLWorkGroupSizeHintAttr>()) {
+    llvm::Optional<llvm::APSInt> XDimVal = A->getXDimVal();
+    llvm::Optional<llvm::APSInt> YDimVal = A->getYDimVal();
+    llvm::Optional<llvm::APSInt> ZDimVal = A->getZDimVal();
+    llvm::SmallVector<llvm::Metadata *, 3> AttrMDArgs;
+
+    // On SYCL target the dimensions are reversed if present.
+    if (ZDimVal)
+      AttrMDArgs.push_back(
+          llvm::ConstantAsMetadata::get(Builder.getInt(*ZDimVal)));
+    if (YDimVal)
+      AttrMDArgs.push_back(
+          llvm::ConstantAsMetadata::get(Builder.getInt(*YDimVal)));
+    AttrMDArgs.push_back(
+        llvm::ConstantAsMetadata::get(Builder.getInt(*XDimVal)));
+
+    Fn->setMetadata("work_group_size_hint",
+                    llvm::MDNode::get(Context, AttrMDArgs));
+  }
+
   if (const ReqdWorkGroupSizeAttr *A = FD->getAttr<ReqdWorkGroupSizeAttr>()) {
     llvm::Metadata *AttrMDArgs[] = {
         llvm::ConstantAsMetadata::get(Builder.getInt32(A->getXDim())),

clang/lib/Driver/Driver.cpp

@@ -9021,7 +9021,7 @@ const char *Driver::GetNamedOutputPath(Compilation &C, const JobAction &JA,
   if (isSaveTempsEnabled()) {
     // If we're saving temps and the temp file conflicts with any
     // input/resulting file, then avoid overwriting.
-    if (!AtTopLevel) {
+    if (!AtTopLevel && NamedOutput == BaseName) {
       bool SameFile = false;
       SmallString<256> Result;
       llvm::sys::fs::current_path(Result);

clang/lib/Sema/SemaDecl.cpp

@@ -2971,8 +2971,8 @@ static bool mergeDeclAttribute(Sema &S, NamedDecl *D,
     NewAttr = S.MergeSYCLIntelForcePow2DepthAttr(D, *A);
   else if (const auto *A = dyn_cast<SYCLIntelInitiationIntervalAttr>(Attr))
     NewAttr = S.MergeSYCLIntelInitiationIntervalAttr(D, *A);
-  else if (const auto *A = dyn_cast<WorkGroupSizeHintAttr>(Attr))
-    NewAttr = S.MergeWorkGroupSizeHintAttr(D, *A);
+  else if (const auto *A = dyn_cast<SYCLWorkGroupSizeHintAttr>(Attr))
+    NewAttr = S.MergeSYCLWorkGroupSizeHintAttr(D, *A);
   else if (const auto *A = dyn_cast<SYCLIntelMaxGlobalWorkDimAttr>(Attr))
     NewAttr = S.MergeSYCLIntelMaxGlobalWorkDimAttr(D, *A);
   else if (const auto *BTFA = dyn_cast<BTFDeclTagAttr>(Attr))