Merge branch 'sycl' into llvmspirv_pulldown

Author: Chenyang-L

.github/workflows/linux_matrix_e2e_on_nightly.yml

@@ -28,7 +28,7 @@ jobs:
             reset_gpu: false
 
           - name: Intel
-            runner: '["Linux", "gen9"]'
+            runner: '["Linux", "gen12"]'
             image: ghcr.io/intel/llvm/sycl_ubuntu2204_nightly:latest
             extra_image_opts:
             extra_cmake_args:

.github/workflows/sycl_detect_changes.yml

@@ -15,7 +15,7 @@ jobs:
     runs-on: [Linux, build]
     timeout-minutes: 3
     outputs:
-      filters: ${{ steps.changes.outputs.changes }}
+      filters: ${{ steps.result.outputs.result }}
     steps:
       - name: Check file changes
         uses: dorny/paths-filter@4512585405083f25c027a35db413c2b3b9006d50
@@ -59,3 +59,18 @@ jobs:
               - *drivers_and_configs
               # Temporary, until plugins are enabled in nightly image.
               - sycl/**
+
+      - name: Set output
+        id: result
+        uses: actions/github-script@v6
+        with:
+          script: |
+            console.log("Number of files changed:");
+            console.log(context.payload.pull_request.changed_files);
+            if (context.payload.pull_request.changed_files < 500) {
+              return '${{ steps.changes.outputs.changes }}';
+            }
+            // Treat everything as changed for huge PRs.
+            return ["llvm", "llvm_spirv", "clang", "sycl_fusion", "xptifw", "libclc", "sycl", "ci", "drivers_and_configs", "test_build"];
+
+      - run: echo '${{ steps.result.outputs.result }}'

.github/workflows/sycl_linux_build_and_test.yml

@@ -234,7 +234,7 @@ jobs:
       target_devices: ${{ matrix.targets }}
       ref: ${{ inputs.build_ref || github.sha }}
       merge_ref: ${{ inputs.merge_ref }}
-      reset_gpu: ${{ contains(matrix.runs-on, 'gen9') && contains(matrix.runs-on, 'Linux') }}
+      reset_gpu: ${{ contains(matrix.runs-on, 'gen12') && contains(matrix.runs-on, 'Linux') }}
 
       sycl_toolchain_artifact: sycl_linux_${{ inputs.build_artifact_suffix }}
       sycl_toolchain_archive: ${{ inputs.artifact_archive_name }}
@@ -257,7 +257,7 @@ jobs:
       options: ${{ matrix.container_options }}
     steps:
     - name: Reset GPU
-      if: ${{ contains(matrix.config, 'gen9') }}
+      if: ${{ contains(matrix.config, 'gen12') }}
       run: |
         sudo mount -t debugfs none /sys/kernel/debug
         sudo bash -c 'echo 1 > /sys/kernel/debug/dri/0/i915_wedged'

.github/workflows/sycl_nightly.yml

@@ -12,8 +12,8 @@ jobs:
     uses: ./.github/workflows/sycl_gen_test_matrix.yml
     with:
       # Restore once plugins are enabled in the build.
-      # lts_config: "hip_amdgpu;ocl_gen9;ocl_x64;l0_gen9;esimd_emu;cuda_aws;win_l0_gen12"
-      lts_config: "ocl_gen9;ocl_x64;l0_gen9;win_l0_gen12"
+      # lts_config: "hip_amdgpu;ocl_gen12;ocl_x64;l0_gen12;esimd_emu;cuda_aws;win_l0_gen12"
+      lts_config: "ocl_gen12;ocl_x64;l0_gen12;win_l0_gen12"
 
   ubuntu2204_build_test:
     if: github.repository == 'intel/llvm'

.github/workflows/sycl_post_commit.yml

@@ -5,6 +5,7 @@ on:
     branches:
     - sycl
     - sycl-devops-pr/**
+    - llvmspirv_pulldown
 
 jobs:
   # This job generates matrix of tests for SYCL End-to-End tests
@@ -13,7 +14,7 @@ jobs:
     if: github.repository == 'intel/llvm'
     uses: ./.github/workflows/sycl_gen_test_matrix.yml
     with:
-      lts_config: "l0_gen9;win_l0_gen12"
+      lts_config: "l0_gen12;win_l0_gen12"
   linux_self_prod:
     name: Linux (Self build + shared libraries + no-assertions)
     if: github.repository == 'intel/llvm'

.github/workflows/sycl_precommit_linux.yml

@@ -5,6 +5,7 @@ on:
     branches:
     - sycl
     - sycl-devops-pr/**
+    - llvmspirv_pulldown
     # Do not run builds if changes are only in the following locations
     paths-ignore:
     - '.github/ISSUE_TEMPLATE/**'
@@ -28,6 +29,7 @@ jobs:
   lint:
     needs: [detect_changes]
     if: |
+      github.event.pull_request.head.repo.full_name == 'intel/llvm' ||
       !contains(needs.detect_changes.outputs.filters, 'ci')
     runs-on: [Linux, build]
     container:
@@ -61,6 +63,7 @@ jobs:
   test_matrix:
     needs: [detect_changes]
     if: |
+      github.event.pull_request.head.repo.full_name == 'intel/llvm' ||
       !contains(needs.detect_changes.outputs.filters, 'ci')
     name: Generate Test Matrix
     uses: ./.github/workflows/sycl_gen_test_matrix.yml
@@ -76,8 +79,8 @@ jobs:
     if: |
       always()
       && (success() || contains(github.event.pull_request.labels.*.name, 'ignore-lint'))
-      && contains(needs.detect_changes.outputs.filters, 'test_build')
-      && !contains(needs.detect_changes.outputs.filters, 'ci')
+      && (github.event.pull_request.head.repo.full_name == 'intel/llvm'
+          || !contains(needs.detect_changes.outputs.filters, 'ci'))
     uses: ./.github/workflows/sycl_linux_build_and_test.yml
     secrets: inherit
     with:
@@ -90,14 +93,3 @@ jobs:
       lts_matrix: ${{ needs.test_matrix.outputs.lts_lx_matrix }}
       lts_aws_matrix: ${{ needs.test_matrix.outputs.lts_aws_matrix }}
       check_filters: ${{ needs.detect_changes.outputs.filters }}
-
-  linux_e2e_on_nightly:
-    name: Linux SYCL E2E on Nightly
-    needs: [detect_changes]
-    if: |
-      !contains(needs.detect_changes.outputs.filters, 'test_build')
-      && !contains(needs.detect_changes.outputs.filters, 'ci')
-    uses: ./.github/workflows/linux_matrix_e2e_on_nightly.yml
-    secrets: inherit
-    with:
-      ref: ${{ github.event.pull_request.head.sha }}

.github/workflows/sycl_precommit_windows.yml

@@ -5,6 +5,7 @@ on:
     branches:
     - sycl
     - sycl-devops-pr/**
+    - llvmspirv_pulldown
     # Do not run builds if changes are only in the following locations
     paths-ignore:
     - '.github/ISSUE_TEMPLATE/**'
@@ -28,6 +29,7 @@ jobs:
   lint:
     needs: [detect_changes]
     if: |
+      github.event.pull_request.head.repo.full_name == 'intel/llvm' ||
       !contains(needs.detect_changes.outputs.filters, 'ci')
     runs-on: [Linux, build]
     container:
@@ -61,6 +63,7 @@ jobs:
   test_matrix:
     needs: [detect_changes]
     if: |
+      github.event.pull_request.head.repo.full_name == 'intel/llvm' ||
       !contains(needs.detect_changes.outputs.filters, 'ci')
     name: Generate Test Matrix
     uses: ./.github/workflows/sycl_gen_test_matrix.yml
@@ -75,7 +78,8 @@ jobs:
       always()
       && (success() || contains(github.event.pull_request.labels.*.name, 'ignore-lint'))
       && github.repository == 'intel/llvm'
-      && !contains(needs.detect_changes.outputs.filters, 'ci')
+      && (github.event.pull_request.head.repo.full_name == 'intel/llvm'
+          || !contains(needs.detect_changes.outputs.filters, 'ci'))
     uses: ./.github/workflows/sycl_windows_build_and_test.yml
     with:
       lts_matrix: ${{ needs.test_matrix.outputs.lts_wn_matrix }}

clang/docs/ClangOffloadPackager.rst

@@ -112,6 +112,8 @@ the following values for the :ref:`offload kind<table-offload_kind>` and the
     +------------+-------+---------------------------------------+
     | OFK_HIP    | 0x03  | The producer was HIP                  |
     +------------+-------+---------------------------------------+
+    | OFK_SYCL   | 0x04  | The producer was SYCL                 |
+    +------------+-------+---------------------------------------+
 
 The flags are used to signify certain conditions, such as the presence of
 debugging information or whether or not LTO was used. The string entry table is

clang/include/clang/Basic/Builtins.def

@@ -1757,6 +1757,9 @@ BUILTIN(__builtin_ms_va_copy, "vc*&c*&", "n")
 // Arithmetic Fence: to prevent FP reordering and reassociation optimizations
 LANGBUILTIN(__arithmetic_fence, "v.", "tE", ALL_LANGUAGES)
 
+// Builtins for Intel SYCL
+BUILTIN(__builtin_intel_sycl_ptr_annotation, "v.", "nt")
+
 // Builtins for Intel FPGA
 BUILTIN(__builtin_intel_fpga_reg, "v.", "nt")
 BUILTIN(__builtin_intel_fpga_mem, "v.", "nt")

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -167,6 +167,13 @@ def err_intel_fpga_mem_arg_mismatch
             "a pointer"
             "|a non-negative integer constant}0">;
 
+def err_intel_sycl_ptr_annotation_arg_number_mismatch
+    :Error<"number of parameters must be odd number">;
+def err_intel_sycl_ptr_annotation_mismatch
+    : Error<"builtin parameter must be %select{"
+            "a pointer"
+            "|a string literal or constexpr const char*}0">;
+
 // C99 variable-length arrays
 def ext_vla : Extension<"variable length arrays are a C99 feature">,
   InGroup<VLAExtension>;

clang/include/clang/Sema/Sema.h

@@ -14052,6 +14052,9 @@ class Sema final {
   bool CheckIntelFPGARegBuiltinFunctionCall(unsigned BuiltinID, CallExpr *Call);
   bool CheckIntelFPGAMemBuiltinFunctionCall(CallExpr *Call);
 
+  bool CheckIntelSYCLPtrAnnotationBuiltinFunctionCall(unsigned BuiltinID,
+                                                      CallExpr *Call);
+
   bool SemaBuiltinVAStart(unsigned BuiltinID, CallExpr *TheCall);
   bool SemaBuiltinVAStartARMMicrosoft(CallExpr *Call);
   bool SemaBuiltinUnorderedCompare(CallExpr *TheCall);

clang/lib/CodeGen/CGBuiltin.cpp

@@ -5564,7 +5564,8 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
     return EmitIntelFPGARegBuiltin(E, ReturnValue);
   case Builtin::BI__builtin_intel_fpga_mem:
     return EmitIntelFPGAMemBuiltin(E);
-
+  case Builtin::BI__builtin_intel_sycl_ptr_annotation:
+    return EmitIntelSYCLPtrAnnotationBuiltin(E);
   case Builtin::BI__builtin_get_device_side_mangled_name: {
     auto Name = CGM.getCUDARuntime().getDeviceSideName(
         cast<DeclRefExpr>(E->getArg(0)->IgnoreImpCasts())->getDecl());
@@ -22533,6 +22534,12 @@ RValue CodeGenFunction::EmitIntelFPGAMemBuiltin(const CallExpr *E) {
   return RValue::get(Ann);
 }
 
+static bool hasFuncNameRequestedFPAccuracy(StringRef Name,
+                                           const LangOptions &LangOpts) {
+  auto FuncMapIt = LangOpts.FPAccuracyFuncMap.find(Name.str());
+  return (FuncMapIt != LangOpts.FPAccuracyFuncMap.end());
+}
+
 llvm::CallInst *CodeGenFunction::EmitFPBuiltinIndirectCall(
     llvm::FunctionType *IRFuncTy, const SmallVectorImpl<llvm::Value *> &IRArgs,
     llvm::Value *FnPtr, const FunctionDecl *FD) {
@@ -22641,9 +22648,53 @@ llvm::CallInst *CodeGenFunction::EmitFPBuiltinIndirectCall(
   if (!FPAccuracyIntrinsicID)
     return nullptr;
 
-  Func = CGM.getIntrinsic(FPAccuracyIntrinsicID, IRArgs[0]->getType());
-  return CreateBuiltinCallWithAttr(*this, Name, Func, ArrayRef(IRArgs),
-                                   FPAccuracyIntrinsicID);
+  // Create an intrinsic only if it exists in the map, or if there
+  // a TU fp-accuracy requested.
+  const LangOptions &LangOpts = getLangOpts();
+  if (hasFuncNameRequestedFPAccuracy(Name, LangOpts) ||
+      !LangOpts.FPAccuracyVal.empty()) {
+    Func = CGM.getIntrinsic(FPAccuracyIntrinsicID, IRArgs[0]->getType());
+    return CreateBuiltinCallWithAttr(*this, Name, Func, ArrayRef(IRArgs),
+                                     FPAccuracyIntrinsicID);
+  }
+  return nullptr;
+}
+
+RValue CodeGenFunction::EmitIntelSYCLPtrAnnotationBuiltin(const CallExpr *E) {
+  const Expr *PtrArg = E->getArg(0);
+  Value *PtrVal = EmitScalarExpr(PtrArg);
+  auto &Ctx = CGM.getContext();
+
+  // Create the pointer annotation.
+  Function *F = CGM.getIntrinsic(llvm::Intrinsic::ptr_annotation,
+                                 {PtrVal->getType(), CGM.ConstGlobalsPtrTy});
+
+  SmallString<256> AnnotStr;
+  llvm::raw_svector_ostream Out(AnnotStr);
+
+  SmallVector<std::pair<std::string, std::string>, 4> Properties;
+
+  for (unsigned I = 1, N = E->getNumArgs(); I <= N / 2; I++) {
+    auto Arg = E->getArg(I)->IgnoreParenCasts();
+    const StringLiteral *Str = dyn_cast<const StringLiteral>(Arg);
+    Expr::EvalResult Result;
+    if (!Str && Arg->EvaluateAsRValue(Result, Ctx) && Result.Val.isLValue()) {
+      const auto *LVE = Result.Val.getLValueBase().dyn_cast<const Expr *>();
+      Str = dyn_cast<const StringLiteral>(LVE);
+    }
+    assert(Str && "Constant parameter string is invalid?");
+
+    auto IntVal = E->getArg(I + N / 2)->getIntegerConstantExpr(Ctx);
+    assert(IntVal.has_value() &&
+           "Constant integer arg isn't actually constant?");
+
+    Properties.push_back(
+        std::make_pair(Str->getString().str(), toString(IntVal.value(), 10)));
+  }
+
+  llvm::Value *Ann =
+      EmitSYCLAnnotationCall(F, PtrVal, E->getExprLoc(), Properties);
+  return RValue::get(Ann);
 }
 
 Value *CodeGenFunction::EmitRISCVBuiltinExpr(unsigned BuiltinID,

clang/lib/CodeGen/CGCall.cpp

@@ -1907,7 +1907,7 @@ void CodeGenModule::getDefaultFunctionFPAccuracyAttributes(
       StringRef FPAccuracyVal = llvm::fp::getAccuracyForFPBuiltin(
           ID, FuncType, convertFPAccuracy(FuncMapIt->second));
       assert(!FPAccuracyVal.empty() && "A valid accuracy value is expected");
-      FuncAttrs.addAttribute("fpbuiltin-max-error=", FPAccuracyVal);
+      FuncAttrs.addAttribute("fpbuiltin-max-error", FPAccuracyVal);
       MD = llvm::ConstantAsMetadata::get(llvm::ConstantInt::get(
           Int32Ty, convertFPAccuracyToAspect(FuncMapIt->second)));
     }
@@ -1917,7 +1917,7 @@ void CodeGenModule::getDefaultFunctionFPAccuracyAttributes(
       StringRef FPAccuracyVal = llvm::fp::getAccuracyForFPBuiltin(
           ID, FuncType, convertFPAccuracy(getLangOpts().FPAccuracyVal));
       assert(!FPAccuracyVal.empty() && "A valid accuracy value is expected");
-      FuncAttrs.addAttribute("fpbuiltin-max-error=", FPAccuracyVal);
+      FuncAttrs.addAttribute("fpbuiltin-max-error", FPAccuracyVal);
       MD = llvm::ConstantAsMetadata::get(llvm::ConstantInt::get(
           Int32Ty, convertFPAccuracyToAspect(getLangOpts().FPAccuracyVal)));
     }

clang/lib/CodeGen/CodeGenFunction.cpp

@@ -2812,13 +2812,25 @@ Address CodeGenFunction::EmitFieldAnnotations(const FieldDecl *D,
 llvm::Value *CodeGenFunction::EmitSYCLAnnotationCall(
     llvm::Function *AnnotationFn, llvm::Value *AnnotatedVal,
     SourceLocation Location, const SYCLAddIRAnnotationsMemberAttr *Attr) {
+
+  llvm::SmallVector<std::pair<std::string, std::string>, 4>
+      AnnotationNameValPairs =
+          Attr->getFilteredAttributeNameValuePairs(getContext());
+  return EmitSYCLAnnotationCall(AnnotationFn, AnnotatedVal, Location,
+                                AnnotationNameValPairs);
+}
+
+llvm::Value *CodeGenFunction::EmitSYCLAnnotationCall(
+    llvm::Function *AnnotationFn, llvm::Value *AnnotatedVal,
+    SourceLocation Location,
+    SmallVectorImpl<std::pair<std::string, std::string>> &Pair) {
   SmallVector<llvm::Value *, 5> Args = {
       AnnotatedVal,
       Builder.CreateBitCast(CGM.EmitAnnotationString("sycl-properties"),
                             ConstGlobalsPtrTy),
       Builder.CreateBitCast(CGM.EmitAnnotationUnit(Location),
                             ConstGlobalsPtrTy),
-      CGM.EmitAnnotationLineNo(Location), CGM.EmitSYCLAnnotationArgs(Attr)};
+      CGM.EmitAnnotationLineNo(Location), CGM.EmitSYCLAnnotationArgs(Pair)};
   return Builder.CreateCall(AnnotationFn, Args);
 }
 

clang/lib/CodeGen/CodeGenFunction.h

@@ -4333,6 +4333,8 @@ class CodeGenFunction : public CodeGenTypeCache {
                                  ReturnValueSlot ReturnValue);
   RValue EmitIntelFPGAMemBuiltin(const CallExpr *E);
 
+  RValue EmitIntelSYCLPtrAnnotationBuiltin(const CallExpr *E);
+
   llvm::CallInst *
   EmitFPBuiltinIndirectCall(llvm::FunctionType *IRFuncTy,
                             const SmallVectorImpl<llvm::Value *> &IRArgs,
@@ -4601,6 +4603,11 @@ class CodeGenFunction : public CodeGenTypeCache {
                          llvm::Value *AnnotatedVal, SourceLocation Location,
                          const SYCLAddIRAnnotationsMemberAttr *Attr);
 
+  llvm::Value *EmitSYCLAnnotationCall(
+      llvm::Function *AnnotationFn, llvm::Value *AnnotatedVal,
+      SourceLocation Location,
+      llvm::SmallVectorImpl<std::pair<std::string, std::string>> &Pair);
+
   /// Emit sycl field annotations for given field & value. Returns the
   /// annotation result.
   Address EmitFieldSYCLAnnotations(const FieldDecl *D, Address V);

clang/lib/CodeGen/CodeGenModule.cpp

@@ -3484,10 +3484,9 @@ void CodeGenModule::AddGlobalAnnotations(const ValueDecl *D,
 }
 
 llvm::Constant *CodeGenModule::EmitSYCLAnnotationArgs(
-    const SYCLAddIRAnnotationsMemberAttr *Attr) {
-  llvm::SmallVector<std::pair<std::string, std::string>, 4>
-      AnnotationNameValPairs =
-          Attr->getFilteredAttributeNameValuePairs(getContext());
+    llvm::SmallVectorImpl<std::pair<std::string, std::string>>
+        &AnnotationNameValPairs) {
+
   if (AnnotationNameValPairs.empty())
     return llvm::ConstantPointerNull::get(ConstGlobalsPtrTy);
 

clang/lib/CodeGen/CodeGenModule.h

@@ -1382,8 +1382,8 @@ class CodeGenModule : public CodeGenTypeCache {
   void AddGlobalAnnotations(const ValueDecl *D, llvm::GlobalValue *GV);
 
   /// Emit additional args of the annotation.
-  llvm::Constant *
-  EmitSYCLAnnotationArgs(const SYCLAddIRAnnotationsMemberAttr *Attr);
+  llvm::Constant *EmitSYCLAnnotationArgs(
+      SmallVectorImpl<std::pair<std::string, std::string>> &Pairs);
 
   /// Add attributes from add_ir_attributes_global_variable on TND to GV.
   void AddGlobalSYCLIRAttributes(llvm::GlobalVariable *GV,