Merge remote-tracking branch 'origin/sycl' into sycl-graph-develop

.github/CODEOWNERS

@@ -30,7 +30,7 @@ sycl/ @intel/llvm-reviewers-runtime
 # Documentation
 sycl/ReleaseNotes.md @intel/dpcpp-doc-reviewers @tfzhu
 sycl/doc/ @intel/dpcpp-doc-reviewers
-sycl/doc/design/ @intel/dpcpp-specification-reviewers
+sycl/doc/design/ @intel/llvm-reviewers-runtime
 sycl/doc/design/spirv-extensions/ @intel/dpcpp-spirv-doc-reviewers
 sycl/doc/extensions/ @intel/dpcpp-specification-reviewers
 
@@ -44,7 +44,7 @@ sycl/plugins/unified_runtime/ @intel/dpcpp-l0-pi-reviewers
 sycl/plugins/esimd_emulator/ @intel/dpcpp-esimd-reviewers
 
 # CUDA plugin
-sycl/plugins/cuda/ @intel/llvm-reviewers-cuda
+sycl/plugins/**/cuda/ @intel/llvm-reviewers-cuda
 
 # XPTI instrumentation utilities
 xpti/ @intel/llvm-reviewers-runtime
@@ -74,7 +74,7 @@ clang/tools/clang-offload-*/ @intel/dpcpp-tools-reviewers
 ESIMD/ @intel/dpcpp-esimd-reviewers
 esimd/ @intel/dpcpp-esimd-reviewers
 sycl/include/sycl/ext/intel/esimd.hpp @intel/dpcpp-esimd-reviewers
-sycl/doc/extensions/experimental/sycl_ext_intel_esimd/ @intel/dpcpp-esimd-reviewers
+sycl/doc/extensions/**/sycl_ext_intel_esimd/ @intel/dpcpp-esimd-reviewers
 llvm/lib/SYCLLowerIR/CMakeLists.txt @intel/dpcpp-tools-reviewers @intel/dpcpp-esimd-reviewers
 
 # invoke_simd

.github/workflows/aws.yml

@@ -0,0 +1,36 @@
+name: Start/Stop AWS instance
+
+on:
+ workflow_call:
+ inputs:
+ mode:
+ description: "Mode of operation: start or stop"
+ type: string
+ required: true
+
+ runs-on-list:
+ # See devops/actions/aws-ec2/action.yml for more details.
+ description: "JSON string with array of objects with aws-type, runs-on, aws-ami, aws-spot, aws-disk, aws-timebomb, one-job properties"
+ type: string
+ required: true
+
+jobs:
+ aws:
+ runs-on: ubuntu-20.04
+ environment: aws
+ steps:
+ - name: Setup script
+ run: |
+ mkdir -p ./aws-ec2
+ wget raw.githubusercontent.com/intel/llvm/sycl/devops/actions/aws-ec2/action.yml -P ./aws-ec2
+ wget raw.githubusercontent.com/intel/llvm/sycl/devops/actions/aws-ec2/aws-ec2.js -P ./aws-ec2
+ wget raw.githubusercontent.com/intel/llvm/sycl/devops/actions/aws-ec2/package.json -P ./aws-ec2
+ npm install ./aws-ec2
+ - name: Start AWS EC2 runners
+ uses: ./aws-ec2
+ with:
+ mode: ${{ inputs.mode }}
+ runs-on-list: ${{ inputs.runs-on-list }}
+ GH_PERSONAL_ACCESS_TOKEN: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
+ AWS_ACCESS_KEY: ${{ secrets.AWS_ACCESS_KEY }}
+ AWS_SECRET_KEY: ${{ secrets.AWS_SECRET_KEY }}

.github/workflows/sycl_linux_build_and_test.yml

@@ -183,23 +183,11 @@ jobs:
  name: Start AWS
  needs: build
  if: ${{ inputs.lts_aws_matrix != '[]' }}
- runs-on: ubuntu-20.04
- environment: aws
- steps:
- - name: Setup script
- run: |
- mkdir -p ./aws-ec2
- wget raw.githubusercontent.com/intel/llvm/sycl/devops/actions/aws-ec2/action.yml -P ./aws-ec2
- wget raw.githubusercontent.com/intel/llvm/sycl/devops/actions/aws-ec2/aws-ec2.js -P ./aws-ec2
- wget raw.githubusercontent.com/intel/llvm/sycl/devops/actions/aws-ec2/package.json -P ./aws-ec2
- npm install ./aws-ec2
- - name: Start AWS EC2 runners
- uses: ./aws-ec2
- with:
- runs-on-list: ${{ inputs.lts_aws_matrix }}
- GH_PERSONAL_ACCESS_TOKEN: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
- AWS_ACCESS_KEY: ${{ secrets.AWS_ACCESS_KEY }}
- AWS_SECRET_KEY: ${{ secrets.AWS_SECRET_KEY }}
+ uses: ./.github/workflows/aws.yml
+ secrets: inherit
+ with:
+ mode: start
+ runs-on-list: ${{ inputs.lts_aws_matrix }}
 
  e2e-tests:
  needs: [build, aws-start]
@@ -301,21 +289,8 @@ jobs:
  # Always attempt to shutdown AWS instance, even if AWS start was not
  # successful.
  if: ${{ always() && inputs.lts_aws_matrix != '[]' }}
- runs-on: ubuntu-20.04
- environment: aws
- steps:
- - name: Setup script
- run: |
- mkdir -p ./aws-ec2
- wget raw.githubusercontent.com/intel/llvm/sycl/devops/actions/aws-ec2/action.yml -P ./aws-ec2
- wget raw.githubusercontent.com/intel/llvm/sycl/devops/actions/aws-ec2/aws-ec2.js -P ./aws-ec2
- wget raw.githubusercontent.com/intel/llvm/sycl/devops/actions/aws-ec2/package.json -P ./aws-ec2
- npm install ./aws-ec2
- - name: Stop AWS EC2 runners
- uses: ./aws-ec2
- with:
- runs-on-list: ${{ inputs.lts_aws_matrix }}
- mode: stop
- GH_PERSONAL_ACCESS_TOKEN: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
- AWS_ACCESS_KEY: ${{ secrets.AWS_ACCESS_KEY }}
- AWS_SECRET_KEY: ${{ secrets.AWS_SECRET_KEY }}
+ uses: ./.github/workflows/aws.yml
+ secrets: inherit
+ with:
+ mode: stop
+ runs-on-list: ${{ inputs.lts_aws_matrix }}

.github/workflows/sycl_nightly.yml

@@ -24,7 +24,7 @@ jobs:
  with:
  build_cache_root: "/__w/"
  build_artifact_suffix: default-2204
- build_configure_extra_args: ''
+ build_configure_extra_args: '--hip --cuda --enable-esimd-emulator'
 
  ubuntu2204_opaque_pointers_build_test:
  if: github.repository == 'intel/llvm'

.github/workflows/sycl_post_commit.yml

@@ -4,16 +4,7 @@ on:
  push:
  branches:
  - sycl
- pull_request:
- branches:
- - sycl
- paths:
- - .github/workflows/sycl_post_commit.yml
- - .github/workflows/sycl_gen_test_matrix.yml
- - .github/workflows/sycl_linux_build_and_test.yml
- - .github/workflows/sycl_windows_build_and_test.yml
- - .github/workflows/sycl_macos_build_and_test.yml
- workflow_dispatch:
+ - test-devops-pr/**
 
 jobs:
  # This job generates matrix of tests for SYCL End-to-End tests

.github/workflows/sycl_precommit.yml

@@ -31,19 +31,29 @@ jobs:
  uses: ./.github/workflows/sycl_detect_changes.yml
 
  lint:
- runs-on: ubuntu-22.04
+ runs-on: cuda
  container:
  image: ghcr.io/intel/llvm/sycl_ubuntu2204_nightly:no-drivers
+ # actions/checkout fails without "--privileged".
+ options: -u 1001:1001 --privileged
  steps:
- - name: 'PR commits + 1'
- run: echo "PR_FETCH_DEPTH=$(( ${{ github.event.pull_request.commits }} + 1 ))" >> "${GITHUB_ENV}"
  - uses: actions/checkout@v3
  with:
- ref: ${{ github.event.pull_request.head.sha }}
- persist-credentials: false
+ sparse-checkout: |
+ devops/actions/cached_checkout
+ - name: 'PR commits + 2'
+ run: echo "PR_FETCH_DEPTH=$(( ${{ github.event.pull_request.commits }} + 2 ))" >> "${GITHUB_ENV}"
+ - uses: ./devops/actions/cached_checkout
+ with:
+ path: src
  fetch-depth: ${{ env.PR_FETCH_DEPTH }}
+ ref: ${{ github.event.pull_request.head.sha }}
+ cache_path: "/__w/repo_cache/"
+ merge: false
  - name: Run clang-format
- uses: ./devops/actions/clang-format
+ uses: ./src/devops/actions/clang-format
+ with:
+ path: src
 
  # This job generates matrix of tests for SYCL End-to-End tests
  test_matrix:

clang/include/clang/AST/PrettyPrinter.h

@@ -212,6 +212,9 @@ struct PrintingPolicy {
  /// \endcode
  unsigned SuppressTypedefs : 1;
 
+ /// When true, suppress printing final specifier.
+ unsigned SuppressFinalSpecifier : 1;
+
  /// When true, suppresses printing template arguments in names of C++
  /// constructors.
  unsigned SuppressTemplateArgsInCXXConstructors : 1;
@@ -220,9 +223,6 @@ struct PrintingPolicy {
  /// argument for the parameter.
  unsigned SuppressDefaultTemplateArgs : 1;
 
- /// When true, suppress printing final specifier.
- unsigned SuppressFinalSpecifier : 1;
-
  /// Whether we can use 'bool' rather than '_Bool' (even if the language
  /// doesn't actually have 'bool', because, e.g., it is defined as a macro).
  unsigned Bool : 1;

clang/include/clang/Basic/CodeGenOptions.def

@@ -373,6 +373,10 @@ CODEGENOPT(VirtualFunctionElimination, 1, 0) ///< Whether to apply the dead
  /// virtual function elimination
  /// optimization.
 
+/// Whether accuracy levels for math library functions are requested by the
+/// user. These accuracy levels will then be expressed in terms of ULPs.
+CODEGENOPT(FPAccuracy, 1, 0)
+
 /// Whether to use public LTO visibility for entities in std and stdext
 /// namespaces. This is enabled by clang-cl's /MT and /MTd flags.
 CODEGENOPT(LTOVisibilityPublicStd, 1, 0)

clang/include/clang/Basic/DiagnosticCommonKinds.td

@@ -301,6 +301,9 @@ def warn_stack_clash_protection_inline_asm : Warning<
 def warn_slh_does_not_support_asm_goto : Warning<
  "speculative load hardening does not protect functions with asm goto">,
  InGroup<DiagGroup<"slh-asm-goto">>;
+
+def err_drv_incompatible_options : Error<
+ "the combination of '%0' and '%1' is incompatible">;
 }
 
 // Sema && Serialization

clang/include/clang/Basic/DiagnosticDriverKinds.td

@@ -63,6 +63,10 @@ def err_drv_no_cuda_libdevice : Error<
  "via '--cuda-path', or pass '-nocudalib' to build without linking with "
  "libdevice">;
 
+def warn_function_fp_accuracy_already_set : Warning <
+ "floating point accuracy value of '%0' has already been assigned to "
+ "function '%1'">,
+ InGroup<DiagGroup<"fp-accuracy-already-set">>;
 def err_drv_no_rocm_device_lib : Error<
  "cannot find ROCm device library%select{| for %1|for ABI version %1}0; provide its path via "
  "'--rocm-path' or '--rocm-device-lib-path', or pass '-nogpulib' to build "
@@ -141,8 +145,9 @@ def err_drv_invalid_unwindlib_name : Error<
  "invalid unwind library name in argument '%0'">;
 def err_drv_incompatible_unwindlib : Error<
  "--rtlib=libgcc requires --unwindlib=libgcc">;
-def err_drv_incompatible_options : Error<
- "the combination of '%0' and '%1' is incompatible">;
+def err_drv_incompatible_fp_accuracy_options : Error<
+ "floating point accuracy requirements cannot be guaranteed when '-fmath-errno' "
+ "is enabled; use '-fno-math-errno' to enable floating point accuracy control">;
 def err_drv_invalid_stdlib_name : Error<
  "invalid library name in argument '%0'">;
 def err_drv_invalid_output_with_multiple_archs : Error<
@@ -355,6 +360,8 @@ def err_drv_expecting_fsycl_with_sycl_opt : Error<
  "'%0' must be used in conjunction with '-fsycl' to enable offloading">;
 def err_drv_fsycl_with_c_type : Error<
  "'%0' must not be used in conjunction with '-fsycl', which expects C++ source">;
+def err_drv_fsycl_with_pch : Error<
+ "Precompiled header generation is not supported with '-fsycl'">;
 def err_drv_fsycl_unsupported_with_opt
  : Error<"'%0' is not supported with '-fsycl'">;
 def err_drv_sycl_missing_amdgpu_arch : Error<

clang/include/clang/Basic/FPOptions.def

@@ -26,5 +26,6 @@ OPTION(AllowReciprocal, bool, 1, NoSignedZero)
 OPTION(AllowApproxFunc, bool, 1, AllowReciprocal)
 OPTION(FPEvalMethod, LangOptions::FPEvalMethodKind, 2, AllowApproxFunc)
 OPTION(Float16ExcessPrecision, LangOptions::ExcessPrecisionKind, 2, FPEvalMethod)
-OPTION(BFloat16ExcessPrecision, LangOptions::ExcessPrecisionKind, 2, FPEvalMethod)
+OPTION(BFloat16ExcessPrecision, LangOptions::ExcessPrecisionKind, 2, Float16ExcessPrecision)
+OPTION(FPAccuracy, LangOptions::FPAccuracyKind, 3, BFloat16ExcessPrecision)
 #undef OPTION

clang/include/clang/Basic/LangOptions.def

@@ -338,6 +338,7 @@ BENIGN_ENUM_LANGOPT(FPExceptionMode, FPExceptionModeKind, 2, FPE_Default, "FP Ex
 BENIGN_ENUM_LANGOPT(FPEvalMethod, FPEvalMethodKind, 2, FEM_UnsetOnCommandLine, "FP type used for floating point arithmetic")
 ENUM_LANGOPT(Float16ExcessPrecision, ExcessPrecisionKind, 2, FPP_Standard, "Intermediate truncation behavior for Float16 arithmetic")
 ENUM_LANGOPT(BFloat16ExcessPrecision, ExcessPrecisionKind, 2, FPP_Standard, "Intermediate truncation behavior for BFloat16 arithmetic")
+BENIGN_ENUM_LANGOPT(FPAccuracy, FPAccuracyKind, 3, FPA_Default, "Accuracy for floating point operations and library functions")
 LANGOPT(NoBitFieldTypeAlign , 1, 0, "bit-field type alignment")
 LANGOPT(HexagonQdsp6Compat , 1, 0, "hexagon-qdsp6 backward compatibility")
 LANGOPT(ObjCAutoRefCount , 1, 0, "Objective-C automated reference counting")

clang/include/clang/Basic/LangOptions.h

@@ -303,6 +303,15 @@ class LangOptions : public LangOptionsBase {
 
  enum ExcessPrecisionKind { FPP_Standard, FPP_Fast, FPP_None };
 
+ enum FPAccuracyKind {
+ FPA_Default,
+ FPA_High,
+ FPA_Medium,
+ FPA_Low,
+ FPA_Sycl,
+ FPA_Cuda,
+ };
+
  /// Possible exception handling behavior.
  enum class ExceptionHandlingKind { None, SjLj, WinEH, DwarfCFI, Wasm };
 
@@ -509,6 +518,10 @@ class LangOptions : public LangOptionsBase {
  /// records.
  std::string OptRecordFile;
 
+ std::string FPAccuracyVal;
+ using FPAccuracyFuncMapTy = std::map<std::string, std::string>;
+ FPAccuracyFuncMapTy FPAccuracyFuncMap;
+
  LangOptions();
 
  /// Set language defaults for the given input language and

clang/include/clang/Driver/Action.h

@@ -79,7 +79,6 @@ class Action {
  OffloadPackagerJobClass,
  OffloadDepsJobClass,
  SPIRVTranslatorJobClass,
- SPIRCheckJobClass,
  SYCLPostLinkJobClass,
  BackendCompileJobClass,
  FileTableTformJobClass,
@@ -750,21 +749,6 @@ class SPIRVTranslatorJobAction : public JobAction {
  }
 };
 
-// Provides a check of the given input file for the existence of SPIR kernel
-// code. This is currently only used for FPGA specific tool chains and can
-// be expanded to perform other SPIR checks if needed.
-// TODO: No longer being used for FPGA (or elsewhere), cleanup needed.
-class SPIRCheckJobAction : public JobAction {
- void anchor() override;
-
-public:
- SPIRCheckJobAction(Action *Input, types::ID OutputType);
-
- static bool classof(const Action *A) {
- return A->getKind() == SPIRCheckJobClass;
- }
-};
-
 class SYCLPostLinkJobAction : public JobAction {
  void anchor() override;
 

clang/include/clang/Driver/Options.td

@@ -1729,6 +1729,13 @@ def ffp_exception_behavior_EQ : Joined<["-"], "ffp-exception-behavior=">, Group<
  Values<"ignore,maytrap,strict">, NormalizedValuesScope<"LangOptions">,
  NormalizedValues<["FPE_Ignore", "FPE_MayTrap", "FPE_Strict"]>,
  MarshallingInfoEnum<LangOpts<"FPExceptionMode">, "FPE_Default">;
+def ffp_accuracy_EQ : Joined<["-"], "ffp-accuracy=">, Group<f_Group>, Flags<[CC1Option]>,
+ HelpText<"Specifies the required accuracy for floating-point operations and library calls.">,
+ Values<"default,high,medium,low,sycl,cuda">, NormalizedValuesScope<"LangOptions">,
+ NormalizedValues<["FPA_Default", "FPA_High", "FPA_Medium", "FPA_Low", "FPA_Sycl", "FPA_Cuda"]>,
+ MarshallingInfoEnum<LangOpts<"FPAccuracy">, "FPA_Default">;
+def ffp_builtin_accuracy_EQ : Joined<["-"], "ffp-builtin-accuracy=">, Group<f_Group>, Flags<[CC1Option]>;
+
 defm fast_math : BoolFOption<"fast-math",
  LangOpts<"FastMath">, DefaultFalse,
  PosFlag<SetTrue, [CC1Option, CoreOption, FC1Option, FlangOption], "Allow aggressive, lossy floating-point optimizations",
@@ -7020,6 +7027,14 @@ class CLRemainingArgsJoined<string name> : Option<["/", "-"], name,
 // (We don't put any of these in cl_compile_Group as the options they alias are
 // already in the right group.)
 
+// INTEL_CUSTOMIZATION
+def _SLASH_Qfp_accuracy_EQ : CLJoined<"Qfp-accuracy=">,
+ Alias<ffp_accuracy_EQ>;
+def _SLASH_Qfp_accuracy_COL : CLJoined<"Qfp-accuracy:">,
+ Alias<ffp_accuracy_EQ>,HelpText<"Specifies the required accuracy for "
+ "floating-point operations and library calls.">;
+// END INTEL_CUSTOMIZATION
+
 def _SLASH_Brepro : CLFlag<"Brepro">,
  HelpText<"Do not write current time into COFF output (breaks link.exe /incremental)">,
  Alias<mno_incremental_linker_compatible>;

clang/include/clang/Driver/ToolChain.h

@@ -160,7 +160,6 @@ class ToolChain {
  mutable std::unique_ptr<Tool> OffloadPackager;
  mutable std::unique_ptr<Tool> OffloadDeps;
  mutable std::unique_ptr<Tool> SPIRVTranslator;
- mutable std::unique_ptr<Tool> SPIRCheck;
  mutable std::unique_ptr<Tool> SYCLPostLink;
  mutable std::unique_ptr<Tool> BackendCompiler;
  mutable std::unique_ptr<Tool> AppendFooter;
@@ -180,7 +179,6 @@ class ToolChain {
  Tool *getOffloadPackager() const;
  Tool *getOffloadDeps() const;
  Tool *getSPIRVTranslator() const;
- Tool *getSPIRCheck() const;
  Tool *getSYCLPostLink() const;
  Tool *getBackendCompiler() const;
  Tool *getAppendFooter() const;

clang/include/clang/Frontend/CompilerInvocation.h

@@ -276,6 +276,9 @@ class CompilerInvocation : public CompilerInvocationRefBase,
  std::vector<std::string> &Includes,
  DiagnosticsEngine &Diags);
 
+ static void ParseFpAccuracyArgs(LangOptions &Opts, llvm::opt::ArgList &Args,
+ DiagnosticsEngine &Diags);
+
  /// Generate command line options from LangOptions.
  static void GenerateLangArgs(const LangOptions &Opts,
  SmallVectorImpl<const char *> &Args,