Merge branch 'sycl' into steffen/stream_property_lists

Author: steffenlarsen

.github/workflows/clang-format.yml

@@ -0,0 +1,36 @@
+name: SYCL
+
+on:
+  pull_request:
+    branches:
+    - sycl
+
+jobs:
+  lint:
+    # TODO use nightly builds of SYCL to get clang-format
+    runs-on: ubuntu-latest
+    steps:
+    - name: Get clang-format first
+      run: sudo apt-get install -yqq clang-format
+
+    - uses: actions/checkout@v2
+      with:
+        fetch-depth: 2
+
+    - name: Run clang-format
+      uses: ./devops/actions/clang-format
+
+  linux_default:
+    name: Default Linux
+    # Only build and test patches, that have passed all linter checks, because
+    # the next commit is likely to be a follow-up on that job.
+    needs: lint
+    if: always() && (success() || contains(github.event.pull_request.labels.*.name, 'ignore-lint'))
+    uses: intel/llvm/.github/workflows/sycl_linux_build_and_test.yml@sycl
+    with:
+      build_runs_on: "sycl-precommit-linux"
+      build_cache_root: "/__w/"
+      build_cache_size: "8G"
+      build_artifact_suffix: "default"
+      build_cache_suffix: "default"
+

.github/workflows/sycl_precommit.yml

@@ -4,8 +4,8 @@ ocl_cpu_rt_ver=2021.12.9.0.24
 # https://github.com/intel/llvm/releases/download/2021-WW40/win-oclcpuexp-2021.12.9.0.24_rel.zip
 ocl_cpu_rt_ver_win=2021.12.9.0.24
 # Same GPU driver supports Level Zero and OpenCL
-# https://github.com/intel/compute-runtime/releases/tag/21.37.20939
-ocl_gpu_rt_ver=21.37.20939
+# https://github.com/intel/compute-runtime/releases/tag/21.44.21506
+ocl_gpu_rt_ver=21.44.21506
 # Same GPU driver supports Level Zero and OpenCL
 # https://downloadmirror.intel.com/677976/igfx_win_100.9955.zip
 ocl_gpu_rt_ver_win=30.0.100.9955
@@ -30,7 +30,7 @@ ocloc_ver_win=27.20.100.9168
 [DRIVER VERSIONS]
 cpu_driver_lin=2021.12.9.0.24
 cpu_driver_win=2021.12.9.0.24
-gpu_driver_lin=21.37.20939
+gpu_driver_lin=21.44.21506
 gpu_driver_win=30.0.100.9955
 fpga_driver_lin=2021.12.9.0.24
 fpga_driver_win=2021.12.9.0.24

buildbot/dependency.conf

@@ -11451,7 +11451,8 @@ def err_invalid_std_type_in_sycl_kernel : Error<"%0 is an invalid kernel name, "
                                           "%q1 is declared in the 'std' namespace ">;
 
 def err_sycl_kernel_incorrectly_named : Error<
-  "%select{%1 should be globally visible"
+  "%select{%1 is invalid; kernel name should be forward declarable "
+  "at namespace scope"
   "|unscoped enum %1 requires fixed underlying type"
   "|unnamed type %1 is invalid; provide a kernel name, or use "
   "'-fsycl-unnamed-lambda' to enable unnamed kernel lambdas"

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -937,7 +937,7 @@ def offload_arch_EQ : Joined<["--"], "offload-arch=">, Flags<[NoXarchOption]>,
            "device architecture followed by target ID features delimited by a colon. Each target ID feature "
            "is a pre-defined string followed by a plus or minus sign (e.g. gfx908:xnack+:sramecc-).  May be "
            "specified more than once.">;
-def cuda_gpu_arch_EQ : Joined<["--"], "cuda-gpu-arch=">, Flags<[NoXarchOption]>,
+def cuda_gpu_arch_EQ : Joined<["--"], "cuda-gpu-arch=">, Flags<[NoXarchOption, CoreOption]>,
   Alias<offload_arch_EQ>;
 def hip_link : Flag<["--"], "hip-link">,
   HelpText<"Link clang-offload-bundler bundles for HIP">;
@@ -954,7 +954,7 @@ def no_cuda_version_check : Flag<["--"], "no-cuda-version-check">,
   HelpText<"Don't error out if the detected version of the CUDA install is "
            "too low for the requested CUDA gpu architecture.">;
 def no_cuda_noopt_device_debug : Flag<["--"], "no-cuda-noopt-device-debug">;
-def cuda_path_EQ : Joined<["--"], "cuda-path=">, Group<i_Group>,
+def cuda_path_EQ : Joined<["--"], "cuda-path=">, Flags<[CoreOption]>, Group<i_Group>,
   HelpText<"CUDA installation path">;
 def cuda_path_ignore_env : Flag<["--"], "cuda-path-ignore-env">, Group<i_Group>,
   HelpText<"Ignore environment variables to detect CUDA installation">;
@@ -2691,7 +2691,7 @@ def fsycl_help : Flag<["-"], "fsycl-help">, Alias<fsycl_help_EQ>,
   "from all of the offline compilation tools">;
 def fsycl_libspirv_path_EQ : Joined<["-"], "fsycl-libspirv-path=">,
   Flags<[CC1Option, CoreOption]>, HelpText<"Path to libspirv library">;
-def fno_sycl_libspirv : Flag<["-"], "fno-sycl-libspirv">, HelpText<"Disable check for libspirv">;
+def fno_sycl_libspirv : Flag<["-"], "fno-sycl-libspirv">, Flags<[CoreOption]>, HelpText<"Disable check for libspirv">;
 def fsycl_host_compiler_EQ : Joined<["-"], "fsycl-host-compiler=">,
   Flags<[CoreOption]>, HelpText<"Specify C++ compiler binary to perform host "
   "compilation with during SYCL offload compiles.">;
@@ -3843,7 +3843,7 @@ def nogpuinc : Flag<["-"], "nogpuinc">, HelpText<"Do not add include paths for C
 def : Flag<["-"], "nocudainc">, Alias<nogpuinc>;
 def nogpulib : Flag<["-"], "nogpulib">,
   HelpText<"Do not link device library for CUDA/HIP device compilation">;
-def : Flag<["-"], "nocudalib">, Alias<nogpulib>;
+def : Flag<["-"], "nocudalib">, Flags<[CoreOption]>, Alias<nogpulib>;
 def nodefaultlibs : Flag<["-"], "nodefaultlibs">;
 def nofixprebinding : Flag<["-"], "nofixprebinding">;
 def nolibc : Flag<["-"], "nolibc">;

clang/include/clang/Driver/Options.td

@@ -10593,9 +10593,7 @@ class Sema final {
                                           Expr *E);
 
   bool checkNSReturnsRetainedReturnType(SourceLocation loc, QualType type);
-  bool checkAllowedSYCLInitializer(VarDecl *VD,
-                                   bool CheckValueDependent = false);
-
+  bool checkAllowedSYCLInitializer(VarDecl *VD);
   //===--------------------------------------------------------------------===//
   // C++ Coroutines TS
   //

clang/include/clang/Sema/Sema.h

@@ -1540,7 +1540,7 @@ void CodeGenFunction::GenerateCode(GlobalDecl GD, llvm::Function *Fn,
   StartFunction(GD, ResTy, Fn, FnInfo, Args, Loc, BodyRange.getBegin());
 
   SyclOptReportHandler &SyclOptReport = CGM.getDiags().getSYCLOptReport();
-  if (SyclOptReport.HasOptReportInfo(FD)) {
+  if (Fn && SyclOptReport.HasOptReportInfo(FD)) {
     llvm::OptimizationRemarkEmitter ORE(Fn);
     for (auto ORI : llvm::enumerate(SyclOptReport.GetInfo(FD))) {
       llvm::DiagnosticLocation DL =

clang/lib/CodeGen/CodeGenFunction.cpp

@@ -4113,7 +4113,10 @@ class OffloadingActionBuilder final {
             Args.getLastArg(options::OPT__SLASH_EP, options::OPT__SLASH_P) ||
             Args.getLastArg(options::OPT_M, options::OPT_MM);
         if (IsPreprocessOnly) {
-          for (Action *&A : SYCLDeviceActions) {
+          for (auto TargetActionInfo :
+               llvm::zip(SYCLDeviceActions, SYCLTargetInfoList)) {
+            Action *&A = std::get<0>(TargetActionInfo);
+            auto &TargetInfo = std::get<1>(TargetActionInfo);
             A = C.getDriver().ConstructPhaseAction(C, Args, CurPhase, A,
                                                    AssociatedOffloadKind);
             if (SYCLDeviceOnly)
@@ -4122,7 +4125,7 @@ class OffloadingActionBuilder final {
             // header.
             Action *CompileAction =
                 C.MakeAction<CompileJobAction>(A, types::TY_Nothing);
-            DA.add(*CompileAction, *ToolChains.front(), nullptr,
+            DA.add(*CompileAction, *TargetInfo.TC, TargetInfo.BoundArch,
                    Action::OFK_SYCL);
           }
           return SYCLDeviceOnly ? ABRT_Ignore_Host : ABRT_Success;
@@ -4147,6 +4150,8 @@ class OffloadingActionBuilder final {
               continue;
             }
           }
+          if (Args.hasArg(options::OPT_fsyntax_only))
+            OutputType = types::TY_Nothing;
           A = C.MakeAction<CompileJobAction>(A, OutputType);
           DeviceCompilerInput = A;
         }
@@ -4875,11 +4880,14 @@ class OffloadingActionBuilder final {
       bool SYCLfpgaTriple = false;
       bool ShouldAddDefaultTriple = true;
       bool GpuInitHasErrors = false;
-      if (SYCLTargets || SYCLAddTargets) {
-        if (SYCLTargets) {
+      bool HasSYCLTargetsOption =
+          SYCLAddTargets || SYCLTargets || SYCLLinkTargets;
+      if (HasSYCLTargetsOption) {
+        if (SYCLTargets || SYCLLinkTargets) {
+          Arg *SYCLTargetsValues = SYCLTargets ? SYCLTargets : SYCLLinkTargets;
           // Fill SYCLTripleList
           llvm::StringMap<StringRef> FoundNormalizedTriples;
-          for (const char *Val : SYCLTargets->getValues()) {
+          for (const char *Val : SYCLTargetsValues->getValues()) {
             llvm::Triple TT(C.getDriver().MakeSYCLDeviceTriple(Val));
             std::string NormalizedName = TT.normalize();
 

clang/lib/Driver/Driver.cpp

@@ -1219,7 +1219,7 @@ llvm::opt::DerivedArgList *ToolChain::TranslateOffloadTargetArgs(
         A->getOption().matches(options::OPT_Xsycl_frontend);
       if (A->getOption().matches(options::OPT_Xsycl_frontend_EQ)) {
         // Passing device args: -Xsycl-target-frontend=<triple> -opt=val.
-        if (A->getValue(0) == getTripleString())
+        if (getDriver().MakeSYCLDeviceTriple(A->getValue(0)) == getTriple())
           Index = Args.getBaseArgs().MakeIndex(A->getValue(1));
         else
           continue;

clang/lib/Driver/ToolChain.cpp

@@ -8886,7 +8886,8 @@ void SPIRVTranslator::ConstructJob(Compilation &C, const JobAction &JA,
       // Don't enable several freshly added extensions on FPGA H/W
       ExtArg += ",+SPV_INTEL_token_type"
                 ",+SPV_INTEL_bfloat16_conversion"
-                ",+SPV_INTEL_joint_matrix";
+                ",+SPV_INTEL_joint_matrix"
+                ",+SPV_INTEL_hw_thread_queries";
     TranslatorArgs.push_back(TCArgs.MakeArgString(ExtArg));
   }
   for (auto I : Inputs) {
@@ -9036,8 +9037,9 @@ void SYCLPostLink::ConstructJob(Compilation &C, const JobAction &JA,
       addArgs(CmdArgs, TCArgs, {"-split=auto"});
     else { // Device code split is off
     }
-  } else {
-    // auto is the default split mode
+  } else if (getToolChain().getTriple().getArchName() != "spir64_fpga") {
+    // for FPGA targets, off is the default split mode,
+    // otherwise auto is the default split mode
     addArgs(CmdArgs, TCArgs, {"-split=auto"});
   }
 

clang/lib/Driver/ToolChains/Clang.cpp

@@ -249,8 +249,7 @@ bool Sema::DiagnoseUseOfDecl(NamedDecl *D, ArrayRef<SourceLocation> Locs,
       // Disallow const statics and globals that are not zero-initialized
       // or constant-initialized.
       else if (IsRuntimeEvaluated && IsConst && VD->hasGlobalStorage() &&
-               !VD->isConstexpr() &&
-               !checkAllowedSYCLInitializer(VD, /*CheckValueDependent =*/true))
+               !VD->isConstexpr() && !checkAllowedSYCLInitializer(VD))
         SYCLDiagIfDeviceCode(*Locs.begin(), diag::err_sycl_restrict)
             << Sema::KernelConstStaticVariable;
     } else if (auto *FDecl = dyn_cast<FunctionDecl>(D)) {

clang/lib/Sema/SemaExpr.cpp

@@ -3508,13 +3508,12 @@ class SYCLKernelNameTypeVisitor
           }
           // Check if the declaration is completely defined within a
           // function or class/struct.
-
           if (Tag->isCompleteDefinition()) {
             S.Diag(KernelInvocationFuncLoc,
                    diag::err_sycl_kernel_incorrectly_named)
-                << /* kernel name should be globally visible */ 0
-                << KernelNameType;
-
+                << /* kernel name should be forward declarable at namespace
+                      scope */
+                0 << KernelNameType;
             IsInvalid = true;
           } else {
             S.Diag(KernelInvocationFuncLoc, diag::warn_sycl_implicit_decl);
@@ -4132,15 +4131,15 @@ void Sema::finalizeSYCLDelayedAnalysis(const FunctionDecl *Caller,
   }
 }
 
-bool Sema::checkAllowedSYCLInitializer(VarDecl *VD, bool CheckValueDependent) {
+bool Sema::checkAllowedSYCLInitializer(VarDecl *VD) {
   assert(getLangOpts().SYCLIsDevice &&
          "Should only be called during SYCL compilation");
 
   if (VD->isInvalidDecl() || !VD->hasInit() || !VD->hasGlobalStorage())
     return true;
 
   const Expr *Init = VD->getInit();
-  bool ValueDependent = CheckValueDependent && Init->isValueDependent();
+  bool ValueDependent = Init->isValueDependent();
   bool isConstantInit =
       Init && !ValueDependent && Init->isConstantInitializer(Context, false);
   if (!VD->isConstexpr() && Init && !ValueDependent && !isConstantInit)

clang/lib/Sema/SemaSYCL.cpp

@@ -111,6 +111,7 @@ if( NOT CLANG_BUILT_STANDALONE )
   list(APPEND CLANG_TEST_DEPS
     llvm-config
     FileCheck count not
+    append-file
     llc
     llvm-ar
     llvm-as

clang/test/CMakeLists.txt

@@ -4,10 +4,12 @@
 /// 1. A SPIR-V-based environment must be targetted
 /// 2. The corresponding Driver option must be enabled explicitly
 
-// RUN: %clangxx -fsycl -fsycl-instrument-device-code --sysroot=%S/Inputs/SYCL -fsycl-targets=spir64 -### %s 2>&1 \
+/// FIXME: Force linux targets to allow for the libraries to be found.  Dummy
+/// inputs for --sysroot should be updated to work better for Windows.
+// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -fsycl-instrument-device-code --sysroot=%S/Inputs/SYCL -fsycl-targets=spir64 -### %s 2>&1 \
 // RUN: | FileCheck -check-prefixes=CHECK-SPIRV,CHECK-HOST %s
 /// -fno-sycl-device-lib mustn't affect the linkage of ITT libraries
-// RUN: %clangxx -fsycl -fsycl-instrument-device-code --sysroot=%S/Inputs/SYCL -fno-sycl-device-lib=all -fsycl-targets=spir64 -### %s 2>&1 \
+// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -fsycl-instrument-device-code --sysroot=%S/Inputs/SYCL -fno-sycl-device-lib=all -fsycl-targets=spir64 -### %s 2>&1 \
 // RUN: | FileCheck -check-prefixes=CHECK-SPIRV %s
 
 // CHECK-SPIRV: "-cc1"{{.*}} "-fsycl-is-device"{{.*}} "-fsycl-instrument-device-code"