Merge branch 'sycl' of https://github.com/intel/llvm into properties

Author: KornevNikita

.github/CODEOWNERS

@@ -60,6 +60,10 @@ sycl/include/sycl/ext/intel/esimd.hpp @intel/dpcpp-esimd-reviewers
 sycl/doc/extensions/experimental/sycl_ext_intel_esimd/ @intel/dpcpp-esimd-reviewers
 llvm/lib/SYCLLowerIR/CMakeLists.txt @intel/dpcpp-tools-reviewers @intel/dpcpp-esimd-reviewers
 
+# invoke_simd
+sycl/include/sycl/ext/oneapi/experimental/invoke_simd.hpp @rolandschulz @kbobrovs
+sycl/include/std/experimental/simd.hpp @rolandschulz @kbobrovs
+
 # DevOps configs
 .github/workflows/ @intel/dpcpp-devops-reviewers
 buildbot/ @intel/dpcpp-devops-reviewers

buildbot/configure.py

@@ -44,6 +44,10 @@ def do_configure(args):
     sycl_enable_xpti_tracing = 'ON'
     xpti_enable_werror = 'OFF'
 
+    # lld is needed on Windows or for the HIP plugin on AMD
+    if platform.system() == 'Windows' or (args.hip and args.hip_platform == 'AMD'):
+        llvm_enable_projects += ';lld'
+
     # replace not append, so ARM ^ X86
     if args.arm:
         llvm_targets_to_build = 'ARM;AArch64'
@@ -65,8 +69,6 @@ def do_configure(args):
             llvm_targets_to_build += ';AMDGPU'
             libclc_targets_to_build += libclc_amd_target_names
 
-            # The HIP plugin for AMD uses lld for linking
-            llvm_enable_projects += ';lld'
         elif args.hip_platform == 'NVIDIA' and not args.cuda:
             llvm_targets_to_build += ';NVPTX'
             libclc_targets_to_build += libclc_nvidia_target_names

buildbot/dependency.conf

@@ -4,8 +4,8 @@ ocl_cpu_rt_ver=2022.14.8.0.04
 # https://github.com/intel/llvm/releases/download/2022-WW33/win-oclcpuexp-2022.14.8.0.04_rel.zip
 ocl_cpu_rt_ver_win=2022.14.8.0.04
 # Same GPU driver supports Level Zero and OpenCL
-# https://github.com/intel/compute-runtime/releases/tag/22.31.23852
-ocl_gpu_rt_ver=22.31.23852
+# https://github.com/intel/compute-runtime/releases/tag/22.35.24055
+ocl_gpu_rt_ver=22.35.24055
 # Same GPU driver supports Level Zero and OpenCL
 # https://downloadmirror.intel.com/723911/igfx_win_101.1404.zip
 ocl_gpu_rt_ver_win=101.1404
@@ -31,7 +31,7 @@ ocloc_ver_win=101.1404
 [DRIVER VERSIONS]
 cpu_driver_lin=2022.14.8.0.04
 cpu_driver_win=2022.14.8.0.04
-gpu_driver_lin=22.31.23852
+gpu_driver_lin=22.35.24055
 gpu_driver_win=101.1404
 fpga_driver_lin=2022.14.8.0.04
 fpga_driver_win=2022.14.8.0.04

clang/include/clang/Sema/Sema.h

@@ -13982,7 +13982,7 @@ class Sema final {
     KernelConstStaticVariable
   };
 
-  bool isKnownGoodSYCLDecl(const Decl *D);
+  bool isDeclAllowedInSYCLDeviceCode(const Decl *D);
   void checkSYCLDeviceVarDecl(VarDecl *Var);
   void copySYCLKernelAttrs(const CXXRecordDecl *KernelObj);
   void ConstructOpenCLKernel(FunctionDecl *KernelCallerFunc, MangleContext &MC);

clang/lib/CodeGen/CodeGenFunction.cpp

@@ -1600,7 +1600,7 @@ void CodeGenFunction::GenerateCode(GlobalDecl GD, llvm::Function *Fn,
   StartFunction(GD, ResTy, Fn, FnInfo, Args, Loc, BodyRange.getBegin());
   if (!getLangOpts().OptRecordFile.empty()) {
     SyclOptReportHandler &SyclOptReport = CGM.getDiags().getSYCLOptReport();
-    if (Fn && SyclOptReport.HasOptReportInfo(FD)) {
+    if (SyclOptReport.HasOptReportInfo(FD)) {
       llvm::OptimizationRemarkEmitter ORE(Fn);
       for (auto ORI : llvm::enumerate(SyclOptReport.GetInfo(FD))) {
         llvm::DiagnosticLocation DL =

clang/lib/Driver/ToolChains/MSVC.cpp

@@ -131,10 +131,11 @@ void visualstudio::Linker::ConstructJob(Compilation &C, const JobAction &JA,
     CmdArgs.push_back("-defaultlib:oldnames");
   }
 
-  if ((!C.getDriver().IsCLMode() && !Args.hasArg(options::OPT_nostdlib) &&
-       Args.hasArg(options::OPT_fsycl) &&
+  if ((!C.getDriver().IsCLMode() && Args.hasArg(options::OPT_fsycl) &&
        !Args.hasArg(options::OPT_nolibsycl)) ||
       Args.hasArg(options::OPT_fsycl_host_compiler_EQ)) {
+    CmdArgs.push_back(Args.MakeArgString(std::string("-libpath:") +
+                                         TC.getDriver().Dir + "/../lib"));
     if (Args.hasArg(options::OPT__SLASH_MDd))
       CmdArgs.push_back("-defaultlib:sycld.lib");
     else

clang/lib/Sema/SemaChecking.cpp

@@ -5967,8 +5967,8 @@ void Sema::checkCall(NamedDecl *FDecl, const FunctionProtoType *Proto,
     CheckSYCLKernelCall(FD, Range, Args);
 
   // Diagnose variadic calls in SYCL.
-  if (FD && FD ->isVariadic() && getLangOpts().SYCLIsDevice &&
-      !isUnevaluatedContext() && !isKnownGoodSYCLDecl(FD))
+  if (FD && FD->isVariadic() && getLangOpts().SYCLIsDevice &&
+      !isUnevaluatedContext() && !isDeclAllowedInSYCLDeviceCode(FD))
     SYCLDiagIfDeviceCode(Loc, diag::err_sycl_restrict)
         << Sema::KernelCallVariadicFunction;
 }

clang/lib/Sema/SemaSYCL.cpp

@@ -364,9 +364,15 @@ static bool IsSyclMathFunc(unsigned BuiltinID) {
   return true;
 }
 
-bool Sema::isKnownGoodSYCLDecl(const Decl *D) {
+bool Sema::isDeclAllowedInSYCLDeviceCode(const Decl *D) {
   if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(D)) {
     const IdentifierInfo *II = FD->getIdentifier();
+
+    // Allow __builtin_assume_aligned to be called from within device code.
+    if (FD->getBuiltinID() &&
+        FD->getBuiltinID() == Builtin::BI__builtin_assume_aligned)
+      return true;
+
     // Allow to use `::printf` only for CUDA.
     if (Context.getTargetInfo().getTriple().isNVPTX()) {
       if (FD->getBuiltinID() == Builtin::BIprintf)

clang/test/Driver/sycl-offload.c

@@ -634,6 +634,11 @@
 // CHECK-LD-NOLIBSYCL: "{{.*}}ld{{(.exe)?}}"
 // CHECK-LD-NOLIBSYCL-NOT: "-lsycl"
 
+/// Check no SYCL runtime is linked with -nostdlib
+// RUN: %clang -fsycl -nostdlib -target x86_64-unknown-linux-gnu %s -o %t -### 2>&1 | FileCheck -check-prefix=CHECK-LD-NOSTDLIB %s
+// CHECK-LD-NOSTDLIB: "{{.*}}ld{{(.exe)?}}"
+// CHECK-LD-NOSTDLIB-NOT: "-lsycl"
+
 /// Check for default linking of sycl.lib with -fsycl usage
 // RUN: %clang -fsycl -target x86_64-unknown-windows-msvc %s -o %t -### 2>&1 | FileCheck -check-prefix=CHECK-LINK-SYCL %s
 // RUN: %clang_cl -fsycl %s -o %t -### 2>&1 | FileCheck -check-prefix=CHECK-LINK-SYCL-CL %s
@@ -643,11 +648,19 @@
 
 /// Check no SYCL runtime is linked with -nolibsycl
 // RUN: %clang -fsycl -nolibsycl -target x86_64-unknown-windows-msvc %s -o %t -### 2>&1 | FileCheck -check-prefix=CHECK-LINK-NOLIBSYCL %s
-// RUN: %clang_cl -fsycl -nolibsycl %s -o %t -### 2>&1 | FileCheck -check-prefix=CHECK-LINK-NOLIBSYCL %s
-// CHECK-LINK-NOLIBSYCL-NOT: "--dependent-lib=sycl"
+// RUN: %clang_cl -fsycl -nolibsycl %s -o %t -### 2>&1 | FileCheck -check-prefix=CHECK-LINK-NOLIBSYCL-CL %s
+// CHECK-LINK-NOLIBSYCL-CL-NOT: "--dependent-lib=sycl"
 // CHECK-LINK-NOLIBSYCL: "{{.*}}link{{(.exe)?}}"
 // CHECK-LINK-NOLIBSYCL-NOT: "-defaultlib:sycl.lib"
 
+/// Check SYCL runtime is linked despite -nostdlib on Windows, this is
+/// necessary for the Windows Clang CMake to work
+// RUN: %clang -fsycl -nostdlib -target x86_64-unknown-windows-msvc %s -o %t -### 2>&1 | FileCheck -check-prefix=CHECK-LINK-NOSTDLIB %s
+// RUN: %clang_cl -fsycl -nostdlib %s -o %t -### 2>&1 | FileCheck -check-prefix=CHECK-LINK-NOSTDLIB-CL %s
+// CHECK-LINK-NOSTDLIB-CL: "--dependent-lib=sycl"
+// CHECK-LINK-NOSTDLIB: "{{.*}}link{{(.exe)?}}"
+// CHECK-LINK-NOSTDLIB: "-defaultlib:sycl.lib"
+
 /// Check sycld.lib is chosen with /MDd
 // RUN:  %clang_cl -fsycl /MDd %s -o %t -### 2>&1 | FileCheck -check-prefix=CHECK-LINK-SYCL-DEBUG %s
 // CHECK-LINK-SYCL-DEBUG: "--dependent-lib=sycld"

clang/test/SemaSYCL/remove-restriction-builtin-assume-aligned.cpp

@@ -0,0 +1,21 @@
+// RUN: %clang_cc1 -fsycl-is-device -internal-isystem %S/Inputs -sycl-std=2020 -verify -fsyntax-only %s
+// This test checks if __builtin_assume_aligned does not throw an error when
+// called from within device code.
+
+#include "sycl.hpp"
+
+using namespace sycl;
+queue q;
+
+int main() {
+  int *Ptr[2];
+  // expected-no-diagnostics
+  q.submit([&](handler &h) {
+    h.single_task<class kernelA>([=]() {
+    int *APtr = (int *)__builtin_assume_aligned(Ptr, 32);
+    *APtr = 42;
+    });
+  });
+  return 0;
+}
+

devops/actions/aws-ec2/package.json

@@ -2,7 +2,7 @@
   "name": "aws-ec2",
   "description": "Start AWS EC2 spot instances with Github actions runner agent in it",
   "dependencies": {
-    "@actions/core": "1.9.0",
+    "@actions/core": "1.9.1",
     "@actions/github": "5.0.3",
     "aws-sdk": "2.1179.0"
   }

devops/dependencies.json

@@ -1,9 +1,9 @@
 {
   "linux": {
     "compute_runtime": {
-      "github_tag": "22.32.23937",
-      "version": "22.32.23937",
-      "url": "https://github.com/intel/compute-runtime/releases/tag/22.32.23937",
+      "github_tag": "22.35.24055",
+      "version": "22.35.24055",
+      "url": "https://github.com/intel/compute-runtime/releases/tag/22.35.24055",
       "root": "{DEPS_ROOT}/opencl/runtime/linux/oclgpu"
     },
     "igc": {

devops/scripts/test-commit-sycl-bisect.bash

@@ -129,6 +129,7 @@ fi
 cd "$TEST_WD"
 if [[ "$COMMAND_ALLOW_BISECT_CODES" == "" ]]; then
   bash -c "$COMMAND" || exit 1
+  exit
 fi
 
 # Otherwise, just run the command.

llvm/include/llvm/SYCLLowerIR/SYCLPropagateAspectsUsage.h

@@ -0,0 +1,28 @@
+//===---- SYCLPropagateAspectsUsage.cpp - SYCLPropagateAspectsUsage Pass --===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// Pass propagates optional kernel features metadata through a module call graph
+//
+//===----------------------------------------------------------------------===//
+//
+#ifndef LLVM_SYCL_PROPAGATE_ASPECTS_USAGE_H
+#define LLVM_SYCL_PROPAGATE_ASPECTS_USAGE_H
+
+#include "llvm/IR/PassManager.h"
+
+namespace llvm {
+
+class SYCLPropagateAspectsUsagePass
+    : public PassInfoMixin<SYCLPropagateAspectsUsagePass> {
+public:
+  PreservedAnalyses run(Module &M, ModuleAnalysisManager &);
+};
+
+} // namespace llvm
+
+#endif // LLVM_SYCL_PROPAGATE_ASPECTS_USAGE_H

llvm/lib/Passes/PassBuilder.cpp

@@ -84,6 +84,7 @@
 #include "llvm/SYCLLowerIR/LowerWGLocalMemory.h"
 #include "llvm/SYCLLowerIR/LowerWGScope.h"
 #include "llvm/SYCLLowerIR/MutatePrintfAddrspace.h"
+#include "llvm/SYCLLowerIR/SYCLPropagateAspectsUsage.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"

llvm/lib/Passes/PassRegistry.def

@@ -136,6 +136,7 @@ MODULE_PASS("SPIRITTAnnotations", SPIRITTAnnotationsPass())
 MODULE_PASS("deadargelim-sycl", DeadArgumentEliminationSYCLPass())
 MODULE_PASS("sycllowerwglocalmemory", SYCLLowerWGLocalMemoryPass())
 MODULE_PASS("lower-esimd-kernel-attrs", SYCLFixupESIMDKernelWrapperMDPass())
+MODULE_PASS("sycl-propagate-aspects-usage", SYCLPropagateAspectsUsagePass())
 #undef MODULE_PASS
 
 #ifndef MODULE_PASS_WITH_PARAMS

llvm/lib/SYCLLowerIR/CMakeLists.txt

@@ -58,6 +58,7 @@ add_llvm_component_library(LLVMSYCLLowerIR
   LowerWGScope.cpp
   LowerWGLocalMemory.cpp
   MutatePrintfAddrspace.cpp
+  SYCLPropagateAspectsUsage.cpp
 
   LocalAccessorToSharedMemory.cpp
   GlobalOffset.cpp

llvm/lib/SYCLLowerIR/LowerInvokeSimd.cpp

@@ -250,14 +250,15 @@ bool collectUsesLookTrhoughMemAndCasts(Value *V,
   for (const Use *U : TmpVUses) {
     User *UU = U->getUser();
     assert(!isCast(UU));
+
     auto *St = dyn_cast<StoreInst>(UU);
 
     if (!St) {
       Uses.insert(U);
       continue;
     }
     // Current user is a store (of V) instruction, see if...
-    assert((V = St->getValueOperand()) &&
+    assert((V == St->getValueOperand()) &&
            "bad V param in collectUsesLookTrhoughMemAndCasts");
     Value *Addr = stripCasts(St->getPointerOperand());