Merge branch 'sycl' into llvmspirv_pulldown

Author: jsji

.github/CODEOWNERS

@@ -103,7 +103,7 @@ buildbot/ @intel/dpcpp-devops-reviewers
 devops/ @intel/dpcpp-devops-reviewers
 
 # Kernel fusion JIT compiler
-sycl-fusion/ @intel/dpcpp-kernel-fusion-reviewers
+sycl-jit/ @intel/dpcpp-kernel-fusion-reviewers
 sycl/doc/design/KernelFusionJIT.md @intel/dpcpp-kernel-fusion-reviewers
 sycl/doc/extensions/experimental/sycl_ext_codeplay_kernel_fusion.asciidoc @intel/dpcpp-kernel-fusion-reviewers
 sycl/include/sycl/ext/codeplay/experimental/fusion_properties.hpp  @intel/dpcpp-kernel-fusion-reviewers

.github/workflows/sycl-aws.yml

@@ -14,22 +14,18 @@ on:
         # 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
+        default: '[{"runs-on":"aws_cuda-${{ github.run_id }}-${{ github.run_attempt }}","aws-ami":"ami-01cb0573cb039ab24","aws-type":["g5.2xlarge","g5.4xlarge"],"aws-disk":"/dev/sda1:64","aws-spot":"false"}]'
 
 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
+      - uses: actions/checkout@v4
+        with:
+          sparse-checkout: devops/actions/aws-ec2
+      - run: npm install ./devops/actions/aws-ec2
+      - uses: ./devops/actions/aws-ec2
         with:
           mode: ${{ inputs.mode }}
           runs-on-list: ${{ inputs.runs-on-list }}

.github/workflows/sycl-detect-changes.yml

@@ -30,9 +30,9 @@ jobs:
             clang: &clang
               - *llvm
               - 'clang/**'
-            sycl_fusion: &sycl-fusion
+            sycl_jit: &sycl-jit
               - *llvm
-              - 'sycl-fusion/**'
+              - 'sycl-jit/**'
             xptifw: &xptifw
               - 'xptifw/**'
             libclc: &libclc
@@ -41,7 +41,7 @@ jobs:
               - 'libclc/**'
             sycl: &sycl
               - *clang
-              - *sycl-fusion
+              - *sycl-jit
               - *llvm_spirv
               - *xptifw
               - *libclc
@@ -84,7 +84,7 @@ jobs:
               return '${{ steps.changes.outputs.changes }}';
             }
             // Treat everything as changed for huge PRs.
-            return ["llvm", "llvm_spirv", "clang", "sycl_fusion", "xptifw", "libclc", "sycl", "ci", "esimd"];
+            return ["llvm", "llvm_spirv", "clang", "sycl_jit", "xptifw", "libclc", "sycl", "ci", "esimd"];
 
       - run: echo '${{ steps.result.outputs.result }}'
 

.github/workflows/sycl-linux-precommit-aws.yml

@@ -1,5 +1,7 @@
 name: E2E on AWS CUDA
 run-name: E2E on AWS CUDA - ${{ github.event.workflow_run.display_title }}
+# Note: This workflow is currently disabled in the Github Actions UI because
+# we run CUDA testing on the self-hosted runners.
 # We have to keep pre-commit AWS CUDA testing in a separate workflow because we
 # need access to AWS secret and that isn't available on pull_request jobs for
 # PRs from forks. And github's "require approval for all outside collaborators"

.github/workflows/sycl-linux-precommit.yml

@@ -81,11 +81,6 @@ jobs:
             image: ghcr.io/intel/llvm/ubuntu2204_build:latest
             image_options: -u 1001 --gpus all --cap-add SYS_ADMIN
             target_devices: ext_oneapi_cuda:gpu
-          - name: AMD/HIP
-            runner: '["Linux", "amdgpu"]'
-            image: ghcr.io/intel/llvm/ubuntu2204_build:latest-0300ac924620a51f76c4929794637b82790f12ab
-            image_options: -u 1001 --device=/dev/dri --device=/dev/kfd
-            target_devices: ext_oneapi_hip:gpu
           - name: Intel
             runner: '["Linux", "gen12"]'
             image: ghcr.io/intel/llvm/ubuntu2204_intel_drivers:latest

.github/workflows/sycl-nightly.yml

@@ -37,13 +37,6 @@ jobs:
             target_devices: ext_oneapi_hip:gpu
             tests_selector: e2e
 
-          - name: NVIDIA/CUDA
-            runner: '["Linux", "cuda"]'
-            image: ghcr.io/intel/llvm/ubuntu2204_build:latest
-            image_options: -u 1001 --gpus all --cap-add SYS_ADMIN
-            target_devices: ext_oneapi_cuda:gpu
-            tests_selector: e2e
-
           - name: Intel L0 GPU
             runner: '["Linux", "gen12"]'
             image: ghcr.io/intel/llvm/ubuntu2204_intel_drivers:latest
@@ -135,6 +128,39 @@ jobs:
       sycl_toolchain_archive: ${{ needs.build-win.outputs.artifact_archive_name }}
       extra_lit_opts: --param gpu-intel-gen12=True
 
+  cuda-aws-start:
+    needs: [ubuntu2204_build]
+    if: ${{ always() && !cancelled() && needs.ubuntu2204_build.outputs.build_conclusion == 'success' }}
+    uses: ./.github/workflows/sycl-aws.yml
+    secrets: inherit
+    with:
+      mode: start
+
+  cuda-run-tests:
+    needs: [ubuntu2204_build, cuda-aws-start]
+    if: ${{ always() && !cancelled() && needs.ubuntu2204_build.outputs.build_conclusion == 'success' }}
+    uses: ./.github/workflows/sycl-linux-run-tests.yml
+    with:
+      name: CUDA E2E
+      runner: '["aws_cuda-${{ github.run_id }}-${{ github.run_attempt }}"]'
+      image: ghcr.io/intel/llvm/ubuntu2204_build:latest-0300ac924620a51f76c4929794637b82790f12ab
+      image_options: -u 1001 --gpus all --cap-add SYS_ADMIN --env NVIDIA_DISABLE_REQUIRE=1
+      target_devices: ext_oneapi_cuda:gpu
+      ref: ${{ github.sha }}
+      merge_ref: ''
+
+      sycl_toolchain_artifact: sycl_linux_default
+      sycl_toolchain_archive: ${{ needs.ubuntu2204_build.outputs.artifact_archive_name }}
+      sycl_toolchain_decompress_command: ${{ needs.ubuntu2204_build.outputs.artifact_decompress_command }}
+
+  cuda-aws-stop:
+    needs: [cuda-aws-start, cuda-run-tests]
+    if: always()
+    uses: ./.github/workflows/sycl-aws.yml
+    secrets: inherit
+    with:
+      mode: stop
+
   nightly_build_upload:
     name: Nightly Build Upload
     if: ${{ github.ref_name == 'sycl' }}

.github/workflows/sycl-post-commit.yml

@@ -56,6 +56,11 @@ jobs:
           - name: Intel Arc A-Series Graphics with Level Zero
             runner: '["Linux", "arc"]'
             extra_lit_opts: --param matrix-xmx8=True --param gpu-intel-dg2=True
+          - name: AMD/HIP
+            runner: '["Linux", "amdgpu"]'
+            image: ghcr.io/intel/llvm/ubuntu2204_build:latest-0300ac924620a51f76c4929794637b82790f12ab
+            image_options: -u 1001 --device=/dev/dri --device=/dev/kfd
+            target_devices: ext_oneapi_hip:gpu
           # Performance tests below. Specifics:
           #  - only run performance tests (use LIT_FILTER env)
           #  - ask llvm-lit to show all the output, even for PASS (-a)

buildbot/configure.py

@@ -31,10 +31,10 @@ def do_configure(args):
     libclc_amd_target_names = ";amdgcn--amdhsa"
     libclc_nvidia_target_names = ";nvptx64--nvidiacl"
 
-    sycl_enable_fusion = "OFF"
-    if not args.disable_fusion:
-        llvm_external_projects += ";sycl-fusion"
-        sycl_enable_fusion = "ON"
+    sycl_enable_jit = "OFF"
+    if not args.disable_jit:
+        llvm_external_projects += ";sycl-jit"
+        sycl_enable_jit = "ON"
 
     if args.llvm_external_projects:
         llvm_external_projects += ";" + args.llvm_external_projects.replace(",", ";")
@@ -45,7 +45,7 @@ def do_configure(args):
     xpti_dir = os.path.join(abs_src_dir, "xpti")
     xptifw_dir = os.path.join(abs_src_dir, "xptifw")
     libdevice_dir = os.path.join(abs_src_dir, "libdevice")
-    fusion_dir = os.path.join(abs_src_dir, "sycl-fusion")
+    jit_dir = os.path.join(abs_src_dir, "sycl-jit")
     llvm_targets_to_build = args.host_target
     llvm_enable_projects = "clang;" + llvm_external_projects
     libclc_build_native = "OFF"
@@ -174,7 +174,7 @@ def do_configure(args):
         "-DXPTI_SOURCE_DIR={}".format(xpti_dir),
         "-DLLVM_EXTERNAL_XPTIFW_SOURCE_DIR={}".format(xptifw_dir),
         "-DLLVM_EXTERNAL_LIBDEVICE_SOURCE_DIR={}".format(libdevice_dir),
-        "-DLLVM_EXTERNAL_SYCL_FUSION_SOURCE_DIR={}".format(fusion_dir),
+        "-DLLVM_EXTERNAL_SYCL_JIT_SOURCE_DIR={}".format(jit_dir),
         "-DLLVM_ENABLE_PROJECTS={}".format(llvm_enable_projects),
         "-DSYCL_BUILD_PI_HIP_PLATFORM={}".format(sycl_build_pi_hip_platform),
         "-DLLVM_BUILD_TOOLS=ON",
@@ -189,7 +189,7 @@ def do_configure(args):
         "-DXPTI_ENABLE_WERROR={}".format(xpti_enable_werror),
         "-DSYCL_CLANG_EXTRA_FLAGS={}".format(sycl_clang_extra_flags),
         "-DSYCL_ENABLE_PLUGINS={}".format(";".join(set(sycl_enabled_plugins))),
-        "-DSYCL_ENABLE_KERNEL_FUSION={}".format(sycl_enable_fusion),
+        "-DSYCL_ENABLE_EXTENSION_JIT={}".format(sycl_enable_jit),
         "-DSYCL_ENABLE_MAJOR_RELEASE_PREVIEW_LIB={}".format(sycl_preview_lib),
         "-DBUG_REPORT_URL=https://github.com/intel/llvm/issues",
     ]
@@ -379,9 +379,9 @@ def main():
         help="Disable building of the SYCL runtime major release preview library",
     )
     parser.add_argument(
-        "--disable-fusion",
+        "--disable-jit",
         action="store_true",
-        help="Disable the kernel fusion JIT compiler",
+        help="Disable the kernel JIT compiler for AMD and Nvidia",
     )
     parser.add_argument(
         "--add_security_flags",

llvm/lib/SYCLLowerIR/SYCLCreateNVVMAnnotations.cpp

@@ -55,7 +55,7 @@ decomposeWGMetadata(const MDNode &Node, const Function &F) {
     NDim = getSingleIntMetadata(*NDimMD).value_or(3);
   assert(NDim >= 1 && NDim <= 3 && "Invalid work-group dimensionality");
 
-  std::array<std::optional<size_t>, 3> Ops;
+  std::array<std::optional<uint64_t>, 3> Ops;
   for (unsigned I = 0, E = std::min(Node.getNumOperands(), NDim); I != E; I++) {
     if (auto *C = mdconst::dyn_extract<ConstantInt>(Node.getOperand(I)))
       Ops[I] = C->getZExtValue();

llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp

@@ -3731,8 +3731,12 @@ void FunctionStackPoisoner::copyArgsPassedByValToAllocas() {
       const Align Alignment =
           DL.getValueOrABITypeAlignment(Arg.getParamAlign(), Ty);
 
+      unsigned int AS = Triple(F.getParent()->getTargetTriple()).isSPIROrSPIRV()
+                            ? Arg.getType()->getPointerAddressSpace()
+                            : DL.getAllocaAddrSpace();
+
       AllocaInst *AI = IRB.CreateAlloca(
-          Ty, nullptr,
+          Ty, AS, nullptr,
           (Arg.hasName() ? Arg.getName() : "Arg" + Twine(Arg.getArgNo())) +
               ".byval");
       AI->setAlignment(Alignment);

llvm/test/Instrumentation/AddressSanitizer/SPIRV/byval_arg.ll

@@ -0,0 +1,15 @@
+; RUN: opt < %s -passes=asan -asan-instrumentation-with-call-threshold=0 -asan-stack=0 -asan-globals=0 -asan-use-after-return=never -asan-stack-dynamic-alloca=0 -asan-mapping-scale=4 -S | FileCheck %s
+
+target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64-G1"
+target triple = "spir64-unknown-unknown"
+
+%struct.Input = type { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 }
+
+; Function Attrs: sanitize_address
+define spir_func void @test(ptr addrspace(4) byval(%struct.Input) %input) #0 {
+entry:
+  ; CHECK: inttoptr i64 %1 to ptr addrspace(4)
+  ret void
+}
+
+attributes #0 = { sanitize_address }

sycl-fusion/CMakeLists.txt renamed to sycl-jit/CMakeLists.txt

@@ -10,15 +10,15 @@ set(SYCL_JIT_BASE_DIR ${CMAKE_CURRENT_SOURCE_DIR})
 set(LLVM_SPIRV_INCLUDE_DIRS "${LLVM_MAIN_SRC_DIR}/../llvm-spirv/include")
 
 # Set library-wide warning options.
-set(SYCL_FUSION_WARNING_FLAGS -Wall -Wextra)
+set(SYCL_JIT_WARNING_FLAGS -Wall -Wextra)
 
-option(SYCL_FUSION_ENABLE_WERROR "Treat all warnings as errors in SYCL kernel fusion library" ON)
-if(SYCL_FUSION_ENABLE_WERROR)
-  list(APPEND SYCL_FUSION_WARNING_FLAGS -Werror)
-endif(SYCL_FUSION_ENABLE_WERROR)
+option(SYCL_JIT_ENABLE_WERROR "Treat all warnings as errors in SYCL kernel JIT library" ON)
+if(SYCL_JIT_ENABLE_WERROR)
+  list(APPEND SYCL_JIT_WARNING_FLAGS -Werror)
+endif(SYCL_JIT_ENABLE_WERROR)
 
 if(WIN32)
-  message(WARNING "Kernel fusion not yet supported on Windows")
+  message(WARNING "Kernel JIT not yet supported on Windows")
 else(WIN32)
   add_subdirectory(common)
   add_subdirectory(jit-compiler)

sycl-fusion/README.md renamed to sycl-jit/README.md

@@ -1,35 +1,35 @@
-add_llvm_library(sycl-fusion-common
+add_llvm_library(sycl-jit-common
   lib/NDRangesHelper.cpp
 
   LINK_COMPONENTS
    Support
 )
 
-target_compile_options(sycl-fusion-common PRIVATE ${SYCL_FUSION_WARNING_FLAGS})
+target_compile_options(sycl-jit-common PRIVATE ${SYCL_JIT_WARNING_FLAGS})
 
 # Mark LLVM headers as system headers to ignore warnigns in them. This
 # classification remains intact even if the same path is added as a normal
 # include path in GCC and Clang.
-target_include_directories(sycl-fusion-common
+target_include_directories(sycl-jit-common
   SYSTEM PRIVATE
   ${LLVM_MAIN_INCLUDE_DIR}
 )
-target_include_directories(sycl-fusion-common
+target_include_directories(sycl-jit-common
   PUBLIC
   ${CMAKE_CURRENT_SOURCE_DIR}/include
   ${CMAKE_CURRENT_SOURCE_DIR}/lib
 )
 
-add_dependencies(sycl-fusion-common sycl-headers)
+add_dependencies(sycl-jit-common sycl-headers)
 
 if (BUILD_SHARED_LIBS)
   if(NOT MSVC AND NOT APPLE)
     # Manage symbol visibility through the linker to make sure no LLVM symbols
     # are exported and confuse the drivers.
     set(linker_script "${CMAKE_CURRENT_SOURCE_DIR}/ld-version-script.txt")
     target_link_libraries(
-      sycl-fusion-common PRIVATE "-Wl,--version-script=${linker_script}")
-    set_target_properties(sycl-fusion-common
+      sycl-jit-common PRIVATE "-Wl,--version-script=${linker_script}")
+    set_target_properties(sycl-jit-common
       PROPERTIES
       LINK_DEPENDS
       ${linker_script})

sycl-fusion/common/CMakeLists.txt renamed to sycl-jit/common/CMakeLists.txt

@@ -1,5 +1,5 @@
 
-add_llvm_library(sycl-fusion
+add_llvm_library(sycl-jit
    lib/KernelFusion.cpp
    lib/translation/KernelTranslation.cpp
    lib/translation/SPIRVLLVMTranslation.cpp
@@ -31,17 +31,17 @@ add_llvm_library(sycl-fusion
    ${LLVM_TARGETS_TO_BUILD}
 )
 
-target_compile_options(sycl-fusion PRIVATE ${SYCL_FUSION_WARNING_FLAGS})
+target_compile_options(sycl-jit PRIVATE ${SYCL_JIT_WARNING_FLAGS})
 
 # Mark LLVM and SPIR-V headers as system headers to ignore warnigns in them.
 # This classification remains intact even if the same paths are added as normal
 # include paths in GCC and Clang.
-target_include_directories(sycl-fusion
+target_include_directories(sycl-jit
   SYSTEM PRIVATE
   ${LLVM_MAIN_INCLUDE_DIR}
   ${LLVM_SPIRV_INCLUDE_DIRS}
 )
-target_include_directories(sycl-fusion
+target_include_directories(sycl-jit
   PUBLIC
   $<INSTALL_INTERFACE:include>
   $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>
@@ -52,29 +52,29 @@ target_include_directories(sycl-fusion
 
 find_package(Threads REQUIRED)
 
-target_link_libraries(sycl-fusion
+target_link_libraries(sycl-jit
   PRIVATE
-  sycl-fusion-common
+  sycl-jit-common
   LLVMSPIRVLib
-  SYCLKernelFusionPasses
+  SYCLKernelJITPasses
   ${CMAKE_THREAD_LIBS_INIT}
 )
 
-add_dependencies(sycl-fusion sycl-headers)
+add_dependencies(sycl-jit sycl-headers)
 
 if("NVPTX" IN_LIST LLVM_TARGETS_TO_BUILD)
-  target_compile_definitions(sycl-fusion PRIVATE FUSION_JIT_SUPPORT_PTX)
+  target_compile_definitions(sycl-jit PRIVATE JIT_SUPPORT_PTX)
 endif()
 
 if("AMDGPU" IN_LIST LLVM_TARGETS_TO_BUILD)
-  target_compile_definitions(sycl-fusion PRIVATE FUSION_JIT_SUPPORT_AMDGCN)
+  target_compile_definitions(sycl-jit PRIVATE JIT_SUPPORT_AMDGCN)
 endif()
 
 if(NOT MSVC AND NOT APPLE)
   # Manage symbol visibility through the linker to make sure no LLVM symbols
   # are exported and confuse the drivers.
   set(linker_script "${CMAKE_CURRENT_SOURCE_DIR}/ld-version-script.txt")
   target_link_libraries(
-    sycl-fusion PRIVATE "-Wl,--version-script=${linker_script}")
-  set_target_properties(sycl-fusion PROPERTIES LINK_DEPENDS ${linker_script})
+    sycl-jit PRIVATE "-Wl,--version-script=${linker_script}")
+  set_target_properties(sycl-jit PROPERTIES LINK_DEPENDS ${linker_script})
 endif()