[SYCL][Graph] 3D kernel update regression test

.github/CODEOWNERS

@@ -126,6 +126,7 @@ sycl/test-e2e/KernelFusion @intel/dpcpp-kernel-fusion-reviewers
 sycl/include/sycl/ext/oneapi/matrix/ @intel/sycl-matrix-reviewers
 sycl/test-e2e/Matrix @intel/sycl-matrix-reviewers
 sycl/test/matrix @intel/sycl-matrix-reviewers
+sycl/test/check_device_code/matrix @intel/sycl-matrix-reviewers
 
 # Native CPU
 llvm/**/*SYCLNativeCPU* @intel/dpcpp-nativecpu-pi-reviewers 

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

@@ -64,7 +64,7 @@ jobs:
  with:
  name: CUDA E2E
  runner: '["aws_cuda-${{ github.event.workflow_run.id }}-${{ github.event.workflow_run.run_attempt }}"]'
- image: ghcr.io/intel/llvm/ubuntu2204_build:latest
+ 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
  # No idea why but that seems to work and be in sync with the main

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

@@ -46,6 +46,7 @@ jobs:
  build_artifact_suffix: "default"
  build_cache_suffix: "default"
  changes: ${{ needs.detect_changes.outputs.filters }}
+ build_image: "ghcr.io/intel/llvm/ubuntu2204_build:latest-0300ac924620a51f76c4929794637b82790f12ab"
 
  determine_arc_tests:
  name: Decide which Arc tests to run
@@ -77,7 +78,7 @@ jobs:
  include:
  - name: AMD/HIP
  runner: '["Linux", "amdgpu"]'
- image: ghcr.io/intel/llvm/ubuntu2204_build:latest
+ 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

.github/workflows/sycl-nightly.yml

@@ -74,13 +74,6 @@ jobs:
  target_devices: opencl:cpu
  tests_selector: e2e
 
- - name: Self-hosted 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: SYCL-CTS on OCL CPU
  runner: '["Linux", "gen12"]'
  image: ghcr.io/intel/llvm/ubuntu2204_intel_drivers:latest

clang/lib/AST/ASTContext.cpp

@@ -12035,10 +12035,10 @@ bool ASTContext::DeclMustBeEmitted(const Decl *D) {
  // or `indirectly_callable' attribute must be emitted regardless of number
  // of actual uses
  if (LangOpts.SYCLIsDevice && isa<CXXMethodDecl>(D)) {
- if (auto *A = D->getAttr<SYCLDeviceIndirectlyCallableAttr>())
- return !A->isImplicit();
- if (auto *A = D->getAttr<SYCLDeviceAttr>())
- return !A->isImplicit();
+ if (D->hasAttr<SYCLDeviceIndirectlyCallableAttr>())
+ return true;
+ if (D->hasAttr<SYCLDeviceAttr>())
+ return true;
  }
 
  GVALinkage Linkage = GetGVALinkageForFunction(FD);

clang/lib/Driver/ToolChains/Clang.cpp

@@ -5843,10 +5843,15 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
  CmdArgs.push_back("-emit-llvm-uselists");
 
  if (IsUsingLTO) {
+ bool IsUsingOffloadNewDriver =
+ Args.hasFlag(options::OPT_offload_new_driver,
+ options::OPT_no_offload_new_driver, false);
+ bool IsSYCLLTOSupported = JA.isDeviceOffloading(Action::OFK_SYCL) &&
+ Triple.isSPIROrSPIRV() &&
+ IsUsingOffloadNewDriver;
  if (IsDeviceOffloadAction && !JA.isDeviceOffloading(Action::OFK_OpenMP) &&
- !Args.hasFlag(options::OPT_offload_new_driver,
- options::OPT_no_offload_new_driver, false) &&
- !Triple.isAMDGPU()) {
+ !IsUsingOffloadNewDriver && !Triple.isAMDGPU() &&
+ !IsSYCLLTOSupported) {
  D.Diag(diag::err_drv_unsupported_opt_for_target)
  << Args.getLastArg(options::OPT_foffload_lto,
  options::OPT_foffload_lto_EQ)
@@ -10423,6 +10428,7 @@ static void getOtherSPIRVTransOpts(Compilation &C,
  ",+SPV_INTEL_fpga_invocation_pipelining_attributes"
  ",+SPV_INTEL_fpga_latency_control"
  ",+SPV_INTEL_task_sequence"
+ ",+SPV_KHR_shader_clock"
  ",+SPV_INTEL_bindless_images";
  ExtArg = ExtArg + DefaultExtArg + INTELExtArg;
  if (C.getDriver().IsFPGAHWMode())

clang/test/CodeGenSYCL/force-emit-device-virtual-funcs.cpp

@@ -0,0 +1,47 @@
+// RUN: %clang_cc1 -internal-isystem %S/Inputs -triple spir64-unknown-unknown -fsycl-is-device \
+// RUN: -fsycl-allow-virtual-functions -emit-llvm %s -o %t.ll
+// RUN: FileCheck %s --input-file=%t.ll --implicit-check-not _ZN7Derived3baz \
+// RUN: --implicit-check-not _ZN4Base4baz --implicit-check-not _ZN4Base3foo
+//
+// Some SYCL properties may be turned into 'sycl_device' attribute implicitly
+// and we would like to ensure that functions like this (at the moment those
+// would be virtual member functions only) are forcefully emitted into device
+// code.
+
+class Base {
+ virtual void foo() {}
+
+ virtual void baz();
+
+ [[__sycl_detail__::add_ir_attributes_function("indirectly-callable", "a")]]
+ virtual void bar();
+};
+
+void Base::bar() {}
+
+void Base::baz() {}
+
+class Derived : public Base {
+public:
+ [[__sycl_detail__::add_ir_attributes_function("indirectly-callable", "b")]]
+ void foo() override;
+
+ [[__sycl_detail__::add_ir_attributes_function("indirectly-callable", "c")]]
+ void bar() override final;
+
+ [[__sycl_detail__::add_ir_attributes_function("not-indirectly-callable", "c")]]
+ void baz() override final;
+};
+
+void Derived::foo() {}
+
+void Derived::bar() {}
+
+void Derived::baz() {}
+
+// CHECK: define {{.*}}spir_func void @_ZN4Base3bar{{.*}} #[[#AttrA:]]
+// CHECK: define {{.*}}spir_func void @_ZN7Derived3foo{{.*}} #[[#AttrB:]]
+// CHECK: define {{.*}}spir_func void @_ZN7Derived3bar{{.*}} #[[#AttrC:]]
+// CHECK: attributes #[[#AttrA]] = {{.*}} "indirectly-callable"="a"
+// CHECK: attributes #[[#AttrB]] = {{.*}} "indirectly-callable"="b"
+// CHECK: attributes #[[#AttrC]] = {{.*}} "indirectly-callable"="c"

clang/test/Driver/sycl-lto.cpp

@@ -0,0 +1,9 @@
+// Verify the usage of -foffload-lto with SYCL.
+
+// Verify we error when using the old offload driver.
+// RUN: not %clangxx -fsycl -foffload-lto=thin %s -### 2>&1 | FileCheck -check-prefix=CHECK_ERROR %s
+// CHECK_ERROR: unsupported option '-foffload-lto=thin' for target 'spir64-unknown-unknown'
+
+// Verify there's no error and we see the expected cc1 flags with the new offload driver.
+// RUN: %clangxx -fsycl --offload-new-driver -foffload-lto=thin %s -### 2>&1 | FileCheck -check-prefix=CHECK_SUPPORTED %s
+// CHECK_SUPPORTED: clang{{.*}} "-cc1" "-triple" "spir64-unknown-unknown" {{.*}} "-flto=thin" "-flto-unit"

clang/test/Driver/sycl-spirv-ext.c

@@ -48,6 +48,7 @@
 // CHECK-DEFAULT-SAME:,+SPV_INTEL_fpga_invocation_pipelining_attributes
 // CHECK-DEFAULT-SAME:,+SPV_INTEL_fpga_latency_control
 // CHECK-DEFAULT-SAME:,+SPV_INTEL_task_sequence
+// CHECK-DEFAULT-SAME:,+SPV_KHR_shader_clock
 // CHECK-DEFAULT-SAME:,+SPV_INTEL_bindless_images
 // CHECK-DEFAULT-SAME:,+SPV_INTEL_token_type
 // CHECK-DEFAULT-SAME:,+SPV_INTEL_bfloat16_conversion
@@ -125,4 +126,3 @@
 // CHECK-CPU-SAME:,+SPV_KHR_non_semantic_info
 // CHECK-CPU-SAME:,+SPV_KHR_cooperative_matrix
 // CHECK-CPU-SAME:,+SPV_INTEL_fp_max_error"
-

devops/containers/ubuntu2204_build.Dockerfile

@@ -1,4 +1,4 @@
-FROM nvidia/cuda:12.1.0-devel-ubuntu22.04
+FROM nvidia/cuda:12.5.0-devel-ubuntu22.04
 
 ENV DEBIAN_FRONTEND=noninteractive
 

devops/dependencies-igc-dev.json

@@ -1,10 +1,10 @@
 {
  "linux": {
  "igc_dev": {
- "github_tag": "igc-dev-4627f1f",
- "version": "4627f1f",
- "updated_at": "2024-05-26T23:48:05Z",
- "url": "https://api.github.com/repos/intel/intel-graphics-compiler/actions/artifacts/1539236241/zip",
+ "github_tag": "igc-dev-8b999ec",
+ "version": "8b999ec",
+ "updated_at": "2024-05-30T02:09:07Z",
+ "url": "https://api.github.com/repos/intel/intel-graphics-compiler/actions/artifacts/1550749489/zip",
  "root": "{DEPS_ROOT}/opencl/runtime/linux/oclgpu"
  }
  }

libdevice/cmake/modules/SYCLLibdevice.cmake

@@ -91,7 +91,7 @@ function(add_devicelib_obj obj_filename)
 
  set(devicelib-obj-file-new-offload ${obj_new_offload_binary_dir}/${obj_filename}.${new-offload-lib-suffix})
  add_custom_command(OUTPUT ${devicelib-obj-file-new-offload}
- COMMAND ${clang} -fsycl -c --offload-new-driver
+ COMMAND ${clang} -fsycl -c --offload-new-driver -foffload-lto=thin
  ${compile_opts} ${sycl_targets_opt} ${OBJ_EXTRA_ARGS}
  ${CMAKE_CURRENT_SOURCE_DIR}/${OBJ_SRC}
  -o ${devicelib-obj-file-new-offload}
@@ -270,7 +270,7 @@ add_custom_command(OUTPUT ${obj_binary_dir}/libsycl-fallback-imf.${lib-suffix}
  VERBATIM)
 
 add_custom_command(OUTPUT ${obj_binary_dir}/libsycl-fallback-imf.${new-offload-lib-suffix}
- COMMAND ${clang} -fsycl -c --offload-new-driver
+ COMMAND ${clang} -fsycl -c --offload-new-driver -foffload-lto=thin
  ${compile_opts} ${sycl_targets_opt}
  ${imf_fp32_fallback_src} -I ${CMAKE_CURRENT_SOURCE_DIR}/imf
  -o ${obj_binary_dir}/libsycl-fallback-imf.${new-offload-lib-suffix}
@@ -286,7 +286,7 @@ add_custom_command(OUTPUT ${obj_binary_dir}/fallback-imf-fp32-host.${lib-suffix}
  VERBATIM)
 
 add_custom_command(OUTPUT ${obj_binary_dir}/fallback-imf-fp32-host.${new-offload-lib-suffix}
- COMMAND ${clang} ${imf_host_cxx_flags} --offload-new-driver
+ COMMAND ${clang} ${imf_host_cxx_flags} --offload-new-driver -foffload-lto=thin
  -I ${CMAKE_CURRENT_SOURCE_DIR}/imf
  ${imf_fp32_fallback_src}
  -o ${obj_binary_dir}/fallback-imf-fp32-host.${new-offload-lib-suffix}
@@ -321,7 +321,7 @@ add_custom_command(OUTPUT ${obj_binary_dir}/libsycl-fallback-imf-fp64.${lib-suff
 
 add_custom_command(OUTPUT ${obj_binary_dir}/libsycl-fallback-imf-fp64.${new-offload-lib-suffix}
  COMMAND ${clang} -fsycl -c -I ${CMAKE_CURRENT_SOURCE_DIR}/imf
- --offload-new-driver
+ --offload-new-driver -foffload-lto=thin
  ${compile_opts} ${sycl_targets_opt}
  ${imf_fp64_fallback_src}
  -o ${obj_binary_dir}/libsycl-fallback-imf-fp64.${new-offload-lib-suffix}
@@ -337,7 +337,7 @@ add_custom_command(OUTPUT ${obj_binary_dir}/fallback-imf-fp64-host.${lib-suffix}
  VERBATIM)
 
 add_custom_command(OUTPUT ${obj_binary_dir}/fallback-imf-fp64-host.${new-offload-lib-suffix}
- COMMAND ${clang} ${imf_host_cxx_flags} --offload-new-driver
+ COMMAND ${clang} ${imf_host_cxx_flags} --offload-new-driver -foffload-lto=thin
  -I ${CMAKE_CURRENT_SOURCE_DIR}/imf
  ${imf_fp64_fallback_src}
  -o ${obj_binary_dir}/fallback-imf-fp64-host.${new-offload-lib-suffix}
@@ -372,7 +372,7 @@ add_custom_command(OUTPUT ${obj_binary_dir}/libsycl-fallback-imf-bf16.${lib-suff
 
 add_custom_command(OUTPUT ${obj_binary_dir}/libsycl-fallback-imf-bf16.${new-offload-lib-suffix}
  COMMAND ${clang} -fsycl -c -I ${CMAKE_CURRENT_SOURCE_DIR}/imf
- --offload-new-driver
+ --offload-new-driver -foffload-lto=thin
  ${compile_opts} ${sycl_targets_opt}
  ${imf_bf16_fallback_src}
  -o ${obj_binary_dir}/libsycl-fallback-imf-bf16.${new-offload-lib-suffix}
@@ -388,7 +388,7 @@ add_custom_command(OUTPUT ${obj_binary_dir}/fallback-imf-bf16-host.${lib-suffix}
  VERBATIM)
 
 add_custom_command(OUTPUT ${obj_binary_dir}/fallback-imf-bf16-host.${new-offload-lib-suffix}
- COMMAND ${clang} ${imf_host_cxx_flags} --offload-new-driver
+ COMMAND ${clang} ${imf_host_cxx_flags} --offload-new-driver -foffload-lto=thin
  -I ${CMAKE_CURRENT_SOURCE_DIR}/imf
  ${imf_bf16_fallback_src}
  -o ${obj_binary_dir}/fallback-imf-bf16-host.${new-offload-lib-suffix}
@@ -437,7 +437,7 @@ add_custom_command(OUTPUT ${obj_binary_dir}/imf-fp32-host.${lib-suffix}
  VERBATIM)
 
 add_custom_command(OUTPUT ${obj_binary_dir}/imf-fp32-host.${new-offload-lib-suffix}
- COMMAND ${clang} ${imf_host_cxx_flags} --offload-new-driver
+ COMMAND ${clang} ${imf_host_cxx_flags} --offload-new-driver -foffload-lto=thin
  ${CMAKE_CURRENT_SOURCE_DIR}/imf_wrapper.cpp
  -o ${obj_binary_dir}/imf-fp32-host.${new-offload-lib-suffix}
  MAIN_DEPENDENCY ${CMAKE_CURRENT_SOURCE_DIR}/imf_wrapper.cpp
@@ -453,7 +453,7 @@ add_custom_command(OUTPUT ${obj_binary_dir}/imf-fp64-host.${lib-suffix}
  VERBATIM)
 
 add_custom_command(OUTPUT ${obj_binary_dir}/imf-fp64-host.${new-offload-lib-suffix}
- COMMAND ${clang} ${imf_host_cxx_flags} --offload-new-driver
+ COMMAND ${clang} ${imf_host_cxx_flags} --offload-new-driver -foffload-lto=thin
  ${CMAKE_CURRENT_SOURCE_DIR}/imf_wrapper_fp64.cpp
  -o ${obj_binary_dir}/imf-fp64-host.${new-offload-lib-suffix}
  MAIN_DEPENDENCY ${CMAKE_CURRENT_SOURCE_DIR}/imf_wrapper_fp64.cpp
@@ -469,7 +469,7 @@ add_custom_command(OUTPUT ${obj_binary_dir}/imf-bf16-host.${lib-suffix}
  VERBATIM)
 
 add_custom_command(OUTPUT ${obj_binary_dir}/imf-bf16-host.${new-offload-lib-suffix}
- COMMAND ${clang} ${imf_host_cxx_flags} --offload-new-driver
+ COMMAND ${clang} ${imf_host_cxx_flags} --offload-new-driver -foffload-lto=thin
  ${CMAKE_CURRENT_SOURCE_DIR}/imf_wrapper_bf16.cpp
  -o ${obj_binary_dir}/imf-bf16-host.${new-offload-lib-suffix}
  MAIN_DEPENDENCY ${CMAKE_CURRENT_SOURCE_DIR}/imf_wrapper_bf16.cpp

llvm/include/llvm/SYCLLowerIR/ModuleSplitter.h

@@ -19,6 +19,7 @@
 #include "llvm/ADT/StringRef.h"
 #include "llvm/IR/Function.h"
 #include "llvm/Support/Error.h"
+#include "llvm/Support/PropertySetIO.h"
 
 #include <memory>
 #include <string>
@@ -196,6 +197,8 @@ class ModuleDesc {
 
  ModuleDesc clone() const;
 
+ std::string makeSymbolTable() const;
+
  const SYCLDeviceRequirements &getOrComputeDeviceRequirements() const {
  if (!Reqs.has_value())
  Reqs = computeDeviceRequirements(*this);
@@ -270,6 +273,33 @@ void dumpEntryPoints(const Module &M, bool OnlyKernelsAreEntryPoints = false,
  const char *msg = "", int Tab = 0);
 #endif // NDEBUG
 
+struct SplitModule {
+ std::string ModuleFilePath;
+ util::PropertySetRegistry Properties;
+ std::string Symbols;
+
+ SplitModule() = default;
+ SplitModule(const SplitModule &) = default;
+ SplitModule &operator=(const SplitModule &) = default;
+ SplitModule(SplitModule &&) = default;
+ SplitModule &operator=(SplitModule &&) = default;
+
+ SplitModule(std::string_view File, util::PropertySetRegistry Properties,
+ std::string Symbols)
+ : ModuleFilePath(File), Properties(std::move(Properties)),
+ Symbols(std::move(Symbols)) {}
+};
+
+struct ModuleSplitterSettings {
+ IRSplitMode Mode;
+ bool OutputAssembly = false; // Bitcode or LLVM IR.
+ StringRef OutputPrefix;
+};
+
+/// Splits the given module \p M according to the given \p Settings.
+Expected<std::vector<SplitModule>>
+splitSYCLModule(std::unique_ptr<Module> M, ModuleSplitterSettings Settings);
+
 } // namespace module_split
 
 } // namespace llvm

llvm/include/llvm/SYCLLowerIR/SYCLDeviceRequirements.h

@@ -30,7 +30,19 @@ class PropertyValue;
 }
 
 struct SYCLDeviceRequirements {
- std::set<uint32_t> Aspects;
+ struct AspectNameValuePair {
+ llvm::SmallString<64> Name;
+ uint32_t Value;
+ AspectNameValuePair(StringRef Name, uint32_t Value)
+ : Name(Name), Value(Value) {}
+ bool operator<(const AspectNameValuePair &rhs) const {
+ return Value < rhs.Value;
+ }
+ bool operator==(const AspectNameValuePair &rhs) const {
+ return Value == rhs.Value;
+ }
+ };
+ std::set<AspectNameValuePair> Aspects;
  std::set<uint32_t> FixedTarget;
  std::optional<llvm::SmallVector<uint64_t, 3>> ReqdWorkGroupSize;
  std::optional<uint32_t> WorkGroupNumDim;

llvm/lib/SYCLLowerIR/CMakeLists.txt

@@ -88,10 +88,12 @@ add_llvm_component_library(LLVMSYCLLowerIR
  LLVMDemangle
  LLVMTargetParser
  LLVMTransformUtils
- 
+
  LINK_COMPONENTS
  Analysis
+ BitWriter
  Core
+ IRPrinter
  Support
  ipo
  )