Merge branch 'sycl' into przemek/bindless-img-host-usm-1d-test

Author: przemektmalon

.github/workflows/sycl-ur-perf-benchmarking.yml

@@ -0,0 +1,12 @@
+name: Benchmarks
+
+# This workflow is a WIP: this workflow file acts as a placeholder.
+
+on: [ workflow_dispatch ]
+
+jobs:
+  do-nothing:
+    runs-on: ubuntu-latest
+    steps:
+      - run: echo 'This workflow is a WIP.'
+

.github/workflows/ur-benchmarks-reusable.yml

@@ -1,6 +1,8 @@
 name: Benchmarks Reusable
 
 # This workflow is a WIP: This workflow file acts as a placeholder.
+#
+# This workflow is set to be merged into benchmark.yml
 
 on: [ workflow_call ]
 

.github/workflows/ur-benchmarks.yml

@@ -1,6 +1,8 @@
 name: Benchmarks
 
 # This workflow is a WIP: this workflow file acts as a placeholder.
+#
+# This workflow is set to be merged into benchmark.yml
 
 on: [ workflow_dispatch ]
 

clang/test/Driver/linker-wrapper-sycl.cpp

@@ -47,11 +47,6 @@
 // RUN: clang-linker-wrapper -sycl-device-libraries=%t.devicelib.o -sycl-post-link-options="SYCL_POST_LINK_OPTIONS" -llvm-spirv-options="LLVM_SPIRV_OPTIONS" "--host-triple=x86_64-unknown-linux-gnu" "--triple=spir64" "--linker-path=/usr/bin/ld" -shared "--" HOST_LINKER_FLAGS "-dynamic-linker" HOST_DYN_LIB "-o" "a.out" HOST_LIB_PATH HOST_STAT_LIB %t.o --dry-run 2>&1 | FileCheck -check-prefix=CHK-SHARED %s
 // CHK-SHARED: "{{.*}}clang"{{.*}} -fPIC
 
-// RUN: rm %T/linker_wrapper_dump || true
-// RUN: clang-linker-wrapper -sycl-dump-device-code=%T/linker_wrapper_dump -sycl-device-libraries=%t.devicelib.o "--host-triple=x86_64-unknown-linux-gnu" "--triple=spir64" "--linker-path=/usr/bin/ld" -shared "--" HOST_LINKER_FLAGS "-dynamic-linker" HOST_DYN_LIB "-o" "a.out" HOST_LIB_PATH HOST_STAT_LIB %t.o --dry-run
-// RUN: ls %T/linker_wrapper_dump | FileCheck -check-prefix=CHK-SYCL-DUMP-DEVICE %s
-// CHK-SYCL-DUMP-DEVICE: {{.*}}.spv
-
 /// Check for list of commands for standalone clang-linker-wrapper run for sycl (AOT for Intel GPU)
 // -------
 // Generate .o file as linker wrapper input.

clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp

@@ -151,7 +151,7 @@ static std::optional<llvm::module_split::IRSplitMode> SYCLModuleSplitMode;
 
 static bool UseSYCLPostLinkTool;
 
-static SmallString<128> SPIRVDumpDir;
+static SmallString<128> OffloadImageDumpDir;
 
 using OffloadingImage = OffloadBinary::OffloadingImage;
 
@@ -923,29 +923,6 @@ static Expected<StringRef> runLLVMToSPIRVTranslation(StringRef File,
   if (Error Err = executeCommands(*LLVMToSPIRVPath, CmdArgs))
     return std::move(Err);
 
-  if (!SPIRVDumpDir.empty()) {
-    std::error_code EC =
-        llvm::sys::fs::create_directory(SPIRVDumpDir, /*IgnoreExisting*/ true);
-    if (EC)
-      return createStringError(
-          EC,
-          formatv("failed to create dump directory. path: {0}, error_code: {1}",
-                  SPIRVDumpDir, EC.value()));
-
-    StringRef Sep = llvm::sys::path::get_separator();
-    StringRef Path = *TempFileOrErr;
-    StringRef Filename = Path.rsplit(Sep).second;
-    SmallString<128> CopyPath = SPIRVDumpDir;
-    CopyPath.append(Filename);
-    EC = llvm::sys::fs::copy_file(Path, CopyPath);
-    if (EC)
-      return createStringError(
-          EC,
-          formatv(
-              "failed to copy file. original: {0}, copy: {1}, error_code: {2}",
-              Path, CopyPath, EC.value()));
-  }
-
   return *TempFileOrErr;
 }
 
@@ -1103,6 +1080,18 @@ wrapSYCLBinariesFromFile(std::vector<module_split::SplitModule> &SplitModules,
     RegularTarget = "spir64";
 
   for (auto &SI : SplitModules) {
+    if (!OffloadImageDumpDir.empty()) {
+      StringRef CopyFrom = SI.ModuleFilePath;
+      SmallString<128> CopyTo = OffloadImageDumpDir;
+      StringRef Filename = sys::path::filename(CopyFrom);
+      CopyTo.append(Filename);
+      std::error_code EC = sys::fs::copy_file(CopyFrom, CopyTo);
+      if (EC)
+        return createStringError(EC, formatv("failed to copy file. From: "
+                                             "{0} to: {1}, error_code: {2}",
+                                             CopyFrom, CopyTo, EC.value()));
+    }
+
     auto MBOrDesc = MemoryBuffer::getFile(SI.ModuleFilePath);
     if (!MBOrDesc)
       return createFileError(SI.ModuleFilePath, MBOrDesc.getError());
@@ -2618,13 +2607,11 @@ int main(int Argc, char **Argv) {
 
   if (Args.hasArg(OPT_sycl_dump_device_code_EQ)) {
     Arg *A = Args.getLastArg(OPT_sycl_dump_device_code_EQ);
-    SmallString<128> Dir(A->getValue());
-    if (Dir.empty())
-      llvm::sys::path::native(Dir = "./");
+    OffloadImageDumpDir = A->getValue();
+    if (OffloadImageDumpDir.empty())
+      sys::path::native(OffloadImageDumpDir = "./");
     else
-      Dir.append(llvm::sys::path::get_separator());
-
-    SPIRVDumpDir = Dir;
+      OffloadImageDumpDir.append(sys::path::get_separator());
   }
 
   {

clang/tools/clang-linker-wrapper/LinkerWrapperOpts.td

@@ -242,7 +242,7 @@ Flags<[WrapperOnlyOption]>,  HelpText<"Embed LLVM IR for runtime kernel fusion">
 
 def sycl_dump_device_code_EQ : Joined<["--", "-"], "sycl-dump-device-code=">,
   Flags<[WrapperOnlyOption]>,
-  HelpText<"Path to the folder where the tool dumps SPIR-V device code. Other formats aren't dumped.">;
+  HelpText<"Directory to dump offloading images to.">;
 
 // Options to enable/disable device dynamic linking.
 def sycl_allow_device_image_dependencies : Flag<["--", "-"], "sycl-allow-device-image-dependencies">,

devops/scripts/benchmarks/benches/base.py

@@ -1,16 +1,37 @@
-# Copyright (C) 2024 Intel Corporation
+# Copyright (C) 2024-2025 Intel Corporation
 # Part of the Unified-Runtime Project, under the Apache License v2.0 with LLVM Exceptions.
 # See LICENSE.TXT
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
+from dataclasses import dataclass
 import os
 import shutil
 from pathlib import Path
-from .result import Result
+from utils.result import BenchmarkMetadata, BenchmarkTag, Result
 from options import options
 from utils.utils import download, run
-import urllib.request
-import tarfile
+
+benchmark_tags = [
+    BenchmarkTag("SYCL", "Benchmark uses SYCL runtime"),
+    BenchmarkTag("UR", "Benchmark uses Unified Runtime API"),
+    BenchmarkTag("L0", "Benchmark uses Level Zero API directly"),
+    BenchmarkTag("UMF", "Benchmark uses Unified Memory Framework directly"),
+    BenchmarkTag("micro", "Microbenchmark focusing on a specific functionality"),
+    BenchmarkTag("application", "Real application-based performance test"),
+    BenchmarkTag("proxy", "Benchmark that simulates real application use-cases"),
+    BenchmarkTag("submit", "Tests kernel submission performance"),
+    BenchmarkTag("math", "Tests math computation performance"),
+    BenchmarkTag("memory", "Tests memory transfer or bandwidth performance"),
+    BenchmarkTag("allocation", "Tests memory allocation performance"),
+    BenchmarkTag("graph", "Tests graph-based execution performance"),
+    BenchmarkTag("latency", "Measures operation latency"),
+    BenchmarkTag("throughput", "Measures operation throughput"),
+    BenchmarkTag("inference", "Tests ML/AI inference performance"),
+    BenchmarkTag("image", "Image processing benchmark"),
+    BenchmarkTag("simulation", "Physics or scientific simulation benchmark"),
+]
+
+benchmark_tags_dict = {tag.name: tag for tag in benchmark_tags}
 
 
 class Benchmark:
@@ -55,19 +76,25 @@ def create_data_path(self, name, skip_data_dir=False):
             data_path = os.path.join(self.directory, name)
         else:
             data_path = os.path.join(self.directory, "data", name)
-            if options.rebuild and Path(data_path).exists():
+            if options.redownload and Path(data_path).exists():
                 shutil.rmtree(data_path)
 
         Path(data_path).mkdir(parents=True, exist_ok=True)
 
         return data_path
 
-    def download(self, name, url, file, untar=False, unzip=False, skip_data_dir=False):
+    def download(
+        self,
+        name,
+        url,
+        file,
+        untar=False,
+        unzip=False,
+        skip_data_dir=False,
+        checksum="",
+    ):
         self.data_path = self.create_data_path(name, skip_data_dir)
-        return download(self.data_path, url, file, untar, unzip)
-
-    def name(self):
-        raise NotImplementedError()
+        return download(self.data_path, url, file, untar, unzip, checksum)
 
     def lower_is_better(self):
         return True
@@ -87,6 +114,30 @@ def stddev_threshold(self):
     def get_suite_name(self) -> str:
         return self.suite.name()
 
+    def name(self):
+        raise NotImplementedError()
+
+    def description(self):
+        return ""
+
+    def notes(self) -> str:
+        return None
+
+    def unstable(self) -> str:
+        return None
+
+    def get_tags(self) -> list[str]:
+        return []
+
+    def get_metadata(self) -> BenchmarkMetadata:
+        return BenchmarkMetadata(
+            type="benchmark",
+            description=self.description(),
+            notes=self.notes(),
+            unstable=self.unstable(),
+            tags=self.get_tags(),
+        )
+
 
 class Suite:
     def benchmarks(self) -> list[Benchmark]:
@@ -97,3 +148,6 @@ def name(self) -> str:
 
     def setup(self):
         return
+
+    def additionalMetadata(self) -> dict[str, BenchmarkMetadata]:
+        return {}