Merge branch 'main' of https://github.com/oneapi-src/unified-runtime …

…into ianayl/2way-prefetch

.github/docker/install_dpcpp.sh

@@ -16,5 +16,5 @@ if [ "${SKIP_DPCPP_BUILD}" ]; then
 fi
 
 mkdir -p ${DPCPP_PATH}/dpcpp_compiler
-wget -O ${DPCPP_PATH}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-09-27/sycl_linux.tar.gz
+wget -O ${DPCPP_PATH}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-12-12/sycl_linux.tar.gz
 tar -xvf ${DPCPP_PATH}/dpcpp_compiler.tar.gz -C ${DPCPP_PATH}/dpcpp_compiler

.github/workflows/benchmarks-reusable.yml

@@ -176,11 +176,25 @@ jobs:
     - name: Build UMF
       run: cmake --build ${{github.workspace}}/umf_build -j $(nproc)
 
+    - name: Compute core range
+      run: |
+        # Compute the core range for the first NUMA node, skipping the first 4 cores.
+        # This is to avoid the first cores that the kernel is likely to schedule more work on.
+        CORES=$(lscpu | awk '
+          /NUMA node0 CPU|On-line CPU/ {line=$0}
+          END {
+            split(line, a, " ")
+            split(a[4], b, ",")
+            sub(/^0/, "4", b[1])
+            print b[1]
+          }')
+        echo "CORES=$CORES" >> $GITHUB_ENV
+
     - name: Run benchmarks
       working-directory: ${{ github.workspace }}/ur-repo/
       id: benchmarks
       run: >
-        numactl -N 0 ${{ github.workspace }}/ur-repo/scripts/benchmarks/main.py
+        taskset -c ${{ env.CORES }} ${{ github.workspace }}/ur-repo/scripts/benchmarks/main.py
         ~/bench_workdir
         --sycl ${{ github.workspace }}/sycl_build
         --ur ${{ github.workspace }}/ur_install

.github/workflows/build-fuzz-reusable.yml

@@ -35,7 +35,7 @@ jobs:
 
     - name: Download DPC++
       run: |
-        wget -O ${{github.workspace}}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-01-29/sycl_linux.tar.gz
+        wget -O ${{github.workspace}}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-12-12/sycl_linux.tar.gz
         mkdir dpcpp_compiler
         tar -xvf ${{github.workspace}}/dpcpp_compiler.tar.gz -C dpcpp_compiler
 

.github/workflows/build-hw-reusable.yml

@@ -77,7 +77,7 @@ jobs:
 
     - name: Download DPC++
       run: |
-        wget -O ${{github.workspace}}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-01-29/sycl_linux.tar.gz
+        wget -O ${{github.workspace}}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-12-12/sycl_linux.tar.gz
         mkdir dpcpp_compiler
         tar -xvf ${{github.workspace}}/dpcpp_compiler.tar.gz -C dpcpp_compiler
 

.github/workflows/cmake.yml

@@ -76,7 +76,7 @@ jobs:
       if: matrix.os == 'ubuntu-22.04'
       run: |
         sudo apt install libncurses5
-        wget -O ${{github.workspace}}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-09-27/sycl_linux.tar.gz
+        wget -O ${{github.workspace}}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-12-12/sycl_linux.tar.gz
         mkdir -p ${{github.workspace}}/dpcpp_compiler
         tar -xvf ${{github.workspace}}/dpcpp_compiler.tar.gz -C ${{github.workspace}}/dpcpp_compiler
 

.github/workflows/multi_device.yml

@@ -33,7 +33,7 @@ jobs:
 
     - name: Download DPC++
       run: |
-        wget -O ${{github.workspace}}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-01-29/sycl_linux.tar.gz
+        wget -O ${{github.workspace}}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-12-12/sycl_linux.tar.gz
         mkdir dpcpp_compiler
         tar -xvf ${{github.workspace}}/dpcpp_compiler.tar.gz -C dpcpp_compiler
 

include/ur_api.h

@@ -8563,6 +8563,7 @@ urCommandBufferReleaseExp(
 ///     - ::UR_RESULT_ERROR_INVALID_NULL_HANDLE
 ///         + `NULL == hCommandBuffer`
 ///     - ::UR_RESULT_ERROR_INVALID_COMMAND_BUFFER_EXP
+///     - ::UR_RESULT_ERROR_INVALID_OPERATION - "If `hCommandBuffer` has already been finalized"
 ///     - ::UR_RESULT_ERROR_OUT_OF_HOST_MEMORY
 ///     - ::UR_RESULT_ERROR_OUT_OF_RESOURCES
 UR_APIEXPORT ur_result_t UR_APICALL

scripts/benchmarks/benches/base.py

@@ -13,8 +13,9 @@
 import tarfile
 
 class Benchmark:
-    def __init__(self, directory):
+    def __init__(self, directory, suite):
         self.directory = directory
+        self.suite = suite
 
     @staticmethod
     def get_adapter_full_path():
@@ -74,9 +75,15 @@ def teardown(self):
     def stddev_threshold(self):
         return None
 
+    def get_suite_name(self) -> str:
+        return self.suite.name()
+
 class Suite:
     def benchmarks(self) -> list[Benchmark]:
         raise NotImplementedError()
 
+    def name(self) -> str:
+        raise NotImplementedError()
+
     def setup(self):
         return

scripts/benchmarks/benches/compute.py

@@ -15,6 +15,9 @@ class ComputeBench(Suite):
     def __init__(self, directory):
         self.directory = directory
 
+    def name(self) -> str:
+        return "Compute Benchmarks"
+
     def setup(self):
         if options.sycl is None:
             return
@@ -90,10 +93,10 @@ def parse_unit_type(compute_unit):
 
 class ComputeBenchmark(Benchmark):
     def __init__(self, bench, name, test):
+        super().__init__(bench.directory, bench)
         self.bench = bench
         self.bench_name = name
         self.test = test
-        super().__init__(bench.directory)
 
     def bin_args(self) -> list[str]:
         return []

scripts/benchmarks/benches/llamacpp.py

@@ -21,6 +21,9 @@ def __init__(self, directory):
 
         self.directory = directory
 
+    def name(self) -> str:
+        return "llama.cpp bench"
+
     def setup(self):
         if options.sycl is None:
             return
@@ -64,8 +67,8 @@ def benchmarks(self) -> list[Benchmark]:
 
 class LlamaBench(Benchmark):
     def __init__(self, bench):
+        super().__init__(bench.directory, bench)
         self.bench = bench
-        super().__init__(bench.directory)
 
     def setup(self):
         self.benchmark_bin = os.path.join(self.bench.build_path, 'bin', 'llama-bench')

scripts/benchmarks/benches/result.py

@@ -25,8 +25,9 @@ class Result:
     # values below should not be set by the benchmark
     name: str = ""
     lower_is_better: bool = True
-    git_hash: str = ''
+    git_hash: str = ""
     date: Optional[datetime] = None
+    suite: str = ""
 
 @dataclass_json
 @dataclass

scripts/benchmarks/benches/syclbench.py

@@ -19,6 +19,9 @@ def __init__(self, directory):
         self.directory = directory
         return
 
+    def name(self) -> str:
+        return "SYCL-Bench"
+
     def setup(self):
         if options.sycl is None:
             return
@@ -87,11 +90,11 @@ def benchmarks(self) -> list[Benchmark]:
 
 class SyclBenchmark(Benchmark):
     def __init__(self, bench, name, test):
+        super().__init__(bench.directory, bench)
         self.bench = bench
         self.bench_name = name
         self.test = test
         self.done = False
-        super().__init__(bench.directory)
 
     def bin_args(self) -> list[str]:
         return []

scripts/benchmarks/benches/umf.py

@@ -22,7 +22,10 @@ def __init__(self, directory):
         self.directory = directory
         if not isUMFAvailable():
             print("UMF not provided. Related benchmarks will not run")
-
+
+    def name(self) -> str:
+        return "UMF"
+
     def setup(self):
         if not isUMFAvailable():
             return []
@@ -40,6 +43,8 @@ def benchmarks(self) -> list[Benchmark]:
 
 class ComputeUMFBenchmark(Benchmark):
     def __init__(self, bench, name):
+        super().__init__(bench.directory, bench)
+
         self.bench = bench
         self.bench_name = name
         self.oneapi = get_oneapi()
@@ -52,8 +57,6 @@ def __init__(self, bench, name):
 
         self.col_statistics_time = None
 
-        super().__init__(bench.directory)
-
     def bin_args(self) -> list[str]:
         return []
 

scripts/benchmarks/benches/velocity.py

@@ -22,6 +22,9 @@ def __init__(self, directory):
 
         self.directory = directory
 
+    def name(self) -> str:
+        return "Velocity Bench"
+
     def setup(self):
         if options.sycl is None:
             return
@@ -46,7 +49,7 @@ def benchmarks(self) -> list[Benchmark]:
 
 class VelocityBase(Benchmark):
     def __init__(self, name: str, bin_name: str, vb: VelocityBench, unit: str):
-        super().__init__(vb.directory)
+        super().__init__(vb.directory, vb)
         self.vb = vb
         self.bench_name = name
         self.bin_name = bin_name

scripts/benchmarks/main.py

@@ -42,6 +42,7 @@ def run_iterations(benchmark: Benchmark, env_vars, iters: int, results: dict[str
 
             bench_result.name = bench_result.label
             bench_result.lower_is_better = benchmark.lower_is_better()
+            bench_result.suite = benchmark.get_suite_name()
 
             if bench_result.label not in results:
                 results[bench_result.label] = []