Merge branch 'main' into urlza-414

Author: winstonzhang-intel

.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

@@ -156,14 +156,49 @@ jobs:
     - name: Install UR
       run: cmake --install ${{github.workspace}}/ur_build
 
+    - name: Checkout UMF
+      uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11 # v4.1.1
+      with:
+        repository: oneapi-src/unified-memory-framework
+        ref: main
+        path: umf-repo
+        fetch-depth: 1
+        fetch-tags: false
+
+    - name: Configure UMF
+      run: >
+        cmake -DCMAKE_BUILD_TYPE=Release
+        -S${{github.workspace}}/umf-repo
+        -B${{github.workspace}}/umf_build
+        -DUMF_BUILD_BENCHMARKS=ON
+        -DUMF_TESTS_FAIL_ON_SKIP=ON
+
+    - 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
+        --umf ${{ github.workspace }}/umf_build
         --adapter ${{ matrix.adapter.str_name }}
         ${{ inputs.upload_report && '--output-html' || '' }}
         ${{ inputs.bench_script_params }}

.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

@@ -233,6 +233,7 @@ typedef enum ur_function_t {
     UR_FUNCTION_ENQUEUE_EVENTS_WAIT_WITH_BARRIER_EXT = 246,               ///< Enumerator for ::urEnqueueEventsWaitWithBarrierExt
     UR_FUNCTION_TENSOR_MAP_ENCODE_IM_2_COL_EXP = 247,                     ///< Enumerator for ::urTensorMapEncodeIm2ColExp
     UR_FUNCTION_TENSOR_MAP_ENCODE_TILED_EXP = 248,                        ///< Enumerator for ::urTensorMapEncodeTiledExp
+    UR_FUNCTION_PHYSICAL_MEM_GET_INFO = 249,                              ///< Enumerator for ::urPhysicalMemGetInfo
     /// @cond
     UR_FUNCTION_FORCE_UINT32 = 0x7fffffff
     /// @endcond
@@ -2525,7 +2526,7 @@ typedef enum ur_mem_type_t {
 ///////////////////////////////////////////////////////////////////////////////
 /// @brief Memory Information type
 typedef enum ur_mem_info_t {
-    UR_MEM_INFO_SIZE = 0,            ///< [size_t] actual size of of memory object in bytes
+    UR_MEM_INFO_SIZE = 0,            ///< [size_t] actual size of the memory object in bytes
     UR_MEM_INFO_CONTEXT = 1,         ///< [::ur_context_handle_t] context in which the memory object was created
     UR_MEM_INFO_REFERENCE_COUNT = 2, ///< [uint32_t] Reference count of the memory object.
                                      ///< The reference count returned should be considered immediately stale.
@@ -4138,6 +4139,50 @@ urPhysicalMemRelease(
     ur_physical_mem_handle_t hPhysicalMem ///< [in][release] handle of the physical memory object to release.
 );
 
+///////////////////////////////////////////////////////////////////////////////
+/// @brief Physical memory range info queries.
+typedef enum ur_physical_mem_info_t {
+    UR_PHYSICAL_MEM_INFO_CONTEXT = 0,         ///< [::ur_context_handle_t] context in which the physical memory object
+                                              ///< was created.
+    UR_PHYSICAL_MEM_INFO_DEVICE = 1,          ///< [::ur_device_handle_t] device associated with this physical memory
+                                              ///< object.
+    UR_PHYSICAL_MEM_INFO_SIZE = 2,            ///< [size_t] actual size of the physical memory object in bytes.
+    UR_PHYSICAL_MEM_INFO_PROPERTIES = 3,      ///< [::ur_physical_mem_properties_t] properties set when creating this
+                                              ///< physical memory object.
+    UR_PHYSICAL_MEM_INFO_REFERENCE_COUNT = 4, ///< [uint32_t] Reference count of the physical memory object.
+                                              ///< The reference count returned should be considered immediately stale.
+                                              ///< It is unsuitable for general use in applications. This feature is
+                                              ///< provided for identifying memory leaks.
+    /// @cond
+    UR_PHYSICAL_MEM_INFO_FORCE_UINT32 = 0x7fffffff
+    /// @endcond
+
+} ur_physical_mem_info_t;
+
+///////////////////////////////////////////////////////////////////////////////
+/// @brief Get information about a physical memory object.
+///
+/// @returns
+///     - ::UR_RESULT_SUCCESS
+///     - ::UR_RESULT_ERROR_UNINITIALIZED
+///     - ::UR_RESULT_ERROR_DEVICE_LOST
+///     - ::UR_RESULT_ERROR_ADAPTER_SPECIFIC
+///     - ::UR_RESULT_ERROR_INVALID_NULL_HANDLE
+///         + `NULL == hPhysicalMem`
+///     - ::UR_RESULT_ERROR_INVALID_ENUMERATION
+///         + `::UR_PHYSICAL_MEM_INFO_REFERENCE_COUNT < propName`
+UR_APIEXPORT ur_result_t UR_APICALL
+urPhysicalMemGetInfo(
+    ur_physical_mem_handle_t hPhysicalMem, ///< [in] handle of the physical memory object to query.
+    ur_physical_mem_info_t propName,       ///< [in] type of the info to query.
+    size_t propSize,                       ///< [in] size in bytes of the memory pointed to by pPropValue.
+    void *pPropValue,                      ///< [out][optional][typename(propName, propSize)] array of bytes holding
+                                           ///< the info. If propSize is less than the real number of bytes needed to
+                                           ///< return the info then the ::UR_RESULT_ERROR_INVALID_SIZE error is
+                                           ///< returned and pPropValue is not used.
+    size_t *pPropSizeRet                   ///< [out][optional] pointer to the actual size in bytes of the queried propName."
+);
+
 #if !defined(__GNUC__)
 #pragma endregion
 #endif
@@ -8517,6 +8562,7 @@ 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
@@ -11317,6 +11363,18 @@ typedef struct ur_physical_mem_release_params_t {
     ur_physical_mem_handle_t *phPhysicalMem;
 } ur_physical_mem_release_params_t;
 
+///////////////////////////////////////////////////////////////////////////////
+/// @brief Function parameters for urPhysicalMemGetInfo
+/// @details Each entry is a pointer to the parameter passed to the function;
+///     allowing the callback the ability to modify the parameter's value
+typedef struct ur_physical_mem_get_info_params_t {
+    ur_physical_mem_handle_t *phPhysicalMem;
+    ur_physical_mem_info_t *ppropName;
+    size_t *ppropSize;
+    void **ppPropValue;
+    size_t **ppPropSizeRet;
+} ur_physical_mem_get_info_params_t;
+
 ///////////////////////////////////////////////////////////////////////////////
 /// @brief Function parameters for urAdapterGet
 /// @details Each entry is a pointer to the parameter passed to the function;

include/ur_api_funcs.def

@@ -96,6 +96,7 @@ _UR_API(urMemImageGetInfo)
 _UR_API(urPhysicalMemCreate)
 _UR_API(urPhysicalMemRetain)
 _UR_API(urPhysicalMemRelease)
+_UR_API(urPhysicalMemGetInfo)
 _UR_API(urAdapterGet)
 _UR_API(urAdapterRelease)
 _UR_API(urAdapterRetain)

include/ur_ddi.h

@@ -978,12 +978,22 @@ typedef ur_result_t(UR_APICALL *ur_pfnPhysicalMemRetain_t)(
 typedef ur_result_t(UR_APICALL *ur_pfnPhysicalMemRelease_t)(
     ur_physical_mem_handle_t);
 
+///////////////////////////////////////////////////////////////////////////////
+/// @brief Function-pointer for urPhysicalMemGetInfo
+typedef ur_result_t(UR_APICALL *ur_pfnPhysicalMemGetInfo_t)(
+    ur_physical_mem_handle_t,
+    ur_physical_mem_info_t,
+    size_t,
+    void *,
+    size_t *);
+
 ///////////////////////////////////////////////////////////////////////////////
 /// @brief Table of PhysicalMem functions pointers
 typedef struct ur_physical_mem_dditable_t {
     ur_pfnPhysicalMemCreate_t pfnCreate;
     ur_pfnPhysicalMemRetain_t pfnRetain;
     ur_pfnPhysicalMemRelease_t pfnRelease;
+    ur_pfnPhysicalMemGetInfo_t pfnGetInfo;
 } ur_physical_mem_dditable_t;
 
 ///////////////////////////////////////////////////////////////////////////////

include/ur_print.h

@@ -594,6 +594,14 @@ UR_APIEXPORT ur_result_t UR_APICALL urPrintPhysicalMemFlags(enum ur_physical_mem
 ///         - `buff_size < out_size`
 UR_APIEXPORT ur_result_t UR_APICALL urPrintPhysicalMemProperties(const struct ur_physical_mem_properties_t params, char *buffer, const size_t buff_size, size_t *out_size);
 
+///////////////////////////////////////////////////////////////////////////////
+/// @brief Print ur_physical_mem_info_t enum
+/// @returns
+///     - ::UR_RESULT_SUCCESS
+///     - ::UR_RESULT_ERROR_INVALID_SIZE
+///         - `buff_size < out_size`
+UR_APIEXPORT ur_result_t UR_APICALL urPrintPhysicalMemInfo(enum ur_physical_mem_info_t value, char *buffer, const size_t buff_size, size_t *out_size);
+
 ///////////////////////////////////////////////////////////////////////////////
 /// @brief Print ur_program_metadata_type_t enum
 /// @returns
@@ -1850,6 +1858,14 @@ UR_APIEXPORT ur_result_t UR_APICALL urPrintPhysicalMemRetainParams(const struct
 ///         - `buff_size < out_size`
 UR_APIEXPORT ur_result_t UR_APICALL urPrintPhysicalMemReleaseParams(const struct ur_physical_mem_release_params_t *params, char *buffer, const size_t buff_size, size_t *out_size);
 
+///////////////////////////////////////////////////////////////////////////////
+/// @brief Print ur_physical_mem_get_info_params_t struct
+/// @returns
+///     - ::UR_RESULT_SUCCESS
+///     - ::UR_RESULT_ERROR_INVALID_SIZE
+///         - `buff_size < out_size`
+UR_APIEXPORT ur_result_t UR_APICALL urPrintPhysicalMemGetInfoParams(const struct ur_physical_mem_get_info_params_t *params, char *buffer, const size_t buff_size, size_t *out_size);
+
 ///////////////////////////////////////////////////////////////////////////////
 /// @brief Print ur_adapter_get_params_t struct
 /// @returns