Revert "Run Compatibility GPU CI jobs only on upstream"

This reverts commit 0d5a89c.

Author: lplewa

.cmake-format

@@ -26,8 +26,7 @@ with section("parse"):
       'kwargs': {
         'NAME': '*',
         'SRCS': '*',
-        'LIBS': '*',
-        'ENVS': '*'}},
+        'LIBS': '*'}},
     'add_umf_library': {
       "pargs": 0,
       "flags": [],

.github/scripts/get_system_info.sh

@@ -15,7 +15,7 @@ function check_L0_version {
     fi
 
     if command -v zypper &> /dev/null; then
-        zypper -n se level-zero || true
+        zypper se level-zero && return
     fi
 
     echo "level-zero not installed"

.github/workflows/reusable_benchmarks.yml

@@ -111,9 +111,7 @@ jobs:
 
     - name: Install benchmarking scripts deps
       run: |
-        python -m venv .venv
-        source .venv/bin/activate
-        pip install -r ${{github.workspace}}/sycl-repo/unified-runtime/third_party/benchmark_requirements.txt
+        pip install --force-reinstall -r ${{github.workspace}}/sycl-repo/unified-runtime/third_party/benchmark_requirements.txt
 
     - name: Set core range and GPU mask
       run: |
@@ -137,7 +135,6 @@ jobs:
       id: benchmarks
       working-directory: ${{env.BUILD_DIR}}
       run: >
-        source ${{github.workspace}}/.venv/bin/activate &&
         taskset -c ${{ env.CORES }} ${{ github.workspace }}/sycl-repo/unified-runtime/scripts/benchmarks/main.py
         ~/bench_workdir_umf
         --umf ${{env.BUILD_DIR}}

.github/workflows/reusable_compatibility.yml

@@ -94,11 +94,9 @@ jobs:
 
     - name: Run "tag" UMF tests with latest UMF libs (warnings enabled)
       working-directory: ${{github.workspace}}/tag_version/build
-      # GTEST_FILTER is used below to skip test that is not compatible
       run: >
         UMF_LOG="level:warning;flush:debug;output:stderr;pid:no"
         LD_LIBRARY_PATH=${{github.workspace}}/latest_version/build/lib/
-        GTEST_FILTER="-*umfIpcTest.GetPoolByOpenedHandle*"
         ctest --verbose
 
   windows:
@@ -183,14 +181,11 @@ jobs:
       working-directory: ${{github.workspace}}/tag_version/build
       run: |
         $env:UMF_LOG="level:warning;flush:debug;output:stderr;pid:no" 
-        $env:GTEST_FILTER="-*umfIpcTest.GetPoolByOpenedHandle*"
         cp ${{github.workspace}}/latest_version/build/bin/Debug/umf.dll ${{github.workspace}}/tag_version/build/bin/Debug/umf.dll
         ctest -C Debug --verbose
 
   gpu:
     name: GPU Ubuntu
-    # run only on upstream; forks will not have the HW
-    if: github.repository == 'oneapi-src/unified-memory-framework'
     strategy:
       matrix:
         provider: ['LEVEL_ZERO', 'CUDA']
@@ -233,10 +228,8 @@ jobs:
 
     - name: Run "tag" UMF tests
       working-directory: ${{github.workspace}}/tag_version/build
-      run: >
-        LD_LIBRARY_PATH=${{github.workspace}}/tag_version/build/lib/
-        GTEST_FILTER="-*umfIpcTest.GetPoolByOpenedHandle*"
-        ctest --output-on-failure
+      run: |
+        LD_LIBRARY_PATH=${{github.workspace}}/tag_version/build/lib/ ctest --output-on-failure
 
     - name: Checkout latest UMF version
       uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11 # v4.1.1
@@ -271,5 +264,4 @@ jobs:
       run: >
         UMF_LOG="level:warning;flush:debug;output:stderr;pid:no"
         LD_LIBRARY_PATH=${{github.workspace}}/latest_version/build/lib/
-        GTEST_FILTER="-*umfIpcTest.GetPoolByOpenedHandle*"
         ctest --verbose -E "not_impl"

.github/workflows/reusable_multi_numa.yml

@@ -1,4 +1,4 @@
-# Runs tests on multi-numa machines
+# Runs tests on multi-numa machine
 name: MultiNuma
 
 on: [workflow_call]
@@ -20,7 +20,7 @@ jobs:
 
     strategy:
       matrix:
-        os: [ubuntu-22.04, rhel-9.1, sles-15]
+        os: [ubuntu-22.04, rhel-9.1]
         build_type: [Debug, Release]
         shared_library: ['ON', 'OFF']
     runs-on:  ["DSS-MULTI-NUMA", "DSS-${{matrix.os}}"]
@@ -31,6 +31,9 @@ jobs:
         with:
           fetch-depth: 0
 
+      - name: Get information about platform
+        run: .github/scripts/get_system_info.sh
+
       - name: Configure build
         run: >
           cmake
@@ -50,24 +53,24 @@ jobs:
         run: cmake --build ${{github.workspace}}/build -j $(nproc)
 
       - name: Run tests
-        if: (matrix.os != 'rhel-9.1') && (matrix.os != 'sles-15')
+        if: matrix.os != 'rhel-9.1'
         working-directory: ${{github.workspace}}/build
         run: ctest --output-on-failure --test-dir test
 
-        # On RHEL/SLES, hwloc version is just a little too low.
+        # On RHEL, hwloc version is just a little too low.
         # Skip some tests until we upgrade hwloc and update CMake to properly handle local hwloc installation.
         # TODO: fix issue #560
         # TODO: add issue for -E test_init_teardown - it is not clear why it fails
-      - name: Run tests (on RHEL/SLES)
-        if: (matrix.os == 'rhel-9.1') || (matrix.os == 'sles-15')
+      - name: Run tests (on RHEL)
+        if: matrix.os == 'rhel-9.1'
         working-directory: ${{github.workspace}}/build
         run: |
           ctest --output-on-failure --test-dir test -E "test_provider_os_memory_multiple_numa_nodes|test_init_teardown"
           ./test/test_provider_os_memory_multiple_numa_nodes \
             --gtest_filter="-*checkModeLocal/*:*checkModePreferredEmptyNodeset/*:testNuma.checkModeInterleave"
 
       - name: Run NUMA tests under valgrind
-        if: (matrix.os != 'rhel-9.1') && (matrix.os != 'sles-15')
+        if: matrix.os != 'rhel-9.1'
         run: |
           ${{github.workspace}}/test/test_valgrind.sh ${{github.workspace}} ${{env.BUILD_DIR}} memcheck "${{env.NUMA_TESTS}}"
           ${{github.workspace}}/test/test_valgrind.sh ${{github.workspace}} ${{env.BUILD_DIR}} drd "${{env.NUMA_TESTS}}"
@@ -88,7 +91,3 @@ jobs:
         with:
           name: ${{env.COVERAGE_NAME}}-${{matrix.os}}-shared-${{matrix.shared_library}}
           path: ${{env.COVERAGE_DIR}}
-
-      - name: Get information about platform
-        if: always()
-        run: .github/scripts/get_system_info.sh

docs/config/api.rst

@@ -168,26 +168,6 @@ IPC API allows retrieving IPC handles for the memory buffers allocated from
 UMF memory pools. The memory provider used by the pool should support IPC 
 operations for this API to work. Otherwise IPC APIs return an error.
 
-IPC caching
-------------------------------------------
-
-UMF employs IPC caching to avoid multiple IPC handles being created for the same 
-coarse-grain memory region allocated by the memory provider. UMF guarantees that 
-for each coarse-grain memory region allocated by the memory provider, only one 
-IPC handle is created when the :any:`umfGetIPCHandle` function is called. All 
-subsequent calls to the :any:`umfGetIPCHandle` function for the pointer to the 
-same memory region will return the entry from the cache.
-
-The same is true for the :any:`umfOpenIPCHandle` function. The actual mapping
-of the IPC handle to the virtual address space is created only once, and all
-subsequent calls to open the same IPC handle will return the entry from the cache.
-The size of the cache for opened IPC handles is controlled by the ``UMF_MAX_OPENED_IPC_HANDLES``
-environment variable. By default, the cache size is unlimited. However, if the environment 
-variable is set and the cache size exceeds the limit, old items will be evicted. UMF tracks 
-the ref count for each entry in the cache and can evict only items with the ref count equal to 0. 
-The ref count is increased when the :any:`umfOpenIPCHandle` function is called and decreased 
-when the :any:`umfCloseIPCHandle` function is called for the corresponding IPC handle.
-
 .. _ipc-api:
 
 IPC API

include/umf/base.h

@@ -47,8 +47,7 @@ typedef enum umf_result_t {
         6, ///< Failure in user provider code (i.e in user provided callback)
     UMF_RESULT_ERROR_DEPENDENCY_UNAVAILABLE =
         7, ///< External required dependency is unavailable or missing
-    UMF_RESULT_ERROR_OUT_OF_RESOURCES = 8, ///< Out of internal resources
-    UMF_RESULT_ERROR_UNKNOWN = 0x7ffffffe  ///< Unknown error
+    UMF_RESULT_ERROR_UNKNOWN = 0x7ffffffe ///< Unknown or internal error
 } umf_result_t;
 
 #ifdef __cplusplus

include/umf/pools/pool_disjoint.h

@@ -100,7 +100,7 @@ umf_result_t umfDisjointPoolParamsSetSharedLimits(
 
 /// @brief Set custom name of the disjoint pool to be used in the traces.
 /// @param hParams handle to the parameters of the disjoint pool.
-/// @param name custom name of the pool. Name longer than 64 characters will be truncated.
+/// @param name custom name of the pool.
 /// @return UMF_RESULT_SUCCESS on success or appropriate error code on failure.
 umf_result_t
 umfDisjointPoolParamsSetName(umf_disjoint_pool_params_handle_t hParams,

src/critnib/critnib.c

@@ -525,9 +525,7 @@ find_predecessor(struct critnib_node *__restrict n) {
     while (1) {
         int nib;
         for (nib = NIB; nib >= 0; nib--) {
-            struct critnib_node *m;
-            utils_atomic_load_acquire_ptr((void **)&n->child[nib], (void **)&m);
-            if (m) {
+            if (n->child[nib]) {
                 break;
             }
         }
@@ -536,12 +534,7 @@ find_predecessor(struct critnib_node *__restrict n) {
             return NULL;
         }
 
-        utils_atomic_load_acquire_ptr((void **)&n->child[nib], (void **)&n);
-
-        if (!n) {
-            return NULL;
-        }
-
+        n = n->child[nib];
         if (is_leaf(n)) {
             return to_leaf(n);
         }
@@ -645,9 +638,7 @@ static struct critnib_leaf *find_successor(struct critnib_node *__restrict n) {
     while (1) {
         unsigned nib;
         for (nib = 0; nib <= NIB; nib++) {
-            struct critnib_node *m;
-            utils_atomic_load_acquire_ptr((void **)&n->child[nib], (void **)&m);
-            if (m) {
+            if (n->child[nib]) {
                 break;
             }
         }
@@ -656,12 +647,7 @@ static struct critnib_leaf *find_successor(struct critnib_node *__restrict n) {
             return NULL;
         }
 
-        utils_atomic_load_acquire_ptr((void **)&n->child[nib], (void **)&n);
-
-        if (!n) {
-            return NULL;
-        }
-
+        n = n->child[nib];
         if (is_leaf(n)) {
             return to_leaf(n);
         }

src/ipc.c

@@ -146,15 +146,19 @@ umf_result_t umfOpenIPCHandle(umf_ipc_handler_handle_t hIPCHandler,
 }
 
 umf_result_t umfCloseIPCHandle(void *ptr) {
-    umf_ipc_info_t ipcInfo;
-    umf_result_t ret = umfMemoryTrackerGetIpcInfo(ptr, &ipcInfo);
+    umf_alloc_info_t allocInfo;
+    umf_result_t ret = umfMemoryTrackerGetAllocInfo(ptr, &allocInfo);
     if (ret != UMF_RESULT_SUCCESS) {
-        LOG_ERR("cannot get IPC info for ptr = %p.", ptr);
+        LOG_ERR("cannot get alloc info for ptr = %p.", ptr);
         return ret;
     }
 
-    return umfMemoryProviderCloseIPCHandle(ipcInfo.provider, ipcInfo.base,
-                                           ipcInfo.baseSize);
+    // We cannot use umfPoolGetMemoryProvider function because it returns
+    // upstream provider but we need tracking one
+    umf_memory_provider_handle_t hProvider = allocInfo.pool->provider;
+
+    return umfMemoryProviderCloseIPCHandle(hProvider, allocInfo.base,
+                                           allocInfo.baseSize);
 }
 
 umf_result_t umfPoolGetIPCHandler(umf_memory_pool_handle_t hPool,