Merge pull request #905 from CHIP-SPV/fix-859-opencl-cpu-usm

skip kernel annotation on CPU
CHIP-SPV · Aug 12, 2024 · a467bc6 · a467bc6
2 parents b3fbd69 + 4c99463
commit a467bc6
Show file tree

Hide file tree

Showing 4 changed files with 44 additions and 14 deletions.
diff --git a/.github/workflows/publish-docker.yml b/.github/workflows/publish-docker.yml
@@ -28,7 +28,6 @@ jobs:
 
   push:
     if: github.event_name == 'push' && github.ref == 'refs/heads/main'
-    needs: build
     runs-on: [self-hosted, Linux, X64]
     steps:
     - uses: actions/checkout@v3
@@ -49,7 +48,7 @@ jobs:
     
 
   cleanup:
-    needs: [build, push]
+    needs: [push]
     runs-on: [self-hosted, Linux, X64]
     steps:
     - name: Clean up old Docker images

diff --git a/scripts/unit_tests.sh b/scripts/unit_tests.sh
@@ -174,7 +174,7 @@ function check_tests {
   if ! grep -q "The following tests FAILED" "$file"; then
     return 0
   else
-    grep -q -E "The following tests FAILED:" -A 1000 "$file" | sed '/^$/q' | tail -n +2
+    grep -E "The following tests FAILED:" -A 1000 "$file" | sed '/^$/q' | tail -n +2
     return 1
   fi
 }
@@ -183,15 +183,25 @@ set +e # disable exit on error
 
 # Run tests for different configurations
 run_tests igpu opencl
-igpu_opencl_result=$(check_tests igpu_opencl_make_check_result.txt)
 if [ "$host" = "salami" ]; then
-  exit $igpu_opencl_result
+  check_tests igpu_opencl_make_check_result.txt
+  igpu_opencl_exit_code=$?
+  exit $igpu_opencl_exit_code
 fi
 run_tests igpu level0
-igpu_level0_result=$(check_tests igpu_level0_make_check_result.txt)
 run_tests dgpu level0
-dgpu_level0_result=$(check_tests dgpu_level0_make_check_result.txt)
 run_tests dgpu opencl
-dgpu_opencl_result=$(check_tests dgpu_opencl_make_check_result.txt)
-
-exit $((igpu_opencl_result || dgpu_opencl_result || igpu_level0_result || dgpu_level0_result))
+run_tests cpu opencl
+
+check_tests igpu_opencl_make_check_result.txt
+igpu_opencl_exit_code=$?
+check_tests cpu_opencl_make_check_result.txt
+cpu_opencl_exit_code=$?
+check_tests dgpu_opencl_make_check_result.txt
+dgpu_opencl_exit_code=$?
+check_tests dgpu_level0_make_check_result.txt
+dgpu_level0_exit_code=$?
+check_tests igpu_level0_make_check_result.txt
+igpu_level0_exit_code=$?
+
+exit $((igpu_opencl_exit_code || dgpu_opencl_exit_code || igpu_level0_exit_code || dgpu_level0_exit_code || cpu_opencl_exit_code))
diff --git a/src/backend/OpenCL/CHIPBackendOpenCL.cc b/src/backend/OpenCL/CHIPBackendOpenCL.cc
@@ -313,7 +313,7 @@ CHIPCallbackDataOpenCL::CHIPCallbackDataOpenCL(hipStreamCallback_t TheCallback,
 
 // EventMonitorOpenCL
 // ************************************************************************
-EventMonitorOpenCL::EventMonitorOpenCL() : chipstar::EventMonitor(){};
+EventMonitorOpenCL::EventMonitorOpenCL() : chipstar::EventMonitor() {};
 
 void EventMonitorOpenCL::monitor() {
   logTrace("EventMonitorOpenCL::monitor()");
@@ -1263,8 +1263,13 @@ CHIPQueueOpenCL::launchImpl(chipstar::ExecItem *ExecItem) {
   LOCK(Backend->DubiousLockOpenCL);
 #endif
 
-  auto AllocationsToKeepAlive = annotateIndirectPointers(
-      *OclContext, Kernel->getModule()->getInfo(), KernelHandle);
+  std::unique_ptr<std::vector<std::shared_ptr<void>>> AllocationsToKeepAlive;
+  // Disable annotation for CPU device. Runtime errror.
+  // https://community.intel.com/t5/OpenCL-for-CPU/OpenCL-Using-USM-on-Intel-CPU-Runtime-along-with/m-p/1621136#M7350
+  if (ChipEnvVars.getDevice().getType() != DeviceType::CPU) {
+    AllocationsToKeepAlive = annotateIndirectPointers(
+        *OclContext, Kernel->getModule()->getInfo(), KernelHandle);
+  }
 
   auto [EventsToWait, EventLocks] = getSyncQueuesLastEvents(LaunchEvent, false);
   std::vector<cl_event> SyncQueuesEventHandles = getOpenCLHandles(EventsToWait);
@@ -1409,7 +1414,7 @@ CHIPQueueOpenCL::memCopyAsyncImpl(void *Dst, const void *Src, size_t Size,
 #ifdef CHIP_DUBIOUS_LOCKS
     LOCK(Backend->DubiousLockOpenCL)
 #endif
-      auto [EventsToWait, EventLocks] = getSyncQueuesLastEvents(Event, false);
+    auto [EventsToWait, EventLocks] = getSyncQueuesLastEvents(Event, false);
     std::vector<cl_event> SyncQueuesEventHandles =
         getOpenCLHandles(EventsToWait);
     auto *Ctx = getContext();

diff --git a/tests/known_failures.yaml b/tests/known_failures.yaml
@@ -412,13 +412,20 @@ ANY:
     CatchMemLeak1: ''
     hipMemset_Unit_hipMemsetAsync_SetMemoryWithOffset_Helgrind: 'False positives from L0 helper thread'
   OPENCL_CPU:
+    hipBlas-sgemm: 'GEMM SYCL exception: Backends mismatch'
     2d_shuffle: ''
     Unit_hipClassKernel_Value: ''
     Unit_hipMemsetFunctional_PartialSet_1D: ''
     fp16: ''
     hipConstantTestDeviceSymbol: ''
     hipDynamicShared2: ''
     unroll: ''
+    Unit_hipMemsetAsync_VerifyExecutionWithKernel: 'Failed'
+    Unit_hipDeviceSynchronize_Functional: 'Failed'
+    Unit_hipTextureObj1DCheckRGBAModes - array: 'Timeout'
+    Unit_hipTextureObj1DCheckRGBAModes - buffer: 'Timeout'
+    Unit_hipTextureObj2DCheckRGBAModes: 'Failed'
+    hipMemset_Unit_hipMemsetAsync_SetMemoryWithOffset_Helgrind: 'Failed'
   OPENCL_GPU:
     Unit_hipEvent: ''
     TestStlFunctionsDouble: ''
@@ -475,6 +482,15 @@ cupcake:
   OPENCL_CPU:
   OPENCL_GPU:
   OPENCL_POCL:
+meatloaf:
+  ALL:
+    Unit_hipMultiStream_sameDevice: 'Started randomly failing on CI node with CL_OUT_OF_RESOURCES'
+    Unit_hipStreamCreate_MultistreamBasicFunctionalities: 'Started randomly failing on CI node with CL_OUT_OF_RESOURCES'
+    Unit_hipDeviceSynchronize_Functional: 'Started randomly failing on CI node with CL_OUT_OF_RESOURCES'
+  LEVEL0_GPU:
+  OPENCL_CPU:
+  OPENCL_GPU:
+  OPENCL_POCL:
 salami:
   ALL:
   LEVEL0_GPU: