[CTS] fix 2 errors in urEnequeuKernelLaunch on PVC

Fix urEnqueueKernelLaunchUSMLinkedList by providing required alignment and calling urKernelSetExecInfo. Also, skip InvalidKernelArgs for L0 as L0 cannot check kernel arguments.
oneapi-src · Oct 18, 2024 · b3bf6f8 · b3bf6f8
1 parent 3a8bf2c
commit b3bf6f8
Show file tree

Hide file tree

Showing 4 changed files with 12 additions and 10 deletions.
diff --git a/.github/workflows/multi_device.yml b/.github/workflows/multi_device.yml
@@ -63,4 +63,4 @@ jobs:
 
     - name: Test adapters
       working-directory: ${{github.workspace}}/build
-      run: env UR_CTS_ADAPTER_PLATFORM="${{matrix.adapter.platform}}" ctest -C ${{matrix.build_type}} --output-on-failure -L "conformance" -E "enqueue|kernel|program|integration|exp_command_buffer|exp_enqueue_native|exp_launch_properties|exp_usm_p2p" --timeout 180
+      run: env UR_CTS_ADAPTER_PLATFORM="${{matrix.adapter.platform}}" ctest -C ${{matrix.build_type}} --output-on-failure -L "conformance" -E "kernel|program|integration|exp_command_buffer|exp_enqueue_native|exp_launch_properties|exp_usm_p2p" --timeout 180
diff --git a/test/conformance/enqueue/enqueue_adapter_level_zero.match b/test/conformance/enqueue/enqueue_adapter_level_zero.match
@@ -3,8 +3,6 @@
 {{OPT}}urEnqueueKernelLaunchTest.InvalidKernelArgs/Intel_R__oneAPI_Unified_Runtime_over_Level_Zero___{{.*}}_
 {{OPT}}urEnqueueKernelLaunchKernelWgSizeTest.Success/Intel_R__oneAPI_Unified_Runtime_over_Level_Zero___{{.*}}_
 {{OPT}}urEnqueueKernelLaunchKernelSubGroupTest.Success/Intel_R__oneAPI_Unified_Runtime_over_Level_Zero___{{.*}}_
-{{OPT}}urEnqueueKernelLaunchUSMLinkedList.Success/Intel_R__oneAPI_Unified_Runtime_over_Level_Zero___{{.*}}__UsePoolEnabled
-{{OPT}}urEnqueueKernelLaunchUSMLinkedList.Success/Intel_R__oneAPI_Unified_Runtime_over_Level_Zero___{{.*}}__UsePoolDisabled
 {{OPT}}urEnqueueMemBufferCopyRectTestWithParam.Success/Intel_R__oneAPI_Unified_Runtime_over_Level_Zero___{{.*}}___copy_2d_3d
 {{OPT}}urEnqueueMemBufferCopyRectTestWithParam.Success/Intel_R__oneAPI_Unified_Runtime_over_Level_Zero___{{.*}}___copy_3d_2d
 {{OPT}}urEnqueueMemBufferReadRectTestWithParam.Success/Intel_R__oneAPI_Unified_Runtime_over_Level_Zero___{{.*}}___write_non_zero_offsets_2D

diff --git a/test/conformance/enqueue/enqueue_adapter_level_zero_v2.match b/test/conformance/enqueue/enqueue_adapter_level_zero_v2.match
@@ -2,8 +2,6 @@
 urEnqueueKernelLaunchTest.InvalidKernelArgs/Intel_R__oneAPI_Unified_Runtime_over_Level_Zero___{{.*}}_
 urEnqueueKernelLaunchKernelWgSizeTest.Success/Intel_R__oneAPI_Unified_Runtime_over_Level_Zero___{{.*}}_
 urEnqueueKernelLaunchWithVirtualMemory.Success/Intel_R__oneAPI_Unified_Runtime_over_Level_Zero___{{.*}}_
-urEnqueueKernelLaunchUSMLinkedList.Success/Intel_R__oneAPI_Unified_Runtime_over_Level_Zero___{{.*}}___UsePoolEnabled
-urEnqueueKernelLaunchUSMLinkedList.Success/Intel_R__oneAPI_Unified_Runtime_over_Level_Zero___{{.*}}___UsePoolDisabled
 {{OPT}}urEnqueueKernelLaunchIncrementTest.Success/Intel_R__oneAPI_Unified_Runtime_over_Level_Zero___{{.*}}___UseEventsEnabled
 {{OPT}}urEnqueueKernelLaunchIncrementTest.Success/Intel_R__oneAPI_Unified_Runtime_over_Level_Zero___{{.*}}___UseEventsDisabled
 {{OPT}}urEnqueueKernelLaunchIncrementMultiDeviceMultiThreadTest.Success/UseEventsNoQueuePerThread

diff --git a/test/conformance/enqueue/urEnqueueKernelLaunch.cpp b/test/conformance/enqueue/urEnqueueKernelLaunch.cpp
@@ -139,8 +139,9 @@ TEST_P(urEnqueueKernelLaunchTest, InvalidKernelArgs) {
                                      nullptr));
 
     if (backend == UR_PLATFORM_BACKEND_CUDA ||
-        backend == UR_PLATFORM_BACKEND_HIP) {
-        GTEST_FAIL() << "AMD and Nvidia can't check kernel arguments.";
+        backend == UR_PLATFORM_BACKEND_HIP ||
+        backend == UR_PLATFORM_BACKEND_LEVEL_ZERO) {
+        GTEST_FAIL() << "AMD, L0 and Nvidia can't check kernel arguments.";
     }
 
     // Enqueue kernel without setting any args
@@ -561,16 +562,16 @@ TEST_P(urEnqueueKernelLaunchUSMLinkedList, Success) {
     }
 
     // Build linked list with USM allocations
-    ASSERT_SUCCESS(urUSMSharedAlloc(context, device, nullptr, pool,
-                                    sizeof(Node),
+    ur_usm_desc_t desc{UR_STRUCTURE_TYPE_USM_DESC, nullptr, 0, alignof(Node)};
+    ASSERT_SUCCESS(urUSMSharedAlloc(context, device, &desc, pool, sizeof(Node),
                                     reinterpret_cast<void **>(&list_head)));
     ASSERT_NE(list_head, nullptr);
     Node *list_cur = list_head;
     for (int i = 0; i < num_nodes; i++) {
         list_cur->num = i * 2;
         if (i < num_nodes - 1) {
             ASSERT_SUCCESS(
-                urUSMSharedAlloc(context, device, nullptr, pool, sizeof(Node),
+                urUSMSharedAlloc(context, device, &desc, pool, sizeof(Node),
                                  reinterpret_cast<void **>(&list_cur->next)));
             ASSERT_NE(list_cur->next, nullptr);
         } else {
@@ -579,6 +580,11 @@ TEST_P(urEnqueueKernelLaunchUSMLinkedList, Success) {
         list_cur = list_cur->next;
     }
 
+    ur_bool_t indirect = true;
+    ASSERT_SUCCESS(urKernelSetExecInfo(kernel,
+                                       UR_KERNEL_EXEC_INFO_USM_INDIRECT_ACCESS,
+                                       sizeof(indirect), nullptr, &indirect));
+
     // Run kernel which will iterate the list and modify the values
     ASSERT_SUCCESS(urKernelSetArgPointer(kernel, 0, nullptr, list_head));
     ASSERT_SUCCESS(urEnqueueKernelLaunch(queue, kernel, 1, &global_offset,