Add CTS to validate device reset ext (#43)

* Add CTS to validate device reset ext Related-to: VLCLJ-2115 Signed-off-by: Vishnu Khanth <[email protected]> * Add CTS to validate device reset ext Related-to: VLCLJ-2115 Signed-off-by: Vishnu Khanth <[email protected]> * Add CTS to validate device reset ext Related-to: VLCLJ-2115 Signed-off-by: Vishnu Khanth <[email protected]> * Add CTS to validate device reset ext Related-to: VLCLJ-2115 Signed-off-by: Vishnu Khanth <[email protected]> --------- Signed-off-by: Vishnu Khanth <[email protected]>
oneapi-src · Jul 4, 2024 · c99886b · c99886b
1 parent 6fe343d
commit c99886b
Show file tree

Hide file tree

Showing 3 changed files with 159 additions and 4 deletions.
diff --git a/conformance_tests/sysman/test_sysman_device/src/test_sysman_device.cpp b/conformance_tests/sysman/test_sysman_device/src/test_sysman_device.cpp
@@ -282,17 +282,16 @@ TEST_F(
   }
 }
 
-TEST_F(
-    SYSMAN_DEVICE_TEST,
-    GivenValidDeviceWhenResettingSysmanDeviceThenSysmanDeviceResetIsSucceded) {
+TEST_F(SYSMAN_DEVICE_TEST,
+       GivenValidDeviceWhenResettingSysmanDeviceThenSysmanDeviceResetSucceeds) {
   for (auto device : devices) {
     lzt::sysman_device_reset(device);
   }
 }
 
 TEST_F(
     SYSMAN_DEVICE_TEST,
-    GivenValidDeviceWhenResettingSysmanDeviceNnumberOfTimesThenSysmanDeviceResetAlwaysSucceded) {
+    GivenValidDeviceWhenResettingSysmanDeviceNnumberOfTimesThenSysmanDeviceResetAlwaysSucceeds) {
   int number_iterations = 2;
   for (int i = 0; i < number_iterations; i++) {
     for (auto device : devices) {
@@ -644,4 +643,147 @@ TEST_F(
   }
 }
 
+TEST_F(SYSMAN_DEVICE_TEST,
+       GivenValidDeviceWhenWarmResettingDeviceThenDeviceResetExtSucceeds) {
+  for (auto device : devices) {
+    lzt::sysman_device_reset_ext(device, false, ZES_RESET_TYPE_WARM);
+  }
+}
+
+TEST_F(SYSMAN_DEVICE_TEST,
+       GivenValidDeviceWhenColdResettingDeviceThenDeviceResetExtSucceeds) {
+  for (auto device : devices) {
+    lzt::sysman_device_reset_ext(device, false, ZES_RESET_TYPE_COLD);
+  }
+}
+
+TEST_F(SYSMAN_DEVICE_TEST,
+       GivenValidDeviceWhenFlrResettingDeviceThenDeviceResetExtSucceeds) {
+  for (auto device : devices) {
+    lzt::sysman_device_reset_ext(device, false, ZES_RESET_TYPE_FLR);
+  }
+}
+
+TEST_F(
+    SYSMAN_DEVICE_TEST,
+    GivenWorkingDeviceHandleWhenWarmResettingSysmanDeviceThenWorkloadExecutionAlwaysSucceedsAfterResetExt) {
+  uint32_t n = 512;
+  std::vector<float> a(n * n, 1);
+  std::vector<float> b(n * n, 1);
+  std::vector<float> c;
+  std::vector<float> c_cpu;
+  c_cpu = perform_matrix_multiplication_on_cpu(a, b, n);
+
+  for (auto device : devices) {
+    // Perform workload execution before reset
+#ifdef USE_ZESINIT
+    auto sysman_device_properties = lzt::get_sysman_device_properties(device);
+    ze_device_handle_t core_device =
+        get_core_device_by_uuid(sysman_device_properties.core.uuid.id);
+    EXPECT_NE(core_device, nullptr);
+    c = submit_workload_for_gpu(a, b, n, core_device);
+#else  // USE_ZESINIT
+    c = submit_workload_for_gpu(a, b, n, device);
+#endif // USE_ZESINIT
+
+    compare_results(c, c_cpu);
+    c.clear();
+    LOG_INFO << "Initiating device reset...\n";
+    // perform device reset
+    lzt::sysman_device_reset_ext(device, false, ZES_RESET_TYPE_WARM);
+    LOG_INFO << "End of device reset...\n";
+
+    // Perform workload execution after reset
+#ifdef USE_ZESINIT
+    c = submit_workload_for_gpu(a, b, n, core_device);
+#else  // USE_ZESINIT
+    c = submit_workload_for_gpu(a, b, n, device);
+#endif // USE_ZESINIT
+
+    compare_results(c, c_cpu);
+    c.clear();
+  }
+}
+
+TEST_F(
+    SYSMAN_DEVICE_TEST,
+    GivenWorkingDeviceHandleWhenColdResettingSysmanDeviceThenWorkloadExecutionAlwaysSucceedsAfterResetExt) {
+  uint32_t n = 512;
+  std::vector<float> a(n * n, 1);
+  std::vector<float> b(n * n, 1);
+  std::vector<float> c;
+  std::vector<float> c_cpu;
+  c_cpu = perform_matrix_multiplication_on_cpu(a, b, n);
+
+  for (auto device : devices) {
+    // Perform workload execution before reset
+#ifdef USE_ZESINIT
+    auto sysman_device_properties = lzt::get_sysman_device_properties(device);
+    ze_device_handle_t core_device =
+        get_core_device_by_uuid(sysman_device_properties.core.uuid.id);
+    EXPECT_NE(core_device, nullptr);
+    c = submit_workload_for_gpu(a, b, n, core_device);
+#else  // USE_ZESINIT
+    c = submit_workload_for_gpu(a, b, n, device);
+#endif // USE_ZESINIT
+
+    compare_results(c, c_cpu);
+    c.clear();
+    LOG_INFO << "Initiating device reset...\n";
+    // perform device reset
+    lzt::sysman_device_reset_ext(device, false, ZES_RESET_TYPE_COLD);
+    LOG_INFO << "End of device reset...\n";
+
+    // Perform workload execution after reset
+#ifdef USE_ZESINIT
+    c = submit_workload_for_gpu(a, b, n, core_device);
+#else  // USE_ZESINIT
+    c = submit_workload_for_gpu(a, b, n, device);
+#endif // USE_ZESINIT
+
+    compare_results(c, c_cpu);
+    c.clear();
+  }
+}
+
+TEST_F(
+    SYSMAN_DEVICE_TEST,
+    GivenWorkingDeviceHandleWhenFlrResettingSysmanDeviceThenWorkloadExecutionAlwaysSucceedsAfterResetExt) {
+  uint32_t n = 512;
+  std::vector<float> a(n * n, 1);
+  std::vector<float> b(n * n, 1);
+  std::vector<float> c;
+  std::vector<float> c_cpu;
+  c_cpu = perform_matrix_multiplication_on_cpu(a, b, n);
+
+  for (auto device : devices) {
+    // Perform workload execution before reset
+#ifdef USE_ZESINIT
+    auto sysman_device_properties = lzt::get_sysman_device_properties(device);
+    ze_device_handle_t core_device =
+        get_core_device_by_uuid(sysman_device_properties.core.uuid.id);
+    EXPECT_NE(core_device, nullptr);
+    c = submit_workload_for_gpu(a, b, n, core_device);
+#else  // USE_ZESINIT
+    c = submit_workload_for_gpu(a, b, n, device);
+#endif // USE_ZESINIT
+
+    compare_results(c, c_cpu);
+    c.clear();
+    LOG_INFO << "Initiating device reset...\n";
+    // perform device reset
+    lzt::sysman_device_reset_ext(device, false, ZES_RESET_TYPE_FLR);
+    LOG_INFO << "End of device reset...\n";
+
+    // Perform workload execution after reset
+#ifdef USE_ZESINIT
+    c = submit_workload_for_gpu(a, b, n, core_device);
+#else  // USE_ZESINIT
+    c = submit_workload_for_gpu(a, b, n, device);
+#endif // USE_ZESINIT
+
+    compare_results(c, c_cpu);
+    c.clear();
+  }
+}
 } // namespace
diff --git a/utils/test_harness/sysman/include/test_harness_sysman_device.hpp b/utils/test_harness/sysman/include/test_harness_sysman_device.hpp
@@ -26,6 +26,9 @@ zes_device_state_t get_device_state(zes_device_handle_t device);
 std::vector<zes_process_state_t> get_processes_state(zes_device_handle_t device,
                                                      uint32_t &count);
 
+void sysman_device_reset_ext(zes_device_handle_t device, ze_bool_t force,
+                             zes_reset_type_t type);
+
 } // namespace level_zero_tests
 
 #endif
diff --git a/utils/test_harness/sysman/src/test_harness_sysman_device.cpp b/utils/test_harness/sysman/src/test_harness_sysman_device.cpp
@@ -57,4 +57,14 @@ zes_device_state_t get_device_state(zes_device_handle_t device) {
   return state;
 }
 
+void sysman_device_reset_ext(zes_device_handle_t device, ze_bool_t force,
+                             zes_reset_type_t type) {
+  zes_reset_properties_t properties{};
+  properties.stype = ZES_STRUCTURE_TYPE_RESET_PROPERTIES;
+  properties.pNext = nullptr;
+  properties.force = force;
+  properties.resetType = type;
+  EXPECT_EQ(ZE_RESULT_SUCCESS, zesDeviceResetExt(device, &properties));
+}
+
 } // namespace level_zero_tests