Merge branch 'sycl' into review/yang/fix_msan_usm

Author: AllanZyne

scripts/benchmarks/output_html.py

@@ -4,6 +4,8 @@
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
 import re
+import os
+from pathlib import Path
 import matplotlib.pyplot as plt
 import mpld3
 from collections import defaultdict
@@ -256,7 +258,9 @@ def generate_html(benchmark_runs: list[BenchmarkRun], github_repo: str, compare_
     suite_names = {t.suite for t in timeseries}
     suite_checkboxes_html = ' '.join(f'<label><input type="checkbox" class="suite-checkbox" data-suite="{suite}" checked> {suite}</label>' for suite in suite_names)
 
-    with open('benchmark_results.html.template', 'r') as file:
+    script_path = os.path.dirname(os.path.realpath(__file__))
+    results_template_path = Path(script_path, 'benchmark_results.html.template')
+    with open(results_template_path, 'r') as file:
         html_template = file.read()
 
     template = Template(html_template)

source/adapters/cuda/tensor_map.cpp

@@ -13,6 +13,24 @@
 
 #include "context.hpp"
 
+#if CUDA_VERSION < 12000
+UR_APIEXPORT ur_result_t UR_APICALL urTensorMapEncodeIm2ColExp(
+    ur_device_handle_t, ur_exp_tensor_map_data_type_flags_t, uint32_t, void *,
+    const uint64_t *, const uint64_t *, const int *, const int *, uint32_t,
+    uint32_t, const uint32_t *, ur_exp_tensor_map_interleave_flags_t,
+    ur_exp_tensor_map_swizzle_flags_t, ur_exp_tensor_map_l2_promotion_flags_t,
+    ur_exp_tensor_map_oob_fill_flags_t, ur_exp_tensor_map_handle_t *) {
+  return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
+}
+UR_APIEXPORT ur_result_t UR_APICALL urTensorMapEncodeTiledExp(
+    ur_device_handle_t, ur_exp_tensor_map_data_type_flags_t, uint32_t, void *,
+    const uint64_t *, const uint64_t *, const uint32_t *, const uint32_t *,
+    ur_exp_tensor_map_interleave_flags_t, ur_exp_tensor_map_swizzle_flags_t,
+    ur_exp_tensor_map_l2_promotion_flags_t, ur_exp_tensor_map_oob_fill_flags_t,
+    ur_exp_tensor_map_handle_t *) {
+  return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
+}
+#else
 struct ur_exp_tensor_map_handle_t_ {
   CUtensorMap Map;
 };
@@ -140,3 +158,4 @@ UR_APIEXPORT ur_result_t UR_APICALL urTensorMapEncodeTiledExp(
   }
   return UR_RESULT_SUCCESS;
 }
+#endif

source/adapters/level_zero/context.cpp

@@ -533,6 +533,13 @@ ur_result_t ur_context_handle_t_::getFreeSlotInExistingOrNewPool(
   if (*ZePool == nullptr) {
     ze_event_pool_counter_based_exp_desc_t counterBasedExt = {
         ZE_STRUCTURE_TYPE_COUNTER_BASED_EVENT_POOL_EXP_DESC, nullptr, 0};
+
+    ze_intel_event_sync_mode_exp_desc_t eventSyncMode = {
+        ZE_INTEL_STRUCTURE_TYPE_EVENT_SYNC_MODE_EXP_DESC, nullptr, 0};
+    eventSyncMode.syncModeFlags =
+        ZE_INTEL_EVENT_SYNC_MODE_EXP_FLAG_LOW_POWER_WAIT |
+        ZE_INTEL_EVENT_SYNC_MODE_EXP_FLAG_SIGNAL_INTERRUPT;
+
     ZeStruct<ze_event_pool_desc_t> ZeEventPoolDesc;
     ZeEventPoolDesc.count = MaxNumEventsPerPool;
     ZeEventPoolDesc.flags = 0;
@@ -552,14 +559,11 @@ ur_result_t ur_context_handle_t_::getFreeSlotInExistingOrNewPool(
       }
       logger::debug("ze_event_pool_desc_t counter based flags set to: {}",
                     counterBasedExt.flags);
+      if (InterruptBasedEventEnabled) {
+        counterBasedExt.pNext = &eventSyncMode;
+      }
       ZeEventPoolDesc.pNext = &counterBasedExt;
-    }
-    if (InterruptBasedEventEnabled) {
-      ze_intel_event_sync_mode_exp_desc_t eventSyncMode = {
-          ZE_INTEL_STRUCTURE_TYPE_EVENT_SYNC_MODE_EXP_DESC, nullptr, 0};
-      eventSyncMode.syncModeFlags =
-          ZE_INTEL_EVENT_SYNC_MODE_EXP_FLAG_LOW_POWER_WAIT |
-          ZE_INTEL_EVENT_SYNC_MODE_EXP_FLAG_SIGNAL_INTERRUPT;
+    } else if (InterruptBasedEventEnabled) {
       ZeEventPoolDesc.pNext = &eventSyncMode;
     }
 

source/loader/layers/sanitizer/msan/msan_libdevice.hpp

@@ -52,6 +52,8 @@ struct MsanLaunchInfo {
     uint32_t IsRecover = 0;
 
     MsanErrorReport Report;
+
+    uint8_t CleanShadow[128] = {};
 };
 
 // Based on the observation, only the last 24 bits of the address of the private