Merge branch 'main' into yc/0829-buffer-alignment-ub

Author: yingcong-wu

CMakeLists.txt

@@ -4,7 +4,7 @@
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
 cmake_minimum_required(VERSION 3.20.0 FATAL_ERROR)
-project(unified-runtime VERSION 0.10.0)
+project(unified-runtime VERSION 0.11.0)
 
 # Check if unified runtime is built as a standalone project.
 if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR OR UR_STANDALONE_BUILD)

include/ur_api.h

@@ -7,7 +7,7 @@
  * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
  *
  * @file ur_api.h
- * @version v0.10-r0
+ * @version v0.11-r0
  *
  */
 #ifndef UR_API_H_INCLUDED
@@ -1133,7 +1133,8 @@ typedef enum ur_api_version_t {
     UR_API_VERSION_0_8 = UR_MAKE_VERSION(0, 8),      ///< version 0.8
     UR_API_VERSION_0_9 = UR_MAKE_VERSION(0, 9),      ///< version 0.9
     UR_API_VERSION_0_10 = UR_MAKE_VERSION(0, 10),    ///< version 0.10
-    UR_API_VERSION_CURRENT = UR_MAKE_VERSION(0, 10), ///< latest known version
+    UR_API_VERSION_0_11 = UR_MAKE_VERSION(0, 11),    ///< version 0.11
+    UR_API_VERSION_CURRENT = UR_MAKE_VERSION(0, 11), ///< latest known version
     /// @cond
     UR_API_VERSION_FORCE_UINT32 = 0x7fffffff
     /// @endcond

include/ur_ddi.h

@@ -7,7 +7,7 @@
  * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
  *
  * @file ur_ddi.h
- * @version v0.10-r0
+ * @version v0.11-r0
  *
  */
 #ifndef UR_DDI_H_INCLUDED

include/ur_print.hpp

@@ -7,7 +7,7 @@
  * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
  *
  * @file ur_print.hpp
- * @version v0.10-r0
+ * @version v0.11-r0
  *
  */
 #ifndef UR_PRINT_HPP

scripts/Doxyfile

@@ -38,7 +38,7 @@ PROJECT_NAME           = "Intel One API Unified Runtime API"
 # could be handy for archiving the generated documentation or if some version
 # control system is used.
 
-PROJECT_NUMBER         = v0.10
+PROJECT_NUMBER         = v0.11
 
 # Using the PROJECT_BRIEF tag one can provide an optional one line description
 # for a project that appears at the top of each page and should give viewer a

scripts/core/CONTRIB.rst

@@ -127,9 +127,9 @@ Adapter Change Process
 .. _intel/llvm:
    https://github.com/intel/llvm
 .. _UNIFIED_RUNTIME_REPO:
-   https://github.com/intel/llvm/blob/sycl/sycl/plugins/unified_runtime/CMakeLists.txt#L102
+   https://github.com/intel/llvm/blob/sycl/sycl/cmake/modules/FetchUnifiedRuntime.cmake#L119
 .. _UNIFIED_RUNTIME_TAG:
-   https://github.com/intel/llvm/blob/sycl/sycl/plugins/unified_runtime/CMakeLists.txt#L109
+   https://github.com/intel/llvm/blob/sycl/sycl/cmake/modules/FetchUnifiedRuntime.cmake#L126
 
 Build Environment
 =================

scripts/core/platform.yml

@@ -143,6 +143,9 @@ etors:
     - name: "0_10"
       value: "$X_MAKE_VERSION( 0, 10 )"
       desc: "version 0.10"
+    - name: "0_11"
+      value: "$X_MAKE_VERSION( 0, 11 )"
+      desc: "version 0.11"
 --- #--------------------------------------------------------------------------
 type: function
 desc: "Returns the API version supported by the specified platform"

scripts/parse_specs.py

@@ -21,8 +21,8 @@
 from version import Version
 
 
-default_version = Version("0.10")
-all_versions = [Version(ver) for ver in ["0.6", "0.7", "0.8", "0.9", "0.10"]]
+default_version = Version("0.11")
+all_versions = [Version(ver) for ver in ["0.6", "0.7", "0.8", "0.9", "0.10", "0.11"]]
 
 """
     preprocess object

source/adapters/level_zero/adapter.cpp

@@ -12,14 +12,27 @@
 #include "ur_level_zero.hpp"
 #include <iomanip>
 
+// As windows order of unloading dlls is reversed from linux, windows will call
+// umfTearDown before it could release umf objects in level_zero, so we call
+// umfInit on urAdapterGet and umfAdapterTearDown to enforce the teardown of umf
+// after umf objects are destructed.
+#if defined(_WIN32)
+#include <umf.h>
+#endif
+
 // Due to multiple DLLMain definitions with SYCL, Global Adapter is init at
 // variable creation.
 #if defined(_WIN32)
 ur_adapter_handle_t_ *GlobalAdapter = new ur_adapter_handle_t_();
 #else
 ur_adapter_handle_t_ *GlobalAdapter;
 #endif
-
+// This is a temporary workaround on windows, where UR adapter is teardowned
+// before the UR loader, which will result in access violation when we use print
+// function as the overrided print function was already released with the UR
+// adapter.
+// TODO: Change adapters to use a common sink class in the loader instead of
+// using thier own sink class that inherit from logger::Sink.
 class ur_legacy_sink : public logger::Sink {
 public:
   ur_legacy_sink(std::string logger_name = "", bool skip_prefix = true)
@@ -32,7 +45,11 @@ class ur_legacy_sink : public logger::Sink {
     fprintf(stderr, "%s", msg.c_str());
   }
 
-  ~ur_legacy_sink() = default;
+  ~ur_legacy_sink() {
+#if defined(_WIN32)
+    logger::isTearDowned = true;
+#endif
+  };
 };
 
 ur_result_t initPlatforms(PlatformVec &platforms) noexcept try {
@@ -74,7 +91,14 @@ ur_result_t initPlatforms(PlatformVec &platforms) noexcept try {
   return exceptionToResult(std::current_exception());
 }
 
-ur_result_t adapterStateInit() { return UR_RESULT_SUCCESS; }
+ur_result_t adapterStateInit() {
+
+#if defined(_WIN32)
+  umfInit();
+#endif
+
+  return UR_RESULT_SUCCESS;
+}
 
 ur_adapter_handle_t_::ur_adapter_handle_t_()
     : logger(logger::get_logger("level_zero")) {
@@ -258,6 +282,7 @@ ur_result_t adapterStateTeardown() {
     // Due to multiple DLLMain definitions with SYCL, register to cleanup the
     // Global Adapter after refcnt is 0
 #if defined(_WIN32)
+  umfTearDown();
   std::atexit(globalAdapterOnDemandCleanup);
 #endif
 

source/adapters/level_zero/device.cpp

@@ -1140,13 +1140,15 @@ ur_device_handle_t_::useImmediateCommandLists() {
     return std::atoi(ImmediateCommandlistsSettingStr);
   }();
 
-  if (ImmediateCommandlistsSetting == -1)
-  // Change this to PerQueue as default after more testing.
-#ifdef _WIN32
-    return NotUsed;
-#else
-    return isPVC() ? PerQueue : NotUsed;
-#endif
+  if (ImmediateCommandlistsSetting == -1) {
+    bool isDG2SupportedDriver =
+        this->Platform->isDriverVersionNewerOrSimilar(1, 5, 30820);
+    if ((isDG2SupportedDriver && isDG2()) || isPVC()) {
+      return PerQueue;
+    } else {
+      return NotUsed;
+    }
+  }
   switch (ImmediateCommandlistsSetting) {
   case 0:
     return NotUsed;