Merge branch 'main' into martin/updateCMakeWarningDPCXX

martygrant · web-flow · commit ac6915f2401b · 2024-08-28T12:44:33.000+01:00
diff --git a/CMakeLists.txt b/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)
diff --git a/include/ur_api.h b/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
diff --git a/include/ur_ddi.h b/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
diff --git a/include/ur_print.hpp b/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
diff --git a/scripts/Doxyfile b/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
diff --git a/scripts/core/CONTRIB.rst b/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
 =================
diff --git a/scripts/core/platform.yml b/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"
diff --git a/scripts/parse_specs.py b/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
diff --git a/source/adapters/cuda/image.cpp b/source/adapters/cuda/image.cpp
@@ -961,8 +961,17 @@ UR_APIEXPORT ur_result_t UR_APICALL urBindlessImagesImageGetInfoExp(
     ur_context_handle_t, ur_exp_image_mem_native_handle_t hImageMem,
     ur_image_info_t propName, void *pPropValue, size_t *pPropSizeRet) {
 
+  CUarray hCUarray;
+  CUresult Err = cuMipmappedArrayGetLevel(
+      &hCUarray, reinterpret_cast<CUmipmappedArray>(hImageMem), 0);
+
+  // If cuMipmappedArrayGetLevel failed, hImageMem is already CUarray.
+  if (Err != CUDA_SUCCESS) {
+    hCUarray = reinterpret_cast<CUarray>(hImageMem);
+  }
+
   CUDA_ARRAY3D_DESCRIPTOR ArrayDesc;
-  UR_CHECK_ERROR(cuArray3DGetDescriptor(&ArrayDesc, (CUarray)hImageMem));
+  UR_CHECK_ERROR(cuArray3DGetDescriptor(&ArrayDesc, hCUarray));
   switch (propName) {
   case UR_IMAGE_INFO_WIDTH:
     if (pPropValue) {
diff --git a/source/adapters/level_zero/adapter.cpp b/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
 
diff --git a/source/adapters/level_zero/device.cpp b/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;
diff --git a/source/adapters/opencl/usm.cpp b/source/adapters/opencl/usm.cpp
@@ -87,6 +87,11 @@ urUSMHostAlloc(ur_context_handle_t hContext, const ur_usm_desc_t *pUSMDesc,
   void *Ptr = nullptr;
   uint32_t Alignment = pUSMDesc ? pUSMDesc->align : 0;
 
+  if (pUSMDesc && pUSMDesc->align != 0 &&
+      ((pUSMDesc->align & (pUSMDesc->align - 1)) != 0)) {
+    return UR_RESULT_ERROR_INVALID_VALUE;
+  }
+
   std::vector<cl_mem_properties_intel> AllocProperties;
   if (pUSMDesc && pUSMDesc->pNext) {
     UR_RETURN_ON_FAILURE(usmDescToCLMemProperties(
@@ -130,6 +135,11 @@ urUSMDeviceAlloc(ur_context_handle_t hContext, ur_device_handle_t hDevice,
   void *Ptr = nullptr;
   uint32_t Alignment = pUSMDesc ? pUSMDesc->align : 0;
 
+  if (pUSMDesc && pUSMDesc->align != 0 &&
+      ((pUSMDesc->align & (pUSMDesc->align - 1)) != 0)) {
+    return UR_RESULT_ERROR_INVALID_VALUE;
+  }
+
   std::vector<cl_mem_properties_intel> AllocProperties;
   if (pUSMDesc && pUSMDesc->pNext) {
     UR_RETURN_ON_FAILURE(usmDescToCLMemProperties(
@@ -173,6 +183,11 @@ urUSMSharedAlloc(ur_context_handle_t hContext, ur_device_handle_t hDevice,
   void *Ptr = nullptr;
   uint32_t Alignment = pUSMDesc ? pUSMDesc->align : 0;
 
+  if (pUSMDesc && pUSMDesc->align != 0 &&
+      ((pUSMDesc->align & (pUSMDesc->align - 1)) != 0)) {
+    return UR_RESULT_ERROR_INVALID_VALUE;
+  }
+
   std::vector<cl_mem_properties_intel> AllocProperties;
   if (pUSMDesc && pUSMDesc->pNext) {
     UR_RETURN_ON_FAILURE(usmDescToCLMemProperties(
diff --git a/source/common/CMakeLists.txt b/source/common/CMakeLists.txt
@@ -24,8 +24,8 @@ if (NOT DEFINED UMF_REPO)
 endif()
 
 if (NOT DEFINED UMF_TAG)
-    # v0.9.x 19.08.2024: Merge pull request #688 ...
-    set(UMF_TAG 59c4150b7120a7af5b3c8eb2d9b8bbb5d2e96aa3)
+    # v0.9.x 23.08.2024: Merge pull request #696 ...
+    set(UMF_TAG 3c340e61c197f4f9e29abd947f90ce27c571433e)
 endif()
 
 message(STATUS "Will fetch Unified Memory Framework from ${UMF_REPO}")
diff --git a/source/common/logger/ur_sinks.hpp b/source/common/logger/ur_sinks.hpp
@@ -17,6 +17,10 @@
 
 namespace logger {
 
+#if defined(_WIN32)
+inline bool isTearDowned = false;
+#endif
+
 class Sink {
   public:
     template <typename... Args>
@@ -28,7 +32,21 @@ class Sink {
         }
 
         format(buffer, fmt, std::forward<Args &&>(args)...);
+// 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.
+#if defined(_WIN32)
+        if (isTearDowned) {
+            std::cerr << buffer.str() << "\n";
+        } else {
+            print(level, buffer.str());
+        }
+#else
         print(level, buffer.str());
+#endif
     }
 
     void setFlushLevel(logger::Level level) { this->flush_level = level; }
diff --git a/source/loader/CMakeLists.txt b/source/loader/CMakeLists.txt
@@ -167,6 +167,26 @@ if(UR_ENABLE_SANITIZER)
         )
         target_include_directories(ur_loader PRIVATE ${LLVM_INCLUDE_DIRS})
         target_link_libraries(ur_loader PRIVATE LLVMSupport LLVMSymbolize)
+        # In in-tree build, if LLVM is built with libc++, we also need to build
+        # symbolizer.cpp with libc++ abi and link libc++ in.
+        if(NOT UR_STANDALONE_BUILD AND LLVM_LIBCXX_USED)
+            execute_process(
+                COMMAND ${CMAKE_CXX_COMPILER} --print-file-name=libc++.a
+                OUTPUT_VARIABLE LIBCXX_PATH
+                OUTPUT_STRIP_TRAILING_WHITESPACE)
+            execute_process(
+                COMMAND ${CMAKE_CXX_COMPILER} --print-file-name=libc++abi.a
+                OUTPUT_VARIABLE LIBCXX_ABI_PATH
+                OUTPUT_STRIP_TRAILING_WHITESPACE)
+            set_property(SOURCE
+                ${CMAKE_CURRENT_SOURCE_DIR}/layers/sanitizer/linux/symbolizer.cpp
+                APPEND_STRING PROPERTY COMPILE_FLAGS
+                " -stdlib=libc++ ")
+            if(NOT EXISTS ${LIBCXX_PATH} OR NOT EXISTS ${LIBCXX_ABI_PATH})
+                message(FATAL_ERROR "libc++ is required but can't find the libraries")
+            endif()
+            target_link_libraries(ur_loader PRIVATE ${LIBCXX_PATH} ${LIBCXX_ABI_PATH})
+        endif()
     endif()
 
     target_include_directories(ur_loader PRIVATE
diff --git a/source/loader/layers/sanitizer/linux/symbolizer.cpp b/source/loader/layers/sanitizer/linux/symbolizer.cpp
@@ -31,8 +31,9 @@ llvm::symbolize::PrinterConfig GetPrinterConfig() {
 
 extern "C" {
 
-bool SymbolizeCode(const std::string ModuleName, uint64_t ModuleOffset,
-                   std::string &Result) {
+void SymbolizeCode(const char *ModuleName, uint64_t ModuleOffset,
+                   char *ResultString, size_t ResultSize, size_t *RetSize) {
+    std::string Result;
     llvm::raw_string_ostream OS(Result);
     llvm::symbolize::Request Request{ModuleName, ModuleOffset};
     llvm::symbolize::PrinterConfig Config =
@@ -51,10 +52,16 @@ bool SymbolizeCode(const std::string ModuleName, uint64_t ModuleOffset,
         {ModuleOffset, llvm::object::SectionedAddress::UndefSection});
 
     if (!ResOrErr) {
-        return false;
+        return;
     }
     Printer->print(Request, *ResOrErr);
     ur_sanitizer_layer::GetSymbolizer()->pruneCache();
-    return true;
+    if (RetSize) {
+        *RetSize = Result.size() + 1;
+    }
+    if (ResultString) {
+        std::strncpy(ResultString, Result.c_str(), ResultSize);
+        ResultString[ResultSize - 1] = '\0';
+    }
 }
 }
diff --git a/source/loader/layers/sanitizer/stacktrace.cpp b/source/loader/layers/sanitizer/stacktrace.cpp
@@ -15,9 +15,10 @@
 
 extern "C" {
 
-__attribute__((weak)) bool SymbolizeCode(const std::string ModuleName,
+__attribute__((weak)) void SymbolizeCode(const char *ModuleName,
                                          uint64_t ModuleOffset,
-                                         std::string &Result);
+                                         char *ResultString, size_t ResultSize,
+                                         size_t *RetSize);
 }
 
 namespace ur_sanitizer_layer {
@@ -49,7 +50,7 @@ void ParseBacktraceInfo(BacktraceInfo BI, std::string &ModuleName,
 // Parse symbolizer output in the following formats:
 //   <function_name>
 //   <file_name>:<line_number>[:<column_number>]
-SourceInfo ParseSymbolizerOutput(std::string Output) {
+SourceInfo ParseSymbolizerOutput(const std::string &Output) {
     SourceInfo Info;
     // Parse function name
     size_t End = Output.find_first_of('\n');
@@ -98,8 +99,14 @@ void StackTrace::print() const {
             std::string ModuleName;
             uptr Offset;
             ParseBacktraceInfo(BI, ModuleName, Offset);
-            if (SymbolizeCode(ModuleName, Offset, Result)) {
-                SourceInfo SrcInfo = ParseSymbolizerOutput(std::move(Result));
+            size_t ResultSize = 0;
+            SymbolizeCode(ModuleName.c_str(), Offset, nullptr, 0, &ResultSize);
+            if (ResultSize) {
+                std::vector<char> ResultVector(ResultSize);
+                SymbolizeCode(ModuleName.c_str(), Offset, ResultVector.data(),
+                              ResultSize, nullptr);
+                std::string Result((char *)ResultVector.data());
+                SourceInfo SrcInfo = ParseSymbolizerOutput(Result);
                 if (SrcInfo.file != "??") {
                     getContext()->logger.always(" #{} in {} {}:{}:{}", index,
                                                 SrcInfo.function, SrcInfo.file,
@@ -109,10 +116,10 @@ void StackTrace::print() const {
                                                 SrcInfo.function, ModuleName,
                                                 (void *)Offset);
                 }
-                continue;
             }
+        } else {
+            getContext()->logger.always("  #{} {}", index, BI);
         }
-        getContext()->logger.always("  #{} {}", index, BI);
         ++index;
     }
     getContext()->logger.always("");
diff --git a/source/ur_api.cpp b/source/ur_api.cpp
@@ -7,7 +7,7 @@
  * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
  *
  * @file ur_api.cpp
- * @version v0.10-r0
+ * @version v0.11-r0
  *
  */
 #include "ur_api.h"
diff --git a/test/conformance/enqueue/enqueue_adapter_level_zero.match b/test/conformance/enqueue/enqueue_adapter_level_zero.match

Original file line number	Diff line number	Diff line change
`@@ -7,7 +7,7 @@`
`7`	`7`	`* SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception`
`8`	`8`	`*`
`9`	`9`	`* @file ur_ddi.h`
`10`		`- * @version v0.10-r0`
	`10`	`+ * @version v0.11-r0`
`11`	`11`	`*`
`12`	`12`	`*/`
`13`	`13`	`#ifndef UR_DDI_H_INCLUDED`
Original file line number	Diff line number	Diff line change
`@@ -7,7 +7,7 @@`
`7`	`7`	`* SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception`
`8`	`8`	`*`
`9`	`9`	`* @file ur_print.hpp`
`10`		`- * @version v0.10-r0`
	`10`	`+ * @version v0.11-r0`
`11`	`11`	`*`
`12`	`12`	`*/`
`13`	`13`	`#ifndef UR_PRINT_HPP`