Merge branch 'adapters' into merge-some-main-changes-into-adapters-second-patch

Author: omarahmed1111

.github/workflows/cmake.yml

@@ -196,12 +196,12 @@ jobs:
         -DUR_BUILD_TESTS=ON
         -DUR_BUILD_ADAPTER_${{matrix.adapter.name}}=ON
         -DUR_DPCXX=${{github.workspace}}/dpcpp_compiler/bin/clang++
+        -DUR_SYCL_LIBRARY_DIR=${{github.workspace}}/dpcpp_compiler/lib
         -DUR_CONFORMANCE_TARGET_TRIPLES=${{matrix.adapter.triplet}}
 
     - name: Build
       # This is so that device binaries can find the sycl runtime library
-      run: LD_LIBRARY_PATH=${{github.workspace}}/dpcpp_compiler/lib
-        cmake --build ${{github.workspace}}/build -j $(nproc)
+      run: cmake --build ${{github.workspace}}/build -j $(nproc)
 
     - name: Test adapter specific
       working-directory: ${{github.workspace}}/build

CMakeLists.txt

@@ -42,6 +42,9 @@ option(UR_BUILD_ADAPTER_CUDA "build cuda adapter from SYCL" OFF)
 option(UR_BUILD_ADAPTER_HIP "build hip adapter from SYCL" OFF)
 option(UR_BUILD_EXAMPLE_CODEGEN "Build the codegen example." OFF)
 option(VAL_USE_LIBBACKTRACE_BACKTRACE "enable libbacktrace validation backtrace for linux" OFF)
+set(UR_DPCXX "" CACHE FILEPATH "Path of the DPC++ compiler executable")
+set(UR_SYCL_LIBRARY_DIR "" CACHE PATH
+    "Path of the SYCL runtime library directory")
 
 set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/lib)
 set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/lib)

README.md

@@ -136,6 +136,8 @@ List of options provided by CMake:
 | UR_BUILD_ADAPTER_HIP    | Fetch and use hip adapter from SYCL                    | ON/OFF     | OFF     |
 | UR_HIP_PLATFORM         | Build hip adapter for AMD or NVIDIA platform           | AMD/NVIDIA | AMD     |
 | UR_ENABLE_COMGR         | Enable comgr lib usage           | AMD/NVIDIA | AMD     |
+| UR_DPCXX | Path of the DPC++ compiler executable to build CTS device binaries | File path | `""` |
+| UR_SYCL_LIBRARY_DIR | Path of the SYCL runtime library directory to build CTS device binaries | Directory path | `""` |
 
 ### Additional make targets
 

source/adapters/level_zero/adapter.cpp

@@ -175,17 +175,19 @@ UR_APIEXPORT ur_result_t UR_APICALL urAdapterRetain(ur_adapter_handle_t) {
 }
 
 UR_APIEXPORT ur_result_t UR_APICALL urAdapterGetLastError(
-    ur_adapter_handle_t Adapter, ///< [in] handle of the platform instance
+    [[maybe_unused]] ur_adapter_handle_t
+        AdapterHandle,    ///< [in] handle of the platform instance
     const char **Message, ///< [out] pointer to a C string where the adapter
                           ///< specific error message will be stored.
-    int32_t *Error ///< [out] pointer to an integer where the adapter specific
-                   ///< error code will be stored.
+    [[maybe_unused]] int32_t
+        *Error ///< [out] pointer to an integer where the adapter specific
+               ///< error code will be stored.
 ) {
-  std::ignore = Adapter;
-  std::ignore = Message;
-  std::ignore = Error;
-  urPrint("[UR][L0] %s function not implemented!\n", __FUNCTION__);
-  return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
+  AdapterHandle = &Adapter;
+  *Message = ErrorMessage;
+  Error = &ErrorAdapterNativeCode;
+
+  return ErrorMessageCode;
 }
 
 UR_APIEXPORT ur_result_t UR_APICALL urAdapterGetInfo(ur_adapter_handle_t,

source/adapters/level_zero/command_buffer.cpp

@@ -177,7 +177,7 @@ ur_result_t calculateKernelWorkDimensions(
             Device->ZeDeviceComputeProperties->maxGroupSizeX,
             Device->ZeDeviceComputeProperties->maxGroupSizeY,
             Device->ZeDeviceComputeProperties->maxGroupSizeZ};
-        GroupSize[I] = std::min(size_t(GroupSize[I]), GlobalWorkSize[I]);
+        GroupSize[I] = (std::min)(size_t(GroupSize[I]), GlobalWorkSize[I]);
         while (GlobalWorkSize[I] % GroupSize[I]) {
           --GroupSize[I];
         }

source/adapters/level_zero/common.cpp

@@ -280,13 +280,16 @@ template <> zes_structure_type_t getZesStructureType<zes_mem_properties_t>() {
 // Global variables for ZER_EXT_RESULT_ADAPTER_SPECIFIC_ERROR
 thread_local ur_result_t ErrorMessageCode = UR_RESULT_SUCCESS;
 thread_local char ErrorMessage[MaxMessageSize];
+thread_local int32_t ErrorAdapterNativeCode;
 
 // Utility function for setting a message and warning
-[[maybe_unused]] void setErrorMessage(const char *message,
-                                      ur_result_t error_code) {
-  assert(strlen(message) <= MaxMessageSize);
-  strcpy(ErrorMessage, message);
-  ErrorMessageCode = error_code;
+[[maybe_unused]] void setErrorMessage(const char *pMessage,
+                                      ur_result_t ErrorCode,
+                                      int32_t AdapterErrorCode) {
+  assert(strlen(pMessage) <= MaxMessageSize);
+  strcpy(ErrorMessage, pMessage);
+  ErrorMessageCode = ErrorCode;
+  ErrorAdapterNativeCode = AdapterErrorCode;
 }
 
 ur_result_t zerPluginGetLastError(char **message) {

source/adapters/level_zero/common.hpp

@@ -467,7 +467,9 @@ constexpr char ZE_SUPPORTED_EXTENSIONS[] =
 constexpr size_t MaxMessageSize = 256;
 extern thread_local ur_result_t ErrorMessageCode;
 extern thread_local char ErrorMessage[MaxMessageSize];
+extern thread_local int32_t ErrorAdapterNativeCode;
 
 // Utility function for setting a message and warning
-[[maybe_unused]] void setErrorMessage(const char *message,
-                                      ur_result_t error_code);
+[[maybe_unused]] void setErrorMessage(const char *pMessage,
+                                      ur_result_t ErrorCode,
+                                      int32_t AdapterErrorCode);

source/adapters/level_zero/device.cpp

@@ -74,7 +74,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGet(
 
   uint32_t ZeDeviceCount = MatchedDevices.size();
 
-  auto N = std::min(ZeDeviceCount, NumEntries);
+  auto N = (std::min)(ZeDeviceCount, NumEntries);
   if (Devices)
     std::copy_n(MatchedDevices.begin(), N, Devices);
 
@@ -631,7 +631,8 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(
   case UR_DEVICE_INFO_GLOBAL_MEM_FREE: {
     if (getenv("ZES_ENABLE_SYSMAN") == nullptr) {
       setErrorMessage("Set ZES_ENABLE_SYSMAN=1 to obtain free memory",
-                      UR_RESULT_SUCCESS);
+                      UR_RESULT_ERROR_UNINITIALIZED,
+                      static_cast<int32_t>(ZE_RESULT_ERROR_UNINITIALIZED));
       return UR_RESULT_ERROR_ADAPTER_SPECIFIC;
     }
     // Only report device memory which zeMemAllocDevice can allocate from.
@@ -1239,7 +1240,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceSelectBinary(
   uint32_t *SelectedBinaryInd = SelectedBinary;
 
   // Find the appropriate device image, fallback to spirv if not found
-  constexpr uint32_t InvalidInd = std::numeric_limits<uint32_t>::max();
+  constexpr uint32_t InvalidInd = (std::numeric_limits<uint32_t>::max)();
   uint32_t Spirv = InvalidInd;
 
   for (uint32_t i = 0; i < NumBinaries; ++i) {

source/adapters/level_zero/kernel.cpp

@@ -79,11 +79,11 @@ UR_APIEXPORT ur_result_t UR_APICALL urEnqueueKernelLaunch(
             UR_RESULT_ERROR_INVALID_VALUE);
   if (LocalWorkSize) {
     // L0
-    UR_ASSERT(LocalWorkSize[0] < std::numeric_limits<uint32_t>::max(),
+    UR_ASSERT(LocalWorkSize[0] < (std::numeric_limits<uint32_t>::max)(),
               UR_RESULT_ERROR_INVALID_VALUE);
-    UR_ASSERT(LocalWorkSize[1] < std::numeric_limits<uint32_t>::max(),
+    UR_ASSERT(LocalWorkSize[1] < (std::numeric_limits<uint32_t>::max)(),
               UR_RESULT_ERROR_INVALID_VALUE);
-    UR_ASSERT(LocalWorkSize[2] < std::numeric_limits<uint32_t>::max(),
+    UR_ASSERT(LocalWorkSize[2] < (std::numeric_limits<uint32_t>::max)(),
               UR_RESULT_ERROR_INVALID_VALUE);
     WG[0] = static_cast<uint32_t>(LocalWorkSize[0]);
     WG[1] = static_cast<uint32_t>(LocalWorkSize[1]);
@@ -110,7 +110,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urEnqueueKernelLaunch(
             Queue->Device->ZeDeviceComputeProperties->maxGroupSizeX,
             Queue->Device->ZeDeviceComputeProperties->maxGroupSizeY,
             Queue->Device->ZeDeviceComputeProperties->maxGroupSizeZ};
-        GroupSize[I] = std::min(size_t(GroupSize[I]), GlobalWorkSize[I]);
+        GroupSize[I] = (std::min)(size_t(GroupSize[I]), GlobalWorkSize[I]);
         while (GlobalWorkSize[I] % GroupSize[I]) {
           --GroupSize[I];
         }
@@ -284,8 +284,9 @@ UR_APIEXPORT ur_result_t UR_APICALL urEnqueueDeviceGlobalVariableWrite(
              (Program->ZeModule, Name, &GlobalVarSize, &GlobalVarPtr));
   if (GlobalVarSize < Offset + Count) {
     setErrorMessage("Write device global variable is out of range.",
-                    UR_RESULT_ERROR_INVALID_VALUE);
-    return UR_RESULT_ERROR_UNKNOWN;
+                    UR_RESULT_ERROR_INVALID_VALUE,
+                    static_cast<int32_t>(ZE_RESULT_ERROR_INVALID_ARGUMENT));
+    return UR_RESULT_ERROR_ADAPTER_SPECIFIC;
   }
 
   // Copy engine is preferred only for host to device transfer.
@@ -333,8 +334,9 @@ UR_APIEXPORT ur_result_t UR_APICALL urEnqueueDeviceGlobalVariableRead(
              (Program->ZeModule, Name, &GlobalVarSize, &GlobalVarPtr));
   if (GlobalVarSize < Offset + Count) {
     setErrorMessage("Read from device global variable is out of range.",
-                    UR_RESULT_ERROR_INVALID_VALUE);
-    return UR_RESULT_ERROR_UNKNOWN;
+                    UR_RESULT_ERROR_INVALID_VALUE,
+                    static_cast<int32_t>(ZE_RESULT_ERROR_INVALID_ARGUMENT));
+    return UR_RESULT_ERROR_ADAPTER_SPECIFIC;
   }
 
   // Copy engine is preferred only for host to device transfer.

source/adapters/level_zero/platform.cpp

@@ -121,7 +121,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urPlatformGet(
     if (*NumPlatforms == 0)
       *NumPlatforms = URPlatformsCache->size();
     else
-      *NumPlatforms = std::min(URPlatformsCache->size(), (size_t)NumEntries);
+      *NumPlatforms = (std::min)(URPlatformsCache->size(), (size_t)NumEntries);
   }
 
   return UR_RESULT_SUCCESS;