Merge branch 'adapters' into adapter-batch1

Author: hdelan

.github/workflows/cmake.yml

@@ -163,9 +163,10 @@ jobs:
     strategy:
       matrix:
         adapter: [
-          {name: CUDA, triplet: nvptx64-nvidia-cuda},
-          {name: HIP, triplet: amdgcn-amd-amdhsa}, 
-          {name: L0, triplet: spir64}
+          {name: CUDA, triplet: nvptx64-nvidia-cuda, platform: ""},
+          {name: HIP, triplet: amdgcn-amd-amdhsa, platform: ""},
+          {name: L0, triplet: spir64, platform: ""},
+          {name: OPENCL, triplet: spir64, platform: "Intel(R) OpenCL"}
         ]
         build_type: [Debug, Release]
         compiler: [{c: gcc, cxx: g++}, {c: clang, cxx: clang++}]
@@ -219,7 +220,7 @@ jobs:
     - name: Test adapters
       if: matrix.adapter.name != 'L0'
       working-directory: ${{github.workspace}}/build
-      run: ctest -C ${{matrix.build_type}} --output-on-failure -L "conformance" --timeout 180
+      run: env UR_CTS_ADAPTER_PLATFORM="${{matrix.adapter.platform}}" ctest -C ${{matrix.build_type}} --output-on-failure -L "conformance" --timeout 180
 
   examples-build-hw:
     name: Build - examples on HW

CMakeLists.txt

@@ -4,7 +4,7 @@
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
 cmake_minimum_required(VERSION 3.14.0 FATAL_ERROR)
-project(unified-runtime VERSION 0.7.0)
+project(unified-runtime VERSION 0.9.0)
 
 include(GNUInstallDirs)
 include(CheckCXXSourceCompiles)

include/ur.py

@@ -6,7 +6,7 @@
  SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
  @file ur.py
- @version v0.7-r0
+ @version v0.9-r0
 
  """
 import platform
@@ -573,7 +573,9 @@ def __str__(self):
 class ur_api_version_v(IntEnum):
     _0_6 = UR_MAKE_VERSION( 0, 6 )                  ## version 0.6
     _0_7 = UR_MAKE_VERSION( 0, 7 )                  ## version 0.7
-    CURRENT = UR_MAKE_VERSION( 0, 7 )               ## latest known version
+    _0_8 = UR_MAKE_VERSION( 0, 8 )                  ## version 0.8
+    _0_9 = UR_MAKE_VERSION( 0, 9 )                  ## version 0.9
+    CURRENT = UR_MAKE_VERSION( 0, 9 )               ## latest known version
 
 class ur_api_version_t(c_int):
     def __str__(self):

include/ur_api.h

@@ -7,7 +7,7 @@
  * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
  *
  * @file ur_api.h
- * @version v0.7-r0
+ * @version v0.9-r0
  *
  */
 #ifndef UR_API_H_INCLUDED
@@ -1025,7 +1025,9 @@ urPlatformGetInfo(
 typedef enum ur_api_version_t {
     UR_API_VERSION_0_6 = UR_MAKE_VERSION(0, 6),     ///< version 0.6
     UR_API_VERSION_0_7 = UR_MAKE_VERSION(0, 7),     ///< version 0.7
-    UR_API_VERSION_CURRENT = UR_MAKE_VERSION(0, 7), ///< latest known version
+    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_CURRENT = UR_MAKE_VERSION(0, 9), ///< 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 v0.7-r0
+ * @version v0.9-r0
  *
  */
 #ifndef UR_DDI_H_INCLUDED

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         = v0.7
+PROJECT_NUMBER         = v0.9
 
 # 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/platform.yml

@@ -133,6 +133,12 @@ etors:
     - name: "0_7"
       value: "$X_MAKE_VERSION( 0, 7 )"
       desc: "version 0.7"
+    - name: "0_8"
+      value: "$X_MAKE_VERSION( 0, 8 )"
+      desc: "version 0.8"
+    - name: "0_9"
+      value: "$X_MAKE_VERSION( 0, 9 )"
+      desc: "version 0.9"
 --- #--------------------------------------------------------------------------
 type: function
 desc: "Returns the API version supported by the specified platform"

scripts/parse_specs.py

@@ -18,8 +18,8 @@
 import ctypes
 import itertools
 
-default_version = "0.7"
-all_versions = ["0.6", "0.7"]
+default_version = "0.9"
+all_versions = ["0.6", "0.7", "0.8", "0.9"]
 
 """
     preprocess object

source/adapters/cuda/command_buffer.cpp

@@ -140,8 +140,19 @@ urCommandBufferReleaseExp(ur_exp_command_buffer_handle_t hCommandBuffer) {
 UR_APIEXPORT ur_result_t UR_APICALL
 urCommandBufferFinalizeExp(ur_exp_command_buffer_handle_t hCommandBuffer) {
   try {
+    const unsigned long long flags = 0;
+#if CUDA_VERSION >= 12000
     UR_CHECK_ERROR(cuGraphInstantiate(&hCommandBuffer->CudaGraphExec,
-                                      hCommandBuffer->CudaGraph, 0));
+                                      hCommandBuffer->CudaGraph, flags));
+#elif CUDA_VERSION >= 11040
+    UR_CHECK_ERROR(cuGraphInstantiateWithFlags(
+        &hCommandBuffer->CudaGraphExec, hCommandBuffer->CudaGraph, flags));
+#else
+    // Cannot use flags
+    UR_CHECK_ERROR(cuGraphInstantiate(&hCommandBuffer->CudaGraphExec,
+                                      hCommandBuffer->CudaGraph, nullptr,
+                                      nullptr, 0));
+#endif
   } catch (...) {
     return UR_RESULT_ERROR_UNKNOWN;
   }

source/adapters/cuda/program.cpp

@@ -263,8 +263,8 @@ UR_APIEXPORT ur_result_t UR_APICALL urProgramBuild(ur_context_handle_t hContext,
 }
 
 UR_APIEXPORT ur_result_t UR_APICALL urProgramLinkExp(
-    ur_context_handle_t, uint32_t, const ur_program_handle_t *, uint32_t,
-    ur_device_handle_t *, const char *, ur_program_handle_t *) {
+    ur_context_handle_t, uint32_t, ur_device_handle_t *, uint32_t,
+    const ur_program_handle_t *, const char *, ur_program_handle_t *) {
   return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
 }