Merge branch 'main' into steffen/composite_devices_not_supported_and_empty

Author: steffenlarsen

.github/workflows/e2e_core.yml

@@ -178,20 +178,23 @@ jobs:
       id: tests
       run: ninja -C build-e2e check-sycl-e2e
 
-    - name: Add comment to PR
-      uses: actions/github-script@60a0d83039c74a4aee543508d2ffcb1c3799cdea # v7.0.1
-      if: ${{ always() }}
-      with:
-        script: |
-          const adapter = '${{ matrix.adapter.name }}';
-          const url = '${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}';
-          const test_status = '${{ steps.tests.outcome }}';
-          const job_status = '${{ job.status }}';
-          const body = `E2E ${adapter} build:\n${url}\nJob status: ${job_status}. Test status: ${test_status}`;
-
-          github.rest.issues.createComment({
-            issue_number: context.issue.number,
-            owner: context.repo.owner,
-            repo: context.repo.repo,
-            body: body
-          })
+    # FIXME: Requires pull-request: write permissions but this is only granted
+    # on pull requests from forks if using pull_request_target workflow
+    # trigger but not the pull_request trigger..
+    # - name: Add comment to PR
+    #   uses: actions/github-script@60a0d83039c74a4aee543508d2ffcb1c3799cdea # v7.0.1
+    #   if: ${{ always() }}
+    #   with:
+    #     script: |
+    #       const adapter = '${{ matrix.adapter.name }}';
+    #       const url = '${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}';
+    #       const test_status = '${{ steps.tests.outcome }}';
+    #       const job_status = '${{ job.status }}';
+    #       const body = `E2E ${adapter} build:\n${url}\nJob status: ${job_status}. Test status: ${test_status}`;
+
+    #       github.rest.issues.createComment({
+    #         issue_number: context.issue.number,
+    #         owner: context.repo.owner,
+    #         repo: context.repo.repo,
+    #         body: body
+    #       })

CMakeLists.txt

@@ -25,6 +25,7 @@ set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_CXX_STANDARD_REQUIRED YES)
 
 # Build Options
+option(UR_BUILD_EXAMPLES "Build example applications." ON)
 option(UR_BUILD_TESTS "Build unit tests." ON)
 option(UR_BUILD_TOOLS "build ur tools" ON)
 option(UR_FORMAT_CPP_STYLE "format code style of C++ sources" OFF)
@@ -258,7 +259,9 @@ install(
     EXPORT ${PROJECT_NAME}-targets)
 
 add_subdirectory(source)
-add_subdirectory(examples)
+if(UR_BUILD_EXAMPLES)
+    add_subdirectory(examples)
+endif()
 if(UR_BUILD_TESTS)
     add_subdirectory(test)
 endif()

README.md

@@ -118,6 +118,7 @@ List of options provided by CMake:
 
 | Name | Description | Values | Default |
 | - | - | - | - |
+| UR_BUILD_EXAMPLES | Build example applications | ON/OFF | ON |
 | UR_BUILD_TESTS | Build the tests | ON/OFF | ON |
 | UR_BUILD_TOOLS | Build tools | ON/OFF | ON |
 | UR_FORMAT_CPP_STYLE | Format code style | ON/OFF | OFF |

include/ur_api.h

@@ -1454,9 +1454,6 @@ inline std::ostream &operator<<(std::ostream &os, enum ur_result_t value) {
     case UR_RESULT_ERROR_INVALID_IMAGE_FORMAT_DESCRIPTOR:
         os << "UR_RESULT_ERROR_INVALID_IMAGE_FORMAT_DESCRIPTOR";
         break;
-    case UR_RESULT_ERROR_IMAGE_FORMAT_NOT_SUPPORTED:
-        os << "UR_RESULT_ERROR_IMAGE_FORMAT_NOT_SUPPORTED";
-        break;
     case UR_RESULT_ERROR_MEM_OBJECT_ALLOCATION_FAILURE:
         os << "UR_RESULT_ERROR_MEM_OBJECT_ALLOCATION_FAILURE";
         break;
@@ -5466,9 +5463,6 @@ inline ur_result_t printFlag<ur_mem_flag_t>(std::ostream &os, uint32_t flag) {
 ///     std::ostream &
 inline std::ostream &operator<<(std::ostream &os, enum ur_mem_type_t value) {
     switch (value) {
-    case UR_MEM_TYPE_BUFFER:
-        os << "UR_MEM_TYPE_BUFFER";
-        break;
     case UR_MEM_TYPE_IMAGE2D:
         os << "UR_MEM_TYPE_IMAGE2D";
         break;
@@ -5484,9 +5478,6 @@ inline std::ostream &operator<<(std::ostream &os, enum ur_mem_type_t value) {
     case UR_MEM_TYPE_IMAGE1D_ARRAY:
         os << "UR_MEM_TYPE_IMAGE1D_ARRAY";
         break;
-    case UR_MEM_TYPE_IMAGE1D_BUFFER:
-        os << "UR_MEM_TYPE_IMAGE1D_BUFFER";
-        break;
     case UR_MEM_TYPE_IMAGE_CUBEMAP_EXP:
         os << "UR_MEM_TYPE_IMAGE_CUBEMAP_EXP";
         break;

include/ur_print.hpp

@@ -50,6 +50,7 @@ Runtime:
   * Sampled images
   * Unsampled images
   * Mipmaps
+  * Image arrays
   * Cubemaps
   * USM backed images
 
@@ -207,6 +208,8 @@ Changelog
 +------------------------------------------------------------------------+
 | 11.0     | Added device queries for sampled image fetch capabilities.  |
 +----------+-------------------------------------------------------------+
+| 12.0     | Added image arrays to list of supported bindless images     |
++----------+-------------------------------------------------------------+
 
 Contributors
 --------------------------------------------------------------------------------

scripts/core/EXP-BINDLESS-IMAGES.rst

@@ -277,6 +277,39 @@ native handle to a driver handle. For example, OpenCL platform
 may expose an extension ${x}ProgramCreateWithNativeHandle to retrieve
 a cl_program.
 
+Memory
+======
+
+UR Mem Handles
+--------------
+
+A ${x}_mem_handle_t can represent an untyped memory buffer object, created by
+${x}MemBufferCreate, or a memory image object, created by ${x}MemImageCreate.
+A ${x}_mem_handle_t manages the internal allocation and deallocation of native
+memory objects across all devices in a ${x}_context_handle_t. A
+${x}_mem_handle_t may only be used by queues that share the same
+${x}_context_handle_t. 
+
+If multiple queues in the same ${x}_context_handle_t use the same
+${x}_mem_handle_t across dependent commands, a dependency must be defined by the
+user using the enqueue entry point's phEventWaitList parameter. Provided that
+dependencies are explicitly passed to UR entry points, a UR adapter will manage
+memory migration of native memory objects across all devices in a context, if
+memory migration is indeed necessary in the backend API.
+
+.. parsed-literal::
+
+    // Q1 and Q2 are both in hContext
+    ${x}_mem_handle_t hBuffer;
+    ${x}MemBufferCreate(hContext,,,,&hBuffer);
+    ${x}EnqueueMemBufferWrite(Q1, hBuffer,,,,,,, &outEv);
+    ${x}EnqueueMemBufferRead(Q2, hBuffer,,,,, 1, &outEv /*phEventWaitList*/, );
+
+As such, the buffer written to in ${x}EnqueueMemBufferWrite can be
+successfully read using ${x}EnqueueMemBufferRead from another queue in the same
+context, since the event associated with the write operation has been passed as
+a dependency to the read operation.
+
 Memory Pooling
 ----------------------------------
 

scripts/core/PROG.rst

@@ -205,8 +205,6 @@ etors:
       desc: "Invalid image size"
     - name: ERROR_INVALID_IMAGE_FORMAT_DESCRIPTOR
       desc: "Invalid image format descriptor"
-    - name: ERROR_IMAGE_FORMAT_NOT_SUPPORTED
-      desc: "Image format not supported"
     - name: ERROR_MEM_OBJECT_ALLOCATION_FAILURE
       desc: "Memory object allocation failure"
     - name: ERROR_INVALID_PROGRAM_EXECUTABLE

scripts/core/common.yml

@@ -387,7 +387,7 @@ returns:
     - $X_RESULT_ERROR_INVALID_CONTEXT
     - $X_RESULT_ERROR_INVALID_VALUE
     - $X_RESULT_ERROR_INVALID_IMAGE_FORMAT_DESCRIPTOR:
-      - "`pImageDesc && UR_MEM_TYPE_IMAGE1D_BUFFER < pImageDesc->type`"
+      - "`pImageDesc && UR_MEM_TYPE_IMAGE1D_ARRAY < pImageDesc->type`"
     - $X_RESULT_ERROR_INVALID_IMAGE_SIZE
     - $X_RESULT_ERROR_INVALID_OPERATION
 --- #--------------------------------------------------------------------------
@@ -442,7 +442,7 @@ returns:
     - $X_RESULT_ERROR_INVALID_CONTEXT
     - $X_RESULT_ERROR_INVALID_VALUE
     - $X_RESULT_ERROR_INVALID_IMAGE_FORMAT_DESCRIPTOR:
-      - "`pImageDesc && UR_MEM_TYPE_IMAGE1D_BUFFER < pImageDesc->type`"
+      - "`pImageDesc && UR_MEM_TYPE_IMAGE1D_ARRAY < pImageDesc->type`"
     - $X_RESULT_ERROR_INVALID_IMAGE_SIZE
     - $X_RESULT_ERROR_INVALID_OPERATION
 --- #--------------------------------------------------------------------------
@@ -479,7 +479,7 @@ returns:
     - $X_RESULT_ERROR_INVALID_CONTEXT
     - $X_RESULT_ERROR_INVALID_VALUE
     - $X_RESULT_ERROR_INVALID_IMAGE_FORMAT_DESCRIPTOR:
-      - "`pImageDesc && UR_MEM_TYPE_IMAGE1D_BUFFER < pImageDesc->type`"
+      - "`pImageDesc && UR_MEM_TYPE_IMAGE1D_ARRAY < pImageDesc->type`"
     - $X_RESULT_ERROR_INVALID_IMAGE_SIZE
     - $X_RESULT_ERROR_INVALID_SAMPLER
     - $X_RESULT_ERROR_INVALID_OPERATION
@@ -542,7 +542,7 @@ returns:
     - $X_RESULT_ERROR_INVALID_QUEUE
     - $X_RESULT_ERROR_INVALID_VALUE
     - $X_RESULT_ERROR_INVALID_IMAGE_FORMAT_DESCRIPTOR:
-      - "`pImageDesc && UR_MEM_TYPE_IMAGE1D_BUFFER < pImageDesc->type`"
+      - "`pImageDesc && UR_MEM_TYPE_IMAGE1D_ARRAY < pImageDesc->type`"
     - $X_RESULT_ERROR_INVALID_IMAGE_SIZE
     - $X_RESULT_ERROR_INVALID_OPERATION
 --- #--------------------------------------------------------------------------
@@ -679,7 +679,7 @@ returns:
     - $X_RESULT_ERROR_INVALID_CONTEXT
     - $X_RESULT_ERROR_INVALID_VALUE
     - $X_RESULT_ERROR_INVALID_IMAGE_FORMAT_DESCRIPTOR:
-      - "`pImageDesc && UR_MEM_TYPE_IMAGE1D_BUFFER < pImageDesc->type`"
+      - "`pImageDesc && UR_MEM_TYPE_IMAGE1D_ARRAY < pImageDesc->type`"
     - $X_RESULT_ERROR_INVALID_IMAGE_SIZE
     - $X_RESULT_ERROR_INVALID_OPERATION
     - $X_RESULT_ERROR_OUT_OF_RESOURCES

scripts/core/exp-bindless-images.yml

@@ -900,7 +900,7 @@ returns:
     - $X_RESULT_ERROR_OUT_OF_HOST_MEMORY
 --- #--------------------------------------------------------------------------
 type: function
-desc: "Update a kernel launch command in a finalized command-buffer."
+desc: "Update a kernel launch command in a finalized command-buffer. This entry-point is synchronous and may block if the command-buffer is executing when the entry-point is called."
 class: $xCommandBuffer
 name: UpdateKernelLaunchExp
 params:

scripts/core/exp-command-buffer.yml

@@ -41,8 +41,6 @@ desc: "Memory types"
 class: $xMem
 name: $x_mem_type_t
 etors:
-    - name: BUFFER
-      desc: "Buffer object"
     - name: IMAGE2D
       desc: "2D image object"
     - name: IMAGE3D
@@ -53,8 +51,6 @@ etors:
       desc: "1D image object"
     - name: IMAGE1D_ARRAY
       desc: "1D image array object"
-    - name: IMAGE1D_BUFFER
-      desc: "1D image buffer object"
 --- #--------------------------------------------------------------------------
 type: enum
 desc: "Memory Information type"
@@ -213,6 +209,15 @@ name: ImageCreate
 ordinal: "0"
 analogue:
     - "**clCreateImage**"
+details:
+    - |
+      The primary $x_image_format_t that must be supported by all the adapters are {UR_IMAGE_CHANNEL_ORDER_RGBA, UR_IMAGE_CHANNEL_TYPE_UNORM_INT8}, 
+      {UR_IMAGE_CHANNEL_ORDER_RGBA, UR_IMAGE_CHANNEL_TYPE_UNORM_INT16}, {UR_IMAGE_CHANNEL_ORDER_RGBA, UR_IMAGE_CHANNEL_TYPE_SNORM_INT8}, 
+      {UR_IMAGE_CHANNEL_ORDER_RGBA, UR_IMAGE_CHANNEL_TYPE_SNORM_INT16}, {UR_IMAGE_CHANNEL_ORDER_RGBA, UR_IMAGE_CHANNEL_TYPE_SIGNED_INT8}, 
+      {UR_IMAGE_CHANNEL_ORDER_RGBA, UR_IMAGE_CHANNEL_TYPE_SIGNED_INT16}, {UR_IMAGE_CHANNEL_ORDER_RGBA, UR_IMAGE_CHANNEL_TYPE_SIGNED_INT32}, 
+      {UR_IMAGE_CHANNEL_ORDER_RGBA, UR_IMAGE_CHANNEL_TYPE_UNSIGNED_INT8}, {UR_IMAGE_CHANNEL_ORDER_RGBA, UR_IMAGE_CHANNEL_TYPE_UNSIGNED_INT16}, 
+      {UR_IMAGE_CHANNEL_ORDER_RGBA, UR_IMAGE_CHANNEL_TYPE_UNSIGNED_INT32}, {UR_IMAGE_CHANNEL_ORDER_RGBA, UR_IMAGE_CHANNEL_TYPE_HALF_FLOAT}, 
+      {UR_IMAGE_CHANNEL_ORDER_RGBA, UR_IMAGE_CHANNEL_TYPE_FLOAT}.
 params:
     - type: $x_context_handle_t
       name: hContext
@@ -236,12 +241,13 @@ returns:
     - $X_RESULT_ERROR_INVALID_CONTEXT
     - $X_RESULT_ERROR_INVALID_VALUE
     - $X_RESULT_ERROR_INVALID_IMAGE_FORMAT_DESCRIPTOR:
-      - "`pImageDesc && UR_MEM_TYPE_IMAGE1D_BUFFER < pImageDesc->type`"
+      - "`pImageDesc && UR_MEM_TYPE_IMAGE1D_ARRAY < pImageDesc->type`"
     - $X_RESULT_ERROR_INVALID_IMAGE_SIZE
     - $X_RESULT_ERROR_INVALID_OPERATION
     - $X_RESULT_ERROR_INVALID_HOST_PTR:
       - "`pHost == NULL && (flags & (UR_MEM_FLAG_USE_HOST_POINTER | UR_MEM_FLAG_ALLOC_COPY_HOST_POINTER)) != 0`"
       - "`pHost != NULL && (flags & (UR_MEM_FLAG_USE_HOST_POINTER | UR_MEM_FLAG_ALLOC_COPY_HOST_POINTER)) == 0`"
+    - $X_RESULT_ERROR_UNSUPPORTED_IMAGE_FORMAT
     - $X_RESULT_ERROR_OUT_OF_HOST_MEMORY
     - $X_RESULT_ERROR_OUT_OF_RESOURCES
 --- #--------------------------------------------------------------------------

scripts/core/memory.yml

@@ -32,7 +32,10 @@ etors:
             The reference count returned should be considered immediately stale. 
             It is unsuitable for general use in applications. This feature is provided for identifying memory leaks.
     - name: SIZE
-      desc: "[uint32_t] The size of the queue"
+      desc: |
+            [uint32_t] The size of the queue on the device. Only a valid query
+            if the queue was created with the `ON_DEVICE` queue flag, otherwise
+            `$xQueueGetInfo` will return `$X_RESULT_ERROR_INVALID_QUEUE`.
     - name: EMPTY
       desc: "[$x_bool_t] return true if the queue was empty at the time of the query"
 --- #--------------------------------------------------------------------------
@@ -49,7 +52,7 @@ etors:
       desc: "Enable/disable profiling"
     - name: ON_DEVICE
       value: "$X_BIT(2)"
-      desc: "Is a device queue"
+      desc: "Is a device queue. If this is enabled `OUT_OF_ORDER_EXEC_MODE_ENABLE` must also be enabled."
     - name: ON_DEVICE_DEFAULT
       value: "$X_BIT(3)"
       desc: "Is the default queue for a device"
@@ -108,6 +111,7 @@ returns:
         - "`propSize != 0 && pPropValue == NULL`"
         - "`pPropValue == NULL && pPropSizeRet == NULL`"
     - $X_RESULT_ERROR_INVALID_QUEUE
+        - "If `hQueue` isn't a valid queue handle or if `propName` isn't supported by `hQueue`."
     - $X_RESULT_ERROR_OUT_OF_HOST_MEMORY
     - $X_RESULT_ERROR_OUT_OF_RESOURCES
 --- #--------------------------------------------------------------------------

scripts/core/queue.yml

@@ -215,7 +215,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
     int Major = 0;
     UR_CHECK_ERROR(cuDeviceGetAttribute(
         &Major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, hDevice->get()));
-    uint64_t Capabilities =
+    ur_memory_scope_capability_flags_t Capabilities =
         (Major >= 7) ? UR_MEMORY_SCOPE_CAPABILITY_FLAG_WORK_ITEM |
                            UR_MEMORY_SCOPE_CAPABILITY_FLAG_SUB_GROUP |
                            UR_MEMORY_SCOPE_CAPABILITY_FLAG_WORK_GROUP |
@@ -270,7 +270,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
     int WarpSize = 0;
     UR_CHECK_ERROR(cuDeviceGetAttribute(
         &WarpSize, CU_DEVICE_ATTRIBUTE_WARP_SIZE, hDevice->get()));
-    size_t Sizes[1] = {static_cast<size_t>(WarpSize)};
+    uint32_t Sizes[1] = {static_cast<uint32_t>(WarpSize)};
     return ReturnValue(Sizes, 1);
   }
   case UR_DEVICE_INFO_MAX_CLOCK_FREQUENCY: {
@@ -569,10 +569,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
         ur_queue_flag_t(UR_QUEUE_FLAG_OUT_OF_ORDER_EXEC_MODE_ENABLE |
                         UR_QUEUE_FLAG_PROFILING_ENABLE));
   case UR_DEVICE_INFO_QUEUE_ON_DEVICE_PROPERTIES: {
-    // The mandated minimum capability:
-    ur_queue_flags_t Capability = UR_QUEUE_FLAG_PROFILING_ENABLE |
-                                  UR_QUEUE_FLAG_OUT_OF_ORDER_EXEC_MODE_ENABLE;
-    return ReturnValue(Capability);
+    return ReturnValue(0);
   }
   case UR_DEVICE_INFO_QUEUE_ON_HOST_PROPERTIES: {
     // The mandated minimum capability: