Merge branch 'main' into review/yang/misalign_access

Author: AllanZyne

.github/workflows/cmake.yml

@@ -8,6 +8,7 @@ concurrency:
 
 permissions:
   contents: read
+  pull-requests: write
 
 jobs:
   ubuntu-build:
@@ -191,6 +192,30 @@ jobs:
     with:
       name: NATIVE_CPU
 
+  e2e-level-zero:
+    name: E2E L0
+    permissions:
+      contents: read
+      pull-requests: write
+    needs: [ubuntu-build, level-zero]
+    uses: ./.github/workflows/e2e_level_zero.yml
+
+  e2e-opencl:
+    name: E2E OpenCL
+    permissions:
+      contents: read
+      pull-requests: write
+    needs: [ubuntu-build, opencl]
+    uses: ./.github/workflows/e2e_opencl.yml
+
+  e2e-cuda:
+    name: E2E CUDA
+    permissions:
+      contents: read
+      pull-requests: write
+    needs: [ubuntu-build, cuda]
+    uses: ./.github/workflows/e2e_cuda.yml
+
   windows-build:
     name: Build - Windows
     strategy:

.github/workflows/e2e.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
+    #       })

.github/workflows/e2e_core.yml

@@ -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

@@ -535,7 +535,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urEnqueueCooperativeKernelLaunchExp(
     const size_t *pGlobalWorkOffset, const size_t *pGlobalWorkSize,
     const size_t *pLocalWorkSize, uint32_t numEventsInWaitList,
     const ur_event_handle_t *phEventWaitList, ur_event_handle_t *phEvent) {
-  if (*pGlobalWorkOffset == 0 || pGlobalWorkOffset == nullptr) {
+  if (pGlobalWorkOffset == nullptr || *pGlobalWorkOffset == 0) {
     ur_exp_launch_property_t coop_prop;
     coop_prop.id = UR_EXP_LAUNCH_PROPERTY_ID_COOPERATIVE;
     coop_prop.value.cooperative = 1;

source/adapters/cuda/enqueue.cpp

@@ -131,7 +131,7 @@ struct ContextInfo {
 };
 
 struct USMLaunchInfo {
-    LaunchInfo *Data;
+    LaunchInfo *Data = nullptr;
 
     ur_context_handle_t Context = nullptr;
     ur_device_handle_t Device = nullptr;