-
Notifications
You must be signed in to change notification settings - Fork 125
Pull in intel/llvm changes to main - Fri 21st Feb #2719
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Merged
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Contributor
kbenzie
commented
Feb 21, 2025
- [SYCL][NativeCPU] Fix kernel argument passing. intel/llvm#16995
- [DevMSAN] Support device memory sanitizer for DG2 GPU device intel/llvm#16619
- [SYCL][CUDA] Update UMF in UR to fix issue in LLVM intel/llvm#17034
- [UR] Add remaining calls shared with queue in level-zero v2 adapter intel/llvm#17061
- [UR] Make command-buffer creation descriptor mandatory intel/llvm#17058
- [UR] In-order path for OpenCL command-buffers intel/llvm#17056
- [UR] Don't set -pie on shared objects intel/llvm#16880
- [DeviceASAN] Fix throw "UR_RESULT_ERROR_INVALID_ARGUMENT" exception when catching free related error intel/llvm#16706
- [UR] Make each profiling info variant for urEventGetProfilingInfo optional and improve its conformance test intel/llvm#17067
- [UR][L0] Fix issue with command-buffer local mem update intel/llvm#17069
- [SYCL][NativeCPU] Fix MSVC build. intel/llvm#17090
- [UR][CI] add manually triggered benchmark action intel/llvm#17088
- [SYCL][CUDA] Use UMF Proxy pool manager with UMF CUDA memory provider in UR intel/llvm#17015
We were reading the kernel arguments at kernel execution time, but kernel arguments are allowed to change between enqueuing and executing. Make sure to create a copy of kernel arguments ahead of time. This was previously approved as a unified-runtime PR: oneapi-src#2700
Update UMF to the commit: ``` commit 5a515c56c92be75944c8246535c408cee7711114 Author: Lukasz Dorau <[email protected]> Date: Mon Feb 17 10:56:05 2025 +0100 Merge pull request oneapi-src#1086 from vinser52/svinogra_l0_linking ``` to fix the issue in LLVM (SYCL/CUDA): intel/llvm#16944 [SYCL][CUDA] Nsys profiling broken after memory providers change Moved from: oneapi-src#2708 Fixes: intel/llvm#16944 Signed-off-by: Lukasz Dorau <[email protected]>
Adds implements calls shared between command buffer and queue in unified-runtime level-zero v2 adapter and moves the shared code to `command_list_manager.cpp`
As discussed in oneapi-src#2670 (comment) the `pCommandBufferDesc` parameter to `urCommandBufferCreateExp` is optional. However, the UR spec doesn't state what the configuration of the created command-buffer is when this isn't passed, and being optional is also inconsistent with the description parameters to urSamplerCreate & urMemImageCreate which are not optional. This PR updates the descriptor parameter to command-buffer creation to be mandatory to address these concerns. Closes oneapi-src#2673 **Note**: This UR patch was previously approved and ready-to-merge in oneapi-src#2676 prior to the repo move
After the [spec bump of cl_khr_command_buffer to 0.9.7](https://github.com/KhronosGroup/OpenCL-Docs/), in the OpenCL adapter we no longer need to worry about the in-order/out-of-order property of the internal queue used on command-command-buffer creation matching the queue used to enqueue the command-buffer. We can therefore take advantage of the in-order flag passed on UR command-buffer creation to use an in-order queue for command-buffer creation, and omit using sync points. **Note:** This UR patch was previously approved and ready-to-merge prior to the UR repo move in oneapi-src#2681
Fixes #16677 by only setting `-pie` linker option in Release builds on executables rather than on any type of target.
…free related error (#16706) UR: oneapi-src#2592 --------- Co-authored-by: Kenneth Benzie (Benie) <[email protected]>
… and improve its conformance test (#17067) Migrated from oneapi-src#2533 This patch turns all of the values returned by urEventGetProfilingInfo to be optional and updates adapters to handle this by returning the appropriate enum when it is not supported. The tests have also been updated, to ensure that returning a counter of "0" or values equal to the previous profiling event is no longer considered a failure.
- Fix group count not being recalculated when a user only passes a new local work size and no new global size - Remove CTS test skips for local update on L0
MSVC warns about a possible uninitialized variable. This is a false positive but explicitly initializing always is harmless, so do this.
This is a first step towards reenabling UR performance testing CI. This introduces the reusable yml workflow and a way to trigger it manually. Here's an example how it looks: pbalcer/llvm#2 (comment)
Use UMF Proxy pool manager with UMF CUDA memory provider in UR. UMF Proxy pool manager is just a wrapper for the UMF memory provider (CUDA memory provider in this case) plus it adds also tracking of memory allocations. Moved from: oneapi-src#2659 Signed-off-by: Lukasz Dorau <[email protected]>
martygrant
approved these changes
Feb 21, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Labels
command-buffer
Command Buffer feature addition/changes/specification
common
Changes or additions to common utilities
conformance
Conformance test suite issues.
cuda
CUDA adapter specific issues
experimental
Experimental feature additions/changes/specification
hip
HIP adapter specific issues
level-zero
L0 adapter specific issues
loader
Loader related feature/bug
native-cpu
Native CPU adapter specific issues
opencl
OpenCL adapter specific issues
specification
Changes or additions to the specification
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.