[LLVMGPU] Set prefetching on translation info #17744
Merged
Conversation
This file contains 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
Groverkss
requested review from
MaheshRavishankar,
qedawkins and
kuhar
as code owners
qedawkins
approved these changes
Jun 26, 2024
|
@kuhar You would need to change your tuning script to explicitly put prefetching in tuning config after this patch |
|
cc: @RattataKing ☝️ |
|
FYI this improved the SDXL benchmarks we run in this repo (on mi250 using rocm): (we set
|
This would mean that not using prefetching on the tuned configurations is actually better. Adding a parameter for prefetching to the tuning script might be a good idea @kuhar @RattataKing |
ScottTodd
added a commit
that referenced
this pull request
Jul 2, 2024
Benchmark metrics improved (when using `--iree-llvmgpu-enable-prefetch=true`), so locking in the improvements. Context: #17744 (comment) Presubmit results: https://github.com/iree-org/iree/actions/runs/9765047731/attempts/1#summary-26955236756
LLITCHEV
pushed a commit
to LLITCHEV/iree
that referenced
this pull request
Jul 30, 2024
This patch makes prefetch_shared_memory part of translation_info config dictionary, allowing us to control prefetching at dispatch level, instead of globally turning it on/off. Prefetching is still off by default, the flag makes KernelConfig add prefetch_shared_memory unit attribute to config dictionary. Signed-off-by: Lubo Litchev <[email protected]>
LLITCHEV
pushed a commit
to LLITCHEV/iree
that referenced
this pull request
Jul 30, 2024
Benchmark metrics improved (when using `--iree-llvmgpu-enable-prefetch=true`), so locking in the improvements. Context: iree-org#17744 (comment) Presubmit results: https://github.com/iree-org/iree/actions/runs/9765047731/attempts/1#summary-26955236756 Signed-off-by: Lubo Litchev <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
This patch makes prefetch_shared_memory part of translation_info config dictionary, allowing us to control prefetching at dispatch level, instead of globally turning it on/off. Prefetching is still off by default, the flag makes KernelConfig add prefetch_shared_memory unit attribute to config dictionary.