[Codegen][Tuner] Allow tuning specs in the LLVMGPU pipeline #19359
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
kuhar
requested review from
MaheshRavishankar,
qedawkins,
Groverkss and
hanhanW
as code owners
This adds the `materialize-tuning-specs` pass to the LLVMGPU executable configuration pipelines. Add a test that shows that the tuning spec gets applied and picked up in the ROCDL pipeline. Also, replace the print-based checks in existing tests with op remarks on transform strategy application in `materialize-user-configs`. Signed-off-by: Jakub Kuderski <[email protected]>
kuhar
force-pushed
the
tuning-spec-pipeline
branch
from
3b95e16 to
b98f645
Compare
Signed-off-by: Jakub Kuderski <[email protected]>
Groverkss
approved these changes
Dec 4, 2024
Comment on lines
-28
to
+34
| // PARENT-LABEL: [ IR printer: Hello Tuning Spec top-level ] | ||
| // PARENT-NEXT: func.func @main_0 | ||
| // | ||
| // PARENT-LABEL: module @parent attributes { | ||
| // PARENT-SAME: iree_codegen.tuning_spec_mlirbc = dense< | ||
| // PARENT-LABEL: module @child { | ||
| // PARENT: func.func @main_0 | ||
|
|
||
| module @parent { | ||
| module @child { | ||
| // expected-remark@+1 {{Applied transform configuration strategy @iree_linked_tuning_spec::@__kernel_config}} |
8 tasks
| @@ -1183,6 +1183,7 @@ static void buildLLVMGPUCodegenConfigurationPassPipelineImpl( | |||
| funcPassManager.addPass(createConfigTrackingCanonicalizerPass); | |||
| funcPassManager.addPass(createCSEPass); | |||
| } | |||
| modulePassManager.addPass(createMaterializeTuningSpecsPass()); | |||
There was a problem hiding this comment.
I think the pass ordering might be off... Do we want to do user configs first and then tuning spec. User config should be the ultimate override.
There was a problem hiding this comment.
User configs do come first. It's documented in the materialize user configs pass. Materialize tuning specs doesn't run the transform dialect interpreter and it's up to the materialize user configs to decide which transforms library to run.
There was a problem hiding this comment.
Maybe I need to come up with a different name to better disambiguate these two passes...
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 adds the
materialize-tuning-specspass to the LLVMGPU executable configuration pipelines.Add a test that shows that the tuning spec gets applied and picked up in the ROCDL pipeline.
Also, replace the print-based checks in existing tests with op remarks on transform strategy application in
materialize-user-configs.Issue: #19214