[TLE][Ascend][XCS][FEAT] Add enable-legacy-insert-load-store-for-mix-cv kernel lau… - #1132
Open
huanghaoXcore wants to merge 1 commit into
Open
huanghaoXcore wants to merge 1 commit into
huanghaoXcore wants to merge 1 commit into
Conversation
github-actions
Bot
added
DOC
…cv kernel launch option Add an `enable_legacy_insert_load_store_for_mix_cv` field to NPUOptions so the flag can be passed as a per-kernel launch kwarg, and forward it to bishengir-compile as `--enable-legacy-insert-load-store-for-mix-cv` in both the 910_95 and A2/A3 paths, following the existing `disable_auto_cv_work_space_manage` pattern. The option rolls the `InsertLoadStoreForMixCV` pass back to its pre-refactor implementation, avoiding the refactored pass misinferring memscope/coreType for PIPE_MTE2 custom ops (out planned into GM workspace plus spurious cbuf-to-cbuf loads that conflict with the `__cbuf__`/`__ubuf__` C++ ABI). Switch the gather_gm_to_l1 / gather_gm_to_ub examples in test_custom_ops.py from the `disable_auto_cv_work_space_manage=True` workaround to the new option, and update CUSTOM_OP_USAGE.md accordingly. The old kwarg disabled mix-CV workspace management wholesale (also disabling multi-buffering / CV pipelining), while the new one only rolls back this single pass. The kwarg can be dropped once bishengir fixes the custom-op inference bug in the refactored pass.
huanghaoXcore
force-pushed
the
hh_3.5_dev
branch
from
September 9, 2026 08:02
b5e4abe to
00f23d1
Compare
huanghaoXcore
changed the title
Collaborator
|
LGTM |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Add an
enable_legacy_insert_load_store_for_mix_cvfield to NPUOptions so the flag can be passed as a per-kernel launch kwarg, and forward it to bishengir-compile as--enable-legacy-insert-load-store-for-mix-cvin both the 910_95 and A2/A3 paths, following the existingdisable_auto_cv_work_space_managepattern. The option rolls theInsertLoadStoreForMixCVpass back to its pre-refactor implementation, avoiding the refactored pass misinferring memscope/coreType for PIPE_MTE2 custom ops (out planned into GM workspace plus spurious cbuf-to-cbuf loads that conflict with the__cbuf__/__ubuf__C++ ABI).Switch the gather_gm_to_l1 / gather_gm_to_ub examples in test_custom_ops.py from the
disable_auto_cv_work_space_manage=Trueworkaround to the new option, and update CUSTOM_OP_USAGE.md accordingly. The old kwarg disabled mix-CV workspace management wholesale (also disabling multi-buffering / CV pipelining), while the new one only rolls back this single pass. The kwarg can be dropped once bishengir fixes the custom-op inference bug in the refactored pass.