[Relay][Strategy] Use x86 dense schedules for arm_cpu #15470
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
|
Thanks for contributing to TVM! Please refer to the contributing guidelines https://tvm.apache.org/docs/contribute/ for useful information and tips. Please request code reviews from Reviewers by @-ing them in a comment. Generated by tvm-bot |
ekalda
approved these changes
Aug 4, 2023
Currently the fallback used when compiling a dense operation with targets such as `llvm -device=arm_cpu` is `dense.generic`. This results very poor performance. Although apache#13775 meant that x86 schedules are used in cases where no strategy is provided by arm_cpu, the dense strategy is registered due to the existance of specialized schedules for arm_cpu e.g. a schedule for embedded devices. This commit ensures x86 schedules are used inplace of a generic schedule which yeilds much better performance. The commit also follows the same approach for the `dense.generic` schedule as the x86 strategy. This will only be used when autoscheduler is enabled. A test has been added to check the intended schedules are picked when compiling with `arm_cpu`. Change-Id: I8697f630d4acfab71a9626cf9e0dc3086987f163
lhutton1
added a commit
to lhutton1/tvm
that referenced
this pull request
Aug 8, 2023
Similar to apache#15470, x86 schedules are used in place of generic schedules to improve performance. Since the pooling strategy does not use `OpStrategy`, mocking is used to ensure the relevant `schedule_pool` function is called when lowing a Relay pooling operation with respect to a given target. Change-Id: I782fe00e29f9c9cf41b3405d33a82a79cd85a99b
lhutton1
added a commit
that referenced
this pull request
Aug 9, 2023
Similar to #15470, x86 schedules are used in place of generic schedules to improve performance. Since the pooling strategy does not use `OpStrategy`, mocking is used to ensure the relevant `schedule_pool` function is called when lowing a Relay pooling operation with respect to a given target.
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.
Currently the fallback used when compiling a dense operation with targets such as
llvm -device=arm_cpuisdense.generic. This results in very poor performance. Although #13775 meant that x86 schedules are used in cases where no strategy is provided by arm_cpu, the dense strategy is registered due to the existence of specialized schedules for arm_cpu e.g. a schedule for embedded devices. This commit ensures x86 schedules are used inplace of a generic schedule which yields much better performance.The commit also follows the same approach for the
dense.genericschedule as the x86 strategy. This will only be used when auto-scheduler is enabled.A test has been added to check the intended schedules are picked when compiling with
arm_cpu.cc @ekalda @neildhickey