-
Notifications
You must be signed in to change notification settings - Fork 3.5k
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
[Target] Make key=arm_cpu
--> key=arm_cpu,cpu
on AArch64
#13775
[Target] Make key=arm_cpu
--> key=arm_cpu,cpu
on AArch64
#13775
Conversation
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 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Argh, this was an unintended side effect of #12474, thanks for fixing this @AndrewZhaoLuo!
key=arm_cpu
--> key=arm_cpu,cpu
key=arm_cpu
--> key=arm_cpu,cpu
on AArch64
…ncatenate NPU used generic implementation for concatenate before cpu schedules were made the default fallback schedules. This leads to performance degradation as this blocks fusion with nearby ops. This commit adds Relay op strategy for arm_cpu implementation which makes it use arm_cpu schedule before cpu one. Reference: apache#13775 Co-authored-by: Luke Hutton <luke.hutton@arm.com> Change-Id: If6e65d74309702daf1f837a24e7b19c2912d0d9c
…atenate NPU used generic implementation for concatenate before cpu schedules were made the default fallback schedules. This leads to performance degradation as this blocks fusion with nearby ops. This commit adds Relay op strategy for arm_cpu implementation which makes it use arm_cpu schedule before cpu one. Reference: apache#13775 Co-authored-by: Luke Hutton <luke.hutton@arm.com> Change-Id: If6e65d74309702daf1f837a24e7b19c2912d0d9c
…atenate NPU used generic implementation for concatenate before cpu schedules were made the default fallback schedules. This leads to performance degradation as this blocks fusion with nearby ops. This commit adds Relay op strategy for arm_cpu implementation which makes it use arm_cpu schedule before cpu one. Reference: apache#13775 Co-authored-by: Luke Hutton <luke.hutton@arm.com>
…oncat (#14270) Previously used generic implementation for concatenate before cpu schedules were made the default fallback schedules. This leads to performance degradation as this blocks fusion with nearby ops. This commit adds Relay op strategy for arm_cpu implementation which makes it use arm_cpu schedule before cpu one. Reference: #13775 Co-authored-by: Luke Hutton <luke.hutton@arm.com>
…13775) * arm cpu is cpu * init commit * fix test
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
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
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 #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
The
key
in a target is used in dispatching to different strategies.In most of the code base
arm_cpu
implies also havingcpu
. E.g. inpython/tvm/target/target.py
:In
src/target/parsers/mprofile.cc
:However A-series targets did not have both
arm_cpu
andcpu
keys. Rather they only hadarm_cpu
keys. This PR makes it so it includes both, matching everywhere else in codebase.This also may lead to increased performance speedups as previously A-series targets would dispatch to generic strategy instead of cpu optimized strategy. With this, they should properly dispatch to
arm_cpu
strategy, and fallback tocpu
strategy.