-
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
[Auto Scheduler] Mali Support #7132
Conversation
cc @antinucleon Apple's M1 GPU is more like Mali (such as unified memory) compared with nv gpu, maybe you could modify a little code and try. |
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.
Looks good to me. It should be fine after you addressing @comaniac 's comments.
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.
Thanks for the amazing PR @FrozenGene ! We are definitely interested, even though it's not in our immediate plans. I attempted a review and I left few minor comments.
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.
Looks good.
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.
LGTM. Thanks.
node->sketch_rules.push_back(&rule_multi_level_tiling); | ||
node->sketch_rules.push_back(&rule_skip_stage); | ||
if (node->search_task->target->GetAttr<String>("device", "") == "mali") { | ||
node->sketch_rules.push_back(&rule_always_inline); |
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.
@FrozenGene @merrymercy I think it worth to export sketch API to Python, make it more configurable. For a given workload, we may want to try multiple sketches pattern, exporting will enable these experiments without rebuild.
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.
i agree with you
* [Auto Scheduler] Mali Support * Fix doc * fix lint * address comments * fix doc
* [Auto Scheduler] Mali Support * Fix doc * fix lint * address comments * fix doc
* [Auto Scheduler] Mali Support * Fix doc * fix lint * address comments * fix doc
* [Auto Scheduler] Mali Support * Fix doc * fix lint * address comments * fix doc
* [Auto Scheduler] Mali Support * Fix doc * fix lint * address comments * fix doc
This is the auto sheduler support for mali gpu.
On one T880-MP4 mali gpu, the performance of TFLite mobilenet v2:
As AutoTVM need much time, I don't test it. But according to previous experience and we beyond one good accelerate framework MNN, I am confident that AutoScheduler's performance is better than AutoTVM.