You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@tvm.apache.org by GitBox <gi...@apache.org> on 2020/04/15 08:30:24 UTC
[GitHub] [incubator-tvm] anijain2305 opened a new issue #5340: Schedule
Transferability between Intel and ARM CPU targets
anijain2305 opened a new issue #5340: Schedule Transferability between Intel and ARM CPU targets
URL: https://github.com/apache/incubator-tvm/issues/5340
Relevant discuss post - https://discuss.tvm.ai/t/topi-using-x86-schedules-for-arm-conv2d/6365
Currently, TVM has different schedules for ARM and Intel for conv2d operators. The discuss post listed above shows that Intel NCHWc schedule on ARM performs better end-to-end latency compared to ARM NCHWC conv2d spatial pack for many TFLite networks.
However, this is just one opportunity and there are also some more ideas that we should pursue. This issue lists those potential issues and anybody interested can pick them up. This list is a result of discussions from the above post.
- [ ] Try ARM winograd NCHW schedule on Intel servers
- [ ] Write NCHWc winograd for Intel servers to comply with existing NCHWc data layout ([PR](https://github.com/apache/incubator-tvm/pull/2111))
- [ ] Work on NHWC ARM schedule - tuning and optimization - work started by @jackwish
- [ ] Investigate NHWC vs NCHWc schedule - NCHWc can bring data layouts transform. Check if NHWC can achieve same performance as NCHWc data layout.
@FrozenGene @masahi @tqchen
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-tvm] jackwish commented on issue #5340: Schedule
Transferability between Intel and ARM CPU targets
Posted by GitBox <gi...@apache.org>.
jackwish commented on issue #5340: Schedule Transferability between Intel and ARM CPU targets
URL: https://github.com/apache/incubator-tvm/issues/5340#issuecomment-615631555
That is really good idea that we can share schedules across *backends*!
A few monthes ago, I had once thought that, maybe we can modularize TOPI such that some *well known* schedules can be shared as much as possible. For example, we can have schedules for CPU and GPU which could shared between x86/ARM/... and CUDA/OpenCL respectively. Though I am not actively working on TVM nowdays, hoping to contribute someday still.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
With regards,
Apache Git Services