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 2022/08/05 15:25:34 UTC
[GitHub] [tvm] yuanfz98 opened a new pull request, #12321: [Topi] add x86 schedule for batch_norm
yuanfz98 opened a new pull request, #12321:
URL: https://github.com/apache/tvm/pull/12321
Follow up of https://github.com/apache/tvm/pull/9694.
1. batch_norm returns a list and it might be tricky to
- apply autotvm and add tiling stages. Need to unpack list object. (here node is a list)
https://github.com/apache/tvm/blob/ca46f21f5168f937cf9b1d205118c770c46aa8c5/python/tvm/autotvm/task/topi_integration.py#L168
- fuse with other kElemwise ops. I found that when batch_norm+relu, all 3 outputs [output, moving_mean, moving_var] are fused with relu and the latter 2 make no sense. Is it possible to select which node to fuse?
2. observed better perf with larger workloads due to parallism (Intel(R) Xeon(R) CPU @ 2.30GHz, 8cores)
```
shape=160, key=generic= {'mean': 0.48806950988364406, 'median': 0.02051979972748086, 'std': 1.402288376912746}
shape=160, key=cpu= {'mean': 0.49737747001927346, 'median': 0.04554405022645369, 'std': 1.3545420627182678}
shape=1600, key=generic= {'mean': 4.354905650106957, 'median': 0.03478030048427172, 'std': 12.959578536273051}
shape=1600, key=cpu= {'mean': 1.4397647402074654, 'median': 0.0432861503213644, 'std': 4.19046676508887}
shape=16000, key=generic= {'mean': 2.161924099782482, 'median': 0.051916949450969696, 'std': 6.330056129490704}
shape=16000, key=cpu= {'mean': 1.2180674902629107, 'median': 0.054580300638917834, 'std': 3.488364687811132}
shape=160000, key=generic= {'mean': 2.184136709984159, 'median': 0.23043325054459274, 'std': 5.860211833531447}
shape=160000, key=cpu= {'mean': 1.2610028796189, 'median': 0.1792462993762456, 'std': 3.2444646099107004}
```
--
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.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] masahi merged pull request #12321: [Topi] add x86 schedule for batch_norm
Posted by GitBox <gi...@apache.org>.
masahi merged PR #12321:
URL: https://github.com/apache/tvm/pull/12321
--
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.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org