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 2022/07/27 14:20:19 UTC
[GitHub] [tvm] echuraev opened a new pull request, #12202: [Adreno] Fix winograd tests and accuracy
echuraev opened a new pull request, #12202:
URL: https://github.com/apache/tvm/pull/12202
Thanks for contributing to TVM! Please refer to guideline https://tvm.apache.org/docs/contribute/ for useful information and tips. After the pull request is submitted, please request code reviews from [Reviewers](https://github.com/apache/incubator-tvm/blob/master/CONTRIBUTORS.md#reviewers) by @ them in the pull request thread.
--
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] echuraev commented on a diff in pull request #12202: [Adreno] Fix winograd tests and accuracy
Posted by GitBox <gi...@apache.org>.
echuraev commented on code in PR #12202:
URL: https://github.com/apache/tvm/pull/12202#discussion_r931174305
##########
python/tvm/topi/adreno/conv2d_winograd_common.py:
##########
@@ -494,16 +494,17 @@ def schedule_conv2d_winograd(cfg, s, output, pre_computed):
s[OL].set_scope("local")
output = s.outputs[0]
- m = alpha - 3 + 1
if len(s[output].op.axis) == 4:
n, co, h, w = s[output].op.axis
else:
- n, co, h, w, _ = s[output].op.axis
- ho, wo, hi, wi = s[output].tile(h, w, m, m)
+ n, co, h, w, cb = s[output].op.axis
inverse_scope, n = s[output].split(n, nparts=1)
- fused = s[output].fuse(n, co, ho, wo)
+ fused = s[output].fuse(n, co, h, w)
bb, tt = s[output].split(fused, 128)
+ if len(s[output].op.axis) > 4:
Review Comment:
Fixed. Replaced to check on `None`
--
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] echuraev commented on pull request #12202: [Adreno] Fix winograd tests and accuracy
Posted by GitBox <gi...@apache.org>.
echuraev commented on PR #12202:
URL: https://github.com/apache/tvm/pull/12202#issuecomment-1196832224
cc: @elvin-n, @masahi
--
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 #12202: [Adreno] Fix winograd tests and accuracy
Posted by GitBox <gi...@apache.org>.
masahi merged PR #12202:
URL: https://github.com/apache/tvm/pull/12202
--
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] elvin-n commented on a diff in pull request #12202: [Adreno] Fix winograd tests and accuracy
Posted by GitBox <gi...@apache.org>.
elvin-n commented on code in PR #12202:
URL: https://github.com/apache/tvm/pull/12202#discussion_r931143961
##########
python/tvm/topi/adreno/conv2d_winograd_common.py:
##########
@@ -494,16 +494,17 @@ def schedule_conv2d_winograd(cfg, s, output, pre_computed):
s[OL].set_scope("local")
output = s.outputs[0]
- m = alpha - 3 + 1
if len(s[output].op.axis) == 4:
n, co, h, w = s[output].op.axis
else:
- n, co, h, w, _ = s[output].op.axis
- ho, wo, hi, wi = s[output].tile(h, w, m, m)
+ n, co, h, w, cb = s[output].op.axis
inverse_scope, n = s[output].split(n, nparts=1)
- fused = s[output].fuse(n, co, ho, wo)
+ fused = s[output].fuse(n, co, h, w)
bb, tt = s[output].split(fused, 128)
+ if len(s[output].op.axis) > 4:
Review Comment:
I don;t like split of init cb and this verification but cannot propose how to do this better without code duplication
--
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