comaniac commented on a change in pull request #10177:
URL: https://github.com/apache/tvm/pull/10177#discussion_r801005738
##########
File path: python/tvm/contrib/cutlass/gen_conv2d.py
##########
@@ -252,6 +252,8 @@ def select_op(
lambda align: all([dim % align == 0 for dim in [IC, OC]]),
use_3xtf32,
profile_all_alignments,
+ # Use fp32 accumulation for wgrad to align with cuDNN
+ accumlator_dtype="float32" if conv_kind == ConvKind.Wgrad else
out_dtype,
Review comment:
I see. Make sense.
We discussed about the accum dtype before when @AndrewZhaoLuo was working on
the ToMixedPrecision pass, but just like you pointed out, this will involve
lots of TOPI changes.
--
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: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
