[TE][Fix] Comparison of the output tensor #9829
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
leeexyz
requested review from
areusch,
comaniac,
icemelon,
jroesch,
junrushao,
kparzysz-quic,
masahi,
merrymercy,
tqchen,
vinx13,
yzhliu and
ZihengJiang
as code owners
|
CC @Hzfengsy would you like to take a look? Thanks! |
|
Could you please rebase it and make CI green? |
|
@Hzfengsy It seems this modification impacts the hybrid op, which needs a different output tensor every time. |
leeexyz
force-pushed
the
te_fix
branch
2 times, most recently
from
0fcbee2 to
a96e47d
Compare
|
@Hzfengsy Would you mind taking a look? I fixed the Hybrid Op issue. :) |
Hzfengsy
approved these changes
Feb 20, 2022
|
There appears to be a memory leak from this when tuning some models on gpu. I'm taking a look at this rn |
AndrewZhaoLuo
added a commit
that referenced
this pull request
Mar 9, 2022
This reverts commit 73cf51b.
|
Please review #10540 |
AndrewZhaoLuo
added a commit
that referenced
this pull request
Mar 9, 2022
This reverts commit 73cf51b.
masahi
pushed a commit
that referenced
this pull request
Mar 9, 2022
ziqiangxu8457
pushed a commit
to ziqiangxu8457/tvm
that referenced
this pull request
Mar 9, 2022
…che#10540) This reverts commit 73cf51b.
ziqiangxu8457
pushed a commit
to ziqiangxu8457/tvm
that referenced
this pull request
Mar 12, 2022
…che#10540) This reverts commit 73cf51b.
pfk-beta
pushed a commit
to pfk-beta/tvm
that referenced
this pull request
Apr 11, 2022
* [TE][Fix] Comparison of the output tensor * fix hybrid op issue * fix tensor replacement in schedule ops * fix compute inline
pfk-beta
pushed a commit
to pfk-beta/tvm
that referenced
this pull request
Apr 11, 2022
…che#10540) This reverts commit 73cf51b.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Hi All,
I encountered an issue that the comparison between the output tensor and itself.
Rank-0Tensor. Do not support comparison even if compare to the tensor itself. There is a use case that tensorize a no keep-dims reduction compute to an HW intrinsic, and it is will be failed in the buffer bind step.tensor.op.output(i), the address of the tensor is charged, it is a little bit confusing. The result is that we cannot simply usesame_asto check a tensor.So I added the
outputsmember inclass Operation, and recorded the output tensors on the first invocation ofOperation::output. This is a minor change and would be the simplest way to fix it. :)What is more, I did not modify the functions
output_dtypeandoutput_shapeaccordingly.