Skip to content
This repository has been archived by the owner on Nov 17, 2023. It is now read-only.

Add support for boolean inputs to FusedOp #16796

Merged
merged 7 commits into from
Nov 23, 2019
Merged

Conversation

ptrendx
Copy link
Member

@ptrendx ptrendx commented Nov 12, 2019

Description

Fixes #16723

This PR fixes 2 problems:

  • FusedOp did not support boolean input type as its inputs.
  • During testing the fix for the first problem another problem was spotted (which I believe was previously fixed during the process of working on the fusion PR, but I guess some merge undid it) - the FusedOp did not recompile the code when there was a change of inputs/outputs number of dimensions, which is wrong.

This PR fixes both of those problems and introduces tests for them.
@sxjscience Please validate the fix.

@ptrendx
Copy link
Member Author

ptrendx commented Nov 21, 2019

@marcoabreu @larroy Could you tell me what is the configuration of the unix-gpu test runners? They make TVM error out and I cannot reproduce it locally.

@ptrendx
Copy link
Member Author

ptrendx commented Nov 22, 2019

@sxjscience Could you validate the fix in your original code?

@sxjscience
Copy link
Member

@ptrendx I'm testing it locally

@sxjscience
Copy link
Member

@ptrendx Thanks for the fix. I've confirmed that it solves my previous failure.

@ptrendx ptrendx merged commit e3a7141 into apache:master Nov 23, 2019
ptrendx added a commit to ptrendx/mxnet that referenced this pull request Nov 23, 2019
* Support bool in fusion

* Added tests

* Use use_np decorator vs calling the numpy convention by hand

* Debug TVM failure

* Fix

* Do not invoke TVM broadcast kernel
@ptrendx ptrendx mentioned this pull request Nov 23, 2019
ptrendx added a commit that referenced this pull request Nov 23, 2019
* Support bool in fusion

* Added tests

* Use use_np decorator vs calling the numpy convention by hand

* Debug TVM failure

* Fix

* Do not invoke TVM broadcast kernel
@larroy
Copy link
Contributor

larroy commented Nov 27, 2019

@ptrendx is in the container in ci/ folder. You can use the exact same run by using ci/build.py you have the same instruction in the output of CI run either at the top or the bottom.

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
Projects
None yet
Development

Successfully merging this pull request may close these issues.

[Bug] fused_op does not support boolean type
3 participants