Make broadcast recursion in flatten structural
#29816
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
The inference enhancements in #29294 work quite well to prevent limiting on many kinds of code. However, targetting TPUs, one code pattern it struggeled with was a fairly large broadcast fusion in Flux: λ.(reshape(γ, affine_shape...) .* ((x .- μ) ./ σ) .+ reshape(β, affine_shape...)) The reason #29294 is because the make_makeargs function used by the implementation of Broadcast.flatten (which the TPU backend uses) had a non-decreasing first argument (passing the return value of a previous invocation of make_makeargs back in as the first argument). However, that's not a fundamental limitation of the operation, but rather an implementation choice. This PR switches that function's recursion pattern to be purely structural, allowing inference to infer through it (with the changes in #29294). As a result, ResNet50 infers properly.
|
@nanosoldier |
nalimilan
reviewed
Oct 26, 2018
Co-Authored-By: mbauman <[email protected]>
mbauman
approved these changes
Oct 26, 2018
|
Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan |
|
This PR seems to have broken Flux. cc @MikeInnes |
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Outdated suggestions cannot be applied.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.
The inference enhancements in #29294 work quite well to prevent limiting
on many kinds of code. However, targetting TPUs, one code pattern it
struggeled with was a fairly large broadcast fusion in Flux:
The reason #29294 doesn't trigger is that the make_makeargs function used by the
implementation of Broadcast.flatten (which the TPU backend uses) had
a non-decreasing first argument (passing the return value of a previous
invocation of make_makeargs back in as the first argument). However,
that's not a fundamental limitation of the operation, but rather an
implementation choice. This PR switches that function's recursion pattern
to be purely structural, allowing inference to infer through it (with
the changes in #29294). As a result, ResNet50 infers properly.