Add new simplification rule for invariant loop parameters. #1990
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
|
@coancea You're very welcome to suggest ways to speed this up. |
athas
force-pushed
the
invariant-loop-simplify
branch
3 times, most recently
from
bbe184f
to
ff4969e
Compare
This was suggested by Cosmin to address some of the code produced by AD. The idea is that for a loop of the form loop p = x ... ...stms... in res we construct and simplify the body let p = x ...stms... in res and if that simplifies to 'x', then we conclude that the loop parameter 'p' must be invariant to the loop and simply bind it (and the loop result) to 'x'. Complication: for multi-parameter loops, we must also check that the *original* computation of 'res' does *only* depends on other invariant loop parameters. Currently we do this only for loops that have a constant as one of their initial loop parameter values. The main downside of this rule is that doing recursive simplification is quite expensive. Especially after sequentialisation, pretty much every 'reduce' will have been turned into a loop that triggers this rule (although the rule itself will fail in most cases, after doing the simplification). Therefore I'm a bit hesitant to enable it as is. Sure, the Futhark compiler is slow and it was never meant to be fast, but it is still quite easy for the compiler to become *uselessly slow* if we are not careful. E.g. on OptionPricing, this rule itself makes compilation 10% slower (and does not actually optimise anything - this is purely the cost of failing checks).
athas
force-pushed
the
invariant-loop-simplify
branch
from
ff4969e
to
3127997
Compare
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.
This was suggested by Cosmin to address some of the code produced by AD.
The idea is that for a loop of the form
we construct and simplify the body
and if that simplifies to 'x', then we conclude that the loop parameter 'p' must be invariant to the loop and simply bind it (and the loop result) to 'x'.
Complication: for multi-parameter loops, we must also check that the original computation of 'res' does only depends on other invariant loop parameters.
Currently we do this only for loops that have a constant as one of their initial loop parameter values.
The main downside of this rule is that doing recursive simplification is quite expensive. Especially after sequentialisation, pretty much every 'reduce' will have been turned into a loop that triggers this rule (although the rule itself will fail in most cases, after doing the simplification). Therefore I'm a bit hesitant to enable it as is. Sure, the Futhark compiler is slow and it was never meant to be fast, but it is still quite easy for the compiler to become uselessly slow if we are not careful. E.g. on OptionPricing, this rule itself makes compilation 10% slower (and does not actually optimise anything - this is purely the cost of failing checks).