inference performance: drop some unnecessary edges, cache more #29795
Conversation
This file contains hidden or 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
When the return value of the call is statically unused, we can discard the result earlier and avoid involving it in cycle resolution or backedge computation (since it has no effect).
We typically expect that all members in a cycle have the same attributes. It may not be necessary, but that is how we record and track the items in a cycle.
vtjnash
added
performance
:( |
|
Oops, sorry Kristoffer. I have no idea why I implicated some entity named Simon, when you did all of the work. |
|
Kristoffer "Simon" Carlsson. I like that. |
|
Why is this backportable but something like #29551 is not? |
|
Since no reason was ever given for not backporting #29551, I've marked it for backport. |
|
Actually, I thought you had already marked that for backporting, and was just following suit here. |
This was referenced Oct 29, 2018
|
This caused the regressions shown in #29444 (comment). Dropping from backporting. |
|
@nanosoldier |
vtjnash
added a commit
that referenced
this pull request
Oct 31, 2018
vtjnash
added a commit
that referenced
this pull request
Oct 31, 2018
|
Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan |
vtjnash
added a commit
that referenced
this pull request
Nov 2, 2018
tkf
pushed a commit
to tkf/julia
that referenced
this pull request
Nov 21, 2018
KristofferC
pushed a commit
that referenced
this pull request
Nov 28, 2018
KristofferC
pushed a commit
that referenced
this pull request
Dec 12, 2018
KristofferC
pushed a commit
that referenced
this pull request
Feb 11, 2019
KristofferC
pushed a commit
that referenced
this pull request
Feb 20, 2020
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.
As noted by Simon in analyzing #27874, the compiler performance can become unstable in the presence of inference-limiting recursion. This is normally rare in most code, but is actually extremely common in the IO show code. Now that we have fully linearized IR, we can easily detect some of these cases and avoid some computational work and rework.
Note that backedges can't represent the full suite of information that was contained in this forward edge, so we're still forced to make an over-approximation (until we switch that design).