Make the fast inverse test throughput-limited rather than latency-limited #7958
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
vksnk
approved these changes
Nov 21, 2023
|
This does indeed fix that test on the new arm bot (though another test is still failing) |
ardier
pushed a commit
to ardier/Halide-mutation
that referenced
this pull request
Mar 3, 2024
…ited (halide#7958) Co-authored-by: Steven Johnson <srj@google.com>
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Outdated suggestions cannot be applied.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 test is currently failing on a Cortex a76 buildbot, because it's a recursive update definition so it ends up limited by instruction latencies rather than throughputs. On an a76 (which is a reasonable CPU to assume for a generic ARM target), if you multiply by a fast inverse the total latency is frecpe + frecps + fmul = 11, whereas the Cortex a76 optimization guide says the latency of an fdiv instruction is 7-10. The cycle costs (sum of inverse throughput) however, are 3 and 8 respectively, so fast_inverse is still a good idea for most imaging workloads that aren't the goofy recursive thing in the test. So hopefully if I just change the test to be thoughput-limited, it'll fix it.
Still disabled on M1, because fdiv there has a throughput of 1?!