efficiently partitionable threefry PRNG 32-bit random bits implementation #13012
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
mattjj
approved these changes
Oct 27, 2022
google-ml-butler
bot
added
kokoro:force-run
pull ready
froystig
force-pushed
the
rng-part-overgenerate
branch
from
e49948c to
60e4d55
Compare
froystig
force-pushed
the
rng-part-overgenerate
branch
from
60e4d55 to
3b8de0d
Compare
mattjj
force-pushed
the
rng-part-overgenerate
branch
2 times, most recently
from
b42f7c7 to
24127f0
Compare
the cost is 2x overgeneration of bits Co-authored-by: Matthew Johnson <[email protected]>
mattjj
force-pushed
the
rng-part-overgenerate
branch
from
24127f0 to
c8b9280
Compare
froystig
changed the title
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Outdated suggestions cannot be applied.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.
Puzzled over and co-authored with @mattjj!
This essentially makes the implementation of Threefry 32-bit random bits into a pure map (in a way that the compiler notices that it is).
The cost relative to the existing implementation is 2x over-generation of bits, but in return we are bit-equivalent between the two implementations.
This sits behind a flag for now. If/when we verify no major (unpartitioned) performance regressions, we can swap this in as the default.