Update PhiloxRNG.jl to v1.1.1 and fix counter overflow with UInt64#717
Open
nhz2 wants to merge 3 commits intoJuliaGPU:masterfrom
Open
Update PhiloxRNG.jl to v1.1.1 and fix counter overflow with UInt64#717nhz2 wants to merge 3 commits intoJuliaGPU:masterfrom
nhz2 wants to merge 3 commits intoJuliaGPU:masterfrom
Conversation
maleadt
reviewed
Apr 30, 2026
Comment on lines
+62
to
+64
| @inline function u01(::Type{F}, u::UInt64)::F where F | ||
| fma(F(u), F(2)^Int32(-64), F(2)^Int32(-65)) | ||
| end |
Member
There was a problem hiding this comment.
Isn't this strictly more expensive than the bit-pattern there was before?
Author
There was a problem hiding this comment.
I think so, but it has a much larger dynamic range.
See the comment about this in https://github.com/DEShawResearch/random123/blob/9545ff6413f258be2f04c1d319d99aaef7521150/include/Random123/u01fixedpt.h#L43-L52
I also don't see a big impact on benchmarks, https://github.com/medyan-dev/PhiloxRNG.jl#gpu--nvidia-geforce-rtx-3080-nsvalue-n--100000000 RTX 3080 is still almost hitting its memory bandwidth. Does this show up as a problem on other devices?
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fixes #713
This PR uses the following scheme to avoid rng stream collisions.
Philox4x32-10 takes a 128-bit counter
(ctr0, ctr1)and a 64-bit key. The scheme is:gid + nthreads * localcounter, varying inside the kernel:gidpartitions across threads,localcounterincrements per Philox call from that thread. This allows for over 2^32 threads with fewer Philox calls per thread, or over 2^32 Philox calls per thread with fewer threads without having a collision.ctr1is also randomized when seeding from aRandomDeviceto make collisions less likely.CC @maleadt