Skip to content

Use an extra bit of entropy in the Float64 uniform 0-1 distribution. #712

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
nhz2 wants to merge 1 commit into from
+5 −3
Closed

Use an extra bit of entropy in the Float64 uniform 0-1 distribution. #712

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 5 additions & 3 deletions src/host/random.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,10 +51,12 @@ end

@inline function u01(::Type{Float64}, u::UInt64)
# Bit-pattern construction avoids the expensive Float64(::UInt64) conversion
# and fma on consumer GPUs (where FP64 throughput is as low as 1:64). The
# low bit of the mantissa is forced set so the result is strictly in (0, 1),
# on consumer GPUs (where FP64 throughput is as low as 1:64).
# The output is uniformly sampled from the following 2^52 evenly spaced points.
# 1*2^-53, 3*2^-53, 5*2^-53 ... (2^53-1)*2^-53
# so the result is strictly in (0, 1),
# which is required by Box-Muller's log(u).
reinterpret(Float64, ((u >> 12) | 0x1) | 0x3ff0000000000000) - 1.0
reinterpret(Float64, (u >> 12) | 0x3ff0000000000000) - prevfloat(1.0)
end


Expand Down