New patches in /srv/darcs/git-mirrors/random
commit 48cf0ffa4e70ba4e8f3ce7d434b7bce4918890ca
Author: Ryan Newton <[email protected]>
Date: Fri Jun 24 22:44:54 2011 -0400
Adding a DEVLOG file to keep track of development.
commit 4cfc44a16bd9dea222882e1762e8fb7407411600
Author: Ryan Newton <[email protected]>
Date: Fri Jun 24 22:39:44 2011 -0400
Fix #5133 - also yields this change a 240X speedup in generating random
floats in my tests.
commit d9a5313593b53655f6024105c7122cae71ca4042
Author: Ryan Newton <[email protected]>
Date: Fri Jun 24 18:22:26 2011 -0400
Cleaned up the simple benchmark and added tests for non-Int datatypes. The
results are a bit surprising. Float generation is performing badly.
Here are some results from a 3.33ghz Intel Nehalem:
Cost of rdtsc (ffi call): 75
Approx getCPUTime calls per second: 206,493
Approx clock frequency: 3,336,174,789
First, timing with System.Random interface:
112,276,629 randoms generated [constant zero gen] ~ 29.71
cycles/int
14,289,712 randoms generated [System.Random stdGen] ~ 233
cycles/int
82,546 randoms generated [System.Random Floats] ~ 40,416
cycles/int
83,138 randoms generated [System.Random CFloats] ~ 40,128
cycles/int
2,533,007 randoms generated [System.Random Doubles] ~ 1,317
cycles/int
841,737 randoms generated [System.Random Integers] ~ 3,963
cycles/int
4,704,318 randoms generated [System.Random Bools] ~ 709
cycles/int
commit e7f72c84cb7dec28a417ce24d085d31e9d97160d
Author: Ryan Newton <[email protected]>
Date: Fri Jun 24 14:26:22 2011 -0400
Added some very minor comments and a simple benchmarking script.
Using this script I saw a 13X regression in stdGen in recent revisions
vs the 1.0.0.3 version, dropping to only 900K random ints/sec on my
machine. However, it was some kind of heisenbug because now it's gone
away for unknown reasons. Watch out for it in the future though.
commit 130e421e912d394653a43c987be992b7d3251b05
Author: Ryan Newton <[email protected]>
Date: Thu Jun 23 22:18:13 2011 -0400
minor: Updated maintainer email.
_______________________________________________
Cvs-ghc mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/cvs-ghc
