#2271: floor, ceiling, round :: Double -> Int are awesomely slow
------------------------------------------+---------------------------------
Reporter: dons | Owner: daniel.is.fischer
Type: bug | Status: patch
Priority: low | Milestone: 7.0.1
Component: libraries/base | Version: 7.1
Keywords: performance, math, double | Testcase:
Blockedby: | Difficulty: Unknown
Os: Unknown/Multiple | Blocking:
Architecture: Unknown/Multiple | Failure: None/Unknown
------------------------------------------+---------------------------------
Changes (by daniel.is.fischer):
* status: infoneeded => patch
Comment:
The new patch contains implementations of `rint` and `rintf`, so we don't
depend on the installed glibc version. The double version is somewhat
slower than glibc's `rint`, but still a lot faster than what we have now.
The float version is even a bit faster than glibc's `rintf` here.
Since the bugs for `double rint(double)` seem to have been fixed pretty
quickly, in contrast to the `rintf` bug, we could call out to math.h's
`rint` instead of using our own if we're willing to risk a very few users
with buggy versions and also the rare cases where the platforms rounding
mode has been set to something other than round to nearest, ties to even.
I've tested and benchmarked the patch on my box (32-bit, i686), test also
on 64-bits and other platforms before merging.
Re the rewrite rules for IntN and WordN, any recommendations where to
stick them?
--
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/2271#comment:22>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler
_______________________________________________
Glasgow-haskell-bugs mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
