[MPlayer-cvslog] r19207 - trunk/libmpdemux/freesdp/parser.c

[Rich Felker]

On Sat, Jul 29, 2006 at 12:30:08PM +0200, Reimar Döffinger wrote:
> Hello,
> On Sat, Jul 29, 2006 at 12:49:12AM -0400, Rich Felker wrote:
> > IMO we should work out some sort of policy about this stuff. I can't
> > imagine strtof being implemented anyway but as a call to strtold
> > casting the result down to float.. Are there really platforms we
> > support where float arithmetic is faster than double (or long double)
> > arithmetic?
> 
> strtof/strtod is usually slow as hell anyway, using it in a performance
> critical path is a bug anyway, so I really think this is a non-issue -
> if you don't need things like inf and nan to be supported you can easily
> write a function that is at least 3 - 8 times faster (I know, I had to do that
> once).

How is inf/nan 3-8 times slower?? Does it really take 8 times as long
to check that the first byte of the string is not [iInN]??

I'm quite aware there may be slowness due to special computations to
get the bit-exact correct results (does the naive algo give correct
results? I don't know..), but I don't think inf/nan have anything to
do with it.

Rich