This is the mail archive of the libc-alpha@sources.redhat.com mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: new alpha division routines


On Mon, Mar 29, 2004 at 12:06:18AM +0200, Falk Hueffner wrote:
> Another option would be to just use powers of two as table size and
> use a non-sucking hash function. I think that is already used in some
> of the special-cases hash tables in gcc.

Eh.  Non-sucking hash functions are harder to come by generally.

> Could you perhaps send me the source for that? I'd be interested in
> some __remqu stats.

Attached.  I don't distinguish between the four uses at present,
but clearly you could add that.

> > So yes, Y a power of two does occur moderately frequently.
> 
> I'd expect it to be more frequent for the remainder functions. It's
> probably not worth doing for divisions.

I'd expect about equal uses.

> Well, division by zero is undefined behaviour... do you think there's
> code which relies on it being trappable and resumable?

Definitely trapable; Java relies on this.  I doubt anyone attepts
to resume after the trap.  One thing we could do is offer a compiler
switch that specifically allows int-div-by-zero to be replaced by
fp-div-by-zero.


r~


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]