[PATCH] Handle sqrt() shrinking in SimplifyLibCalls like any other call

[Sanjay Patel]

Thanks, all - checked in r220514.


On Thu, Oct 23, 2014 at 2:32 PM, Hal Finkel <hfinkel at anl.gov> wrote:

> ----- Original Message -----
> > From: "Stephen Canon" <scanon at apple.com>
> > To: "Hal Finkel" <hfinkel at anl.gov>
> > Cc: llvm-commits at cs.uiuc.edu, spatel at rotateright.com, beanz at apple.com,
> > reviews+D5919+public+763adead774985f1 at reviews.llvm.org
> > Sent: Thursday, October 23, 2014 3:26:47 PM
> > Subject: Re: [PATCH] Handle sqrt() shrinking in SimplifyLibCalls like
> any other call
> >
> > On Oct 23, 2014, at 3:57 PM, Hal Finkel < hfinkel at anl.gov > wrote:
> >
> > ----- Original Message -----
> >
> >
> > From: "Hal Finkel" < hfinkel at anl.gov >
> > To: "Stephen Canon" < scanon at apple.com >
> > Cc: llvm-commits at cs.uiuc.edu , spatel at rotateright.com ,
> > beanz at apple.com ,
> > reviews+D5919+public+763adead774985f1 at reviews.llvm.org
> > Sent: Thursday, October 23, 2014 2:51:34 PM
> > Subject: Re: [PATCH] Handle sqrt() shrinking in SimplifyLibCalls like
> > any other call
> >
> > ----- Original Message -----
> >
> >
> > From: "Stephen Canon" < scanon at apple.com >
> > To: "Hal Finkel" < hfinkel at anl.gov >
> > Cc: llvm-commits at cs.uiuc.edu , spatel at rotateright.com ,
> > beanz at apple.com ,
> > reviews+D5919+public+763adead774985f1 at reviews.llvm.org
> > Sent: Thursday, October 23, 2014 2:46:17 PM
> > Subject: Re: [PATCH] Handle sqrt() shrinking in SimplifyLibCalls
> > like any other call
> >
> > (float)sqrt((double)x) —> sqrtf(x) is a valid transformation [See
> > Sam
> > Figueroa’s thesis for details].
> >
> > Does this assume that the sqrtf is correct to <= 0.5ulps?
> >
> > I suppose that C99 F.3 does say that, "The sqrt functions in <math.h>
> > provide the IEC 60559 square root operation.", so perhaps this is a
> > safe assumption on a compliant system.
> >
> > Alright, Sanjay, go ahead. You're not actually changing anything. ;)
> >
> > Yeah, I would say that a system without a correctly rounded sqrtf is
> > broken. (But: Annex F is not required).
>
> True, but a user can always pass -fno-builtin-sqrt if this really bothers
> them.
>
>  -Hal
>
>
> --
> Hal Finkel
> Assistant Computational Scientist
> Leadership Computing Facility
> Argonne National Laboratory
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20141023/1ae78d5d/attachment.html>