[Bug middle-end/105206] New: mis-optimization with -ffast-math and __builtin_powf

[Bug middle-end/105206] New: mis-optimization with -ffast-math and __builtin_powf

[kargl at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105206

            Bug ID: 105206
           Summary: mis-optimization with -ffast-math and __builtin_powf
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: kargl at gcc dot gnu.org
  Target Milestone: ---

[Bug middle-end/105206] mis-optimization with -ffast-math and __builtin_powf

[kargl at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105206

kargl at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |minor

--- Comment #1 from kargl at gcc dot gnu.org ---
Not sure if anyone cares.  I don't use -ffast-math, but this might considered a
mis-optimization with that option.

#include <math.h>

float
foof(float x)
{
   return (powf(10.f,x));
}

double
food(double x)
{
   return (pow(10.,x));
}


-fdump-tree-original shows

;; Function foof (null)
;; enabled by -tree-original

{
  return powf (1.0e+1, x);
}


;; Function food (null)
;; enabled by -tree-original

{
  return pow (1.0e+1, x);
}

Compiling to assembly shows

foof:
.LFB3:
        .cfi_startproc
        movaps  %xmm0, %xmm1
        movss   .LC0(%rip), %xmm0
        jmp     powf
        .cfi_endproc
food:
.LFB4:
        .cfi_startproc
        mulsd   .LC1(%rip), %xmm0
        jmp     exp
        .cfi_endproc

So, the middle-end is converting pow(10.x) to exp(x*log(10.0)) where log(10.0)
is reduced, but the same transformation of powf(10.f,x) still yields a call to
powf.

[Bug middle-end/105206] mis-optimization with -ffast-math and __builtin_powf

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105206

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rguenth at gcc dot gnu.org
           Keywords|                            |missed-optimization
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2022-04-11
             Status|UNCONFIRMED                 |WAITING

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Hm, it works for me on linux with glibc, can you be more specific as to what
OS/libc (version) you are using?

foof:
.LFB0:
        .cfi_startproc
        mulss   .LC0(%rip), %xmm0
        jmp     expf

food:
.LFB1:
        .cfi_startproc
        mulsd   .LC1(%rip), %xmm0
        jmp     exp

[Bug middle-end/105206] mis-optimization with -ffast-math and __builtin_powf

[kargl at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105206

--- Comment #3 from kargl at gcc dot gnu.org ---
(In reply to Richard Biener from comment #2)
> Hm, it works for me on linux with glibc, can you be more specific as to what
> OS/libc (version) you are using?
> 
> foof:
> .LFB0:
>         .cfi_startproc
>         mulss   .LC0(%rip), %xmm0
>         jmp     expf
> 
> food:
> .LFB1:
>         .cfi_startproc
>         mulsd   .LC1(%rip), %xmm0
>         jmp     exp

Richard, I'm using x86_64-*-freebsd.  I just checked with powl(), and got 
fool:
.LFB5:
        .cfi_startproc
        pushq   %rbp
        .cfi_def_cfa_offset 16
        .cfi_offset 6, -16
        movq    %rsp, %rbp
        .cfi_def_cfa_register 6
        subq    $16, %rsp
        flds    .LC0(%rip)
        pushq   24(%rbp)
        pushq   16(%rbp)
        fstpt   -16(%rbp)
        movq    -16(%rbp), %rax
        movl    -8(%rbp), %edx
        pushq   %rdx
        pushq   %rax
        call    powl
        addq    $32, %rsp
        leave
        .cfi_def_cfa 7, 8
        ret
        .cfi_endproc

with -ffast-math.

This might be related to PR89125.  The patch I attached to that PR has never
found its way into the repository.  I'll do some more testing later today.

[Bug middle-end/105206] mis-optimization with -ffast-math and __builtin_powf

[sgk at troutmask dot apl.washington.edu]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105206

--- Comment #4 from Steve Kargl <sgk at troutmask dot apl.washington.edu> ---
On Mon, Apr 11, 2022 at 02:45:22PM +0000, kargl at gcc dot gnu.org wrote:
> 
> This might be related to PR89125.  The patch I attached to that PR has never
> found its way into the repository.  I'll do some more testing later today.
> 

Yep.  This is related to PR89125.  GCC does not use configure
to check for the availability of individual C99 libm functions.
After updating the patch in PR89125 for the *.c to *.cc migration
the patch gcc gives

foof:
.LFB3:
        .cfi_startproc
        mulss   .LC0(%rip), %xmm0
        jmp     expf
        .cfi_endproc

food:
.LFB4:
        .cfi_startproc
        mulsd   .LC1(%rip), %xmm0
        jmp     exp
        .cfi_endproc

fool:
.LFB5:
        .cfi_startproc
        pushq   %rbp
        .cfi_def_cfa_offset 16
        .cfi_offset 6, -16
        movq    %rsp, %rbp
        .cfi_def_cfa_register 6
        fldt    16(%rbp)
        popq    %rbp
        .cfi_def_cfa 7, 8
        fldt    .LC2(%rip)
        fmulp   %st, %st(1)
        fldl2e
        fmulp   %st, %st(1)
        fld     %st(0)
        frndint
        fsubr   %st, %st(1)
        fxch    %st(1)
        f2xm1
        fld1
        faddp   %st, %st(1)
        fscale
        fstp    %st(1)
        ret
        .cfi_endproc

[Bug target/105206] mis-optimization with -ffast-math and __builtin_powf

[kargl at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105206

kargl at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|middle-end                  |target
         Depends on|                            |89125
             Target|                            |x86_64-*-freebsd

--- Comment #5 from kargl at gcc dot gnu.org ---
I've updated the Component to target and added the target triplet.


Referenced Bugs:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89125
[Bug 89125] Misoptimization of converting sin(x) and cos(x) into
sincos(x,&s,&c)

[Bug target/105206] mis-optimization with -ffast-math and __builtin_powf

[rguenther at suse dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105206

--- Comment #6 from rguenther at suse dot de <rguenther at suse dot de> ---
On Mon, 11 Apr 2022, kargl at gcc dot gnu.org wrote:

> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105206
> 
> --- Comment #3 from kargl at gcc dot gnu.org ---
> (In reply to Richard Biener from comment #2)
> > Hm, it works for me on linux with glibc, can you be more specific as to what
> > OS/libc (version) you are using?
> > 
> > foof:
> > .LFB0:
> >         .cfi_startproc
> >         mulss   .LC0(%rip), %xmm0
> >         jmp     expf
> > 
> > food:
> > .LFB1:
> >         .cfi_startproc
> >         mulsd   .LC1(%rip), %xmm0
> >         jmp     exp
> 
> Richard, I'm using x86_64-*-freebsd.  I just checked with powl(), and got 
> fool:
> .LFB5:
>         .cfi_startproc
>         pushq   %rbp
>         .cfi_def_cfa_offset 16
>         .cfi_offset 6, -16
>         movq    %rsp, %rbp
>         .cfi_def_cfa_register 6
>         subq    $16, %rsp
>         flds    .LC0(%rip)
>         pushq   24(%rbp)
>         pushq   16(%rbp)
>         fstpt   -16(%rbp)
>         movq    -16(%rbp), %rax
>         movl    -8(%rbp), %edx
>         pushq   %rdx
>         pushq   %rax
>         call    powl
>         addq    $32, %rsp
>         leave
>         .cfi_def_cfa 7, 8
>         ret
>         .cfi_endproc
> 
> with -ffast-math.
> 
> This might be related to PR89125.  The patch I attached to that PR has never
> found its way into the repository.  I'll do some more testing later today.

Ah, yeah - that old issue, indeed good to revisit that.

[Bug target/105206] mis-optimization with -ffast-math and __builtin_powf

[kargl at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105206

kargl at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |RESOLVED
         Resolution|---                         |DUPLICATE

--- Comment #7 from kargl at gcc dot gnu.org ---
(In reply to rguenther@suse.de from comment #6)
> On Mon, 11 Apr 2022, kargl at gcc dot gnu.org wrote:
>
> > 
> > This might be related to PR89125.  The patch I attached to that PR has never
> > found its way into the repository.  I'll do some more testing later today.
> 
> Ah, yeah - that old issue, indeed good to revisit that.

It's definitely a manifestation of PR89125.  I'll close this
as a duplicate.

*** This bug has been marked as a duplicate of bug 89125 ***

[Bug target/105206] mis-optimization with -ffast-math and __builtin_powf

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105206
Bug 105206 depends on bug 89125, which changed state.

Bug 89125 Summary: Misoptimization of converting sin(x) and cos(x) into sincos(x,&s,&c)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89125

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED