Loop unrolling-related Bench++ regressions?

Loop unrolling-related Bench++ regressions?

[Roger Sayle]

Richard Henderson asked for test cases where loop unrolling fails
where it succeeded before.  A good example comes from test L000003
in the bench++ benchmark suite, which has recently regressed.  My
timings show that its took 0.626 nanoseconds a week or so ago, but
now takes 2.88 nanoseconds [over 4.5 times slower] on 1GHz pentium
III running RedHat linux v7.2.

The core of the test can be reduced to the following code:

const int feature_times = 25;

extern void foo(void);

void
test (void)
{
   int i = 1;

   for (;;)
     {
       foo ();
       i++;
       if (i > feature_times)
         break;
     }
}


with the system compiler, gcc v2.96, the above loop produces the following
code with "-O3 -funroll-all-loops -fomit-frame-pointer -S"

test:
	pushl	%ebx
	subl	$8, %esp
	movl	$24, %ebx
	.p2align 2
.L3:
	call	foo
	call	foo
	call	foo
	call	foo
	call	foo
	subl	$5, %ebx
	jns	.L3
	addl	$8, %esp
	popl	%ebx
	ret


but with the same command line options and the current mainline, v3.1
the following code is generated:

test:
	pushl	%ebx
	movl	$1, %ebx
	subl	$8, %esp
.L2:
	call	foo
	leal	1(%ebx), %edx
	cmpl	$25, %edx
	jg	.L6
	call	foo
	leal	2(%ebx), %edx
	cmpl	$25, %edx
	jg	.L6
	call	foo

	...

	addl	$8, %ebx
	call	foo
	cmpl	$25, %ebx
	jle	.L2
.L6:
	addl	$8, %esp
	popl	%ebx
	ret


So although the loop is being unrolled, its actually unrolled eight times
in the second version compared to just five in the first, the compiler is
unable to determine that the loop iterates a fixed number of times, and
inserts termination checks between each call to foo.


I hope this helps.  I was actually investigating the 45% performance
improvement on some bench++ tests from re-enabling g++ builtins, and
happened to stumble across this performance regression by accident.

Roger
--
Roger Sayle,                         E-mail: roger@eyesopen.com
OpenEye Scientific Software,         WWW: http://www.eyesopen.com/
Suite 1107, 3600 Cerrillos Road,     Tel: (+1) 505-473-7385
Santa Fe, New Mexico, 87507.         Fax: (+1) 505-473-0833

Re: Loop unrolling-related Bench++ regressions?

[Richard Henderson]

On Tue, Feb 05, 2002 at 06:46:43PM -0700, Roger Sayle wrote:
> I hope this helps.  I was actually investigating the 45% performance
> improvement on some bench++ tests from re-enabling g++ builtins, and
> happened to stumble across this performance regression by accident.

Yes, this is exactly the sort of example I was looking for.  I'll
not be able to examine it right away, but it's on my to-do list.


r~

Re: Loop unrolling-related Bench++ regressions?

[Paolo Carlini]

Roger Sayle wrote:

>Richard Henderson asked for test cases where loop unrolling fails
>where it succeeded before.  A good example comes from test L000003
>in the bench++ benchmark suite, which has recently regressed.  My
>timings show that its took 0.626 nanoseconds a week or so ago, but
>now takes 2.88 nanoseconds [over 4.5 times slower] on 1GHz pentium
>III running RedHat linux v7.2.
>
Thanks Roger!!!

Ciao, Paolo.

Re: Loop unrolling-related Bench++ regressions?

[Roger Sayle]

On Tue, 5 Feb 2002, Richard Henderson wrote:
> Yes, this is exactly the sort of example I was looking for.  I'll
> not be able to examine it right away, but it's on my to-do list.

With the recent discussions of 3.1 performance regressions, I thought
I'd check whether the loop unrolling regressions that I pointed out
in http://gcc.gnu.org/ml/gcc/2002-02/msg00307.html had been fixed.
Unfortunately, it looks like its slipped through the cracks, despite
having a very simple testcase.

Roger
--
Roger Sayle,                         E-mail: roger@eyesopen.com
OpenEye Scientific Software,         WWW: http://www.eyesopen.com/
Suite 1107, 3600 Cerrillos Road,     Tel: (+1) 505-473-7385
Santa Fe, New Mexico, 87507.         Fax: (+1) 505-473-0833

Re: Loop unrolling-related Bench++ regressions?

[Gerald Pfeifer]

On Sun, 21 Apr 2002, Roger Sayle wrote:
> With the recent discussions of 3.1 performance regressions, I thought
> I'd check whether the loop unrolling regressions that I pointed out
> in http://gcc.gnu.org/ml/gcc/2002-02/msg00307.html had been fixed.
> Unfortunately, it looks like its slipped through the cracks, despite
> having a very simple testcase.

In case you haven't filed a PR yet, please do so.

(In general, whenever a regression or other problem is not solved within
one, two weeks, one should always file a PR to avoid that important bugs
are missed or dealt with only very late in the release cycle.)

Gerald
-- 
Gerald "Jerry" pfeifer@dbai.tuwien.ac.at http://www.dbai.tuwien.ac.at/~pfeifer/

Re: Loop unrolling-related Bench++ regressions?

[Roger Sayle]

> On Sun, 21 Apr 2002, Roger Sayle wrote:
> > With the recent discussions of 3.1 performance regressions, I thought
> > I'd check whether the loop unrolling regressions that I pointed out
> > in http://gcc.gnu.org/ml/gcc/2002-02/msg00307.html had been fixed.
> > Unfortunately, it looks like its slipped through the cracks, despite
> > having a very simple testcase.
>
> In case you haven't filed a PR yet, please do so.

I've just done this.  Its PR opt/6405.

> (In general, whenever a regression or other problem is not solved within
> one, two weeks, one should always file a PR to avoid that important bugs
> are missed or dealt with only very late in the release cycle.)

Its true that I've been remiss about following up on bugs/PRs.  I've
just noticed whilst filing the above that atleast one of the PRs I've
fixed for 3.1 is still open.

Roger
--
Roger Sayle,                         E-mail: roger@eyesopen.com
OpenEye Scientific Software,         WWW: http://www.eyesopen.com/
Suite 1107, 3600 Cerrillos Road,     Tel: (+1) 505-473-7385
Santa Fe, New Mexico, 87507.         Fax: (+1) 505-473-0833