Re: [PATCH] Fix PR83089

[Christophe Lyon <christophe.lyon@linaro.org>]

On 23 November 2017 at 09:22, Richard Biener <rguenther@suse.de> wrote:
> On Wed, 22 Nov 2017, Christophe Lyon wrote:
>
>> On 22 November 2017 at 09:44, Richard Biener <rguenther@suse.de> wrote:
>> >
>> > The following fixes if-conversion to free SCEV/niter estimates because
>> > it DCEs stmts.
>> >
>> > Bootstrapped and tested on x86_64-unknown-linux-gnu, applied.
>> >
>> > Richard.
>> >
>> > 2017-11-22  Richard Biener  <rguenther@suse.de>
>> >
>> >         PR tree-optimization/83089
>> >         * tree-if-conv.c (pass_if_conversion::execute): If anything
>> >         changed reset SCEV and free the number of iteration estimates.
>> >
>> >         * gcc.dg/pr83089.c: New testcase.
>> >
>> > Index: gcc/tree-if-conv.c
>> > ===================================================================
>> > --- gcc/tree-if-conv.c  (revision 254990)
>> > +++ gcc/tree-if-conv.c  (working copy)
>> > @@ -2959,6 +2959,12 @@ pass_if_conversion::execute (function *f
>> >             && !loop->dont_vectorize))
>> >        todo |= tree_if_conversion (loop);
>> >
>> > +  if (todo)
>> > +    {
>> > +      free_numbers_of_iterations_estimates (fun);
>> > +      scev_reset ();
>> > +    }
>> > +
>> >    if (flag_checking)
>> >      {
>> >        basic_block bb;
>> > Index: gcc/testsuite/gcc.dg/pr83089.c
>> > ===================================================================
>> > --- gcc/testsuite/gcc.dg/pr83089.c      (nonexistent)
>> > +++ gcc/testsuite/gcc.dg/pr83089.c      (working copy)
>> > @@ -0,0 +1,27 @@
>> > +/* { dg-do compile } */
>> > +/* { dg-options "-O2 -ftree-loop-if-convert -ftree-parallelize-loops=2" } */
>> > +
>> > +int rl, s8;
>> > +
>> > +void
>> > +it (int zy, short int el)
>> > +{
>> > +  int hb;
>> > +
>> > +  while (el != 0)
>> > +    {
>> > +      hb = el;
>> > +      for (rl = 0; rl < 200; ++rl)
>> > +       {
>> > +         for (s8 = 0; s8 < 2; ++s8)
>> > +           {
>> > +           }
>> > +         if (s8 < 3)
>> > +           zy = hb;
>> > +         if (hb == 0)
>> > +           ++s8;
>> > +         zy += (s8 != -1);
>> > +       }
>> > +      el = zy;
>> > +    }
>> > +}
>>
>> I see the new testcase failing to compile on bare-metal targets
>> (aarch64-elf, arm-eabi):
>> spawn -ignore SIGHUP
>> /aci-gcc-fsf/builds/gcc-fsf-gccsrc/obj-aarch64-none-elf/gcc3/gcc/xgcc
>> -B/aci-gcc-fsf/builds/gcc-fsf-gccsrc/obj-aarch64-none-elf/gcc3/gcc/
>> /gcc/testsuite/gcc.dg/pr83089.c -fno-diagnostics-show-caret
>> -fdiagnostics-color=never -O2 -ftree-loop-if-convert
>> -ftree-parallelize-loops=2 -S -specs=aem-ve.specs -o pr83089.s
>> xgcc: error: unrecognized command line option '-pthread'
>> compiler exited with status 1
>> FAIL: gcc.dg/pr83089.c (test for excess errors)
>>
>> Where does that -pthread come from?
>
> From some specs processing I guess.  Didn't think that it could
> happen for a compile only testcase.
>

In fact, it's triggered by ftree-parallelize-loops, because in
gcc/gcc.c we have:
#define GOMP_SELF_SPECS \
  "%{fopenacc|fopenmp|%:gt(%{ftree-parallelize-loops=*:%*} 1): " \
  "-pthread}"

> Fixed with
>
> 2017-11-23  Richard Biener  <rguenther@suse.de>
>
>         PR tree-optimization/83089
>         * gcc.dg/pr83089.c: Require pthread.
>

Thanks,

Christophe

> Index: gcc/testsuite/gcc.dg/pr83089.c
> ===================================================================
> --- gcc/testsuite/gcc.dg/pr83089.c      (revision 255060)
> +++ gcc/testsuite/gcc.dg/pr83089.c      (working copy)
> @@ -1,4 +1,5 @@
>  /* { dg-do compile } */
> +/* { dg-require-effective-target pthread } */
>  /* { dg-options "-O2 -ftree-loop-if-convert -ftree-parallelize-loops=2" } */
>
>  int rl, s8;