[Bug middle-end/67754] New: [graphite] Many testsuite failures

[Bug middle-end/67754] New: [graphite] Many testsuite failures

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67754

            Bug ID: 67754
           Summary: [graphite] Many testsuite failures
           Product: gcc
           Version: 6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: hjl.tools at gmail dot com
  Target Milestone: ---

On Linux/x86, r228220 gave

FAIL: gcc.dg/graphite/interchange-10.c execution test
FAIL: gcc.dg/graphite/interchange-11.c execution test
FAIL: gcc.dg/graphite/interchange-1.c execution test
FAIL: gcc.dg/graphite/interchange-3.c execution test
FAIL: gcc.dg/graphite/interchange-4.c execution test
FAIL: gcc.dg/graphite/interchange-7.c execution test
FAIL: gcc.dg/graphite/pr46185.c execution test
FAIL: gcc.dg/graphite/uns-block-1.c execution test
FAIL: gcc.dg/graphite/uns-interchange-12.c execution test
FAIL: gcc.dg/graphite/uns-interchange-14.c execution test
FAIL: gcc.dg/graphite/uns-interchange-15.c execution test
FAIL: gcc.dg/graphite/uns-interchange-9.c execution test
FAIL: gcc.dg/graphite/uns-interchange-mvt.c execution test
FAIL: gfortran.dg/graphite/block-1.f90   -O  (internal compiler error)
FAIL: gfortran.dg/graphite/block-1.f90   -O  (test for excess errors)

[Bug middle-end/67754] [graphite] Many testsuite failures

[spop at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67754

Sebastian Pop <spop at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |spop at gcc dot gnu.org

--- Comment #1 from Sebastian Pop <spop at gcc dot gnu.org> ---
All these tests are passing with ISL-0.15.

I also tried with ISL-0.14.1 and most of the tests are failing as reported,
with the exception of gfortran.dg/graphite/block-1.f90 that is passing on my
side.  This may be related to the fact that we have not built a bootstrap
compiler with ISL-0.14.1.

We will investigate first the wrong code for the interchange-* files, and then
we will try to reproduce the error in block-1.f90.

[Bug middle-end/67754] [graphite] Many testsuite failures

[spop at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67754

--- Comment #2 from Sebastian Pop <spop at gcc dot gnu.org> ---
We diffed the output of -fdump-tree-graphite-all for interchange-1.c and the
problem seems to be related to the ISL scheduler: I would highly recommend that
people stop using outdated ISL versions.

with ISL 0.15, the code generated looks like this:

ISL AST generated by ISL: 
{
  for (int c1 = 0; c1 <= 1334; c1 += 51)
    for (int c2 = 0; c2 <= 1334; c2 += 51)
      for (int c3 = c1; c3 <= min(1334, c1 + 50); c3 += 1)
        for (int c4 = c2; c4 <= min(1334, c2 + 50); c4 += 1)
          S_4(c3, c4);
  for (int c1 = 0; c1 <= 1334; c1 += 1)
    S_10(c1);
  S_8();
}

with ISL 0.14:

ISL AST generated by ISL: 
{
  for (int c0 = -1377; c0 < 0; c0 += 51)
    for (int c1 = -1326; c1 <= c0 + 70; c1 += 51)
      for (int c2 = max(max(-1334, c0), c1 - 20); c2 <= c0 + 50; c2 += 1)
        for (int c3 = max(-1314, c1); c3 <= min(c1 + 50, c2 + 20); c3 += 1)
          S_4(c2 - c3 + 20, -c3 + 20);
  for (int c1 = -1326; c1 < 0; c1 += 51) {
    for (int c3 = max(-1314, c1); c3 <= c1 + 50; c3 += 1) {
      S_4(-c3 + 20, -c3 + 20);
      S_10(-c3 + 20);
    }
    for (int c2 = 1; c2 <= min(50, c1 + 1364); c2 += 1)
      for (int c3 = max(c1, c2 - 1314); c3 <= c1 + 50; c3 += 1)
        S_4(c2 - c3 + 20, -c3 + 20);
  }
  S_4(20, 20);
  S_10(20);
  S_8();
  for (int c3 = 1; c3 <= 20; c3 += 1) {
    S_4(-c3 + 20, -c3 + 20);
    S_10(-c3 + 20);
  }
  for (int c2 = 1; c2 <= 50; c2 += 1)
    for (int c3 = 0; c3 <= 20; c3 += 1)
      S_4(c2 - c3 + 20, -c3 + 20);
  for (int c0 = 51; c0 <= 1334; c0 += 51)
    for (int c1 = c0 - 1326; c1 <= 20; c1 += 51)
      for (int c2 = c0; c2 <= min(min(1334, c0 + 50), c1 + 1364); c2 += 1)
        for (int c3 = max(c1, c2 - 1314); c3 <= min(20, c1 + 50); c3 += 1)
          S_4(c2 - c3 + 20, -c3 + 20);
}

[Bug middle-end/67754] [graphite] Many testsuite failures

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67754

--- Comment #3 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to Sebastian Pop from comment #2)
> We diffed the output of -fdump-tree-graphite-all for interchange-1.c and the
> problem seems to be related to the ISL scheduler: I would highly recommend
> that people stop using outdated ISL versions.
> 
> with ISL 0.15, the code generated looks like this:

Should we require ISL 0.15 for trunk?

[Bug middle-end/67754] [graphite] Many testsuite failures

[spop at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67754

--- Comment #4 from Sebastian Pop <spop at gcc dot gnu.org> ---
The release managers will decide the requirements and when lib versions will be
deprecated.

I am still investigating a fix to continue supporting ISL-0.14: we will add
code to configure.ac to detect ISL version number based on which we will
disable functionality that we know is broken for older ISL versions.

[Bug middle-end/67754] [graphite] Many testsuite failures

[spop at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67754

--- Comment #5 from Sebastian Pop <spop at gcc dot gnu.org> ---
Author: spop
Date: Tue Sep 29 22:15:08 2015
New Revision: 228268

URL: https://gcc.gnu.org/viewcvs?rev=228268&root=gcc&view=rev
Log:
use MIN fusion for ISL-14

This patch fixes PR66754 by reverting an earlier unintended change.
We now generate a much simpler AST for interchange-1.c:

ISL AST generated by ISL:
{
  for (int c1 = 0; c1 <= 1334; c1 += 1) {
    S_7(c1);
    for (int c3 = 0; c3 <= 1334; c3 += 1)
      S_4(c1, c3);
    S_5(c1);
  }
  for (int c1 = 0; c1 <= 1334; c1 += 1)
    S_10(c1);
  S_8();
}

Bootstrap and check pass on x86_64-linux with isl-0.14.1

          PR tree-optimization/67754
          * graphite-optimize-isl.c (optimize_isl): Call
          isl_options_set_schedule_fuse with ISL_SCHEDULE_FUSE_MIN for ISL-14.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/graphite-optimize-isl.c

[Bug middle-end/67754] [graphite] Many testsuite failures

[spop at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67754

Sebastian Pop <spop at gcc dot gnu.org> changed:

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

--- Comment #6 from Sebastian Pop <spop at gcc dot gnu.org> ---
Fixed.

[Bug middle-end/67754] [graphite] Many testsuite failures

[dominiq at lps dot ens.fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67754

--- Comment #7 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
Note that the failures

FAIL: gfortran.dg/graphite/block-1.f90   -O  (internal compiler error)
FAIL: gfortran.dg/graphite/block-1.f90   -O  (test for excess errors)

with -m32 are still there with isl-0.15 and r228270:

[Book15] f90/bug% gfc
/opt/gcc/_clean/gcc/testsuite/gfortran.dg/graphite/block-1.f90 -O2
-floop-nest-optimize -m32
/opt/gcc/_clean/gcc/testsuite/gfortran.dg/graphite/block-1.f90:1:0:

 subroutine matrix_multiply(a,b,c,n)
^
internal compiler error: in extract_affine_chrec, at
graphite-sese-to-poly.c:605

[Bug middle-end/67754] [graphite] Many testsuite failures

[spop at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67754

Sebastian Pop <spop at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |ASSIGNED
   Last reconfirmed|                            |2015-09-29
         Resolution|FIXED                       |---
     Ever confirmed|0                           |1

--- Comment #8 from Sebastian Pop <spop at gcc dot gnu.org> ---
As I do not see the errors in my builds, I will investigate.
Thanks for letting me know that this testcase is still failing.

[Bug middle-end/67754] [graphite] Many testsuite failures

[dominiq at lps dot ens.fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67754

--- Comment #9 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
> As I do not see the errors in my builds, I will investigate.
> Thanks for letting me know that this testcase is still failing.

Are you testing with -m32?

[Bug middle-end/67754] [graphite] Many testsuite failures

[spop at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67754

--- Comment #10 from Sebastian Pop <spop at gcc dot gnu.org> ---
Yes, with -m32 it fails with an ICE.  Thanks for pointing this out.

[Bug middle-end/67754] [graphite] Many testsuite failures

[spop at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67754

--- Comment #11 from Sebastian Pop <spop at gcc dot gnu.org> ---
Author: spop
Date: Thu Oct  1 15:17:51 2015
New Revision: 228346

URL: https://gcc.gnu.org/viewcvs?rev=228346&root=gcc&view=rev
Log:
add recursion on the inner loops

We now check that all data references in the current loop and inner loops
contained within loop are valid in an outer region before declaring that the
outer loop is a valid scop.

2015-09-30  Sebastian Pop  <s.pop@samsung.com>
            Aditya Kumar  <aditya.k7@samsung.com>

        PR tree-optimization/67754
        * graphite-scop-detection.c (loop_body_is_valid_scop): Add missing
        recursion on the inner loops.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/graphite-scop-detection.c

[Bug middle-end/67754] [graphite] Many testsuite failures

[spop at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67754

--- Comment #12 from Sebastian Pop <spop at gcc dot gnu.org> ---
Author: spop
Date: Thu Oct  1 15:17:58 2015
New Revision: 228347

URL: https://gcc.gnu.org/viewcvs?rev=228347&root=gcc&view=rev
Log:
call scev analysis in scop-detection as in sese-to-poly

Before our rewrite of the scop detection, we used to not have a valid SESE
region under hand, and so we used to do more ad-hoc analysis of data references
by trying to prove that at all levels of a loop nest the data references would
be still valid.

Now that we have a valid SESE region, we can call the scev analysis in the same
way on the same computed loop nest in the scop-detection as in the
sese-to-poly.

Next step will be to cache the data references analyzed in the scop detection
and not compute the same info in sese-to-poly.

The patch fixes block-1.f90 that used to ICE on x86_64-linux when compiled with
-m32.  Patch passed bootstrap with BOOT_CFLAGS="-g -O2 -fgraphite-identity
-floop-nest-optimize" and check on x86_64-linux using ISL-0.15.

2015-09-28  Sebastian Pop  <s.pop@samsung.com>
        Aditya Kumar  <aditya.k7@samsung.com>

        PR tree-optimization/67754
        * graphite-scop-detection.c (stmt_has_simple_data_refs_p): Call
        scev analysis on the same loop nest as analyze_drs_in_stmts.
        * graphite-sese-to-poly.c (outermost_loop_in_sese_1): Moved and
renamed...
        (try_generate_gimple_bb): Call outermost_loop_in_sese.
        (analyze_drs_in_stmts): Same.
        * sese.c (outermost_loop_in_sese): ...here.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/graphite-scop-detection.c
    trunk/gcc/graphite-sese-to-poly.c
    trunk/gcc/sese.c

[Bug middle-end/67754] [graphite] Many testsuite failures

[spop at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67754

Sebastian Pop <spop at gcc dot gnu.org> changed:

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

--- Comment #13 from Sebastian Pop <spop at gcc dot gnu.org> ---
Fixed.

[Bug middle-end/67754] [graphite] Many testsuite failures

[dominiq at lps dot ens.fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67754

--- Comment #14 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
>From https://gcc.gnu.org/ml/gcc-testresults/2015-10/msg00092.html

FAIL: gfortran.dg/graphite/block-1.f90   -O   scan-tree-dump-times graphite
"number of SCoPs: 1" 1
FAIL: gfortran.dg/graphite/block-2.f   -O   scan-tree-dump-times graphite
"number of SCoPs: 2" 1

[Bug middle-end/67754] [graphite] Many testsuite failures

[spop at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67754

Sebastian Pop <spop at gcc dot gnu.org> changed:

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

--- Comment #15 from Sebastian Pop <spop at gcc dot gnu.org> ---
Thanks for letting me know, I will investigate more why this still fails.

[Bug middle-end/67754] [graphite] Many testsuite failures

[spop at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67754

Sebastian Pop <spop at gcc dot gnu.org> changed:

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

--- Comment #16 from Sebastian Pop <spop at gcc dot gnu.org> ---
Fixed in r228402