[Bug tree-optimization/91069] [10 Regression] Miscompare of 453.povray since r272843

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91069

--- Comment #11 from Jakub Jelinek  ---
Author: jakub
Date: Thu Jul  4 05:10:52 2019
New Revision: 273039

URL: https://gcc.gnu.org/viewcvs?rev=273039&root=gcc&view=rev
Log:
PR middle-end/91069
* gcc.dg/pr91069.c (v2df): Use 2 * sizeof (double) instead of
hardcoded 16 for better portability.
(v2di): Change from long vector to long long vector.  Use
2 * sizeof (long long) instead of hardcoded 16.

Modified:
trunk/gcc/testsuite/ChangeLog
trunk/gcc/testsuite/gcc.dg/pr91069.c

[Bug tree-optimization/91069] [10 Regression] Miscompare of 453.povray since r272843

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91069

--- Comment #10 from Richard Biener  ---
Author: rguenth
Date: Wed Jul  3 12:47:07 2019
New Revision: 273007

URL: https://gcc.gnu.org/viewcvs?rev=273007&root=gcc&view=rev
Log:
2019-07-03  Richard Biener  

PR middle-end/91069
* match.pd (vec_perm -> bit_insert): Fix element read from
first vector.

* gcc.dg/pr91069.c: New testcase.

Added:
trunk/gcc/testsuite/gcc.dg/pr91069.c
Modified:
trunk/gcc/ChangeLog
trunk/gcc/match.pd
trunk/gcc/testsuite/ChangeLog

[Bug tree-optimization/91069] [10 Regression] Miscompare of 453.povray since r272843

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91069

Richard Biener  changed:

   What|Removed |Added

 Status|ASSIGNED|RESOLVED
 Resolution|--- |FIXED

--- Comment #9 from Richard Biener  ---
Fixed.

[Bug tree-optimization/91069] [10 Regression] Miscompare of 453.povray since r272843

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91069

--- Comment #8 from Richard Biener  ---
It "mismatches" in

if (sel.series_p (1, 1, nelts + 1, 1))
  {
/* After canonicalizing the first elt to come from the
   first vector we only can insert the first elt from
   the first vector.  */
at = 0;
if ((ins = fold_read_from_vector (cop0, 0)))
  op0 = op1;
  }

it should read from element sel[0], not 0.  Testing patch.

[Bug tree-optimization/91069] [10 Regression] Miscompare of 453.povray since r272843

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91069

Richard Biener  changed:

   What|Removed |Added

 Status|NEW |ASSIGNED
   Assignee|unassigned at gcc dot gnu.org  |rguenth at gcc dot 
gnu.org

--- Comment #7 from Richard Biener  ---
Testcase:

typedef double v2df __attribute__((vector_size(16)));
typedef long v2di __attribute__((vector_size(16)));

void foo (v2df *res, v2df *src)
{
  v2df x = *src;
  *res = __builtin_shuffle ((v2df) { 1.0, 0.0 }, x, (v2di) { 1, 3 });
}

int main()
{
  v2df x = (v2df) { 0.0, 2.0 };
  foo (&x, &x);
  if (x[0] != 0.0 || x[1] != 2.0)
__builtin_abort ();
  return 0;
}

[Bug tree-optimization/91069] [10 Regression] Miscompare of 453.povray since r272843

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91069

--- Comment #6 from Richard Biener  ---
-march=x86-64 -mtune=generic (aka "default") is also broken, haswell (AVX2) is
fine.  -O3 is enough to trigger the issue for that.  Disabling either
of loop or SLP vectorization hides the issue.

Inlining can be tuned down for Compute_Rotation_Transform by not inlining
MTimesA, MTranspose (MATRIX *) and MTimesB.

Disabling the BIT_INSERT_EXPR generation from match.pd seems to fix it though.

[Bug tree-optimization/91069] [10 Regression] Miscompare of 453.povray since r272843

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91069

--- Comment #5 from Martin Liška  ---
Created attachment 46553
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=46553&action=edit
FRE4 dump file with -O3 for the problematic function

[Bug tree-optimization/91069] [10 Regression] Miscompare of 453.povray since r272843

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91069

--- Comment #4 from Martin Liška  ---
Created attachment 46552
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=46552&action=edit
FRE4 dump file with -Ofast

[Bug tree-optimization/91069] [10 Regression] Miscompare of 453.povray since r272843

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91069

--- Comment #3 from Martin Liška  ---
Created attachment 46551
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=46551&action=edit
pre-processed source file

[Bug tree-optimization/91069] [10 Regression] Miscompare of 453.povray since r272843

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91069

--- Comment #2 from Martin Liška  ---
And the problematic function in the file is:
void Compute_Rotation_Transform (TRANSFORM *transform, VECTOR vector)

[Bug tree-optimization/91069] [10 Regression] Miscompare of 453.povray since r272843

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91069

--- Comment #1 from Martin Liška  ---
So the problematic file is:

$ g++ -c -o matrices.o -DSPEC_CPU -DNDEBUG   -Ofast -march=native -g
-fpermissive   -DSPEC_CPU_LP64   matrices.cpp

[Bug tree-optimization/91069] [10 Regression] Miscompare of 453.povray since r272843

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91069

Martin Liška  changed:

   What|Removed |Added

 Status|UNCONFIRMED |NEW
   Last reconfirmed||2019-07-03
 CC||rguenth at gcc dot gnu.org
  Known to work||9.1.0
   Target Milestone|--- |10.0
 Ever confirmed|0   |1
  Known to fail||10.0