[Bug target/84201] 549.fotonik3d_r from SPEC2017 fails verification with -mprefer-vector-width=256 or 512 on Zen

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84201

--- Comment #3 from Martin Liška  ---
If I'm correct it's also problematic on Haswell with -march=native.
Particularly the only affected file is power.fppized.o.
I also tried -mno-fma4 and -mno-fma and it does not help. From the assembly
diff, there is additional vectorization happening.

[Bug target/84201] 549.fotonik3d_r from SPEC2017 fails verification with -mprefer-vector-width=256 or 512 on Zen

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84201

Martin Liška  changed:

   What|Removed |Added

 Status|UNCONFIRMED |NEW
   Last reconfirmed||2018-09-09
 CC||marxin at gcc dot gnu.org
 Ever confirmed|0   |1
  Known to fail||6.4.0, 7.3.0, 8.2.0, 9.0

--- Comment #2 from Martin Liška  ---
Same happens on an Intel machine with -march=skylake. The issue is as old as
the skylake option (GCC 6.1+), bisection will not help us here. I can reproduce
that on a Haswell machine with the march option.

[Bug target/84201] 549.fotonik3d_r from SPEC2017 fails verification with -mprefer-vector-width=256 or 512 on Zen

[jamborm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84201

--- Comment #1 from Martin Jambor  ---
When benchmarking GCC 8 on an older Ivy Bridge Xeon, I also got 549.fotonik3d_r
verification error just with -Ofast -g -march=native -mtune=native