[Bug middle-end/117801] [15 regression] aarch64: 20% regression in TSVC s278 since r15-3509-gd34cda72098867

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117801

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|ra                          |
          Component|target                      |middle-end

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Trunk:
  vect__79.41_96 = .FMA (vect__ifc__144.33_89, vect_pretmp_54.10_137,
vect__ifc__147.37_21);
  MEM <vector(4) float> [(float *)&c + ivtmp.48_31 * 1] = vect__ifc__144.33_89;
  MEM <vector(4) float> [(float *)&b + ivtmp.48_31 * 1] = vect__ifc__147.37_21;
  MEM <vector(4) float> [(float *)&a + ivtmp.48_31 * 1] = vect__79.41_96;


GCC 14:
  MEM <vector(4) float> [(float *)&c + ivtmp.48_31 * 1] = vect__ifc__144.33_45;
  _3 = -vect__ifc__141.28_26;
  vect__71.36_92 = .FMA (vect_pretmp_55.16_69, vect_pretmp_54.13_140, _3);
  vect__ifc__147.37_93 = .VCOND_MASK (mask__100.31_89, vect__ifc__141.28_26,
vect__71.36_92);
  MEM <vector(4) float> [(float *)&b + ivtmp.48_31 * 1] = vect__ifc__147.37_93;
  vect__79.41_5 = .FMA (vect__ifc__144.33_45, vect_pretmp_54.13_140,
vect__ifc__147.37_93);
  MEM <vector(4) float> [(float *)&a + ivtmp.48_31 * 1] = vect__79.41_5;



So yes it is a scheduling issue because FMA is destructive.