[Bug rtl-optimization/42802] New: FMAs not exploited

[rguenth at gcc dot gnu dot org]

double bar(double x, double y)
{
  double tmp = 0.1234 * y;
  return ((x + tmp) * (x - tmp));
}

GCC should use multiply-add and multiply-sub when that is cheaper than
one multiplication and two additions.

With -mfma4 on x86_64 instead of

        vmulsd  .LC0(%rip), %xmm1, %xmm1
        vaddsd  %xmm1, %xmm0, %xmm2
        vsubsd  %xmm1, %xmm0, %xmm0
        vmulsd  %xmm0, %xmm2, %xmm0

it should generate

        vmovsd  .LC0(%rip), %xmm3
        vfmaddsd        %xmm0, %xmm3, %xmm1, %xmm2
        vfnmaddsd       %xmm0, %xmm3, %xmm1, %xmm0
        vmulsd  %xmm0, %xmm2, %xmm0

See also PR19988.

FMA opportunities of this kind should probably be detected during RTL
expansion, similar to widening multiplications.


-- 
           Summary: FMAs not exploited
           Product: gcc
           Version: 4.5.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: enhancement
          Priority: P3
         Component: rtl-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: rguenth at gcc dot gnu dot org
GCC target triplet: powerpc64-*-*, x86_64-*-*
OtherBugsDependingO 19988
             nThis:


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42802