[Bug target/94308] New: [10 Regression] ICE in final_scan_insn_1 with vzeroupper

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94308

            Bug ID: 94308
           Summary: [10 Regression] ICE in final_scan_insn_1 with
                    vzeroupper
           Product: gcc
           Version: 10.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: jakub at gcc dot gnu.org
  Target Milestone: ---

/* { dg-do compile } */
/* { dg-options "-O2 -mfpmath=sse -mavx2 -mfma" } */

#include <x86intrin.h>

void
foo (float *x, const float *y, const float *z, unsigned int w)
{
  unsigned int a;
  const unsigned int b = w / 8;
  const float *c = y;
  const float *d = z;
  __m256 e = _mm256_setzero_ps ();
  __m256 f, g;
  for (a = 0; a < b; a++)
    {
      f = _mm256_loadu_ps (c);
      g = _mm256_loadu_ps (d);
      c += 8;
      d += 8;
      e = _mm256_fmadd_ps (f, g, e);
    }
  __attribute__ ((aligned (32))) float h[8];
  _mm256_storeu_ps (h, e);
  _mm256_zeroupper ();
  float i = h[0] + h[1] + h[2] + h[3] + h[4] + h[5] + h[6] + h[7];
  for (a = b * 8; a < w; a++)
    i += (*c++) * (*d++);
  *x = i;
}

ICEs on i686-linux or x86_64-linux with -m32.
The problem is that the vzeroupper pass in this case fills in sets for all
xmm0..xmm7 regs, but doesn't force rerecognition of the insn, so it is still
considered avx_vzeroupper_1, but the splitter doesn't trigger for it.