[Bug tree-optimization/115841] 521.wrf_r ICEs when building with -march=znver4 -Ofast -flto --param vect-partial-vector-usage=1

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115841

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
  <bb 8> [local count: 955630228]:
  # sfn11s_17 = PHI <sfn11s_156(69), 0.0(7)>
  # sfn12s_2 = PHI <sfn12s_133(69), 0.0(7)>
  # kr_4 = PHI <kr_118(69), 1(7)>
  # ivtmp_966 = PHI <ivtmp_965(69), 32(7)>
  _420 = kr_4 + -1;
  _205 = (long int) _420;
  fk_206 = MEM <float[0:D.7142]> [(float[0:D.7057] *)&A.170][_205];
  _171 = MEM <float[0:D.7166]> [(float[0:D.7061] *)&xl][_205];
  _266 = _171 * fk_206;
  fun_163 = _266 * 6.93149983882904052734375e-1;
  b11_164 = MEM <float[0:D.7145]> [(float[0:D.7067] *)&A.166][_205];
  b12_161 = MEM <float[0:D.7148]> [(float[0:D.7071] *)&A.167][_205];
  _162 = fun_163 * b11_164;
  sfn11s_156 = sfn11s_17 + _162;
  _152 = b12_161 * fun_163;
  sfn12s_133 = sfn12s_2 + _152;
  kr_118 = kr_4 + 1;
  ivtmp_965 = ivtmp_966 - 1;
  if (ivtmp_965 == 0)
    goto <bb 9>; [12.36%]
  else
    goto <bb 69>; [87.64%]

  <bb 69> [local count: 837518626]:
  goto <bb 8>; [100.00%]

  <bb 9> [local count: 118111601]:
  # sfn12s_990 = PHI <sfn12s_133(8)>
  # sfn11s_989 = PHI <sfn11s_156(8)>



is the loop vectorized where we somehow peel the prologue and the main
vect loop (which isn't masked) vectorized epilog has a broken
prologue (continuing the vect reduction with the vector defs) and that
epilogue is masked.