[Bug tree-optimization/66856] [6 Regression] ICE in compute_live_loop_exits, at tree-ssa-loop-manip.c:234

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66856

--- Comment #8 from Richard Biener <rguenth at gcc dot gnu.org> ---
Vectorizer generates

  <bb 16>:
  vect_cst__109 = { 3, 5, 3, 5, 3, 5, 3, 5 };
  vect_cst__110 = { 3, 5, 3, 5, 3, 5, 3, 5 };
  vect_cst__113 = {g_14(D), 1, g_14(D), 1, g_14(D), 1, g_14(D), 1};
  vect_cst__114 = {g_14(D), 1, g_14(D), 1, g_14(D), 1, g_14(D), 1};
  vect_cst__122 = { 5, 3, 5, 3, 5, 3, 5, 3 };
  vect_cst__123 = { 5, 3, 5, 3, 5, 3, 5, 3 };
  vect_cst__126 = {1, _21, 1, _21, 1, _21, 1, _21};
  vect_cst__127 = {1, _21, 1, _21, 1, _21, 1, _21};
  vect_cst__139 = {g_14(D), g_14(D), g_14(D), g_14(D), g_14(D), g_14(D),
g_14(D), g_14(D)};
  vect_cst__140 = {g_14(D), g_14(D), g_14(D), g_14(D), g_14(D), g_14(D),
g_14(D), g_14(D)};

  <bb 5>:
  # e_38 = PHI <&c(16), e_18(7)>
  # f_39 = PHI <f_4(D)(16), f_19(7)>
  # prephitmp_46 = PHI <_36(16), _7(7)>
  # vectp_f.19_104 = PHI <f_4(D)(16), vectp_f.19_105(7)>
  # vectp_f.25_117 = PHI <f_4(D)(16), vectp_f.25_118(7)>
  # vectp_f.32_132 = PHI <f_4(D)(16), vectp_f.32_133(7)>
  # vectp_c.41_148 = PHI <&c(16), vectp_c.41_149(7)>
  # ivtmp_151 = PHI <0(16), ivtmp_152(7)>
  _13 = *f_39;
  _25 = _13 >> 3;
...
  _21 = _20 >> 3;
  vect__30.30_128 = vect__29.29_124 & vect_cst__127;
  vect__30.30_129 = vect__29.29_125 & vect_cst__126;


where _21 is defined after its use.