https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116842
--- Comment #1 from ktkachov at gcc dot gnu.org ---
(In reply to ktkachov from comment #0)
> short a, b, c;
> unsigned d(unsigned, int e) { return e; }
> void f(bool g, short e[][3][3][3][3], unsigned h[][3][3], char i[][8],
> short j[][18][18][18], short k[][18][18][18], short l[][8][8][8][8]) {
> for (char m;;) {
> for (short n = 0; n < 8; n += 5)
> a = j[m][6][2][m];
> for (short o(l[m][m][m][m][m] / i[m][m] ?: e[m][m][4][m][2]); o; o = g)
> for (char p; p < (c && i[g]) + 7; p += 2)
> b = d(h[6][g][2], k[m][5][g][2] != m);
> }
> }
>
> at -O3 with aarch64 ICEs:
> 3 | void f(bool g, short e[][3][3][3][3], unsigned h[][3][3], char
> i[][8],
> | ^
> for SSA_NAME: vect__105.20_159 in statement:
> mask__106.21_163 = _153 != vect__105.20_159;
> during GIMPLE pass: vect
> func.ii:3:6: internal compiler error: verify_ssa failed
> 0x25220ab internal_error(char const*, ...)
> $SRC/gcc/diagnostic-global-context.cc:517
> 0x156a7e3 verify_ssa(bool, bool)
> $SRC/gcc/tree-ssa.cc:1203
> 0x118e61f execute_function_todo
> $SRC/gcc/passes.cc:2108
> 0x118ebef execute_todo
> $SRC/gcc/passes.cc:2155
> Please submit a full bug report, with preprocessed source (by using
> -freport-bug).
> Please include the complete backtrace with any bug report.
> See <https://gcc.gnu.org/bugs/> for instructions.
>
> Could be the various phiopt improvements landing recently
Or maybe not. The original testcase had some MAX_EXPRs that were appearing, but
the reduced one seems to be more vectoriser-related
