https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63520
--- Comment #3 from Markus Trippelsdorf <trippels at gcc dot gnu.org> ---
Unfortunately during bisection another ICE pops up:
decNumber.i: In function ‘fn1’:
decNumber.i:9:9: internal compiler error: in expand_expr_real_1, at expr.c:9487
a -= 1;
^
0x1038feab expand_expr_real_1(tree_node*, rtx_def*, machine_mode,
expand_modifier, rtx_def**)
../../gcc/gcc/expr.c:9487
0x104b4743 expand_expr
../../gcc/gcc/expr.h:453
0x104b4743 ubsan_expand_si_overflow_addsub_check(tree_code,
gimple_statement_base*)
../../gcc/gcc/internal-fn.c:180
0x104b5883 expand_internal_call(gimple_statement_base*)
../../gcc/gcc/internal-fn.c:482
0x10277c9b expand_call_stmt
../../gcc/gcc/cfgexpand.c:2185
0x10277c9b expand_gimple_stmt_1
../../gcc/gcc/cfgexpand.c:3154
0x10277c9b expand_gimple_stmt
../../gcc/gcc/cfgexpand.c:3306
0x10279683 expand_gimple_basic_block
../../gcc/gcc/cfgexpand.c:5146
0x1027bb2b gimple_expand_cfg
../../gcc/gcc/cfgexpand.c:5712
0x1027bb2b execute
../../gcc/gcc/cfgexpand.c:5932
And because one cannot be sure if it hides the original ICE,
there is no exact bisection result.
The bug was introduced between r205684 (bad) to 205802 (good)
