https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88870
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
fast_dce gathers postorder and friends computing all_blocks before it
eventually does
if (global_changed)
{
/* Turn off the RUN_DCE flag to prevent recursive calls to
dce. */
int old_flag = df_clear_flags (DF_LR_RUN_DCE);
/* So something was deleted that requires a redo. Do it on
the cheap. */
delete_unmarked_insns ();
^^^
which alters the CFG, removing edges/blocks here:
#0 remove_edge (e=<edge 0x7ffff69d82d0 (3 -> 5)>)
at /space/rguenther/src/svn/gcc-8-branch/gcc/cfghooks.c:423
#1 0x0000000000a16458 in purge_dead_edges (
bb=<basic_block 0x7ffff68852d8 (3)>)
at /space/rguenther/src/svn/gcc-8-branch/gcc/cfgrtl.c:3080
#2 0x0000000000a0fcb9 in delete_insn_and_edges (insn=0x7ffff69d51f8)
at /space/rguenther/src/svn/gcc-8-branch/gcc/cfgrtl.c:233
#3 0x0000000001b8f314 in delete_unmarked_insns ()
at /space/rguenther/src/svn/gcc-8-branch/gcc/dce.c:636
where we'd eventually alos purge unreachable blocks.
but then we simply continue
bitmap_clear (marked);
bitmap_clear (processed);
bitmap_clear (redo_out);
/* We do not need to rescan any instructions. We only need
to redo the dataflow equations for the blocks that had a
change at the top of the block. Then we need to redo the
iteration. */
if (word_level)
df_analyze_problem (df_word_lr, all_blocks, postorder, n_blocks);
else
df_analyze_problem (df_lr, all_blocks, postorder, n_blocks);
and here things blow up because DF is still set up on the old CFG or
we are at least passing in old CFG state (all_blocks, postorder, n_blocks)
here.
