[Bug target/124454] [14/15/16 Regression] during RTL pass: sched2 ICE: in operator[], at vec.h:911 with custom flags since r14-4944

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124454

--- Comment #7 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
So yes REG_EH_REGION and the corresponding eh edge is being removed (inside
split3) but while calling find_many_sub_basic_blocks which calls
purge_dead_edges. And that leaves behind the unreachable block.

I am trying to figure if we should just call cleanup_cfg unconditionally if
something changed after the call to find_many_sub_basic_blocks or not.


But the real issue is inside copy_hardreg. We are just delaying the who might
catch it.

Where we change:
(insn 69 62 183 6 (set (reg:SI 1 x1 [orig:107 _12 ] [107])
        (zero_extend:SI (mem:QI (reg/f:DI 1 x1 [150]) [0 MEM[(char *)_4]+0 S1
A64]))) "/app/example.cpp":8:7 discrim 1 149 {*zero_extendqisi2_aarch64}
     (expr_list:REG_EH_REGION (const_int 2 [0x2])
        (nil)))

into:
(insn 69 62 183 6 (set (reg:SI 1 x1 [orig:107 _12 ] [107])
        (zero_extend:SI (mem:QI (plus:DI (reg/f:DI 31 sp)
                    (const_int 56 [0x38])) [0 MEM[(char *)_4]+0 S1 A64])))
"/app/example.cpp":8:7 discrim 1 149 {*zero_extendqisi2_aarch64}
     (expr_list:REG_EH_REGION (const_int 2 [0x2])
        (nil)))

And that no longer traps.