[Bug rtl-optimization/108463] [13 Regression] ICE: in cselib_subst_to_values, at cselib.cc:2148 with -O2 -fsched2-use-superblocks -g

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108463

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
So, seems during the code added by above patch without the 0 && new_cselib_val
creates a value on DEBUG_INSNs for:
1) (mem:SI (plus:DI (reg/f:DI 7 sp)
        (const_int NN)) [1  S4 A64])
   for NN 76 to 132 inclusive in steps of 4, except for 104 and 120
2) (plus:DI (reg/f:DI 7 sp) (const_int NN))
   for NN 72 to 132 inclusive in steps of 4, and for 200 and 264
3) (symbol_ref:DI ("a") [flags 0x2]  <var_decl 0x7fffea13ad80 a>)
4) complex expressions like:
(plus:DI (plus:DI (ashift:DI (zero_extend:DI (and:SI (mem:SI (plus:DI (reg/f:DI
7 sp)
                            (const_int 132 [0x84])) [1  S4 A32])
                    (const_int 15 [0xf])))
            (const_int 2 [0x2]))
        (reg/f:DI 7 sp))
    (const_int 8 [0x8]))

Now, I strongly doubt anything looks up for normal insns the large expressions
in 4) category, because they aren't valid in normal insns, so it will be 1), 2)
or 3) that affect the scheduling.