[Bug target/119340] [14 regression] ICE when building gegl-0.4.52 on ppc64

[xry111 at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119340

--- Comment #7 from Xi Ruoyao <xry111 at gcc dot gnu.org> ---
Well, I think this is just PR116550.  Before LRA:

(jump_insn 930 383 1043 73 (parallel [
            (set (pc)
                (if_then_else (ne (reg:DI 592 [424])
                        (const_int 1 [0x1]))
                    (label_ref:DI 1081) 
                    (pc)))
            (set (reg:DI 592 [424])
                (plus:DI (reg:DI 592 [424])
                    (const_int -1 [0xffffffffffffffff])))
            (clobber (scratch:CC))
            (clobber (scratch:DI))
        ]) "pr119340.i":37:16 874 {bdnz_di}
     (int_list:REG_BR_PROB 941032196 (nil))
 -> 1081) 

Then LRA substitutes scratch with pesudos:

      Creating newreg=635
Removing SCRATCH to p635 in insn #930 (nop 3)
      Creating newreg=636
Removing SCRATCH to p636 in insn #930 (nop 4)
rescanning insn with uid = 930.

And then LRA tries to spill 635:

(jump_insn 930 383 1043 79 (parallel [
            (set (pc)
                (if_then_else (ne (reg:DI 97 ctr [424])
                        (const_int 1 [0x1]))
                    (label_ref:DI 1081)
                    (pc)))
            (set (reg:DI 97 ctr [424])
                (plus:DI (reg:DI 97 ctr [424])
                    (const_int -1 [0xffffffffffffffff])))
            (clobber (mem/c:CC (plus:DI (reg/f:DI 110 sfp)
                        (const_int 108 [0x6c])) [6 %sfp+108 S4 A32]))
            (clobber (scratch:DI))
        ]) "pr119340.i":37:16 874 {bdnz_di}
     (int_list:REG_BR_PROB 941032196 (nil))
 -> 1081)

And per PR116550 comment 4 doing so is not valid.

Thus the fix seems backporting r15-4406?