http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49423
--- Comment #26 from jules at gcc dot gnu.org ---
The testcase I previously had for this bug no longer reproduces since LRA was
enabled by default on ARM. So, it's possible the bug is now dormant, or indeed
fixed. The difference in the postreload dump around the insn which previously
failed (136) is, using -mno-lra/-mlra on current trunk:
-(insn 136 7 12 2 (set (reg:SI 11 fp [orig:231 D.6590 ] [231])
- (zero_extend:SI (mem/u/c:QI (symbol_ref/u:SI ("*.LC0") [flags 0x2]) [0
S1 A8]))) ../../aes/aeskey.c:509 182 {*arm_zero_extendqisi2_v6}
+(insn 952 7 136 2 (set (reg:SI 12 ip [1488])
+ (const_int 0 [0])) ../../aes/aeskey.c:509 666 {*arm_movsi_vfp}
+ (nil))
+(insn 136 952 12 2 (set (reg:SI 11 fp [orig:231 D.6590 ] [231])
+ (zero_extend:SI (reg:QI 12 ip [1488]))) ../../aes/aeskey.c:509 182
{*arm_zero_extendqisi2_v6}
Or, does anyone know of a testcase which still causes this ICE with -mlra
enabled? If not, we might be able to consider this fixed. (Insn 136 -- of the
form which causes breakage -- was generated by reload.)
Thanks,
Julian
