Aurelien Buhrig <aurelien.buhrig.gcc@xxxxxxxxx> writes: > Is there a way to find which define_insn/define_expand/define_split is > responsible for this NULL mem ? Nothing other that debugging it. Using -da to dump all the RTL passes may help you identify where it first appears. You'll probably have to look closely, though, as a NULL rtx will most likely simply be omitted from the dump. Ian > 2011/6/18 Ian Lance Taylor <iant@xxxxxxxxxx>: >> Aurelien BUHRIG <aurelien.buhrig.gcc@xxxxxxxxx> writes: >> >>> #0 0x000000000050b9e5 in copy_to_mode_reg (mode=PSImode, x=0x0) at >>> <...>/src/gcc/explow.c:621 >> >> x can not be NULL there. >> >>> #1 0x00000000005206d3 in set_storage_via_libcall (object=0x8a15ec0, >>> size=0x88a1480, val=0x88a1400, tailcall=0 '\000') at >>> <...>/src/gcc/expr.c:2644 >> >> The line is >> object = copy_to_mode_reg (Pmode, XEXP (object, 0)); >> so it looks like XEXP (object, 0) is NULL, which is invalid. Here >> object is a MEM. So you need to find out how you got a MEM with a NULL >> address. >> >> Ian >>
