Le 05/05/2021 à 18:43, Laurent Vivier a écrit : > Le 05/05/2021 à 18:03, Philippe Mathieu-Daudé a écrit : >> The ABCD / SBCD memory opcodes (introduced in commit fb5543d8200) >> don't use their "addr" variable. >> >> Remove the unused variable and pass a NULL argument instead to >> gen_ea_mode(). This fixes warnings generated when building with >> CFLAGS=-O3 (using GCC 10.2.1 20201125): >> >> target/m68k/translate.c: In function ‘disas_sbcd_mem’: >> target/m68k/translate.c:897:13: warning: ‘addr’ may be used uninitialized >> in this function [-Wmaybe-uninitialized] >> 897 | delay_set_areg(s, reg0, tmp, false); >> | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ >> target/m68k/translate.c:1866:21: note: ‘addr’ was declared here >> 1866 | TCGv src, dest, addr; >> | ^~~~ >> >> target/m68k/translate.c: In function ‘disas_abcd_mem’: >> target/m68k/translate.c:897:13: warning: ‘addr’ may be used uninitialized >> in this function [-Wmaybe-uninitialized] >> 897 | delay_set_areg(s, reg0, tmp, false); >> | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ >> target/m68k/translate.c:1829:21: note: ‘addr’ was declared here >> 1829 | TCGv src, dest, addr; >> | ^~~~ >> > > It's really strange because ABCD and SBCD support indirect predecrement (mode > 4, "-(Ay),-(Ax)"), and > if you look into gen_ea_mode() &addr (addrp) is used with mode 4, it is > initialized on EA_LOADU to > be reused on EA_STORE. > > The bug is somewhere else... >
I think I see what is the problem: as the mode is indirect pre-decrement, the register doesn't need to be updated and thus the addr is not needed. But if we replace addrp by NULL, gen_lea_mode() will be called twice and the register will be decremented twice (on load and store, rather than only on load). Thanks, Laurent
