Le 11/09/2012 15:25, Ian Lance Taylor a écrit : > On Tue, Sep 11, 2012 at 2:15 AM, Aurelien Buhrig > <aurelien.buhrig.gcc@xxxxxxxxx> wrote: >> >> >> >>> Hi, >>> >>> The gcc.c-torture/execute/nestfunc-3.c test fails on a private target. I >>> suspect this is because the STATIC_CHAIN_REGNUM, which is not a call >>> used register, is used by the caller function (main), modified by the >>> trampoline, but not restored upon nested function return. >>> >>> Should the STATIC_CHAIN_REGNUM be defined so that >>> CALL_USED_REGISTERS[STATIC_CHAIN_REGNUM] == 0? >> >> The question is rather: >> Should the STATIC_CHAIN_REGNUM be defined so that >> CALL_USED_REGISTERS[STATIC_CHAIN_REGNUM] == 1? Or is it possible to make >> nested functions work if CALL_USED_REGISTERS[STATIC_CHAIN_REGNUM] == 0? > > The static chain register is normally a call-used register or a fixed > register or a stack location. It may be possible to make it work with > a call-saved register, but clearly it is essential that the call to > the nested function restore the original value of the register before > returning. > > Ian > Thank you Ian for your reply. I think I will use last call-used register that was used for arg passing before. So nested functions will have less arg registers, but a better epilogue. Aurélien
