On Wed, Jan 18, 2012 at 8:44 PM, Patrick Marlier <patrick.marl...@gmail.com> wrote:
>> There was some discussion on where to pass arguments to regparm >> decorated vararg functions. Well, as the ABI is pretty clear - regparm >> should be ignored in this case, so all function arguments have to be >> passed in memory, even if that means that the value is kicked to the >> memory before the call, and pulled back into the register in >> _ITM_beginTransaction. > My main concern here is performance... Indeed, in case of libitm using > Hardware Transactional Memory, it could be great to use registers for > parameters. I would prefer to remove the variadic function as proposed here: > http://gcc.gnu.org/ml/gcc-patches/2011-12/msg01784.html Please note that all recent x86 processors implement store forwarding, so passing arguments through memory is mostly a non-issue nowadays. > As Torvald wrote, it was in case for hypothetical future parameters. So I > would agree to do: > extern uint32_t _ITM_beginTransaction(uint32_t,uint32_t) ITM_REGPARM; > > At least, it provides a new parameter for future use and do not use the > stack for parameters. > > Other thoughts? IMO, whatever the future decision would be, we shouldn't leave one part of the compiler out-of-sync from the other. Proposed patch fixes _current_ situation, where in the future, it is expected that compiler and library changes in sync... Uros.
