On Wed, 2012-01-18 at 22:25 +0100, Uros Bizjak wrote: > On Wed, Jan 18, 2012 at 10:16 PM, Patrick Marlier > <patrick.marl...@gmail.com> wrote: > > >> 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... > > > > > > So in order to keep them in sync, this should be also applied to libitm if > > your solution is chosen (At least to avoid confusion). > > If the Intel TM-ABI (no idea what's the status of this specification) > > specifies variadic function and regparm, it should be changed too. > > > > Index: libitm.h > > =================================================================== > > --- libitm.h (revision 183273) > > +++ libitm.h (working copy) > > @@ -136,7 +136,7 @@ typedef uint64_t _ITM_transactionId_t; /* Transact > > > > extern _ITM_transactionId_t _ITM_getTransactionId(void) ITM_REGPARM; > > > > -extern uint32_t _ITM_beginTransaction(uint32_t, ...) ITM_REGPARM; > > +extern uint32_t _ITM_beginTransaction(uint32_t, ...); > > > > extern void _ITM_abortTransaction(_ITM_abortReason) ITM_REGPARM > > ITM_NORETURN; > > The spec does say that all function should be regparm(2), but I agree > that the above is less confusing. The attribute is ignored, but > perhaps a comment would clear this confusion even more.
Uros, thanks for spotting the vararg issue. This looks okay to me, but Richard Henderson will have to OK this. If regparm(2) cannot work with variadic functions on x86, then I'd prefer removing the regparm. beginTransaction was switched to being variadic to allow communicating which kinds of versions a compiler has generated for the transaction's code (besides the default instrumentation that we have right now). I'd believe Ulrich Drepper's experience that making this variadic is better than restricting this to 64b (minus 10 bits or so already in use). BTW, would regparm(2) optimize on any arch/platform besides 32b x86? What about x32? Note that if we remove the regparm, we should also remove it on the other functions associated with txn begin (GTM_beginTransaction etc.). Torvald
