On Thu, Oct 20, 2005 at 07:02:45PM +0200, Kurt Roeckx wrote: > > BTW: Upstream does not define the RC4_CHAR flag for this architecture. > > It also does not define the flag for ia64 where we do define it. We > > might have the same problem with ia64. > > Right, so it's a good idea to drop it in the future. > > The IA64 version atleast seems to have support for using > different sizes of the code, unfortuantly, it does: > #define SZ 4 // this is set to sizeof(RC4_INT) > > Instead of just setting it to the sizeof() itself. > > So this might require changing too. I'll try and test this on > the ia64 porter machine later.
ia64 actually seems to be using the proper SZ of 1 looking at the generated assembler. From the Makefile: rc4-ia64.s: asm/rc4-ia64.S @case `awk '/^#define RC4_INT/{print$$NF}' $(TOP)/include/openssl/opensslconf.h` in \ int) set -x; $(CC) $(CFLAGS) -DSZ=4 -E asm/rc4-ia64.S > $@ ;; \ char) set -x; $(CC) $(CFLAGS) -DSZ=1 -E asm/rc4-ia64.S > $@ ;; \ *) exit 1 ;; \ esac It also seems to be passing the regression tests, so it looks like it's working proplery. But we should consider dropping the RC4_CHAR when the soname changes again, should give 40% increase in performance. Kurt -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]