If the results of my playing around with gcc options is remotely applicable to SunCC, then specifying a definite target architecture should result in faster AND compacter code. At least for 32bit output (64bit output is already implying an AMD64 baseline). Targetting the generic architecture in gcc can easily blow up the code to three times what it'd be targetting a specific architecture (reference is Freetype here, 1.8MB vs. 650KB). Disregarding specific SIMD instructions, targetting a specific architecture should allow the use of other x86 instructions that otherwise wouldn't be used.
-mg > I can do a Native build of firefox " -xtarget =opteron -xarch=sse2a " > > But would it benefit the end user ? > > Is there any way to find out , except dissasembling the resulting > binary > if any major bits of graphic processing has been converted into SIMD > instructions ? > > Firefox is ofcourse one of the main applications run on a Sun Ray > based Info Kiosk. > And I need it as snappy as possible.
