On Sunday 26 Feb 2006 03:02, cdr wrote:
on win32 with MinGW i've had pd crash with 'bad instruction' errors by compiling with -march=athlon or i686, running on a Celeron II 500 (Coppermine). it is praobly either a bug in the audio interface code, or the aggressive optimization flags from Hans that caused the bug that started this thead..
Indeed, optimisation is never flawless. Although I don't see why under any circumstances gcc would spit out instructions with -march=i686 that at Celeron II can't handle.
-march implies -mtune and other stuff so its unnecessary to specify
-march=i586 implies -mtune=i586, but in reality most people will be running processors that are more similar to the pentium3, so 'tuning' the code to something the pentium3 will pipeline better and will work better with its cache arrangement will give most people a slight performance boost while allowing i586 users to still run it. Which is why the original flags are probably a good compromise. Though I'm really not sure about O3.
This is offtopic and unproductive, so I'll quieten down.
robert