On Tue, Apr 20, 2004 at 04:30:42PM -0400, Dan Sugalski wrote: : At 4:20 PM -0400 4/20/04, Aaron Sherman wrote: : >On Tue, 2004-04-20 at 11:53, Dan Sugalski wrote: : > : >> Y'know... let's just go all the way with this, since we're going to have : >> to. : >> : >> We'll add a hyper version of all the vtable entries. : > : >Another of those darned "I don't get it" posts, but I'll keep this one : >short. : > : >Why does Parrot need this? What's so special about hyper operations that : >makes Parrot want to take them on? : : Because they can be overridden separately from the regular version of : the operation.
Well, for the record I've never said that they can be, but... : Because if they're separate, people can do Really Evil Things with : them. (Like, say, custom matrix manipulation code with assembly : routines that use MMX, SSE, or Altivec functionality) : : Because the things being operated on may give the compilers : insufficient information to generate at compile time the proper array : iteration code. I don't know if these rate a bunch of vtable entries, though. If they're a little slower to dispatch it's no biggie, since even with hardware acceleration the actual vector ops will tend to dominate the timings. Larry
