From: michael g. <mic...@gm...> - 2007-12-22 02:34:49
|
Ok, So I've been playing with sb-simd. Maybe its out of date, maybe not. The patches are against the x86 version, and I run x86-64 so doesn't matter for me. I started out on the task of bringing the MMX and SSE vector definitions over from the sb-simd patch, but I found native SSE support already for ordinary floating point.. presumably because x87 code is somewhat deprecated. Some of the code collides (not just contextually), and it would be much cleaner to add SIMD vector support on top of the code that is already there, I think. Though I'm still not entirely comfortable with the code. Is anyone familiar with these patches? From the SBCL developer's standpoint, what is needed to get primitive SIMD support rolled into the mainline code? I'm something of a math geek and I've got lots of code just begging for loop optimization. -M |