I wonder if there are still CPUs being used these days that don't have SSE, but do have MMX or that run the MMX codepath better than the SSE one...Is MMX still relevant?
Phoronix: Pixman Keeps On Pushing & Optimizing Pixels
Pixman 0.27.2 has been released and it offers up more features...
http://www.phoronix.com/vr.php?view=MTE1MjA
I wonder if there are still CPUs being used these days that don't have SSE, but do have MMX or that run the MMX codepath better than the SSE one...Is MMX still relevant?
The reason I was working on the MMX code paths were for Loongson and ARM/iwMMXt.
http://mattst88.com/blog/2012/05/17/...s_and_Results/
http://mattst88.com/blog/2012/07/06/...O_1,75_laptop/
Yes, the Xcore86 cpus. They only have MMX.
But they can't run pixman's MMX paths, because that's actually SSE-enhanced MMX, in layman terms![]()
Not a huge market share no, but they're used in thin clients / signage / point of sale / etc.
Btw, it's quite interesting to have these things happen, just like once the 486 patents expired 486 clones started appearing from non-intel-amd-via manufacturers. 586 and MMX are free to go now it seems, and we're seeing 1GHz low power cpus with those.
curaga, do you have an Xcore86 cpu yourself? It is just difficult to maintain pure legacy MMX code and make sure that it always works properly (no stray SSE-enhanced instructions accidentally mixed in) when there is no hardware available to run tests.
No, but several of Tiny Core users have such. It seems TC is well suited for that hw
They tend to be pretty cheap though, I don't know how pixman does donations, but a complete xcore86 system is often < 200$.
I should add that while Qemu is badly suited for that (it allows more advanced code even when you specify 586), Bochs is more strict.
So you could set up a Bochs VM to run your test suite, set to emulate a P5 MMX, and it should blow up if there's any SSE additions.