Would implementing SSE4/AVX optimizations speed up dolphin?

drhycodan · 07-31-2013, 05:30 AM #1

I noticed that Dolphin only requires SSE2, which every CPU has in the past 15 years. Why not add support for AVX optimizations?

haddockd · 07-31-2013, 05:53 AM #2

I am by no means an expert, but I did see similar thread over at PCSX2. Preliminary results were not promising (but far from conclusive):
http://forums.pcsx2.net/Thread-SSE4-vs-AVX

**NaturalViolence** · 07-31-2013, 06:37 AM #3

drhycodan Wrote:Why not add support for AVX optimizations?

Short Answer: Because it would be stupid and pointless. Those new instructions are completely useless for emulated a cpu that doesn't have any similar instructions.

While SSE2 is the minimum requirement for dolphin the program does in fact use newer SSE instructions when it detects that your cpu supports them (SSE3, SSSE3, SSE4).

dryhycodan Wrote:I noticed that Dolphin only requires SSE2, which every CPU has in the past 15 years.

The very first SSE2 capable cpus were launched 12.5 years ago. They didn't become common until 10 years ago.

lamedude · 07-31-2013, 07:32 PM #4

Any dev's willing to throw out a number on much the JIT engine could benefit from AVX's 3 operand encoding? How much time is wasted moving data around?

**delroth** · 08-01-2013, 12:06 AM #5

Very little.

**degasus** · 08-01-2013, 05:21 AM #6

Newer simd instructions like sse and avx aren't faster, they can only process more information at once. But as we emulate raw ppc instructions, there is no need for bigger simd than the original ppc one.