Use faster CheckGatherPipe from JIT. #1988
Conversation
We don't need to profile explicit calls to CheckGatherPipe: we aleady know where they are. Skip profiling, and explciitly insert an exception check. While I'm here, get rid of unused code in JitIL.
|
This looks similar to this: #1624 , it might be useful to look at some of it. I encountered some tricky bits doing this though (it felt /very/ easy to break things), and I wasn't able to measure a huge perf gain, but maybe you can do better. |
|
Mmm, yes, #1624 is pretty similar, although this is a somewhat less aggressive. I was looking into this because CompileExceptionCheck() was showing up on profiles on dynamic-bat... as the most frequent caller of HostRead_U32, which was rather confusing. |
|
Yeah, that was basically the same reason I looked into it, even. A lot of this needs cleaning up, even if the cleanup isn't as aggressive as what I tried there. |
|
2-3 fps gain in ZWW intro on Armv7, thanks a lot Edit : 2-3 fps gain in Mario Sunshine, 2 fps in game Zelda Wind Waker 2-3 fps in Luigi's Manson, make the game really playable (20-30) fps. 3 fps in Mario Party 4. 5 fps in MKDD in Championship. Really good fix ;) |
|
Ping. |
|
Definitely faster in Master in every game I tried. I did check a few games more thoroughly though. It was a good 10 - 15% faster in spots of Wind Waker, where as Melee tends to be around 3 - 5% faster depending on stage. |
Use faster CheckGatherPipe from JIT.
We don't need to profile explicit calls to CheckGatherPipe: we aleady know
where they are. Skip profiling, and explicitly insert an exception check.
While I'm here, get rid of unused code in JitIL.
Haven't benchmarked yet; I'm hoping this is a little faster than master.