+- Dolphin, the GameCube and Wii emulator - Forums (https://forums.dolphin-emu.org)
+-- Forum: Dolphin Emulator Discussion and Support (https://forums.dolphin-emu.org/Forum-dolphin-emulator-discussion-and-support)
+--- Forum: Development Discussion (https://forums.dolphin-emu.org/Forum-development-discussion)
+--- Thread: Builds with texture decode Improvement SSSE3, SSE2 , SSE4 (/Thread-builds-with-texture-decode-improvement-ssse3-sse2-sse4)
Builds with texture decode Improvement SSSE3, SSE2 , SSE4 - IceD_X - 01-08-2011
Hi guys.
First of all congratulations for the magnificent dolphin community, and especially compliments to the devs for their continued and relentless work
I have doubts about the latest release builds ...
Especially the latest builds that involves an increase in decoding performance of the texture with implementations SSSE3 I4 I8, SSE2, SSE4.
My question is this ... These builds are already optimized in SSS3, SSE2, SSE4, or are "Regular builds?
Excuse my lack of competence in the field ...
PS: I tried the r6772-r6774 builds with Donkey Kong Returns, there is a performance improvement of about 4-5 fps compared to r5186, but there is an issue.
When I press the Home Wii button, the texts are unreadable ... you see black and red lines in the vicinity of the texts. The problem presents itself under DX9, and it is solved by activating OpenCL option.
Thank you in advance, and sorry for my bad English, google translator docet
RE: Builds with texture decode Improvement SSSE3, SSE2 , SSE4 - StripTheSoul - 01-09-2011
It's solved by activating OpenCL? That's odd. Do you happen to know which revision exactly introduced the issue in the home menu? If you could find it, that would help the devs get rid of the regressions.
And about your question: These optimizations are in the code and usable with every kinda build...you don't need a special optimized build, that's for something else. You can just get the revs from here: http://www.dolphin-emu.org/download.html
RE: Builds with texture decode Improvement SSSE3, SSE2 , SSE4 - IceD_X - 01-09-2011
Hi Strip!
Thanks alot for the explanation about the builds
For some reason unknown to me, the latest build does not have this problem. The Wii menu is perfect.
Besides, in Donkey Kong returns I noticed a slight improvement of about 1-2 fps with r6784.
Unfortunately, and I apologize if I go in OT, Donkey Kong Returns ranging from FPS Min 44 (Map 1-2 waterfall area) to Max FPS 59, FPS avarage 49-52.
I have used all possible optimizations, but the situation does not change. It can change only 2 fps + -
Mmmmm ... Do I have to overclock my CPU? or look for further improvements?
Thanks again Strip! And sorry to all for my OT...
RE: Builds with texture decode Improvement SSSE3, SSE2 , SSE4 - diegochiha - 01-09-2011
an overclock to 3.6 GHZ could give you full speed i guess, and besides with a powerfull cpu like yours you should overclock it
RE: Builds with texture decode Improvement SSSE3, SSE2 , SSE4 - Squall Leonhart - 01-09-2011
go back to catalyst 10.10 (10.4 is still the best though)
RE: Builds with texture decode Improvement SSSE3, SSE2 , SSE4 - NaturalViolence - 01-09-2011
When you download an "optimized build" the difference is it was COMPILED using newer sse instructions. That does not guarantee that the compiler will know how to use sse instructions for something like the texture decoder, in fact I can guarantee that it won't. What the developers are doing here is actually recoding the texture decoders to use sse2/sse3/ssse3.
RE: Builds with texture decode Improvement SSSE3, SSE2 , SSE4 - fredx21 - 01-09-2011
NV, I am not sure to follow you. There is #if _M_SSE >= 0x301 in the code which encapsulate SSE3-and-greater code implementation.
What do you mean by "That does not guarantee that the compiler will know how to use sse instructions for something like the texture decoder"? How could the compiler not know how to use instructions? To me, that would imply that the program would not link, so if it compiles, it uses the optimized instructions.
Fred
RE: Builds with texture decode Improvement SSSE3, SSE2 , SSE4 - Metzelmaennchen - 01-09-2011
If you are having an old compiler it is possible that he doesn't know the new calls.
So if the compiler is capable of building SSE code, it will do it. Therefore the Macros to check if the compiler is capable (SSE is available in different versions).
Afterwards the executable checks on runtime if the CPU has the corresponding feature and decides which code to run, the SSE or fallback version.
If you know that you wll execute the code on machines not having SSSE3 for example and do your own build, why having the code in there
So you can switch off this features for the compiler and it will not build them into the executable.
RE: Builds with texture decode Improvement SSSE3, SSE2 , SSE4 - fredx21 - 01-09-2011
If the software has auto-fallback, why not always compile using the latest SSE macro and stop bothering about it?
What is the SSE define used by the dolphin-emu.org SVN builds?
Fred