Dolphin, the GameCube and Wii emulator - Forums

Both of these have my patch applied which fixes a shader related bug.
3.0-96
3.0-96 /w Mylek's Xenoblade Patch 3.1

-----------------------------------------------
There seems to be a shortage of recent ICC builds, so I made some for myself and the posters on a different forum. Figured I would share the builds here, if people want them.

All of these are x64 builds compiled with /Ox, with inter-procedural optimization and optimized headers turned on. The SSE3 builds are compiled with /arch:SSE3 to maintain AMD processor compatibility, while the others use /Qx with the mentioned SSE version.

These builds have no special patches.
3.0-71 SSE3
3.0-71 SSE4.1
3.0-71 SSE4.2

These builds include version 3.1 of Mylek's Xenoblade HLE audio patch. If you are using LLE for sound, you should not use these.
3.0-71 SSE3
3.0-71 SSE4.1

I've done some rough benchmarking in Xenoblade by turning off the frame limiter and audio throttle to let the game speed run uncapped. The ICC builds are about eight to ten percent faster than a VC++ compiled build on my system. That is no guarantee that they will be equally fast on your system, of course.

Many ICC builds I've seen posted on this forum do not properly set the _M_SSE define to enable the hand-coded SSE optimizations in the VideoCommon library. The define is currently set automatically when making a VC++ build, but not when using ICC. I made this mistake myself at first. Without the hand-coded SSE optimizations, the ICC builds perform, at the very best, equal to a VC++ build. In some situations, they perform worse. Needless to say, my builds above have the optimizations included.

The difference in performance between the builds using different compiler generated SSE instruction sets in minimal in my testing. The SSE3 build performs just about as well as the SSE4.2 build on my 2600k. Regardless, I made separate versions for people to use. In the future, I may just use /arch:SSE3 /QaXSSE4.2 to generate a single binary with SSE3 as the baseline and alternate SSE4.2 codepaths.

You should tell lectrode how you managed to compile on ICC

How to "set the _M_SSE define to enable the hand-coded SSE optimizations in the VideoCommon library."

can you teach me

Checking this out and so far I likey like like. First build post 3.0 that has all my games running again on my comp. In the preliminary testings I have found an increase in speed for me in hyrule field and other parts on twilight princess. I have been comparing to the version listed in my SIG. I have noticed less fluctuating of fps on some other games I checked on. I've only played about 5 min on each popping to different save areas and whatnot.

I always loved it dirty, but I am feeling filthy tonight!

using 3.0-71 SSE4.2

btw, don't use that "r71" format, it's "3.0-71" ... this will get confusing as hell once we release 3.1 >_>

(09-02-2011, 09:54 AM)disraelie Wrote: [ -> ]How to "set the _M_SSE define to enable the hand-coded SSE optimizations in the VideoCommon library."

can you teach me

See the optional section in the Windows Build page of the wiki. The only addendum I would have to that is that it is only necessary for ICC builds. 0x402 will enable all hand-coded SSE optimizations using SSE4.2 or lower. The code path is chosen at runtime, so there is no danger in putting the hand-coded SSE4.2 optimizations in a build you are targeting for SSE3 processors, say. Might as well just include them all.

(09-02-2011, 03:28 PM)Trunk8 Wrote: [ -> ]I always loved it dirty, but I am feeling filthy tonight!

Glad to hear it is working well for you. The speed increase is definitely not imagined. I tested it by turning of the frame limiter and audio throttle, and loading a save state to the same spot in xenoblade different emulator builds. My VC++ reference build would stabilize around 47 FPS, while my ICC builds stabilize around 52 FPS. ICC builds I tried made by others were occasionally slower than the VC++ counterparts.

(09-02-2011, 06:52 PM)neobrain Wrote: [ -> ]btw, don't use that "r71" format, it's "3.0-71" ... this will get confusing as hell once we release 3.1 >_>

I've changed the thread title and links to reflect this.

(09-02-2011, 06:52 PM)neobrain Wrote: [ -> ]btw, don't use that "r71" format, it's "3.0-71" ... this will get confusing as hell once we release 3.1 >_>

There's going to be a 3.1?

There's going to be a 3.1 in the sense that a 3.0 was comming right after the 2.0 release... :/

(09-02-2011, 09:45 PM)neobrain Wrote: [ -> ]There's going to be a 3.1 in the sense that a 3.0 was comming right after the 2.0 release... :/

I take that as a no

Unpatched 3.0-77 build added. Seems to fix the minor graphical glitches with the railings at Alcamoth in Xenoblade.