DISCLAIMER: this post does not contain any assertions about quality!
I was about to do a little listening test determine, if the SSE optimized version sounded the same as the regular one (for the upcoming 64k multi format test). I couldn't abx a first sample so I went the wave subtraction route, to establish if there was an actual difference.
To my surprise, large portions of the difference file were zero! The SSE version produced bit identical output for the most part (rough estimate: 80-90% identical). But there were also some short blips which were quite audible, meaning the difference was not just in the last bit.
So could it be, that the aotuv tunings were not the same in the versions I tested?
I tested:
Archer RC4, md5: 362b64e40395e670fc5052dc89af383d
Ogg + aotuv b3 from rarewares, md5: 922fed5040ebe839524007b5938f01eb
Full Version: Q: Aotuv b3 regular vs. archer rc4 SSE
Checksums are the same?
QUOTE(imbirek @ Mar 27 2005, 01:51 PM)
Checksums are the same?
Thx, corrected
If I understood it correctly this is exactly the way it is ment to be. Archer is nothing but a patch for AoTuV replacing some C routines with optimized ones written in ASM using SSE instructions.
Personally I doubt that these smalls blips are audible.
Personally I doubt that these smalls blips are audible.
QUOTE(dev0 @ Mar 27 2005, 02:04 PM)
If I understood it correctly this is exactly the way it is ment to be. Archer is nothing but a patch for AoTuV replacing some C routines with optimized ones written in ASM using SSE instructions.
Personally I doubt that these smalls blips are audible.
Personally I doubt that these smalls blips are audible.
From what I understood, people were afraid of using the SSE version because they feared different output.
I have difficulty putting into words what I want to say:
If the overall difference is so small, perhaps it does not stem from the SSE optimizations but from somewhere else? The difference is mostly in frequencies > 2kHz and AFAIK aotuv b3 among other things addresses encoding of transients. So perhaps the blips are a result of different aotuv tunings?
I couldn't abx the one sample I tried.
I ripped a complete wav image of a CD and the result is the same: mostly identical.
Do you have basic C coding skills? YOu could try to do a trial and error and search which parts of SSE optimizationa (mostly) alter the output by just going back to the usual C routine for that parts.
Usually a "#undef __SSE__" or alike will help on top of the .c file.
Usually a "#undef __SSE__" or alike will help on top of the .c file.
Unfortunately my programming skills are less than rudimentary.
This is a "lo-fi" version of our main content. To view the full version with more information, formatting and images, please click here.