I tried three different translators and this was the best one:
Quote:OK I try to be more clear as possible and excuses if me I do not explain it in English it is not the same instead an idea in another language that is not yours
but clarified I have nothing discover the only emulator I say that it lacks time to be 100% profitable in general chewy
Now if I think that require both cpu is wrong and is something that is that is anger running as in the project64 now my processor is not a gem but I have many thing to tell me that it is slow and in the middle of the emualacion the cpu is not using less than 40% of the two core I not going to say that it is not problem of optimization.
be that change with time and a good alternative is the computation in parallel with the gpu and cpu rather support or split the load between the two
It is true I'm not programming am administrator server why mention that not is if it is something difficult to do because it isn't my area
but my lack of comprencion what is the gpcpu and you you are an expert please explain is not of more learn from others
It's still mostly unreadable. Bits and pieces make sense and that's about it.
It's not difficult, just pointless. GPUs are good at performing simple repetitive arithmetic/calculus on large sets of data. Ideal GPGPU applications have large data sets, high parallelism, have minimal dependency between data elements, and are compute intensive.
large data sets = big groups of data that are made of lots of small data elements. For example images are usually made of millions of pixels. An image is a data set, a pixel is a data element.
high parallelism = many things that can be done at the same time (because they don't rely on each others output)
data dependency = an arithmetic operation relies on the output of another arithmetic operation
compute intensive = lots of arithmetic being done on each data element
Dolphin has none of these properties except in the video thread which already uses the gpu and the texture decoder which has an openCL option that uses the gpu. Dolphin spends most of its cpu processing power emulating dsp and cpu instructions that would either be impossible to emulate on a gpu due to architectural limitations or much slower to emulate on the gpu.
Some examples of where using a gpu would be useful would be matrix algebra, image processing, video processing, vertex (geometry) processing, FFT (fast fourier transform), neural networking, cryptography, cryptanalysis, computational finance, speech/image recognition, audio processing, physically based simulations (literally anything to do with physics), and probably some more stuff.
All of these involve doing the same math over and over on many separate pieces of data that are part of a larger set of data with minimal data dependencies. For example images are usually made of millions of pixels and image processing involves running on a series of calculations on each pixel. GPUs can do this in parallel by processing many pixels at the same time. This is done through a programming paradigm called stream processing. You stream a set of data into the gpu (an image for example), give the gpu the kernel to run, the kernel is a small program/function that the gpu runs on each data element in the set, then you stream the resulting data out of the gpu. So in other words:
1. stream data set in
2. run kernel
3. stream data set out
In d3d and openGL shaders are compiled into kernels and the data sets are usually either made of pixels (texture/framebuffers) or vertexes (mesh/model). In GPGPU you write general purpose functions that are compiled into kernels and the data sets can be whatever you want. However these data sets are usually made of vectors.
Emulation cannot use this system in any useful way unless you are emulating a similar architecture, such as another gpu. I can't think of anything dolphin does that is unrelated to graphics where dolphin is trying to go through lots of data in a data set doing repetitive calculations.
