GPU Acceleration for Scan Tailor?

[Anonymous1]

I've been wondering about this after I found out Python had a nice GPU library. Would it be too hard to implement GPU acceleration into Scan Tailor? NVidia's CUDA language syntax is nearly identical to C++, so conversion wouldn't be too hard. Anyone have any ideas?

[Tulon]

Would it be too hard to implement GPU acceleration into Scan Tailor?

It's not that it's hard, it's just time consuming.

NVidia's CUDA language syntax is nearly identical to C++, so conversion wouldn't be too hard.

Even though CUDA is superior to OpenCL, I would use the latter, as it's cross platform.

It won't happen by itself though. Someone will have to put effort into it. Don't count on me, as I've retired from Scan Tailor development.

[Anonymous1]

Good man. You've done your part, and that's all that was needed. I can't even get my OCR Page Naming script to work, so I've "temporarily" suspended it, so I can't imaging how you could code that whole thing almost by yourself...

I'll see what I can do, but I doubt I can do much, as C++ is not my thing.

[Anonymous1]

Tulon, what are the most processor-intensive parts of the whole program? I've compiled it from source a bunch of times, so maybe I can tinker with the GPU acceleration stuff too. Thanks!

[Tulon]

I haven't profiled it for a while. You will have to do it yourself. On Linux you can use sysprof or callgrind + kcachegrind. On Windows you would need an expensive version of Visual Studio (Professional is not enough, I believe you need Team Editition).