- you can't make a DirectCompute 11/11.1 (DirectX 11/11.1) kernel/program faster than the OpenCL version
- you can't speed up poclbm by changing the "front end" (the part that is written in Python or, in my case, in C#).
- I have wasted the money I used to buy a 5970 + power supply... The difficulty is growing too much and in two or three weeks I'll be out-of-the-game, without even recouping the money spent... But at least I have a good graphical adaptor :-)
BUT
There are always persons that LOVE to tinker a little. And C# is probably one of the two/three most "known" languages. So I'm putting on sale the C# program (the source, and the exe/dll if you truly need them). To give some numbers... counting comments and blank lines it's 1500 lines of code, 620 effective LOC using the Code Metrics Tool of VS 2010. I would love a collective offer to open source it (poclbm is MIT licensed, and I think the MIT license is very good).
What I'm "selling":
a complete, working conversion, nearly line-by-line of poclbm PLUS a retooled version that uses multiple threads (so much line-by-line that if you rename the exe to poclbm.exe and put it in the folder of GUIMiner it will run flawlessly). Both of them, on my 4 core intel with an ATI HD 5970 are nearly as much fast as the original poclbm (my computer does 380 mhash/sec) (I think my multithreaded version is around 0.15%-0.20% faster than poclbm, but I could be wrong, and my version uses 1%-2% more CPU). I'm even working on a multi-kernel version (extracting the code that "speaks" to the kernel in a different module)
If a single person wants to buy it, we will discuss what he is buying (probably a one-year exclusive license and then I can open source it, or something similar... a longer license if he feels he truly need it. We aren't speaking of the secret sauce for the most good cake on Earth... We are speaking of a C# conversion of a working program). If it's bought by a collective group, it can be licensed as you prefer.
The target price... I don't know... In the end it won't be much (especially if we compare it to the days it took me to convert it). First I want to see if there is an "interest", then someone can throw a number :-) Otherwise I'll throw it in the bin of useless projects...
Ok... this is quite stupid... And I know it... And I don't even truly need the money... It's more a matter of principle...