Enable algo options and full code omissions of disabled algos;
Reactivates CUDA 8.0
Allows for completely shutting off code that won't compile on CUDA 8.0 (RandomX, AstroBWT, and KawPow) while the resulting plugin works great for the other possible algos.
Compiling with CUDA 8.0 (or other versions) was accomplished in the added cmake/CUDA-Version.cmake which moves the CUDA Toolkit detections a step earlier so the main CMakeLists.txt can make decisions and route the user more appropriately to a working result. For example, compiling with an arch 20 option with some newer CUDA will explain why it won't work.
Any disabled algos will throw an unsupported condition so the plugin API still works appropriately, even with chunks of code carved out. Very similar to how the main miner does algo selections.
Have used very similar patch for quite some time now to compile working miners for arch-20 (or 21) devices. It is well tested in all my use cases.
Thank you, I need some time to review, however changes in Algorithm.cpp are not required, plugin always deal with canonical algorithm names, there is no need for aliases.
Thank you.
Allows for completely shutting off code that won't compile on CUDA 8.0 (RandomX, AstroBWT, and KawPow) while the resulting plugin works great for the other possible algos.
Compiling with CUDA 8.0 (or other versions) was accomplished in the added
cmake/CUDA-Version.cmake
which moves the CUDA Toolkit detections a step earlier so the mainCMakeLists.txt
can make decisions and route the user more appropriately to a working result. For example, compiling with an arch 20 option with some newer CUDA will explain why it won't work.Any disabled algos will throw an unsupported condition so the plugin API still works appropriately, even with chunks of code carved out. Very similar to how the main miner does algo selections.
Have used very similar patch for quite some time now to compile working miners for arch-20 (or 21) devices. It is well tested in all my use cases.