I have the access to both NVIDIA and AMD GPU's in different systems. The model i use is Nvidia Quadro 410 and AMD Sapphire R9290x TRI-X OC .
I wanted to fetch the core information on which a particular thread was running. I came across INLINE PTX and was able to fetch information like the stream multiprocessor on which the thread ran, the warp id and the warp lane the thread uses. This all was possible on my NVIDIA card. BUT, this did not work on my AMD card(which is a different computer with very high configuration).
Now I post these questions:-
1. Why was I able to fetch the hardware info of nvidia card but failed to do so on AMD card, since OPENCL should be able to run on any opencl devices
2. The max compute unit in my nvidia is 1 even though it has 192 cuda cores. so using inline ptx the smid prints 0 for all threads . How can i get the cuda core info on which the thread runs.
I hope you come up with the solution for the above problems, however vague it is plz do share your knowledge.