Hi everyone ! I am trying to programming CUDA directly based on PTX assembly language, not C\C++, because I want to compile each little program in Runtime to do some part of work, not everying.
How should I ? It seems that the PTX’s document is too short External Image
Write some stuff in nvcc, and look at the generated PTX. This helps to avoid internal errors (e.g. st.global.xxx [reg],imm;).
Also, it may help to write PTX in SSA style, since ptxas is in charge of register allocation.
hey jedimaster,
any progress here ? did you achive to call a prog which is generate at runtime ?
regards,
jj
