I am new with GPU programming and exploring all possible available ways for writing my application. I have looked at the Parallel Thread Execution (PTX) ISA version 1.0 Release 1.0 document. I have understood most of things written in the document, but still not able to write a complete code. Can any one help me out and give me a very basic example code and the compliation intstructions using PTX.
Thanks and regards
San
Why not generate your own, using nvcc -ptx ?
Compilation is easy, that’s where ptxas is for:
ptxas file.ptx -o file.cubin
What’s the next step after you’ve compiled with ptxas from ptx to cubin to link and to have a working program?
You can use Driver API to load end execute .cubin files.
May I ask why you want to write you program in PTX? why not use CUDA? Cuda generates PTX in the background right? Or am I just a stupid moron who doesn’t know anything about GPGPU?
I think their reason might be trying to avoid some optimization bug in nvcc. Also ptx gives more flexible control on memory so that you can force a variable to stay on register and so on. With ptx, you can have your own optimized code, just like ppl do some optimization jobs in traditional assembly.
(So I would say inline ptx assembly within cuda code will be extremely helpful.)
Mark
Ohh ok, now I know that. So it’s basically back to the die hard way of programming you GPU :)