.PTX & how to get it running. How to create a hello world type ptx program, and get it to run.

Hi everyone.

I have read the PTX_ISA_2.3 documentation, and “get” it. i have attempted assembly in the past (HLA) and “get” that the files need to be compiled to executable machine language before execution. but i cannot understand the process of how to take ( even a hello world style ) hand coded .ptx file to an executable stage.

I have downloaded the Nvidia development toolkit 4.0 and begun sifting though the documentation in the doc folder, but i have no underlying c programing experience, so am having difficulty understanding sentences containing multiple terms that i am unfamiliar with ( even after looking up terms, as the descriptions often use several other unfamiliar terms, and well, its doing my head in… 8-) )

I have absolutely no issue with the ptx language, syntax etc. am familiar with console ( cmd prompt ) type operations and calling a .exe file with parameters in order to use ptxas, nvcc and the other /bin/ files in the computing toolkit, but i have no idea as to how to get something to run without first undertaking a bachelors degree and installing c+ on my machine. the language of the documentation is too condensed with jargon and abstracted to far for someone with limited programming experience to understand.

I understand that my lack of background knowledge IS the main bottleneck in my GPU programming learning curve, but any help in regard to the location of good help files, walkthroughs, pdfs, sites or blogs dealing with how to get hand coded ptx files to execute would be great.

Cheers and thanks for your support in advance.

[font=“Courier New”]nvcc[/font] has the [font=“Courier New”]–dryrun[/font] option that prints out the steps it performs instead of executing them. This is helpful to understand the way .cu or .ptx are compiled to executables.

One problem however is is that the host programming still is in C/C++. So even if you can write a PTX “assembly” program for the GPU, you will still need to write the host code to start and invoke it. So I don’t currently see a way to use CUDA without knowing C (or Fortran, if you use third party compilers).