PTX export not working

[A1-TheDarkKnight]

Thank you for the beautiful work.
I would like to report that exporting Nvidia PTX code not working.

• I tried it on the FuildX3D, by uncommenting the "#define PTX" in "opencl.hpp" file. The program compiles fine, but there is no PTX file.
• I tried the same thing on the "OpenCL-Wrapper" sample project. It didn't compile at all because of an error:
  • "Error C3861 'write_file': identifier not found OpenCL-Wrapper ...\OpenCL-Wrapper\src\opencl.hpp 289"
  • "Error (active) E0020 identifier "write_file" is undefined OpenCL-Wrapper ...\OpenCL-Wrapper\src\opencl.hpp 289"

Specs:
NVidia RTX 3090
Win 11 x64
driver : 551.76 WHQL DCH [r551_71-2]
Visual Sudio 2022

Best Regards

[ProjectPhysX]

Hi @A1-TheDarkKnight,

are you running an Nvidia GPU? The PTX export only works with Nvidia hardware.

For the OpenCL-Wrapper, you have to uncomment this line to get it working: https://github.com/ProjectPhysX/OpenCL-Wrapper/blob/master/src/utilities.hpp#L4
I disabled the read_file()/write_file() functions by default to not have the OpenCL-Wrapper include the <fstream>/<filesystem> headers by default, to make it more sleak and reduce compile time.

Kind regards,
Moritz

[A1-TheDarkKnight]

Hi @ProjectPhysX , Yes I am on Nvidia RTX 3090. I forgot to mention the specs. I added the driver specific version too, and visual studio version.
I tried the OpenCL-Wrapper, and It compiled, but still no PTX file.

Best Regards

[ProjectPhysX]

Hi @A1-TheDarkKnight,

is it possible that you compiled with latest official Khronos Group OpenCL headers instead of the older OpenCL headers I had shipped?
I have updated to the latest OpenCL headers in v3.1, and during the update I realized that the API return type of cl_program.getInfo<CL_PROGRAM_BINARIES>() was changed so conversion to string now failed. A fix was necessary to accomodate for new return type. Now it is:

string ptx = (char*)&cl_program.getInfo<CL_PROGRAM_BINARIES>()[0][0];

Kind regards,
Moritz

[A1-TheDarkKnight]

@ProjectPhysX Hi there, sorry for being late.
It works now. Thank you very much 👍 .

By the way for anyone who doesn't see the PTX file after compilation, you will not see it.
You need to run the program once, then you will see a folder called "bin" near the program, inside it you will find the ptx file.

Best Regards

[ProjectPhysX]

Great! Yes the OpenCL kernels are only compiled at runtime, so have to start the executable once to get PTX output.

Kind regards,
Moritz