ptxtract

Simple examples for PTX code extraction from CUDA and OpenCL kernels.

Requirements

Besides of a common C development environment (GCC, Make, etc.), you'll need the CUDA Toolkit.

Building

To build, simply do:

make

Be sure that you have the CUDA Toolkit installed under /usr/local/cuda or the environment variable CUDA_PATH set to the proper path.

Running

The PTX file for the CUDA kernel will be available under the name ptx/add_vectors_cuda.ptx, there's no need to run anything (in fact, no executables will be generated).

For the OpenCL kernel, it's necessary to execute the host-side code, by simply typing:

./bin/add_vectors_opencl

The resulting PTX file is generated under the name ptx/add_vectors_opencl.ptx. Note that you require an NVIDIA GPU as the first OpenCL device, otherwise, the file will be there but it will NOT be a PTX file, instead, it will be a vendor-specific binary/pre-binary file.