如何在linux中读取CUDA .cubin二进制文件？

Question

I'm trying to get the number of registers per thread for each of my kernels in my CUDA program. This will help me use the CUDA Occupancy Calculator (http://developer.download.nvidia.com/compute/cuda/CUDA_Occupancy_calculator.xls) as to determine the highest occupancy of the GPU for my program.

However, I have generated the .cubin file using the -cubin flag but I am unable to read it in vim/othertexteditors, as suggested by NVIDIA (http://forums.nvidia.com/index.php?showtopic=31279). Does anyone know how to get to read it?

Thanks

Answer 1

The easiest solution is to pass -Xptxas -v to nvcc like so

$ nvcc -Xptxas -v foo.cu
ptxas info    : Compiling entry function '_Z9my_kernelPfS_f' for 'sm_10'
ptxas info    : Used 2 registers, 20+16 bytes smem

Alternatively, you can use the cudaFuncGetAttributes API function to obtain the necessary values at runtime.