Search code examples
gcccudanvcc

Finding compute and sm numbers in a binary file

I have a binary file compiled with cuda enabled features and gcc. I want to know which cuda architectures are included in the binary file?

I mean sm and compute compatibility numbers which are -gencode arch=compute_XX,code=sm_XX.

Solution

  • NVIDIA ship binary utilites which can display headers and disassemble the binary payloads in object files, libraries, and executables. cudaobjdump can be used as follows:

    $ nvcc -arch=sm_62 -std=c++11 -I ./ main.cu -o main
$ cuobjdump ./main

Fatbin elf code:
================
arch = sm_62
code version = [1,7]
producer = <unknown>
host = linux
compile_size = 64bit

Fatbin elf code:
================
arch = sm_62
code version = [1,7]
producer = cuda
host = linux
compile_size = 64bit

Fatbin ptx code:
================
arch = sm_62
code version = [5,0]
producer = cuda
host = linux
compile_size = 64bit
compressed

    Which shows both the binary and PTX payloads included in the executable.