/cuda-makefile

Using a makefile to setup cuda on Wits' lab computers.

Primary LanguageCuda

Makefile for Cuda

  1. git clone https://github.com/AngusTheMack/cuda-makefile - clone the repo
  2. cd cuda-makefile - change into the directory
  3. make - run the make command
  4. ./histogram - run the executable

I haven't touched the code in histo_main.cu, so it will say "Test failed". However, the code compiled without having to move files around or adding nvcc to the path - which is what this makefile helps with.

Changes Made

The initial document was histo_main_before.cu, and the problem was that it had dependencies laid out with relative paths - which are tricky to continuously setup when working on different lab machines.

The relative paths are the second and third lines below:

// CUDA Runtime
#include <cuda_runtime.h>
#include "../common/helper_cuda.h"
#include "../common/helper_functions.h"

By changing these to <...> and getting rid of the relative paths, we tell the preprocessor to search the directories we have included with the -I flag. It looks as follows:

// CUDA Runtime
#include <cuda_runtime.h>
#include <helper_cuda.h>
#include <helper_functions.h>

Makefile structure

The makefile sets the compiler (CC) to the nvcc executable on the lab computers. This is a faster than having to add the executable location to the path everytime.

It then sets the library flag (LDFLAGS) to the library files for Cuda on the lab computers.

On line three the IFLAGS, includes the location of the .h files that are required for compiling histo_main.cu.

CC  = /usr/local/cuda-10.0/bin/nvcc
LDFLAGS = -L /usr/local/cuda-10.0/lib64
IFLAGS 	= -I/usr/local/cuda-10.0/samples/common/inc

histogram: histo_main.cu
	$(CC) histo_main.cu $(LDFLAGS) $(IFLAGS) -c $<
	$(CC) histo_main.o  $(LDFLAGS) $(IFLAGS) -o histogram

Then the histogram target is compiled with all the necessary flags, and output (-o) to an executable called histogram. Check out this page if you are confused by makefiles.

Notes

This makefile will only work for the Lab computers at wits, and it won't work for every program. If you want to run a different .cu file you will have to change the related objects and targets. You may also have to change the -I flag (or add more) to point to different header files depending on what the other files have #included.

Acknowledgements

Thanks to Dr Hairong Wang for taking the High Performance Computing course at the University of Witswatersrand. The hist_main.cu file was supplied by her as an example for one of the labs in the course.