This is code project for ECE408(Applied Parallel Programming). And you can use this project as a good start for your own lab implementation. Here's how you finish your own project. You can get all slices here.
templates/mpx_template.cu are template codes and you need to full fill these code files. For me, I add these implementations in mpx_implement.cu.
Once you finish your new implementation code, you need to add it to CmakeLists.txt so it can be compiled.
// mpx_implement.cu is your implementation code file name.
CUDA_ADD_EXECUTABLE(MPX_Implement mpx_implement.cu ${wbhdr} ${wbsrc})
NOTICE: In CmakeLists.txt, this line set GPU arch and code generation configuration. Change it if your GPU arch is not sm_86.
mkdir build
cd build
cmake ..
make
Executable files will be generated in under build/, and any time when you update your implementation you just need to rerun make under build/. But if you update your CmakeLists.txt, you should run cmake .. first and then run make to compile all codes.
datasets for code test are under test_data/, you can use them to test your implementation with
./program -e <expected_output_file> -i <input_file_1>,<input_file_2> -o <output_file> -t <type>
The <expected_output_file> and <input_file_n> are the input and output files provided in the dataset. The <output_file> is the location you’d like to place the output from your program. The<type> is the output file type: vector, matrix, or image. If an MP does not expect an input or output, then pass none as the parameter.
For example, if you want test mp1_implementatiton, you can run
#suppose you are in ./build directory
./MP1_Implement -e ../test_data/mp01/0/output.raw -i ../test_data/mp01/0/input0.raw,../test_data/mp01/0/input1.raw -o ../test_data/mp01/0/res.raw -t vector
And if you see logs like this, then congratulations, you have passed the test!
You can generate your own dataset if you want more test. For example, you can use MP2_Dataset to generate dataset.
MP2_Dataset -d /data/directory/ -ra 100 -ca 128 -rb 128 -cb 256