This is a course project for Introduction to Parallel and Distributed Computation, School of EECS Peking University, 2018 spring
In this project we try to design parallel algorithms for two classical problem, transposed convolution and K-Nearest Neighbors(KNN). We run our programs with GPU and gain obviously better performance than serial ones.
Run following commands to compile & run serial program.
$ gcc -o trans_conv_serial trans_conv_serial.c -l OpenCL -std=gnu99
$ ./trans_conv_serial
Run following commands to compile & run parallel programs. We provide three different versions of implementation. (x = 0,1,2)
$ gcc -o trans_conv_cl_vx trans_conv_cl_vx.c -l OpenCL -std=gnu99
$ ./trans_conv_cl_vx
The serial program outputs total time while parallel ones output running time and total time.
Run generate_sample.py to generate data.
$ python generate_sample.py
Run following commands to compile & run serial program.
$ gcc -o knn_serial.out knn_serial.c
$ ./knn_serial.out
Run following commands to compile & run parallel programs.
$ gcc -o knn_1.out knn_1.c -lOpenCL
$ ./knn_1.out