We present VersaPipe, a framework for pipelined computing on GPU. We propose two execution models (coarse pipeline model and fine pipeline model) and a hybrid model to take advantage of all existing models. The hybrid pipeline model explores both temporal and spatial schedulings. The Versapipe framework implements the hybrid model and provide an auto-tuner that can automatically configure the pipeline with a combination of models.
"VersaPipe: A Versatile Programming Framework for Pipelined Computing on GPU", Zhen Zheng, Chanyoung Oh, Jidong Zhai, Xipeng Shen, Youngmin Yi, Wenguang Chen, The 50th Annual IEEE/ACM International Symposium on Microarchitecture, Boston, USA, Oct 14-18, 2017.
On Linux the CUDA compiler with C++11 support (CUDA 6.5 or later) and cmake are required.
Clone the source tree and build basic examples:
$ cd HybridPipe/examples/demo
$ mkdir build
$ cd build
$ cmake ..
$ make
Three different procedures are defined using proc.cuh. The host control logic is found in test.cu and could be executed via loop binary:
$ ./loop
If launch fails, you may need to add code generation for Compute Capability of your GPU in the file of CMakeLists.txt and recompile.