Skip to content

A pattern-based algorithmic autotuner for graph processing on GPUs.

Notifications You must be signed in to change notification settings

PAA-NCIC/GSWITCH

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

23 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation



DOI

GSWITCH

GSWITCH is a pattern-based algorithmic autotuning system that dynamically switched to the suitable optimization variants with negligible overhead. Specifically, It is a CUDA library targeting the GPU-based graph processing application, it supports both vertex-centric or edge-centric abstractions. By far, GSWITCH can automatically determine the suitable optimization variants in Direction (push, pull), data-structure (Bitmap, Sorted Queue, Unsorted Queue), Load-Balance (TWC, WM, CM, STRICT, 2D-partition), Stepping (Increase, Decrease, Remain), Kernel Fusion(Standalone, Fused). The fast optimization transition of GSWITCH is based on a machine learning model trained from 600+ real graphs from the network repository. The model can be reused by news applications, or be retrained to adapt new architectures. In addition, GSWITCH provides succinct programming interface which hides all low-level tuning details. Developers can implements their graph application with high performance in just ~100 lines of code.

For more details, please visit our website.

Dependency

  • nvcc 7.5+
  • cmake
  • moderngpu

Quickstart

mkdir build
cd build
cmake ..
make

Usage

Here are the basic useages of pre-integrated applications(BFS,CC,PR,SSSP,BC) in GSWITCH.

./EXE <graph_path> [options]
[-r, --src=<int>]         Choose a root vertex. (Default: ch
                          oose randomly).
[-v, --verbose]           Print verbose per iteration info. 
                          (Default: quiet mode)
[-V, --validation]        Process the CPU reference validati
                          on. (Defaule: no validation)
[-H, --with-header]       Input file has header (e.g. nverte
                          xs, nvertexs, nedges, Default: no 
                          header).
[-W, --with-weight]       Input file has weight.(Default: no
                           weight value)
[-i, --ignore-weight]     Ignore the graph weight.(Default: 
                          false)
[-d, --directed]          Graph is directed.(Default: undire
                          cted)
[-c, --configs=Push-Queue-CM-Fused]
                          Set debug strategies, use - to sep
                          arater them (Default: <empty>).
[-D, --device=<int>]      Choose GPU for testing (Default: 0
                          )
[-j, --json=<string>]     set the json path (Default: 0)

Example

Here is a sample codes of BFS for graph soc-orkut. For more details please visit ./application/bfs.cu

#include "gswitch.h"
using G = device_graph_t<CSR, Empty>;

struct BFS:Functor<VC,int,Empty,Empty>{
  __device__ Status filter(int vid, G g){
    int lvl = *wa_of(vid);
    if(lvl == g.get_level()) return Active;
    else if (lvl < 0) return Inactive;
    else return Fixed;
  }
  __device__ int emit(int vid, Empty *w, G g) {return g.get_level();}
  __device__ bool cond(int v, int newv, G g) {return *wa_of(v)==-1;}
  __device__ bool comp(int* v, int newv, G g) {*v=newv; return true;}
  __device__ bool compAtomic(int* v, int newv, G g) {*v=newv; return true;}
};

int main(){
  // load graph
  for(level=0;;level++){
    inspector.inspect(as, g, f, stats, fets, conf);
    if(as.finish(g, f, conf)) break;
    selector.select(stats, fets, conf);
    executor.filter(as, g, f, stats, fets, conf);
    g.update_level();
    executor.expand(as, g, f, stats, fets, conf);
  }
  // copy data back
}

run the with ./BFS soc-orkut.mtx --with-header --src=0 --device=0 --verbose:

run-bfs

Performance

please visit our website

License

All the libraryies, examples, and source codes of GSWITCH are released under Apache 2.0.