TauBench
This is an unstructured grid benchmark. The benchmark mimics the run-time performance of the Tau solver (a Navier Stokes solver which has been developed at the DLR in Germany). Tau is a three-dimensional parallel hybrid multigrid solver, which solves the Reynolds-averaged Navier-Stokes equations by finite volume scheme.Install
$ make
To change the compiler
$ make MPICC=mpicc.mpich
Run
Tau supports both vector and cache colored grids, you might also need to adapt the compiler directives in nodep.h and expand.h. The benchmark itself can be used with two flags, the gridsize per process and the number of pseudo steps:$ mpirun -np 2 ./taubench -n 100000 -s 10
This is TauBench.
Evaluating kernels - please be patient.
..........
- kernel_1_0 : 0.934 secs - 1506.596 mflops
- kernel_1_1 : 0.349 secs - 625.287 mflops
- kernel_2_1 : 0.619 secs - 1603.554 mflops
- kernel_2_2 : 0.809 secs - 1508.574 mflops
- kernel_2_3 : 0.310 secs - 794.652 mflops
- kernel_2_4 : 0.472 secs - 1174.215 mflops
- kernel_3_0 : 3.048 secs - 898.502 mflops
total : 6.703 secs - 1888.867 mflops
points : 100000
steps : 10
procs : 2
comp : 6.351 secs
comm : 0.352 secs
comm ratio : 0.055
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent