Skip to content

Compare MKL inv speed

... to https://github.com/pc2/SubmatrixMethod/blob/master/mkl-matrix-inv.c.

Test runs

Input matrix: sprandsym-s1000-d1-c2-n1.txt

Results:

➜  bauerc@cn-0252 SubmatrixMethod git:(master)  MKL_NUM_THREADS=1 ./mkl-matrix-inv 1000 sprandsym-s1000-d1-c2-n1.txt sprandsym-s1000-d1-c2-n1.out.txt
Reading input matrix took 0.042s
Doing a precise inversion took 0.081s
Writing result matrix took 0.253s

➜  bauerc@cn-0252 SubmatrixMethod git:(master)  MKL_NUM_THREADS=1 ./mkl-matrix-inv 1000 sprandsym-s1000-d1-c2-n1.txt sprandsym-s1000-d1-c2-n1.out.txt
Reading input matrix took 0.027s
Doing a precise inversion took 0.077s
Writing result matrix took 0.296s

➜  bauerc@cn-0252 SubmatrixMethod git:(master)  MKL_NUM_THREADS=1 ./mkl-matrix-inv 1000 sprandsym-s1000-d1-c2-n1.txt sprandsym-s1000-d1-c2-n1.out.txt
Reading input matrix took 0.027s
Doing a precise inversion took 0.077s
Writing result matrix took 0.296s

➜  bauerc@cn-0252 SubmatrixMethod git:(master)  MKL_NUM_THREADS=1 ./mkl-matrix-inv 1000 sprandsym-s1000-d1-c2-n1.txt sprandsym-s1000-d1-c2-n1.out.txt
Reading input matrix took 0.027s
Doing a precise inversion took 0.065s
Writing result matrix took 0.252s

➜  bauerc@cn-0252 SubmatrixMethod git:(master)  MKL_NUM_THREADS=20 ./mkl-matrix-inv 1000 sprandsym-s1000-d1-c2-n1.txt sprandsym-s1000-d1-c2-n1.out.txt
Reading input matrix took 0.027s
Doing a precise inversion took 0.080s
Writing result matrix took 0.257s

➜  bauerc@cn-0252 SubmatrixMethod git:(master)  MKL_NUM_THREADS=20 ./mkl-matrix-inv 1000 sprandsym-s1000-d1-c2-n1.txt sprandsym-s1000-d1-c2-n1.out.txt
Reading input matrix took 0.027s
Doing a precise inversion took 0.054s
Writing result matrix took 0.256s

➜  bauerc@cn-0252 SubmatrixMethod git:(master)  MKL_NUM_THREADS=20 ./mkl-matrix-inv 1000 sprandsym-s1000-d1-c2-n1.txt sprandsym-s1000-d1-c2-n1.out.txt
Reading input matrix took 0.027s
Doing a precise inversion took 0.056s
Writing result matrix took 0.257s

➜  bauerc@cn-0252 SubmatrixMethod git:(master)  MKL_NUM_THREADS=20 ./mkl-matrix-inv 1000 sprandsym-s1000-d1-c2-n1.txt sprandsym-s1000-d1-c2-n1.out.txt
Reading input matrix took 0.027s
Doing a precise inversion took 0.094s
Writing result matrix took 0.261s

Considering the minimum we find 65ms for 1 thread and 54ms for 20 threads.

Edited by Carsten Bauer