A python wrapper for fast svd of 3x3 matrix
see https://github.com/ericjang/svd3 for details
sh make.sh
testing method: average over 100000 randomly initialized matrix
| method | evaluation | numpy with openblas | svd3 w/o patching | svd3 w patching |
|---|---|---|---|---|
| QR decomposition | ||QR - A||_F / ||A||_F | 3.716e-08 | 5.608e-03 | 1.789e-07 |
| singular value decomposition | ||USV* - A||_F / ||A||_F | 5.891e-08 | 1.750e-02 | 3.291e-07 |
| polar decomposition | ||UP - A||_F / ||A||_F | 3.948e-07 | 2.609e-02 | 4.501e-07 |
| method | evaluation | numpy with openblas | svd3 w/o patching | svd3 w patching |
|---|---|---|---|---|
| QR decomposition | average | 43.307us | 2.771us | 2.740us |
| singular value decomposition | average | 21.133us | 4.374us | 4.656us |
| polar decomposition | average | 78.317us | 3.201us | 3.665us |