Add RowVec_8u32f by Nicholas-Ho-arm · Pull Request #20712 · opencv/opencv

Nicholas-Ho-arm · 2021-09-17T10:59:40Z

Improve the performance of RowFilter in imgproc for u8 to f32 data.

Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

I agree to contribute to the project under Apache 2 License.
To the best of my knowledge, the proposed patch is not based on a code under GPL or other license that is incompatible with OpenCV
The PR is proposed to proper branch
There is reference to original bug report and related work
There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
The feature is well documented and sample code can be built with the project CMake

force_builders=Linux AVX2,Custom
buildworker:Custom=linux-3
build_image:Custom=ubuntu:18.04
CPU_BASELINE:Custom=AVX512_SKX
disable_ipp=ON

modules/imgproc/src/filter.simd.hpp

jondea

Great work! Some minor comments to consider.

modules/imgproc/src/filter.simd.hpp

Nicholas-Ho-arm · 2021-09-22T15:33:38Z

This RowVec_8u32f seems to be used in quite a few perf tests in imgproc. I have included here some of the speedup in results run on single thread Cortex-A72.

Geometric mean (ms)

                                    Name of Test                                      imgproc  imgproc  imgproc
                                                                                      myChange master    master
                                                                                                           vs
                                                                                                        imgproc
                                                                                                        myChange
                                                                                                       (x-factor)
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 16SC1, (0, 1), BORDER_CONSTANT)    0.028    0.028     1.00
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 16SC1, (0, 1), BORDER_REPLICATE)   0.027    0.027     0.99
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 16SC1, (0, 2), BORDER_CONSTANT)    0.030    0.029     1.01
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 16SC1, (0, 2), BORDER_REPLICATE)   0.030    0.029     1.03
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 16SC1, (1, 0), BORDER_CONSTANT)    0.028    0.028     1.01
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 16SC1, (1, 0), BORDER_REPLICATE)   0.028    0.027     1.02
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 16SC1, (1, 1), BORDER_CONSTANT)    0.027    0.027     1.03
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 16SC1, (1, 1), BORDER_REPLICATE)   0.026    0.026     1.03
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 16SC1, (2, 0), BORDER_CONSTANT)    0.029    0.029     1.00
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 16SC1, (2, 0), BORDER_REPLICATE)   0.029    0.028     1.03
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 16SC1, (2, 2), BORDER_CONSTANT)    0.029    0.029     1.01
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 16SC1, (2, 2), BORDER_REPLICATE)   0.029    0.028     1.00
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 32FC1, (0, 1), BORDER_CONSTANT)    0.034    0.044     0.77
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 32FC1, (0, 1), BORDER_REPLICATE)   0.033    0.042     0.78
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 32FC1, (0, 2), BORDER_CONSTANT)    0.037    0.046     0.80
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 32FC1, (0, 2), BORDER_REPLICATE)   0.035    0.045     0.78
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 32FC1, (1, 0), BORDER_CONSTANT)    0.036    0.046     0.79
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 32FC1, (1, 0), BORDER_REPLICATE)   0.035    0.045     0.78
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 32FC1, (1, 1), BORDER_CONSTANT)    0.034    0.044     0.78
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 32FC1, (1, 1), BORDER_REPLICATE)   0.033    0.043     0.77
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 32FC1, (2, 0), BORDER_CONSTANT)    0.036    0.046     0.78
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 32FC1, (2, 0), BORDER_REPLICATE)   0.036    0.045     0.80
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 32FC1, (2, 2), BORDER_CONSTANT)    0.036    0.046     0.78
sobelFilter::Size_MatType_dx_dy_Border3x3::(127x61, 32FC1, (2, 2), BORDER_REPLICATE)   0.035    0.045     0.78
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 16SC1, (0, 1), BORDER_CONSTANT)   0.118    0.120     0.98
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 16SC1, (0, 1), BORDER_REPLICATE)  0.120    0.120     1.00
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 16SC1, (0, 2), BORDER_CONSTANT)   0.137    0.137     1.00
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 16SC1, (0, 2), BORDER_REPLICATE)  0.136    0.134     1.01
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 16SC1, (1, 0), BORDER_CONSTANT)   0.127    0.125     1.02
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 16SC1, (1, 0), BORDER_REPLICATE)  0.127    0.125     1.01
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 16SC1, (1, 1), BORDER_CONSTANT)   0.109    0.107     1.02
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 16SC1, (1, 1), BORDER_REPLICATE)  0.107    0.107     1.00
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 16SC1, (2, 0), BORDER_CONSTANT)   0.138    0.140     0.98
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 16SC1, (2, 0), BORDER_REPLICATE)  0.135    0.135     1.01
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 16SC1, (2, 2), BORDER_CONSTANT)   0.141    0.141     1.00
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 16SC1, (2, 2), BORDER_REPLICATE)  0.137    0.139     0.98
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 32FC1, (0, 1), BORDER_CONSTANT)   0.222    0.327     0.68
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 32FC1, (0, 1), BORDER_REPLICATE)  0.222    0.326     0.68
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 32FC1, (0, 2), BORDER_CONSTANT)   0.239    0.343     0.70
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 32FC1, (0, 2), BORDER_REPLICATE)  0.238    0.341     0.70
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 32FC1, (1, 0), BORDER_CONSTANT)   0.242    0.348     0.70
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 32FC1, (1, 0), BORDER_REPLICATE)  0.242    0.348     0.69
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 32FC1, (1, 1), BORDER_CONSTANT)   0.220    0.322     0.68
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 32FC1, (1, 1), BORDER_REPLICATE)  0.221    0.325     0.68
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 32FC1, (2, 0), BORDER_CONSTANT)   0.240    0.343     0.70
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 32FC1, (2, 0), BORDER_REPLICATE)  0.238    0.341     0.70
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 32FC1, (2, 2), BORDER_CONSTANT)   0.241    0.342     0.70
sobelFilter::Size_MatType_dx_dy_Border3x3::(320x240, 32FC1, (2, 2), BORDER_REPLICATE)  0.237    0.341     0.70
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (0, 1), BORDER_CONSTANT)   0.406    0.417     0.97
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (0, 1), BORDER_REPLICATE)  0.417    0.410     1.02
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (0, 2), BORDER_CONSTANT)   0.486    0.478     1.02
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (0, 2), BORDER_REPLICATE)  0.477    0.484     0.99
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (1, 0), BORDER_CONSTANT)   0.439    0.449     0.98
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (1, 0), BORDER_REPLICATE)  0.444    0.448     0.99
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (1, 1), BORDER_CONSTANT)   0.368    0.371     0.99
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (1, 1), BORDER_REPLICATE)  0.367    0.382     0.96
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (2, 0), BORDER_CONSTANT)   0.490    0.479     1.02
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (2, 0), BORDER_REPLICATE)  0.480    0.480     1.00
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (2, 2), BORDER_CONSTANT)   0.492    0.481     1.02
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (2, 2), BORDER_REPLICATE)  0.484    0.472     1.02
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (0, 1), BORDER_CONSTANT)   0.815    1.231     0.66
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (0, 1), BORDER_REPLICATE)  0.814    1.237     0.66
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (0, 2), BORDER_CONSTANT)   0.903    1.316     0.69
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (0, 2), BORDER_REPLICATE)  0.891    1.310     0.68
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (1, 0), BORDER_CONSTANT)   0.909    1.314     0.69
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (1, 0), BORDER_REPLICATE)  0.893    1.315     0.68
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (1, 1), BORDER_CONSTANT)   0.812    1.231     0.66
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (1, 1), BORDER_REPLICATE)  0.810    1.229     0.66
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (2, 0), BORDER_CONSTANT)   0.892    1.321     0.68
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (2, 0), BORDER_REPLICATE)  0.891    1.311     0.68
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (2, 2), BORDER_CONSTANT)   0.895    1.317     0.68
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (2, 2), BORDER_REPLICATE)  0.894    1.310     0.68

Nicholas-Ho-arm · 2021-09-23T09:07:25Z

Gentle ping @asmorkalov. Could you take a look please? Thanks.

alalek

Thank you for contribution!

I have included here some of the speedup in results run on single thread Cortex-A72.

Great results!
"Single thread" reports is the right way for SIMD optimization (use --perf_threads=1)
BTW, we usually use "base"/"patch" columns order (example: #13781).
Also you can use -o markdown with summary.py to get GitHub friendly output.

What compiler was used for benchmarks? -O3?

modules/imgproc/src/filter.simd.hpp

alalek

Thank you 👍

Perf results on i5-6600 (1thread, NOIPP):

Details

Name of Test	base	patch	x-factor
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (0, 1), BORDER_CONSTANT)	0.129	0.125	1.03
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (0, 1), BORDER_REPLICATE)	0.126	0.122	1.04
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (1, 0), BORDER_CONSTANT)	0.118	0.121	0.98
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (1, 0), BORDER_REPLICATE)	0.118	0.121	0.98
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (0, 1), BORDER_CONSTANT)	0.436	0.153	2.85
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (0, 1), BORDER_REPLICATE)	0.446	0.160	2.79
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (1, 0), BORDER_CONSTANT)	0.426	0.169	2.52
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (1, 0), BORDER_REPLICATE)	0.456	0.167	2.73
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (0, 1), BORDER_CONSTANT\|BORDER_ISOLATED)	0.128	0.123	1.04
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (0, 1), BORDER_DEFAULT)	0.125	0.122	1.03
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (0, 1), BORDER_REPLICATE\|BORDER_ISOLATED)	0.125	0.122	1.03
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (1, 0), BORDER_CONSTANT\|BORDER_ISOLATED)	0.118	0.118	1.00
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (1, 0), BORDER_DEFAULT)	0.117	0.117	1.00
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (1, 0), BORDER_REPLICATE\|BORDER_ISOLATED)	0.117	0.118	0.99
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (0, 1), BORDER_CONSTANT\|BORDER_ISOLATED)	0.443	0.149	2.98
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (0, 1), BORDER_DEFAULT)	0.446	0.154	2.90
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (0, 1), BORDER_REPLICATE\|BORDER_ISOLATED)	0.436	0.149	2.93
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (1, 0), BORDER_CONSTANT\|BORDER_ISOLATED)	0.426	0.163	2.61
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (1, 0), BORDER_DEFAULT)	0.457	0.164	2.79
scharrViaSobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (1, 0), BORDER_REPLICATE\|BORDER_ISOLATED)	0.463	0.166	2.78
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (0, 1), BORDER_CONSTANT)	0.111	0.106	1.05
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (0, 1), BORDER_REPLICATE)	0.106	0.102	1.04
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (0, 2), BORDER_CONSTANT)	0.116	0.116	1.00
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (0, 2), BORDER_REPLICATE)	0.122	0.118	1.03
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (1, 0), BORDER_CONSTANT)	0.107	0.107	1.00
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (1, 0), BORDER_REPLICATE)	0.110	0.107	1.02
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (1, 1), BORDER_CONSTANT)	0.098	0.095	1.03
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (1, 1), BORDER_REPLICATE)	0.098	0.095	1.03
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (2, 0), BORDER_CONSTANT)	0.115	0.115	1.00
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (2, 0), BORDER_REPLICATE)	0.116	0.118	0.98
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (2, 2), BORDER_CONSTANT)	0.119	0.116	1.03
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 16SC1, (2, 2), BORDER_REPLICATE)	0.116	0.116	1.00
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (0, 1), BORDER_CONSTANT)	0.418	0.149	2.80
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (0, 1), BORDER_REPLICATE)	0.441	0.153	2.88
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (0, 2), BORDER_CONSTANT)	0.441	0.164	2.69
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (0, 2), BORDER_REPLICATE)	0.459	0.164	2.81
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (1, 0), BORDER_CONSTANT)	0.442	0.177	2.50
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (1, 0), BORDER_REPLICATE)	0.456	0.162	2.82
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (1, 1), BORDER_CONSTANT)	0.443	0.150	2.95
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (1, 1), BORDER_REPLICATE)	0.445	0.149	2.99
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (2, 0), BORDER_CONSTANT)	0.427	0.162	2.64
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (2, 0), BORDER_REPLICATE)	0.451	0.166	2.72
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (2, 2), BORDER_CONSTANT)	0.451	0.174	2.58
sobelFilter::Size_MatType_dx_dy_Border3x3::(640x480, 32FC1, (2, 2), BORDER_REPLICATE)	0.455	0.164	2.77
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (0, 1), BORDER_CONSTANT\|BORDER_ISOLATED)	0.106	0.103	1.02
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (0, 1), BORDER_DEFAULT)	0.104	0.104	1.00
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (0, 1), BORDER_REPLICATE\|BORDER_ISOLATED)	0.108	0.103	1.05
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (0, 2), BORDER_CONSTANT\|BORDER_ISOLATED)	0.115	0.118	0.98
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (0, 2), BORDER_DEFAULT)	0.114	0.115	0.99
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (0, 2), BORDER_REPLICATE\|BORDER_ISOLATED)	0.115	0.117	0.98
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (1, 0), BORDER_CONSTANT\|BORDER_ISOLATED)	0.107	0.108	0.99
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (1, 0), BORDER_DEFAULT)	0.109	0.107	1.01
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (1, 0), BORDER_REPLICATE\|BORDER_ISOLATED)	0.107	0.108	1.00
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (1, 1), BORDER_CONSTANT\|BORDER_ISOLATED)	0.097	0.093	1.05
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (1, 1), BORDER_DEFAULT)	0.097	0.095	1.02
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (1, 1), BORDER_REPLICATE\|BORDER_ISOLATED)	0.097	0.095	1.02
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (2, 0), BORDER_CONSTANT\|BORDER_ISOLATED)	0.115	0.117	0.98
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (2, 0), BORDER_DEFAULT)	0.117	0.115	1.02
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (2, 0), BORDER_REPLICATE\|BORDER_ISOLATED)	0.115	0.115	1.00
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (2, 2), BORDER_CONSTANT\|BORDER_ISOLATED)	0.115	0.116	0.99
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (2, 2), BORDER_DEFAULT)	0.115	0.115	0.99
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 16SC1, (2, 2), BORDER_REPLICATE\|BORDER_ISOLATED)	0.115	0.116	0.99
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (0, 1), BORDER_CONSTANT\|BORDER_ISOLATED)	0.411	0.149	2.76
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (0, 1), BORDER_DEFAULT)	0.436	0.153	2.85
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (0, 1), BORDER_REPLICATE\|BORDER_ISOLATED)	0.427	0.159	2.69
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (0, 2), BORDER_CONSTANT\|BORDER_ISOLATED)	0.444	0.163	2.71
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (0, 2), BORDER_DEFAULT)	0.453	0.163	2.79
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (0, 2), BORDER_REPLICATE\|BORDER_ISOLATED)	0.455	0.164	2.78
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (1, 0), BORDER_CONSTANT\|BORDER_ISOLATED)	0.446	0.163	2.73
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (1, 0), BORDER_DEFAULT)	0.457	0.161	2.83
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (1, 0), BORDER_REPLICATE\|BORDER_ISOLATED)	0.446	0.162	2.75
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (1, 1), BORDER_CONSTANT\|BORDER_ISOLATED)	0.428	0.151	2.84
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (1, 1), BORDER_DEFAULT)	0.443	0.153	2.90
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (1, 1), BORDER_REPLICATE\|BORDER_ISOLATED)	0.427	0.149	2.87
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (2, 0), BORDER_CONSTANT\|BORDER_ISOLATED)	0.444	0.162	2.75
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (2, 0), BORDER_DEFAULT)	0.455	0.163	2.80
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (2, 0), BORDER_REPLICATE\|BORDER_ISOLATED)	0.453	0.162	2.80
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (2, 2), BORDER_CONSTANT\|BORDER_ISOLATED)	0.439	0.163	2.69
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (2, 2), BORDER_DEFAULT)	0.440	0.163	2.70
sobelFilter::Size_MatType_dx_dy_Border3x3ROI::(640x480, 32FC1, (2, 2), BORDER_REPLICATE\|BORDER_ISOLATED)	0.458	0.164	2.80
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (0, 1), BORDER_CONSTANT)	0.541	0.549	0.99
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (0, 1), BORDER_REFLECT101)	0.549	0.556	0.99
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (0, 1), BORDER_REFLECT)	0.545	0.559	0.98
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (0, 1), BORDER_REPLICATE)	0.544	0.558	0.98
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (0, 2), BORDER_CONSTANT)	0.607	0.612	0.99
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (0, 2), BORDER_REFLECT101)	0.607	0.607	1.00
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (0, 2), BORDER_REFLECT)	0.611	0.607	1.01
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (0, 2), BORDER_REPLICATE)	0.607	0.596	1.02
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (1, 0), BORDER_CONSTANT)	0.579	0.578	1.00
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (1, 0), BORDER_REFLECT101)	0.580	0.589	0.98
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (1, 0), BORDER_REFLECT)	0.579	0.582	1.00
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (1, 0), BORDER_REPLICATE)	0.556	0.585	0.95
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (1, 1), BORDER_CONSTANT)	0.518	0.527	0.98
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (1, 1), BORDER_REFLECT101)	0.526	0.523	1.01
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (1, 1), BORDER_REFLECT)	0.516	0.529	0.98
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (1, 1), BORDER_REPLICATE)	0.522	0.534	0.98
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (2, 0), BORDER_CONSTANT)	0.569	0.539	1.06
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (2, 0), BORDER_REFLECT101)	0.571	0.568	1.00
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (2, 0), BORDER_REFLECT)	0.574	0.568	1.01
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 16SC1, (2, 0), BORDER_REPLICATE)	0.572	0.572	1.00
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (0, 1), BORDER_CONSTANT)	0.674	0.286	2.36
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (0, 1), BORDER_REFLECT101)	0.663	0.275	2.41
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (0, 1), BORDER_REFLECT)	0.695	0.267	2.60
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (0, 1), BORDER_REPLICATE)	0.694	0.306	2.27
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (0, 2), BORDER_CONSTANT)	0.698	0.260	2.68
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (0, 2), BORDER_REFLECT101)	0.707	0.274	2.58
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (0, 2), BORDER_REFLECT)	0.712	0.278	2.56
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (0, 2), BORDER_REPLICATE)	0.706	0.258	2.74
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (1, 0), BORDER_CONSTANT)	0.705	0.259	2.72
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (1, 0), BORDER_REFLECT101)	0.714	0.259	2.76
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (1, 0), BORDER_REFLECT)	0.711	0.267	2.66
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (1, 0), BORDER_REPLICATE)	0.714	0.258	2.77
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (1, 1), BORDER_CONSTANT)	0.698	0.260	2.68
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (1, 1), BORDER_REFLECT101)	0.692	0.259	2.67
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (1, 1), BORDER_REFLECT)	0.691	0.265	2.61
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (1, 1), BORDER_REPLICATE)	0.702	0.266	2.64
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (2, 0), BORDER_CONSTANT)	0.720	0.268	2.69
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (2, 0), BORDER_REFLECT101)	0.716	0.284	2.52
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (2, 0), BORDER_REFLECT)	0.716	0.259	2.76
sobelFilter::Size_MatType_dx_dy_Border5x5::(640x480, 32FC1, (2, 0), BORDER_REPLICATE)	0.716	0.262	2.73
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (0, 1), BORDER_CONSTANT\|BORDER_ISOLATED)	0.547	0.549	1.00
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (0, 1), BORDER_DEFAULT)	0.542	0.539	1.01
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (0, 1), BORDER_REFLECT101\|BORDER_ISOLATED)	0.552	0.555	0.99
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (0, 1), BORDER_REFLECT\|BORDER_ISOLATED)	0.544	0.520	1.05
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (0, 1), BORDER_REPLICATE\|BORDER_ISOLATED)	0.701	0.559	1.25
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (0, 2), BORDER_CONSTANT\|BORDER_ISOLATED)	0.618	0.611	1.01
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (0, 2), BORDER_DEFAULT)	0.607	0.606	1.00
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (0, 2), BORDER_REFLECT101\|BORDER_ISOLATED)	0.608	0.609	1.00
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (0, 2), BORDER_REFLECT\|BORDER_ISOLATED)	0.608	0.607	1.00
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (0, 2), BORDER_REPLICATE\|BORDER_ISOLATED)	0.608	0.613	0.99
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (1, 0), BORDER_CONSTANT\|BORDER_ISOLATED)	0.580	0.550	1.05
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (1, 0), BORDER_DEFAULT)	0.579	0.578	1.00
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (1, 0), BORDER_REFLECT101\|BORDER_ISOLATED)	0.580	0.578	1.00
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (1, 0), BORDER_REFLECT\|BORDER_ISOLATED)	0.583	0.582	1.00
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (1, 0), BORDER_REPLICATE\|BORDER_ISOLATED)	0.579	0.579	1.00
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (1, 1), BORDER_CONSTANT\|BORDER_ISOLATED)	0.517	0.526	0.98
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (1, 1), BORDER_DEFAULT)	0.518	0.522	0.99
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (1, 1), BORDER_REFLECT101\|BORDER_ISOLATED)	0.519	0.529	0.98
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (1, 1), BORDER_REFLECT\|BORDER_ISOLATED)	0.518	0.528	0.98
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (1, 1), BORDER_REPLICATE\|BORDER_ISOLATED)	0.518	0.507	1.02
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (2, 0), BORDER_CONSTANT\|BORDER_ISOLATED)	0.569	0.578	0.98
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (2, 0), BORDER_DEFAULT)	0.580	0.571	1.02
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (2, 0), BORDER_REFLECT101\|BORDER_ISOLATED)	0.572	0.569	1.01
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (2, 0), BORDER_REFLECT\|BORDER_ISOLATED)	0.573	0.571	1.00
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 16SC1, (2, 0), BORDER_REPLICATE\|BORDER_ISOLATED)	0.578	0.560	1.03
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (0, 1), BORDER_CONSTANT\|BORDER_ISOLATED)	0.706	0.264	2.67
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (0, 1), BORDER_DEFAULT)	0.704	0.265	2.66
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (0, 1), BORDER_REFLECT101\|BORDER_ISOLATED)	0.693	0.265	2.61
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (0, 1), BORDER_REFLECT\|BORDER_ISOLATED)	0.698	0.267	2.62
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (0, 1), BORDER_REPLICATE\|BORDER_ISOLATED)	0.702	0.259	2.71
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (0, 2), BORDER_CONSTANT\|BORDER_ISOLATED)	0.714	0.267	2.67
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (0, 2), BORDER_DEFAULT)	0.717	0.259	2.77
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (0, 2), BORDER_REFLECT101\|BORDER_ISOLATED)	0.715	0.260	2.75
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (0, 2), BORDER_REFLECT\|BORDER_ISOLATED)	0.706	0.260	2.72
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (0, 2), BORDER_REPLICATE\|BORDER_ISOLATED)	0.713	0.267	2.67
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (1, 0), BORDER_CONSTANT\|BORDER_ISOLATED)	0.712	0.259	2.75
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (1, 0), BORDER_DEFAULT)	0.713	0.262	2.73
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (1, 0), BORDER_REFLECT101\|BORDER_ISOLATED)	0.708	0.260	2.73
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (1, 0), BORDER_REFLECT\|BORDER_ISOLATED)	0.710	0.258	2.75
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (1, 0), BORDER_REPLICATE\|BORDER_ISOLATED)	0.708	0.276	2.57
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (1, 1), BORDER_CONSTANT\|BORDER_ISOLATED)	0.710	0.258	2.75
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (1, 1), BORDER_DEFAULT)	0.700	0.258	2.71
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (1, 1), BORDER_REFLECT101\|BORDER_ISOLATED)	0.698	0.280	2.49
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (1, 1), BORDER_REFLECT\|BORDER_ISOLATED)	0.701	0.266	2.63
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (1, 1), BORDER_REPLICATE\|BORDER_ISOLATED)	0.684	0.267	2.56
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (2, 0), BORDER_CONSTANT\|BORDER_ISOLATED)	0.707	0.259	2.73
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (2, 0), BORDER_DEFAULT)	0.711	0.282	2.52
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (2, 0), BORDER_REFLECT101\|BORDER_ISOLATED)	0.711	0.283	2.51
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (2, 0), BORDER_REFLECT\|BORDER_ISOLATED)	0.714	0.258	2.76
sobelFilter::Size_MatType_dx_dy_Border5x5ROI::(640x480, 32FC1, (2, 0), BORDER_REPLICATE\|BORDER_ISOLATED)	0.708	0.272	2.61

modules/imgproc/src/filter.simd.hpp

Nicholas-Ho-arm · 2021-09-27T10:20:33Z

Thank you 👍

Perf results on i5-6600 (1thread, NOIPP):

Thank you very much!

Add RowVec_8u32f

b9a2833

fpetrogalli suggested changes Sep 17, 2021

View reviewed changes

modules/imgproc/src/filter.simd.hpp Outdated Show resolved Hide resolved

modules/imgproc/src/filter.simd.hpp Outdated Show resolved Hide resolved

modules/imgproc/src/filter.simd.hpp Outdated Show resolved Hide resolved

modules/imgproc/src/filter.simd.hpp Outdated Show resolved Hide resolved

Fix build errors in Linux x64 Debug and armeabi-v7a

ddfa78f

asmorkalov added optimization category: imgproc labels Sep 20, 2021

jondea reviewed Sep 20, 2021

View reviewed changes

Reformat code to make it more clean and conventional

9ff0f46

asmorkalov self-requested a review September 23, 2021 12:18

asmorkalov approved these changes Sep 23, 2021

View reviewed changes

fpetrogalli approved these changes Sep 23, 2021

View reviewed changes

jondea approved these changes Sep 23, 2021

View reviewed changes

alalek reviewed Sep 23, 2021

View reviewed changes

Optimise with vx_load_expand_q()

c2217d4

alalek approved these changes Sep 25, 2021

View reviewed changes

alalek merged commit 236c64a into opencv:3.4 Sep 25, 2021

alalek mentioned this pull request Sep 25, 2021

(4.x) Merge 3.4 #20749

Merged

fpetrogalli reviewed Sep 27, 2021

View reviewed changes

modules/imgproc/src/filter.simd.hpp Show resolved Hide resolved

alalek mentioned this pull request Oct 15, 2021

(5.x) Merge 4.x #20886

Merged

Uh oh!

Conversation

Nicholas-Ho-arm commented Sep 17, 2021 • edited by alalek Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Pull Request Readiness Checklist

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jondea left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Nicholas-Ho-arm commented Sep 22, 2021

Uh oh!

Nicholas-Ho-arm commented Sep 23, 2021

Uh oh!

alalek left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

alalek left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Nicholas-Ho-arm commented Sep 27, 2021

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Nicholas-Ho-arm commented Sep 17, 2021 •

edited by alalek

Loading

jondea left a comment •

edited

Loading

alalek left a comment •

edited

Loading