Enable SIMD_SCALABLE for exp and sqrt by sk1er52 · Pull Request #26886 · opencv/opencv

sk1er52 · 2025-02-07T13:31:26Z

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 another license that is incompatible with OpenCV
The PR is proposed to the proper branch
There is a reference to the 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

CPU - Banana Pi k1, compiler - clang 18.1.4

Geometric mean (ms)

              Name of Test               baseline  hal     ui      hal         ui    
                                                                    vs         vs
                                                                 baseline   baseline
                                                                (x-factor) (x-factor)
Exp::ExpFixture::(127x61, 32FC1)          0.358     --   0.033      --       10.70   
Exp::ExpFixture::(640x480, 32FC1)         14.304    --   1.167      --       12.26   
Exp::ExpFixture::(1280x720, 32FC1)        42.785    --   3.538      --       12.09
Exp::ExpFixture::(1920x1080, 32FC1)       96.206    --   7.927      --       12.14   
Exp::ExpFixture::(127x61, 64FC1)          0.433   0.050  0.098     8.59       4.40   
Exp::ExpFixture::(640x480, 64FC1)         17.315  1.935  3.813     8.95       4.54   
Exp::ExpFixture::(1280x720, 64FC1)        52.181  5.877  11.519    8.88       4.53   
Exp::ExpFixture::(1920x1080, 64FC1)      117.082  13.157 25.854    8.90       4.53

Additionally, this PR brings Sqrt optimization with UI:

Geometric mean (ms)

              Name of Test                     baseline    ui       ui    
                                                                    vs
                                                                 baseline
                                                                (x-factor)
Sqrt::SqrtFixture::(127x61, 5, false)            0.111   0.027     4.11   
Sqrt::SqrtFixture::(127x61, 6, false)            0.149   0.053     2.82   
Sqrt::SqrtFixture::(640x480, 5, false)           4.374   0.967     4.52   
Sqrt::SqrtFixture::(640x480, 6, false)           5.885   2.046     2.88   
Sqrt::SqrtFixture::(1280x720, 5, false)          12.960  2.915     4.45   
Sqrt::SqrtFixture::(1280x720, 6, false)          17.648  6.107     2.89   
Sqrt::SqrtFixture::(1920x1080, 5, false)         29.178  6.524     4.47   
Sqrt::SqrtFixture::(1920x1080, 6, false)         39.709  13.670    2.90

Reference
Muller, J.-M. Elementary Functions: Algorithms and Implementation. 2nd ed. Boston: Birkhäuser, 2006.
https://www.springer.com/gp/book/9780817643720

3rdparty/hal_rvv/hal_rvv_1p0/exp.hpp

asmorkalov · 2025-02-08T06:46:30Z

We have v_exp universal intrinsic introduced in #25881. I propose to integrated it to core first.

Build with Syntacore Clang Run on self-hosted Disable OpenCL summary Run baseline filter from PR body

dkurt · 2025-02-20T16:08:36Z

By changing m8 to m4 performance does not change significantly:

Geometric mean (ms)

              Name of Test                hal    hal      hal    
                                           m8     m4       m4
                                                           vs
                                                          hal
                                                           m8
                                                       (x-factor)
Exp::ExpFixture::(127x61, 64FC1)         0.050  0.051     0.99
Exp::ExpFixture::(640x480, 64FC1)        1.935  1.979     0.98
Exp::ExpFixture::(1280x720, 64FC1)       5.877  5.979     0.98   
Exp::ExpFixture::(1920x1080, 64FC1)      13.157 13.294    0.99

So the main question in algo part. Despite tests passed, vfmacc remain a critical part because the order of accumulation may give different results.

sk1er52 · 2025-02-20T16:12:06Z

regarding the algorithmic part:
I'm still studying the algorithm from the book Muller J. M., Muller J. M. Elementary functions, I think I'll fix it soon

dkurt

@sk1er52, to give you a credit for a done work, I recommend to merge this PR with v_exp universal intrinsics enabled. Let's keep this PR or separate issue for the future discussion. Potential x2 performance improvement should be analyzed more precisely.

dkurt · 2025-02-20T16:17:10Z

I'm still studying the algorithm from the book Muller J. M., Muller J. M. Elementary functions, I think I'll fix it soon

Please add a link to the article in PR description

asmorkalov · 2025-02-21T14:34:58Z

My perf results for Muse Pi v 30 (GCC 14.2):

Exp::ExpFixture::(127x61, 32FC1) 	0.310 	0.038 	8.20
Exp::ExpFixture::(127x61, 64FC1) 	0.368 	0.091 	4.03
Exp::ExpFixture::(640x480, 32FC1) 	12.412 	1.324 	9.37
Exp::ExpFixture::(640x480, 64FC1) 	15.071 	3.664 	4.11
Exp::ExpFixture::(1280x720, 32FC1) 	37.106 	3.999 	9.28
Exp::ExpFixture::(1280x720, 64FC1) 	45.205 	10.493 	4.31
Exp::ExpFixture::(1920x1080, 32FC1) 	83.658 	8.976 	9.32
Exp::ExpFixture::(1920x1080, 64FC1) 	100.830 	23.518 	4.29

Sqrt::SqrtFixture::(127x61, 5, false) 	0.182 	0.028 	6.60
Sqrt::SqrtFixture::(127x61, 5, true) 	0.050 	0.052 	0.98
Sqrt::SqrtFixture::(127x61, 6, false) 	0.220 	0.053 	4.12
Sqrt::SqrtFixture::(127x61, 6, true) 	0.102 	0.102 	0.99
Sqrt::SqrtFixture::(640x480, 5, false) 	7.328 	0.967 	7.58
Sqrt::SqrtFixture::(640x480, 5, true) 	1.883 	1.904 	0.99
Sqrt::SqrtFixture::(640x480, 6, false) 	9.098 	2.246 	4.05
Sqrt::SqrtFixture::(640x480, 6, true) 	3.978 	4.217 	0.94
Sqrt::SqrtFixture::(1280x720, 5, false) 	22.062 	2.921 	7.55
Sqrt::SqrtFixture::(1280x720, 5, true) 	5.691 	5.753 	0.99
Sqrt::SqrtFixture::(1280x720, 6, false) 	27.224 	6.110 	4.46
Sqrt::SqrtFixture::(1280x720, 6, true) 	11.999 	12.023 	1.00
Sqrt::SqrtFixture::(1920x1080, 5, false) 	50.162 	6.575 	7.63
Sqrt::SqrtFixture::(1920x1080, 5, true) 	12.810 	12.946 	0.99
Sqrt::SqrtFixture::(1920x1080, 6, false) 	61.132 	13.722 	4.45
Sqrt::SqrtFixture::(1920x1080, 6, true) 	26.906 	27.025 	1.00

Enable SIMD_SCALABLE for exp and sqrt opencv#26886 ### Pull Request Readiness Checklist See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request - [x] I agree to contribute to the project under Apache 2 License. - [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV - [x] The PR is proposed to the proper branch - [x] There is a reference to the original bug report and related work - [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable Patch to opencv_extra has the same branch name. - [x] The feature is well documented and sample code can be built with the project CMake ``` CPU - Banana Pi k1, compiler - clang 18.1.4 ``` ``` Geometric mean (ms) Name of Test baseline hal ui hal ui vs vs baseline baseline (x-factor) (x-factor) Exp::ExpFixture::(127x61, 32FC1) 0.358 -- 0.033 -- 10.70 Exp::ExpFixture::(640x480, 32FC1) 14.304 -- 1.167 -- 12.26 Exp::ExpFixture::(1280x720, 32FC1) 42.785 -- 3.538 -- 12.09 Exp::ExpFixture::(1920x1080, 32FC1) 96.206 -- 7.927 -- 12.14 Exp::ExpFixture::(127x61, 64FC1) 0.433 0.050 0.098 8.59 4.40 Exp::ExpFixture::(640x480, 64FC1) 17.315 1.935 3.813 8.95 4.54 Exp::ExpFixture::(1280x720, 64FC1) 52.181 5.877 11.519 8.88 4.53 Exp::ExpFixture::(1920x1080, 64FC1) 117.082 13.157 25.854 8.90 4.53 ``` Additionally, this PR brings Sqrt optimization with UI: ``` Geometric mean (ms) Name of Test baseline ui ui vs baseline (x-factor) Sqrt::SqrtFixture::(127x61, 5, false) 0.111 0.027 4.11 Sqrt::SqrtFixture::(127x61, 6, false) 0.149 0.053 2.82 Sqrt::SqrtFixture::(640x480, 5, false) 4.374 0.967 4.52 Sqrt::SqrtFixture::(640x480, 6, false) 5.885 2.046 2.88 Sqrt::SqrtFixture::(1280x720, 5, false) 12.960 2.915 4.45 Sqrt::SqrtFixture::(1280x720, 6, false) 17.648 6.107 2.89 Sqrt::SqrtFixture::(1920x1080, 5, false) 29.178 6.524 4.47 Sqrt::SqrtFixture::(1920x1080, 6, false) 39.709 13.670 2.90 ``` Reference Muller, J.-M. Elementary Functions: Algorithms and Implementation. 2nd ed. Boston: Birkhäuser, 2006. https://www.springer.com/gp/book/9780817643720

dkurt marked this pull request as draft February 7, 2025 13:34

dkurt self-assigned this Feb 7, 2025

dkurt added optimization platform: riscv labels Feb 7, 2025

asmorkalov requested a review from vpisarev February 7, 2025 14:13

dkurt reviewed Feb 7, 2025

View reviewed changes

3rdparty/hal_rvv/hal_rvv_1p0/exp.hpp Outdated Show resolved Hide resolved

dkurt reviewed Feb 7, 2025

View reviewed changes

3rdparty/hal_rvv/hal_rvv_1p0/exp.hpp Outdated Show resolved Hide resolved

dkurt reviewed Feb 7, 2025

View reviewed changes

3rdparty/hal_rvv/hal_rvv_1p0/exp.hpp Outdated Show resolved Hide resolved

dkurt and others added 5 commits February 20, 2025 18:43

Cross compile

bd1919d

Build with Syntacore Clang Run on self-hosted Disable OpenCL summary Run baseline filter from PR body

ver 1.0

3667c8b

fix perf test

c314889

SIMD_SCALABLE for sqrt

d78a6c1

exp SIMD_SCALABLE

af22e47

dkurt force-pushed the feature/exp64f branch from 7af263c to af22e47 Compare February 20, 2025 15:55

dkurt marked this pull request as ready for review February 20, 2025 15:59

dkurt marked this pull request as draft February 20, 2025 15:59

sk1er52 closed this Feb 20, 2025

sk1er52 reopened this Feb 20, 2025

Keep only UI for faster integration

952aecb

dkurt approved these changes Feb 20, 2025

View reviewed changes

dkurt changed the title ~~[HAL] exp64f rvv1.0~~ Enable SIMD_SCALABLE for exp and sqrt Feb 20, 2025

fengyuentau self-requested a review February 20, 2025 17:42

dkurt marked this pull request as ready for review February 21, 2025 06:48

asmorkalov added this to the 4.12.0 milestone Feb 21, 2025

asmorkalov approved these changes Feb 21, 2025

View reviewed changes

asmorkalov merged commit b5f5540 into opencv:4.x Feb 21, 2025
28 checks passed

opencv-alalek added the category: core label Feb 21, 2025

This was referenced Mar 4, 2025

5.x merge 4.x #27009

Merged

[HAL RVV] impl exp and log | add log perf test #27010

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Enable SIMD_SCALABLE for exp and sqrt#26886

Enable SIMD_SCALABLE for exp and sqrt#26886
asmorkalov merged 6 commits intoopencv:4.xfrom
sk1er52:feature/exp64f

sk1er52 commented Feb 7, 2025 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

asmorkalov commented Feb 8, 2025

Uh oh!

dkurt commented Feb 20, 2025

Uh oh!

sk1er52 commented Feb 20, 2025 •

edited

Loading

Uh oh!

dkurt left a comment

Uh oh!

dkurt commented Feb 20, 2025

Uh oh!

asmorkalov commented Feb 21, 2025 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Uh oh!

Conversation

sk1er52 commented Feb 7, 2025 • edited 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!

asmorkalov commented Feb 8, 2025

Uh oh!

dkurt commented Feb 20, 2025

Uh oh!

sk1er52 commented Feb 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dkurt left a comment

Choose a reason for hiding this comment

Uh oh!

dkurt commented Feb 20, 2025

Uh oh!

asmorkalov commented Feb 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

sk1er52 commented Feb 7, 2025 •

edited

Loading

sk1er52 commented Feb 20, 2025 •

edited

Loading

asmorkalov commented Feb 21, 2025 •

edited

Loading