Skip to content

HAL: implemented cv_hal_dotProduct in hal_rvv#27201

Merged
asmorkalov merged 8 commits intoopencv:4.xfrom
fengyuentau:4x/hal_rvv/dotprod
Apr 21, 2025
Merged

HAL: implemented cv_hal_dotProduct in hal_rvv#27201
asmorkalov merged 8 commits intoopencv:4.xfrom
fengyuentau:4x/hal_rvv/dotprod

Conversation

@fengyuentau
Copy link
Copy Markdown
Member

@fengyuentau fengyuentau commented Apr 7, 2025

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

@fengyuentau fengyuentau added this to the 4.12.0 milestone Apr 7, 2025
@fengyuentau fengyuentau requested a review from asmorkalov April 7, 2025 06:55
@fengyuentau
Copy link
Copy Markdown
Member Author

fengyuentau commented Apr 7, 2025

Performance results on K1:

Details

K1, GCC

           Name of Test            base-gcc patch-gcc patch-gcc 
                                                          vs    
                                                       base-gcc 
                                                      (x-factor)
dot::MatType_Length::(8UC1, 32)     0.001     0.001      1.41   
dot::MatType_Length::(8UC1, 64)     0.002     0.001      1.70   
dot::MatType_Length::(8UC1, 128)    0.008     0.005      1.64   
dot::MatType_Length::(8UC1, 256)    0.029     0.024      1.22   
dot::MatType_Length::(8UC1, 512)    0.116     0.117      1.00   
dot::MatType_Length::(8UC1, 1024)   0.459     0.395      1.16   
dot::MatType_Length::(8SC1, 32)     0.001     0.001      1.42   
dot::MatType_Length::(8SC1, 64)     0.002     0.001      1.67   
dot::MatType_Length::(8SC1, 128)    0.008     0.005      1.64   
dot::MatType_Length::(8SC1, 256)    0.029     0.024      1.23   
dot::MatType_Length::(8SC1, 512)    0.119     0.108      1.10   
dot::MatType_Length::(8SC1, 1024)   0.458     0.393      1.16   
dot::MatType_Length::(16UC1, 32)    0.001     0.001      1.54   
dot::MatType_Length::(16UC1, 64)    0.004     0.002      1.74   
dot::MatType_Length::(16UC1, 128)   0.015     0.012      1.24   
dot::MatType_Length::(16UC1, 256)   0.057     0.047      1.21   
dot::MatType_Length::(16UC1, 512)   0.230     0.202      1.14   
dot::MatType_Length::(16UC1, 1024)  0.957     0.855      1.12   
dot::MatType_Length::(16SC1, 32)    0.001     0.001      1.55   
dot::MatType_Length::(16SC1, 64)    0.004     0.002      1.73   
dot::MatType_Length::(16SC1, 128)   0.016     0.012      1.25   
dot::MatType_Length::(16SC1, 256)   0.058     0.047      1.22   
dot::MatType_Length::(16SC1, 512)   0.231     0.204      1.13   
dot::MatType_Length::(16SC1, 1024)  1.023     1.328      0.77   
dot::MatType_Length::(32SC1, 32)    0.003     0.001      2.00   
dot::MatType_Length::(32SC1, 64)    0.011     0.005      2.00   
dot::MatType_Length::(32SC1, 128)   0.036     0.024      1.51   
dot::MatType_Length::(32SC1, 256)   0.148     0.109      1.35   
dot::MatType_Length::(32SC1, 512)   0.587     0.410      1.43   
dot::MatType_Length::(32SC1, 1024)  2.537     2.329      1.09   
dot::MatType_Length::(32FC1, 32)    0.001     0.001      1.12   
dot::MatType_Length::(32FC1, 64)    0.004     0.004      1.11   
dot::MatType_Length::(32FC1, 128)   0.025     0.024      1.05   
dot::MatType_Length::(32FC1, 256)   0.117     0.108      1.08   
dot::MatType_Length::(32FC1, 512)   0.405     0.399      1.01   
dot::MatType_Length::(32FC1, 1024)  1.687     1.847      0.91

K1 vs RK3568

           Name of Test             rk   patch-gcc patch-clang patch-gcc  patch-clang
                                                                   vs         vs     
                                                                   rk         rk     
                                                               (x-factor) (x-factor) 
dot::MatType_Length::(8UC1, 32)    0.001   0.001      0.001       0.96       0.97    
dot::MatType_Length::(8UC1, 64)    0.001   0.001      0.001       1.19       1.21    
dot::MatType_Length::(8UC1, 128)   0.007   0.005      0.005       1.35       1.31    
dot::MatType_Length::(8UC1, 256)   0.025   0.024      0.017       1.06       1.53    
dot::MatType_Length::(8UC1, 512)   0.113   0.117      0.116       0.97       0.98    
dot::MatType_Length::(8UC1, 1024)  0.380   0.395      0.392       0.96       0.97    
dot::MatType_Length::(8SC1, 32)    0.001   0.001      0.001       0.94       0.96    
dot::MatType_Length::(8SC1, 64)    0.001   0.001      0.001       1.17       1.20    
dot::MatType_Length::(8SC1, 128)   0.007   0.005      0.005       1.34       1.34    
dot::MatType_Length::(8SC1, 256)   0.026   0.024      0.017       1.07       1.52    
dot::MatType_Length::(8SC1, 512)   0.112   0.108      0.107       1.04       1.05    
dot::MatType_Length::(8SC1, 1024)  0.380   0.393      0.408       0.97       0.93    
dot::MatType_Length::(16UC1, 32)   0.002   0.001      0.001       2.26       2.28    
dot::MatType_Length::(16UC1, 64)   0.006   0.002      0.002       3.11       3.12    
dot::MatType_Length::(16UC1, 128)  0.025   0.012      0.009       2.03       2.84    
dot::MatType_Length::(16UC1, 256)  0.100   0.047      0.032       2.12       3.09    
dot::MatType_Length::(16UC1, 512)  0.406   0.202      0.213       2.01       1.91    
dot::MatType_Length::(16UC1, 1024) 2.324   0.855      0.870       2.72       2.67    
dot::MatType_Length::(16SC1, 32)   0.002   0.001      0.001       1.83       1.87    
dot::MatType_Length::(16SC1, 64)   0.005   0.002      0.002       2.57       2.58    
dot::MatType_Length::(16SC1, 128)  0.028   0.012      0.013       2.26       2.23    
dot::MatType_Length::(16SC1, 256)  0.085   0.047      0.047       1.80       1.80    
dot::MatType_Length::(16SC1, 512)  0.346   0.204      0.203       1.70       1.70    
dot::MatType_Length::(16SC1, 1024) 2.056   1.328      1.338       1.55       1.54    
dot::MatType_Length::(32SC1, 32)   0.002   0.001      0.001       1.78       1.76    
dot::MatType_Length::(32SC1, 64)   0.010   0.005      0.006       1.85       1.81    
dot::MatType_Length::(32SC1, 128)  0.033   0.024      0.024       1.39       1.37    
dot::MatType_Length::(32SC1, 256)  0.153   0.109      0.107       1.40       1.43    
dot::MatType_Length::(32SC1, 512)  0.537   0.410      0.398       1.31       1.35    
dot::MatType_Length::(32SC1, 1024) 3.015   2.329      2.332       1.29       1.29    
dot::MatType_Length::(32FC1, 32)   0.001   0.001      0.001       1.28       1.25    
dot::MatType_Length::(32FC1, 64)   0.006   0.004      0.004       1.60       1.59    
dot::MatType_Length::(32FC1, 128)  0.023   0.024      0.025       0.96       0.94    
dot::MatType_Length::(32FC1, 256)  0.098   0.108      0.106       0.90       0.93    
dot::MatType_Length::(32FC1, 512)  0.380   0.399      0.396       0.95       0.96    
dot::MatType_Length::(32FC1, 1024) 1.882   1.847      1.885       1.02       1.00

More: perf-dotprod.zip

@asmorkalov
Copy link
Copy Markdown
Contributor

asmorkalov commented Apr 8, 2025

Accuracy test failure on RISC-V (CI):

[ RUN      ] Core_DotProduct.accuracy
/home/ci/opencv/modules/ts/src/ts.cpp:612: Failure
Failed

	failure reason: Bad accuracy
	test case #2
	seed: 00000000000c5a5e
-----------------------------------
	LOG:
output: Too big difference (=5.2399 > 2.5e-05) at element 0
input array 0 type=32fC3, size=(191, 42)
input array 1 type=32fC3, size=(191, 42)
ref output array 0 type=64fC1, size=(1, 4)
test_case_idx = 2

-----------------------------------

[  FAILED  ] Core_DotProduct.accuracy (8 ms)

@asmorkalov
Copy link
Copy Markdown
Contributor

asmorkalov commented Apr 8, 2025

Please rebase and fix conflicts.

@fengyuentau
Copy link
Copy Markdown
Member Author

fengyuentau commented Apr 8, 2025

Accuracy issues are fixed. Updated performance results:

Details

K1, GCC

           Name of Test            base-gcc patch-gcc patch-gcc 
                                                          vs    
                                                       base-gcc 
                                                      (x-factor)
dot::MatType_Length::(8UC1, 32)     0.001     0.001      1.42   
dot::MatType_Length::(8UC1, 64)     0.002     0.001      1.69   
dot::MatType_Length::(8UC1, 128)    0.008     0.005      1.64   
dot::MatType_Length::(8UC1, 256)    0.029     0.017      1.74   
dot::MatType_Length::(8UC1, 512)    0.116     0.122      0.96   
dot::MatType_Length::(8UC1, 1024)   0.459     0.395      1.16   
dot::MatType_Length::(8SC1, 32)     0.001     0.001      1.52   
dot::MatType_Length::(8SC1, 64)     0.002     0.001      1.68   
dot::MatType_Length::(8SC1, 128)    0.008     0.005      1.62   
dot::MatType_Length::(8SC1, 256)    0.029     0.017      1.77   
dot::MatType_Length::(8SC1, 512)    0.119     0.119      1.01   
dot::MatType_Length::(8SC1, 1024)   0.458     0.416      1.10   
dot::MatType_Length::(16UC1, 32)    0.001     0.001      1.56   
dot::MatType_Length::(16UC1, 64)    0.004     0.002      1.76   
dot::MatType_Length::(16UC1, 128)   0.015     0.009      1.68   
dot::MatType_Length::(16UC1, 256)   0.057     0.032      1.78   
dot::MatType_Length::(16UC1, 512)   0.230     0.212      1.08   
dot::MatType_Length::(16UC1, 1024)  0.957     0.886      1.08   
dot::MatType_Length::(16SC1, 32)    0.001     0.001      1.57   
dot::MatType_Length::(16SC1, 64)    0.004     0.002      1.78   
dot::MatType_Length::(16SC1, 128)   0.016     0.013      1.22   
dot::MatType_Length::(16SC1, 256)   0.058     0.047      1.22   
dot::MatType_Length::(16SC1, 512)   0.231     0.211      1.09   
dot::MatType_Length::(16SC1, 1024)  1.023     1.292      0.79   
dot::MatType_Length::(32SC1, 32)    0.003     0.001      2.01   
dot::MatType_Length::(32SC1, 64)    0.011     0.006      1.94   
dot::MatType_Length::(32SC1, 128)   0.036     0.024      1.51   
dot::MatType_Length::(32SC1, 256)   0.148     0.116      1.27   
dot::MatType_Length::(32SC1, 512)   0.587     0.396      1.48   
dot::MatType_Length::(32SC1, 1024)  2.537     2.397      1.06   
dot::MatType_Length::(32FC1, 32)    0.001     0.001      1.14   
dot::MatType_Length::(32FC1, 64)    0.004     0.004      1.10   
dot::MatType_Length::(32FC1, 128)   0.025     0.026      0.98   
dot::MatType_Length::(32FC1, 256)   0.117     0.113      1.04   
dot::MatType_Length::(32FC1, 512)   0.405     0.395      1.02   
dot::MatType_Length::(32FC1, 1024)  1.687     1.899      0.89

K1 vs RK3568

           Name of Test             rk   patch-gcc patch-clang patch-gcc  patch-clang
                                                                   vs         vs     
                                                                   rk         rk     
                                                               (x-factor) (x-factor) 
dot::MatType_Length::(8UC1, 32)    0.001   0.001      0.001       0.97       1.01    
dot::MatType_Length::(8UC1, 64)    0.001   0.001      0.001       1.18       1.20    
dot::MatType_Length::(8UC1, 128)   0.007   0.005      0.005       1.36       1.37    
dot::MatType_Length::(8UC1, 256)   0.025   0.017      0.024       1.52       1.06    
dot::MatType_Length::(8UC1, 512)   0.113   0.122      0.110       0.93       1.03    
dot::MatType_Length::(8UC1, 1024)  0.380   0.395      0.426       0.96       0.89    
dot::MatType_Length::(8SC1, 32)    0.001   0.001      0.001       1.00       0.98    
dot::MatType_Length::(8SC1, 64)    0.001   0.001      0.001       1.17       1.17    
dot::MatType_Length::(8SC1, 128)   0.007   0.005      0.005       1.33       1.33    
dot::MatType_Length::(8SC1, 256)   0.026   0.017      0.024       1.53       1.06    
dot::MatType_Length::(8SC1, 512)   0.112   0.119      0.113       0.95       1.00    
dot::MatType_Length::(8SC1, 1024)  0.380   0.416      0.427       0.91       0.89    
dot::MatType_Length::(16UC1, 32)   0.002   0.001      0.001       2.30       2.32    
dot::MatType_Length::(16UC1, 64)   0.006   0.002      0.002       3.15       3.14    
dot::MatType_Length::(16UC1, 128)  0.025   0.009      0.012       2.75       2.03    
dot::MatType_Length::(16UC1, 256)  0.100   0.032      0.047       3.10       2.10    
dot::MatType_Length::(16UC1, 512)  0.406   0.212      0.206       1.91       1.98    
dot::MatType_Length::(16UC1, 1024) 2.324   0.886      0.849       2.62       2.74    
dot::MatType_Length::(16SC1, 32)   0.002   0.001      0.001       1.86       1.85    
dot::MatType_Length::(16SC1, 64)   0.005   0.002      0.002       2.65       2.63    
dot::MatType_Length::(16SC1, 128)  0.028   0.013      0.011       2.21       2.58    
dot::MatType_Length::(16SC1, 256)  0.085   0.047      0.052       1.80       1.65    
dot::MatType_Length::(16SC1, 512)  0.346   0.211      0.215       1.64       1.61    
dot::MatType_Length::(16SC1, 1024) 2.056   1.292      1.158       1.59       1.78    
dot::MatType_Length::(32SC1, 32)   0.002   0.001      0.001       1.79       1.81    
dot::MatType_Length::(32SC1, 64)   0.010   0.006      0.005       1.80       1.86    
dot::MatType_Length::(32SC1, 128)  0.033   0.024      0.018       1.39       1.88    
dot::MatType_Length::(32SC1, 256)  0.153   0.116      0.109       1.32       1.40    
dot::MatType_Length::(32SC1, 512)  0.537   0.396      0.397       1.35       1.35    
dot::MatType_Length::(32SC1, 1024) 3.015   2.397      2.306       1.26       1.31    
dot::MatType_Length::(32FC1, 32)   0.001   0.001      0.001       1.29       1.29    
dot::MatType_Length::(32FC1, 64)   0.006   0.004      0.004       1.59       1.57    
dot::MatType_Length::(32FC1, 128)  0.023   0.026      0.025       0.89       0.94    
dot::MatType_Length::(32FC1, 256)  0.098   0.113      0.115       0.87       0.85    
dot::MatType_Length::(32FC1, 512)  0.380   0.395      0.396       0.96       0.96    
dot::MatType_Length::(32FC1, 1024) 1.882   1.899      1.955       0.99       0.96

More:
perf-dotprod.zip

@asmorkalov asmorkalov self-assigned this Apr 10, 2025
@asmorkalov asmorkalov merged commit 11e46cd into opencv:4.x Apr 21, 2025
53 of 55 checks passed
@asmorkalov asmorkalov added the backport is needed Label for maintainers. Authors of PR can ignore this label Apr 21, 2025
@fengyuentau fengyuentau deleted the 4x/hal_rvv/dotprod branch April 22, 2025 08:04
@asmorkalov asmorkalov mentioned this pull request Apr 29, 2025
Merged
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@asmorkalov asmorkalov asmorkalov approved these changes

Assignees

@asmorkalov asmorkalov

Labels

backport is needed Label for maintainers. Authors of PR can ignore this category: core optimization platform: riscv

Projects

OpenCV 4.x HAL improvement
Status: Done

Milestone

4.12.0

Development

Successfully merging this pull request may close these issues.

2 participants

@fengyuentau @asmorkalov