Skip to content

opt(RVV): Optimize CV color conversion functions with intrinsics#4079

Merged
wangzhaode merged 2 commits intoalibaba:masterfrom
ihb2032:opt/rvv-c3-nv21
Dec 26, 2025
Merged

opt(RVV): Optimize CV color conversion functions with intrinsics#4079
wangzhaode merged 2 commits intoalibaba:masterfrom
ihb2032:opt/rvv-c3-nv21

Conversation

@ihb2032
Copy link
Contributor

@ihb2032 ihb2032 commented Dec 23, 2025

Summary

Optimize MNNC3 and MNNNV21 related color conversion functions using RVV intrinsics, including C3 to YUV/XYZ/HSV/BGR555/BGR565 and NV21 to RGBA/RGB/BGRA/BGR.

Environment

  • Platform: Banana PI BPI-F3
  • OS: EulixOS 3.0

Benchmark

Click to expand full test logs 
root@yangwang:~/mnntest# ./test_c3_to_yuv
number=1
Scalar time: 0.000003 sec
RVV time    : 0.000022 sec
Speedup     : 0.13x
Test number=1: PASSED
number=3
Scalar time: 0.000004 sec
RVV time    : 0.000007 sec
Speedup     : 0.57x
Test number=3: PASSED
number=8
Scalar time: 0.000005 sec
RVV time    : 0.000006 sec
Speedup     : 0.84x
Test number=8: PASSED
number=33
Scalar time: 0.000019 sec
RVV time    : 0.000005 sec
Speedup     : 3.81x
Test number=33: PASSED
number=100
Scalar time: 0.000051 sec
RVV time    : 0.000009 sec
Speedup     : 5.76x
Test number=100: PASSED
number=1024
Scalar time: 0.000675 sec
RVV time    : 0.000071 sec
Speedup     : 9.50x
Test number=1024: PASSED
number=65536
Scalar time: 0.033821 sec
RVV time    : 0.004405 sec
Speedup     : 7.68x
Test number=65536: PASSED
number=1000000
Scalar time: 0.517025 sec
RVV time    : 0.066878 sec
Speedup     : 7.73x
Test number=1000000: PASSED
number=10000000
Scalar time: 1.987096 sec
RVV time    : 0.262969 sec
Speedup     : 7.56x
Test number=10000000: PASSED
number=50000000
Scalar time: 9.916534 sec
RVV time    : 1.314206 sec
Speedup     : 7.55x
Test number=50000000: PASSED

All tests PASSED 
root@yangwang:~# ./test_c3_to_xyz
number=1
Scalar time: 0.000001 sec
RVV time    : 0.000012 sec
Speedup     : 0.08x
Test number=1: PASSED
number=3
Scalar time: 0.000001 sec
RVV time    : 0.000002 sec
Speedup     : 0.50x
Test number=3: PASSED
number=8
Scalar time: 0.000001 sec
RVV time    : 0.000002 sec
Speedup     : 0.62x
Test number=8: PASSED
number=33
Scalar time: 0.000004 sec
RVV time    : 0.000002 sec
Speedup     : 2.12x
Test number=33: PASSED
number=100
Scalar time: 0.000012 sec
RVV time    : 0.000004 sec
Speedup     : 2.94x
Test number=100: PASSED
number=1024
Scalar time: 0.000122 sec
RVV time    : 0.000027 sec
Speedup     : 4.53x
Test number=1024: PASSED
number=65536
Scalar time: 0.007678 sec
RVV time    : 0.001646 sec
Speedup     : 4.67x
Test number=65536: PASSED
number=1000000
Scalar time: 0.118545 sec
RVV time    : 0.024820 sec
Speedup     : 4.78x
Test number=1000000: PASSED
number=10000000
Scalar time: 1.179671 sec
RVV time    : 0.249469 sec
Speedup     : 4.73x
Test number=10000000: PASSED
number=50000000
Scalar time: 5.890856 sec
RVV time    : 1.239900 sec
Speedup     : 4.75x
Test number=50000000: PASSED

All tests PASSED 
[root@EulixOS ~]# ./test_c3_to_bgr555
[RGB] count=1 | Scalar: 0.000001 s | RVV: 0.000017 s | Speedup: 0.06x
[BGR] count=1 | Scalar: 0.000000 s | RVV: 0.000001 s | Speedup: 0.00x
Test size=1 PASSED

[RGB] count=3 | Scalar: 0.000000 s | RVV: 0.000001 s | Speedup: 0.00x
[BGR] count=3 | Scalar: 0.000001 s | RVV: 0.000001 s | Speedup: 0.80x
Test size=3 PASSED

[RGB] count=8 | Scalar: 0.000000 s | RVV: 0.000001 s | Speedup: 0.00x
[BGR] count=8 | Scalar: 0.000001 s | RVV: 0.000001 s | Speedup: 1.25x
Test size=8 PASSED

[RGB] count=15 | Scalar: 0.000001 s | RVV: 0.000001 s | Speedup: 1.00x
[BGR] count=15 | Scalar: 0.000001 s | RVV: 0.000001 s | Speedup: 1.00x
Test size=15 PASSED

[RGB] count=31 | Scalar: 0.000001 s | RVV: 0.000001 s | Speedup: 1.00x
[BGR] count=31 | Scalar: 0.000002 s | RVV: 0.000001 s | Speedup: 2.00x
Test size=31 PASSED

[RGB] count=32 | Scalar: 0.000001 s | RVV: 0.000001 s | Speedup: 1.00x
[BGR] count=32 | Scalar: 0.000002 s | RVV: 0.000001 s | Speedup: 1.60x
Test size=32 PASSED

[RGB] count=33 | Scalar: 0.000001 s | RVV: 0.000001 s | Speedup: 1.00x
[BGR] count=33 | Scalar: 0.000002 s | RVV: 0.000001 s | Speedup: 2.00x
Test size=33 PASSED

[RGB] count=100 | Scalar: 0.000004 s | RVV: 0.000001 s | Speedup: 4.25x
[BGR] count=100 | Scalar: 0.000005 s | RVV: 0.000001 s | Speedup: 4.20x
Test size=100 PASSED

[RGB] count=1024 | Scalar: 0.000037 s | RVV: 0.000007 s | Speedup: 5.17x
[BGR] count=1024 | Scalar: 0.000056 s | RVV: 0.000006 s | Speedup: 9.40x
Test size=1024 PASSED

[RGB] count=65536 | Scalar: 0.002423 s | RVV: 0.000417 s | Speedup: 5.81x
[BGR] count=65536 | Scalar: 0.003773 s | RVV: 0.000398 s | Speedup: 9.48x
Test size=65536 PASSED

[RGB] count=1000000 | Scalar: 0.037708 s | RVV: 0.006110 s | Speedup: 6.17x
[BGR] count=1000000 | Scalar: 0.054990 s | RVV: 0.006060 s | Speedup: 9.07x
Test size=1000000 PASSED

[RGB] count=10000000 | Scalar: 0.368202 s | RVV: 0.061179 s | Speedup: 6.02x
[BGR] count=10000000 | Scalar: 0.544550 s | RVV: 0.060702 s | Speedup: 8.97x
Test size=10000000 PASSED

[RGB] count=50000000 | Scalar: 1.830795 s | RVV: 0.304496 s | Speedup: 6.01x
[BGR] count=50000000 | Scalar: 2.722448 s | RVV: 0.303552 s | Speedup: 8.97x
Test size=50000000 PASSED

Final Result: ALL TESTS PASSED 
[root@EulixOS ~]# ./test_c3_to_hsv
number=1
Scalar time: 0.000004 sec
RVV time    : 0.000029 sec
Speedup     : 0.14x
Test number=1: PASSED
number=3
Scalar time: 0.000002 sec
RVV time    : 0.000003 sec
Speedup     : 0.75x
Test number=3: PASSED
number=8
Scalar time: 0.000004 sec
RVV time    : 0.000003 sec
Speedup     : 1.42x
Test number=8: PASSED
number=33
Scalar time: 0.000016 sec
RVV time    : 0.000002 sec
Speedup     : 8.38x
Test number=33: PASSED
number=100
Scalar time: 0.000045 sec
RVV time    : 0.000005 sec
Speedup     : 9.00x
Test number=100: PASSED
number=1024
Scalar time: 0.000457 sec
RVV time    : 0.000038 sec
Speedup     : 12.06x
Test number=1024: PASSED
number=65536
Scalar time: 0.028634 sec
RVV time    : 0.002295 sec
Speedup     : 12.48x
Test number=65536: PASSED
number=1000000
Scalar time: 0.432992 sec
RVV time    : 0.034932 sec
Speedup     : 12.40x
Test number=1000000: PASSED
number=10000000
Scalar time: 4.342924 sec
RVV time    : 0.348972 sec
Speedup     : 12.44x
Test number=10000000: PASSED
number=50000000
Scalar time: 21.713869 sec
RVV time    : 1.750756 sec
Speedup     : 12.40x
Test number=50000000: PASSED

All tests PASSED 
[root@EulixOS ~]# ./test_c3_to_bgr565
[RGB] count=1 | Scalar: 0.000000 s | RVV: 0.000019 s | Speedup: 0.00x
[BGR] count=1 | Scalar: 0.000000 s | RVV: 0.000001 s | Speedup: 0.00x
Test size=1 PASSED

[RGB] count=3 | Scalar: 0.000001 s | RVV: 0.000001 s | Speedup: 1.25x
[BGR] count=3 | Scalar: 0.000001 s | RVV: 0.000001 s | Speedup: 1.00x
Test size=3 PASSED

[RGB] count=8 | Scalar: 0.000000 s | RVV: 0.000001 s | Speedup: 0.00x
[BGR] count=8 | Scalar: 0.000000 s | RVV: 0.000001 s | Speedup: 0.00x
Test size=8 PASSED

[RGB] count=15 | Scalar: 0.000001 s | RVV: 0.000001 s | Speedup: 1.00x
[BGR] count=15 | Scalar: 0.000002 s | RVV: 0.000001 s | Speedup: 2.00x
Test size=15 PASSED

[RGB] count=31 | Scalar: 0.000001 s | RVV: 0.000001 s | Speedup: 1.00x
[BGR] count=31 | Scalar: 0.000002 s | RVV: 0.000001 s | Speedup: 2.00x
Test size=31 PASSED

[RGB] count=32 | Scalar: 0.000002 s | RVV: 0.000001 s | Speedup: 2.00x
[BGR] count=32 | Scalar: 0.000002 s | RVV: 0.000001 s | Speedup: 2.00x
Test size=32 PASSED

[RGB] count=33 | Scalar: 0.000002 s | RVV: 0.000001 s | Speedup: 2.00x
[BGR] count=33 | Scalar: 0.000002 s | RVV: 0.000000 s | Speedup: 0.00x
Test size=33 PASSED

[RGB] count=100 | Scalar: 0.000004 s | RVV: 0.000001 s | Speedup: 4.25x
[BGR] count=100 | Scalar: 0.000006 s | RVV: 0.000001 s | Speedup: 6.25x
Test size=100 PASSED

[RGB] count=1024 | Scalar: 0.000037 s | RVV: 0.000007 s | Speedup: 5.34x
[BGR] count=1024 | Scalar: 0.000056 s | RVV: 0.000007 s | Speedup: 7.83x
Test size=1024 PASSED

[RGB] count=65536 | Scalar: 0.002868 s | RVV: 0.000410 s | Speedup: 6.99x
[BGR] count=65536 | Scalar: 0.004017 s | RVV: 0.000426 s | Speedup: 9.43x
Test size=65536 PASSED

[RGB] count=1000000 | Scalar: 0.037677 s | RVV: 0.005883 s | Speedup: 6.40x
[BGR] count=1000000 | Scalar: 0.055934 s | RVV: 0.005829 s | Speedup: 9.60x
Test size=1000000 PASSED

[RGB] count=10000000 | Scalar: 0.373215 s | RVV: 0.058184 s | Speedup: 6.41x
[BGR] count=10000000 | Scalar: 0.564888 s | RVV: 0.062682 s | Speedup: 9.01x
Test size=10000000 PASSED

[RGB] count=50000000 | Scalar: 1.861281 s | RVV: 0.290735 s | Speedup: 6.40x
[BGR] count=50000000 | Scalar: 2.821005 s | RVV: 0.290990 s | Speedup: 9.69x
Test size=50000000 PASSED

Final Result: ALL TESTS PASSED 
[root@EulixOS ~]# ./test_nv21_to_rgba
count=1
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 0.06x
Test count=1: PASSED
count=3
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 0.33x
Test count=3: PASSED
count=8
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 2.00x
Test count=8: PASSED
count=33
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 7.50x
Test count=33: PASSED
count=100
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 6.46x
Test count=100: PASSED
count=1024
Scalar time: 0.0002 sec
RVV time    : 0.0000 sec
Speedup     : 8.68x
Test count=1024: PASSED
count=65536
Scalar time: 0.0131 sec
RVV time    : 0.0016 sec
Speedup     : 8.39x
Test count=65536: PASSED
count=1000000
Scalar time: 0.2007 sec
RVV time    : 0.0227 sec
Speedup     : 8.85x
Test count=1000000: PASSED
count=10000000
Scalar time: 1.9936 sec
RVV time    : 0.2274 sec
Speedup     : 8.77x
Test count=10000000: PASSED
count=50000000
Scalar time: 9.9450 sec
RVV time    : 1.1195 sec
Speedup     : 8.88x
Test count=50000000: PASSED

All tests PASSED 
[root@EulixOS ~]# ./test_nv21_to_rgb
count=1
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 0.11x
Test count=1: PASSED
count=3
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 0.33x
Test count=3: PASSED
count=8
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 2.00x
Test count=8: PASSED
count=33
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 2.78x
Test count=33: PASSED
count=100
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 6.46x
Test count=100: PASSED
count=1024
Scalar time: 0.0002 sec
RVV time    : 0.0000 sec
Speedup     : 8.96x
Test count=1024: PASSED
count=65536
Scalar time: 0.0130 sec
RVV time    : 0.0014 sec
Speedup     : 9.01x
Test count=65536: PASSED
count=1000000
Scalar time: 0.1984 sec
RVV time    : 0.0213 sec
Speedup     : 9.32x
Test count=1000000: PASSED
count=10000000
Scalar time: 1.9873 sec
RVV time    : 0.2127 sec
Speedup     : 9.34x
Test count=10000000: PASSED
count=50000000
Scalar time: 9.9288 sec
RVV time    : 1.0647 sec
Speedup     : 9.33x
Test count=50000000: PASSED

All tests PASSED 
[root@EulixOS ~]# ./test_nv21_to_bgra
count=1
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 0.10x
Test count=1: PASSED
count=3
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 0.31x
Test count=3: PASSED
count=8
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 1.12x
Test count=8: PASSED
count=33
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 3.62x
Test count=33: PASSED
count=100
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 6.77x
Test count=100: PASSED
count=1024
Scalar time: 0.0002 sec
RVV time    : 0.0000 sec
Speedup     : 9.07x
Test count=1024: PASSED
count=65536
Scalar time: 0.0134 sec
RVV time    : 0.0015 sec
Speedup     : 8.76x
Test count=65536: PASSED
count=1000000
Scalar time: 0.2019 sec
RVV time    : 0.0227 sec
Speedup     : 8.91x
Test count=1000000: PASSED
count=10000000
Scalar time: 2.0204 sec
RVV time    : 0.2268 sec
Speedup     : 8.91x
Test count=10000000: PASSED
count=50000000
Scalar time: 10.0888 sec
RVV time    : 1.1324 sec
Speedup     : 8.91x
Test count=50000000: PASSED

All tests PASSED 
[root@EulixOS ~]# ./test_nv21_to_bgr
count=1
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 0.12x
Test count=1: PASSED
count=3
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 0.42x
Test count=3: PASSED
count=8
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 0.31x
Test count=8: PASSED
count=33
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 3.22x
Test count=33: PASSED
count=100
Scalar time: 0.0000 sec
RVV time    : 0.0000 sec
Speedup     : 6.77x
Test count=100: PASSED
count=1024
Scalar time: 0.0002 sec
RVV time    : 0.0000 sec
Speedup     : 8.19x
Test count=1024: PASSED
count=65536
Scalar time: 0.0134 sec
RVV time    : 0.0014 sec
Speedup     : 9.27x
Test count=65536: PASSED
count=1000000
Scalar time: 0.2010 sec
RVV time    : 0.0214 sec
Speedup     : 9.38x
Test count=1000000: PASSED
count=10000000
Scalar time: 1.9991 sec
RVV time    : 0.2125 sec
Speedup     : 9.41x
Test count=10000000: PASSED
count=50000000
Scalar time: 10.0013 sec
RVV time    : 1.0601 sec
Speedup     : 9.43x
Test count=50000000: PASSED

All tests PASSED

</details>

@lyd1992
opt(RVV): Optimize CV color conversion functions with intrinsics
73bfaa4 
Optimize MNNC3 and MNNNV21 related color conversion functions using RVV intrinsics, including C3 to YUV/XYZ/HSV/BGR555/BGR565 and NV21 to RGBA/RGB/BGRA/BGR.

Signed-off-by: ihb2032 <1355790728@qq.com>
Co-authored-by: lyd1992 <liuyudong@iscas.ac.cn>
@ihb2032 ihb2032 mentioned this pull request Dec 23, 2025
Closed
@wangzhaode wangzhaode self-assigned this Dec 26, 2025
@wangzhaode wangzhaode merged commit 07bda85 into alibaba:master Dec 26, 2025
6 checks passed
wangzhaode added a commit that referenced this pull request Dec 26, 2025
@wangzhaode
Merge pull request #4079 from ihb2032/opt/rvv-c3-nv21
c6ea9c4 
opt(RVV): Optimize CV color conversion functions with intrinsics

GitOrigin-RevId: 9c25bfac9121a326b86cd69e67949f742321ee9f
@ihb2032 ihb2032 deleted the opt/rvv-c3-nv21 branch December 27, 2025 02:09
wangzhaode added a commit that referenced this pull request Dec 30, 2025
@wangzhaode
Merge pull request #4079 from ihb2032/opt/rvv-c3-nv21
2703cbe 
opt(RVV): Optimize CV color conversion functions with intrinsics

GitOrigin-RevId: f5d0f2956495380fe72b693ea9a4e322886b4d9f
wangzhaode added a commit that referenced this pull request Dec 30, 2025
@wangzhaode
Merge pull request #4079 from ihb2032/opt/rvv-c3-nv21
fb6048c 
opt(RVV): Optimize CV color conversion functions with intrinsics

GitOrigin-RevId: 1150246ab040f30449b228b1cc9d948de543b7dc
wangzhaode added a commit that referenced this pull request Jan 7, 2026
@wangzhaode
Merge pull request #4079 from ihb2032/opt/rvv-c3-nv21
2fd705d 
opt(RVV): Optimize CV color conversion functions with intrinsics

GitOrigin-RevId: 9a7c2a465c1101b6c0690aef44d502795c0970dc
wangzhaode added a commit that referenced this pull request Jan 8, 2026
@wangzhaode
Merge pull request #4079 from ihb2032/opt/rvv-c3-nv21
1c79d7d 
opt(RVV): Optimize CV color conversion functions with intrinsics

GitOrigin-RevId: 6e09d6eb2b9e8a4c280564dc1d5f6f9ff5e16605
Juude pushed a commit to Juude/MNN that referenced this pull request Jan 14, 2026
@wangzhaode @Juude
Merge pull request alibaba#4079 from ihb2032/opt/rvv-c3-nv21
6a3a9a6 
opt(RVV): Optimize CV color conversion functions with intrinsics

GitOrigin-RevId: 9a7c2a465c1101b6c0690aef44d502795c0970dc
Juude pushed a commit to Juude/MNN that referenced this pull request Jan 14, 2026
@wangzhaode @Juude
Merge pull request alibaba#4079 from ihb2032/opt/rvv-c3-nv21
529b34e 
opt(RVV): Optimize CV color conversion functions with intrinsics

GitOrigin-RevId: 6e09d6eb2b9e8a4c280564dc1d5f6f9ff5e16605
wangzhaode added a commit that referenced this pull request Jan 20, 2026
@wangzhaode
Merge pull request #4079 from ihb2032/opt/rvv-c3-nv21
3b1eee6 
opt(RVV): Optimize CV color conversion functions with intrinsics

GitOrigin-RevId: 62f1133a743912812e28114c7c08821473d19a68
wangzhaode added a commit that referenced this pull request Jan 21, 2026
@wangzhaode
Merge pull request #4079 from ihb2032/opt/rvv-c3-nv21
7e2472a 
opt(RVV): Optimize CV color conversion functions with intrinsics

GitOrigin-RevId: 62f1133a743912812e28114c7c08821473d19a68
wangzhaode added a commit that referenced this pull request Jan 21, 2026
@wangzhaode
Merge pull request #4079 from ihb2032/opt/rvv-c3-nv21
f845393 
opt(RVV): Optimize CV color conversion functions with intrinsics

GitOrigin-RevId: f491541f2b63732eec0b1bd38327b6b3c4191fdb
ORIGINAL_AUTHOR=MNNSyncBot <hi@zhaode.wang>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

@wangzhaode wangzhaode

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@ihb2032 @wangzhaode