Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

riscv general optimization for convolution sgemm and winograd and innerproduct #3857

Merged
merged 12 commits into from May 29, 2022

Conversation

nihui
Copy link
Member

@nihui nihui commented May 29, 2022

StarFive VisionFive V1 RV64GC 1.5GHz x 2

model baseline pr3857 diff
squeezenet 1697.78 609.41 -64.11%
mobilenet 2239.54 993.87 -55.62%
mobilenet_v2 1468.94 728.17 -50.43%
mobilenet_v3 1191.72 674.72 -43.38%
shufflenet 623.26 358.98 -42.40%
shufflenet_v2 608.18 303.73 -50.06%
mnasnet 1379.11 735.94 -46.64%
proxylessnasnet 1523 871.64 -42.77%
efficientnet_b0 2147.2 1355.94 -36.85%
efficientnetv2_b0 3967.31 1325.71 -66.58%
regnety_400m 1718.47 878.48 -48.88%
blazeface 192.36 106.69 -44.54%
googlenet 9846.36 2484.78 -74.76%
resnet18 12075.38 2198.83 -81.79%
alexnet 4575.12 2252.52 -50.77%
vgg16 146658.11 18000.95 -87.73%
resnet50 20093.45 5499.88 -72.63%
squeezenet_ssd 8924.95 1904.07 -78.67%
mobilenet_ssd 4967.77 2230.24 -55.11%
mobilenet_yolo 13190.93 6003.25 -54.49%
mobilenetv2_yolov3 5637.48 2685.14 -52.37%
yolov4-tiny 29967.02 4873.81 -83.74%
nanodet_m 1589.48 836.73 -47.36%
yolo-fastest-1.1 849.2 437.98 -48.42%
yolo-fastestv2 670.07 393.94 -41.21%

@codecov-commenter
Copy link

codecov-commenter commented May 29, 2022

Codecov Report

Merging #3857 (a25bd2e) into master (c3adbcf) will decrease coverage by 0.00%.
The diff coverage is 92.10%.

@@            Coverage Diff             @@
##           master    #3857      +/-   ##
==========================================
- Coverage   93.32%   93.31%   -0.01%     
==========================================
  Files         678      681       +3     
  Lines      174439   175736    +1297     
==========================================
+ Hits       162794   163995    +1201     
- Misses      11645    11741      +96     
Impacted Files Coverage Δ
src/layer/riscv/convolution_3x3.h 87.12% <87.12%> (ø)
src/layer/riscv/convolution_riscv.cpp 99.09% <100.00%> (+0.02%) ⬆️
src/layer/riscv/convolution_sgemm.h 100.00% <100.00%> (ø)
src/layer/riscv/convolution_winograd_transform.h 100.00% <100.00%> (ø)
src/layer/riscv/convolutiondepthwise_3x3.h 100.00% <100.00%> (ø)
src/layer/riscv/convolutiondepthwise_riscv.cpp 98.93% <100.00%> (+0.01%) ⬆️
src/layer/riscv/innerproduct_riscv.cpp 99.54% <100.00%> (+0.04%) ⬆️
src/gpu.cpp 82.84% <0.00%> (+0.10%) ⬆️
... and 1 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c3adbcf...a25bd2e. Read the comment docs.

@nihui nihui changed the title riscv general optimization for convolution sgemm and winograd pack1 riscv general optimization for convolution sgemm and winograd and innerproduct May 29, 2022
@nihui nihui merged commit 667be10 into Tencent:master May 29, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

2 participants