ncnn rvv int8 Convolution/ConvDw/Quantize/Requantize/Dequantize

[MollySophia]

复现环境

• 官方buildroot镜像，55MB 内存
• 编译ncnn：

$ cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/c906-v240.toolchain.cmake -DCMAKE_BUILD_TYPE=release -DNCNN_OPENMP=OFF -DNCNN_THREADS=OFF -DNCNN_RUNTIME_CPU=OFF -DNCNN_RVV=ON -DNCNN_SIMPLEOCV=ON -DNCNN_BUILD_EXAMPLES=OFF -DNCNN_BUILD_TESTS=OFF -DNCNN_BENCHMARK=OFF  ..
$ cmake --build .

测试情况

• 以下耗时在55M内存，加上了milkv qq群里发的固定大核频率patch，耗时可能整体低一些，以组委会复现情况为准

[root@milkv-duo]/mnt/sd/ncnn/benchmark# ../build-duo/benchmark/benchncnn 4 1 0 -
1 1
syscall error -1
loop_count = 4
num_threads = 1
powersave = 0
gpu_device = -1
cooling_down = 1
          squeezenet  min =  236.18  max =  238.41  avg =  237.23
     squeezenet_int8  min =  792.14  max =  797.17  avg =  794.45
           mobilenet  min =  342.71  max =  343.84  avg =  343.32
      mobilenet_int8  min = 1124.57  max = 1134.26  avg = 1130.00
        mobilenet_v2  min =  267.92  max =  270.24  avg =  269.61
        mobilenet_v3  min =  217.07  max =  219.68  avg =  218.28
          shufflenet  min =  383.98  max =  385.90  avg =  385.00
       shufflenet_v2  min =  253.54  max =  255.95  avg =  254.80
             mnasnet  min =  266.44  max =  269.46  avg =  267.87
     proxylessnasnet  min =  309.38  max =  311.88  avg =  310.69
     efficientnet_b0  min =  372.42  max =  372.99  avg =  372.74
   efficientnetv2_b0  min =  527.15  max =  528.06  avg =  527.76
        regnety_400m  min =  436.96  max =  438.73  avg =  438.19
           blazeface  min =  135.91  max =  138.26  avg =  137.13
       mobilenet_ssd  min =  774.66  max =  798.78  avg =  782.37
  mobilenetv2_yolov3  min =  921.68  max =  926.01  avg =  923.53
           nanodet_m  min =  578.88  max =  580.19  avg =  579.65
    yolo-fastest-1.1  min =  307.14  max =  309.53  avg =  308.40
      yolo-fastestv2  min =  205.72  max =  208.24  avg =  206.49
          FastestDet  min =  210.62  max =  213.29  avg =  211.69
[root@milkv-duo]/mnt/sd/ncnn/benchmark#

[shiptux]

尊敬的参赛选手，您好。
本次锦标赛您所提交的 PR 初步复测结果如 https://github.com/plctlab/rvspoc/blob/main/Results/Verifications/S2310/README.md 所示。如有任何异议请回复本条评论。如确认无误请回复 “确认无误”，感谢您的配合。

[MollySophia]

确认无误

…

On Tue, Mar 26, 2024 at 12:55 PM Cheng Li ***@***.***> wrote: 尊敬的参赛选手，您好。 本次锦标赛您所提交的 PR 初步复测结果如 https://github.com/plctlab/rvspoc/blob/main/Results/Verifications/S2310/README.md 所示。如有任何异议请回复本条评论。如确认无误请回复 “确认无误”，感谢您的配合。 — Reply to this email directly, view it on GitHub <#1 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AE6JFBCA7ASKBJV5O2WFZFTY2D5VVAVCNFSM6AAAAABD5JATZ2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDAMJZGM4DSNRTGM> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[shiptux]

感谢您的配合。