Benchmark on Linux_x86_64, Int8 performance down

[hellocsi]

• int8 time is much longer than the original for mobilenet and mobilenet_ssd

  Chang@PC-CHANG:/mnt/d/workspace/ncnn/benchmark$ ../build/benchmark/benchncnn 
  loop_count = 4
  num_threads = 4
  powersave = 0
  gpu_device = -1
        squeezenet  min =   36.35  max =   70.11  avg =   46.18
   squeezenet_int8  min =   56.66  max =   59.84  avg =   57.64
         mobilenet  min =   37.65  max =   41.45  avg =   38.77
    mobilenet_int8  min =  117.75  max =  140.89  avg =  125.39
      mobilenet_v2  min =   32.10  max =   34.74  avg =   33.03
      mobilenet_v3  min =   54.22  max =   65.59  avg =   57.55
        shufflenet  min =   37.56  max =   83.67  avg =   54.27
     shufflenet_v2  min =   14.72  max =   17.60  avg =   15.97
           mnasnet  min =   56.48  max =  235.74  avg =  115.90
   proxylessnasnet  min =   73.03  max =   94.88  avg =   79.04
         googlenet  min =  142.23  max =  145.58  avg =  144.27
    googlenet_int8  min =  210.10  max =  286.43  avg =  235.88
          resnet18  min =  153.02  max =  274.57  avg =  200.75
     resnet18_int8  min =  131.59  max =  157.12  avg =  143.83
           alexnet  min =  109.27  max =  168.34  avg =  133.72
             vgg16  min = 1247.88  max = 1492.38  avg = 1348.06
        vgg16_int8  min =  677.22  max =  721.92  avg =  693.12
          resnet50  min =  314.13  max =  357.75  avg =  327.32
     resnet50_int8  min =  488.33  max =  590.42  avg =  517.96
    squeezenet_ssd  min =  128.09  max =  158.92  avg =  147.91
  squeezenet_ssd_int8  min =   99.45  max =  126.24  avg =  108.95
     mobilenet_ssd  min =   93.30  max =  117.57  avg =  103.74
  mobilenet_ssd_int8  min =  248.40  max =  290.73  avg =  268.24
    mobilenet_yolo  min =  184.16  max =  251.08  avg =  206.49
  mobilenetv2_yolov3  min =  123.99  max =  142.85  avg =  133.62

[nihui]

most x86 code in ncnn is not well optimized for speed. It is generally recommended to use gpu inference for pc usecases