Created by: ysh329
状态：WIP单测出错正在看，等review
更新后
多出以下：
- 打印维度信息（input、filter、output）；
- Kernel具体方法（如direct、gemm、depthwise）；
- Kernel相关的参数信息remark（具体参数）；
- GOPS。
===== Detailed Profiler Summary: N/A, Exclude 0 warm-ups =====
Operator Type Kernel Name Remark Input Filter Output GOPs Avg (ms) Min (ms) Max (ms) Last (ms)
calib 0.000000 0.025390 0.023000 0.051000 0.026000
conv2d direct/arm/int8_t/NCHW 3x3s2p1g1 1x3x128x128 8x3x3x3 1x8x64x64 0.001769 0.144360 0.140000 0.217000 0.140000
conv2d gemm/arm/int8_t/NCHW 1x1s1p0g1 1x8x64x64 8x8x1x1 1x8x64x64 0.000524 0.052840 0.050000 0.129000 0.052000
conv2d depthwise/arm/int8_t/NCHW 3x3s1p1g8 1x8x64x64 8x1x3x3 1x8x64x64 0.000590 0.055860 0.054000 0.066000 0.057000
conv2d gemm/arm/int8_t/NCHW 1x1s1p0g1 1x8x64x64 4x8x1x1 1x4x64x64 0.000262 0.027500 0.027000 0.036000 0.027000
conv2d gemm/arm/int8_t/NCHW 1x1s1p0g1 1x4x64x64 24x4x1x1 1x24x64x64 0.000786 0.122690 0.118000 0.154000 0.122000
conv2d depthwise/arm/int8_t/NCHW 3x3s2p1g24 1x24x64x64 24x1x3x3 1x24x32x32 0.000442 0.056950 0.054000 0.140000 0.056000
conv2d gemm/arm/int8_t/NCHW 1x1s1p0g1 1x24x32x32 6x24x1x1 1x6x32x32 0.000295 0.020630 0.019000 0.030000 0.021000
calib 0.000000 0.003530 0.003000 0.005000 0.004000
conv2d gemm/arm/int8_t/NCHW 1x1s1p0g1 1x6x32x32 36x6x1x1 1x36x32x32 0.000442 0.053050 0.051000 0.108000 0.052000
conv2d depthwise/arm/int8_t/NCHW 3x3s1p1g36 1x36x32x32 36x1x3x3 1x36x32x32 0.000664 0.073620 0.070000 0.129000 0.078000