feat(benchmark/benchncnn.cpp): support user defined case

[tpoisonooo]

benchncnn 支持指定 param 文件

1. 兼容原行为

$ ./benchncnn
$ ./benchncnn  8
..
$ ./benchncnn 1 8 0 -1 1 param=vgg16.param "shape=[112,112,3]"
loop_count = 1
num_threads = 8
powersave = 0
gpu_device = -1
cooling_down = 1
         vgg16.param  min =   28.64  max =   28.64  avg =   28.64

2. 发现输入 -h 或者 --help 打印 usage

$ ./benchncnn -h
Usage: benchncnn [loop count] [num threads] [powersave] [gpu device] [cooling down] [(key=value)...]
  param=model.param
  shape=[227,227,3],...

$ ./benchncnn 12 --help
Usage: benchncnn [loop count] [num threads] [powersave] [gpu device] [cooling down] [(key=value)...]
  param=model.param
  shape=[227,227,3],...

3. 增加指定 param 和 shape

shape 用 whc 格式

$ ./benchncnn 1 8 0 0 1 param=squeezenet.param "shape=[224,224,3]"
loop_count = 1
num_threads = 8
powersave = 0
gpu_device = 0
cooling_down = 1
    squeezenet.param  min =   11.40  max =   11.40  avg =   11.40

4. 异常处理：没输入shape

$ ./benchncnn 1 8 0 0 1 param=squeezenet.param                    
input tensor shape empty!

5. 异常处理：shape 输入了 4 维

$ ./benchncnn 1 8 0 0 1 param=squeezenet.param "shape=[224,224,3,1]"
unsupported input shape size 4
input tensor shape empty!

6. 模型只需要 1 个输入，输入了 2 个 shape

正常运行，此时取第一个 shape

$ ./benchncnn 1 8 0 0 1 param=squeezenet.param "shape=[224,224,3],[666]" 
loop_count = 1
num_threads = 8
powersave = 0
gpu_device = 0
cooling_down = 1
    squeezenet.param  min =    9.09  max =    9.09  avg =    9.09

[tpoisonooo]

rebase 一下 master. done.

[nihui]

参考 pnnx main.cpp，额外的高级参数采用 [(key=value)...] 的形式，有更好的扩展性，比如多输入

./benchncnn 20 1 1 -1 1
./benchncnn 20 1 1 -1 1 param=/home/khj/ncnn/benchmark/squeezenet.param shape=[227,227,3]
./benchncnn 20 1 1 -1 1 param=/home/khj/ncnn/benchmark/squeezenet.param shape=[227,227,3],[10],[1,1,1]

[tpoisonooo]

done.

[tpoisonooo]

@nihui 每日一催

[tpoisonooo]

@nihui 每周一催

[tpoisonooo]

@nihui 每月一催

[tpoisonooo]

@nihui 日催

[nihui]

Thanks for your contribution !

@tpoisonooo 合并啦！