基于PYNQ的神经网络硬件加速（一）pool实现

 

pool_core.h

#ifndef __POOL_CORE_H__
#define __POOL_CORE_H__

#include 
#include 

typedef float Dtype_f;

void Pool(ap_uint<16> CHin,ap_uint<16> Hin,ap_uint<16> Win,
		ap_uint<8> Kx,ap_uint<8> Ky,ap_uint<2> mode,
		Dtype_f feature_in[],Dtype_f feature_out[]
	);//mode: 0:MEAN, 1:MIN, 2:MAX

#endif

pool_core.cpp

#include "pool_core.h"

#define max(a,b) ((a>b)?a:b)
#define min(a,b) ((a>b)?b:a)

void Pool(ap_uint<16> CHin,ap_uint<16> Hin,ap_uint<16> Win,
		ap_uint<8> Kx,ap_uint<8> Ky,ap_uint<2> mode,
		Dtype_f feature_in[],Dtype_f feature_out[]
	)//mode: 0:MEAN, 1:MIN, 2:MAX
{
	#pragma HLS INTERFACE m_axi depth=4294967295 port=feature_out offset=slave
	#pragma HLS INTERFACE m_axi depth=4294967295 port=feature_in offset=slave
	#pragma HLS INTERFACE s_axilite port=Win
	#pragma HLS INTERFACE s_axilite port=Kx
	#pragma HLS INTERFACE s_axilite port=Hin
	#pragma HLS INTERFACE s_axilite port=mode
	#pragma HLS INTERFACE s_axilite port=Ky
	#pragma HLS INTERFACE s_axilite port=CHin
	#pragma HLS INTERFACE s_axilite port=return
	ap_uint<16> Hout,Wout;
	Wout=Win/Kx;
	Hout=Hin/Ky;

	for(int c=0;c h=i*Ky+ii;
						ap_int<16> w=j*Kx+jj;
						switch(mode)
						{
							case 0:{sum+=feature_in[h*CHin*Win+w*CHin+c];break;}
							case 1:{sum=min(sum,feature_in[h*CHin*Win+w*CHin+c]);break;}
							case 2:{sum=max(sum,feature_in[h*CHin*Win+w*CHin+c]);break;}
							default:break;
						}
					}
				if(mode==0)
					sum=sum/(Kx*Ky);
				feature_out[i*Wout*CHin+j*CHin+c]=sum;
			}
}