Vector implementation for int8 BCHW maxpool with 2x2 kernel, requires INP_W%16==0, OUT_W%8==0, KH==KW==2, TT==int8_t or uint8_t, Maxpool2x2Int8BCHW::filter<24,32,16,1,6> total = 324.

#include <pool.h>

Public Member Functions
void	filter (input_window< TT > in_window, output_window< TT > out_window)

Static Public Member Functions
static void	registerKernelClass ()

Member Function Documentation

◆ filter()

template<typename TT , int INP_H, int INP_W, int OUT_H, int OUT_W, int B, int C, int KH, int KW, int STEP_H, int STEP_W>

void Maxpool2x2Int8BCHW< TT, INP_H, INP_W, OUT_H, OUT_W, B, C, KH, KW, STEP_H, STEP_W >::filter	(	input_window< TT > *	in_window,
		output_window< TT > *	out_window
	)

max32 (v64int16 xbuff, int xstart, unsigned int xoffsets, unsigned int xoffsets_hi, unsigned int xsquare, int ystart, unsigned int yoffsets, unsigned int yoffsets_hi, unsigned int ysquare)

0x06...00, 0x0e...08 => 0 1 2 3 ... 12 13 14 15, 16 17 18 19 ... 28 29 30 31 max32(v, 0, 0x06040200, 0x0e0c0a08, 0x3210, 32, 0x06040200, 0x0e0c0a08, 0x3210); // first 32 with next 32 problem: offsets index <= 32, each 4b selects 2 adjacent lanes

128 int16 max

The documentation for this class was generated from the following files:

design/aie_src/pool.h
design/aie_src/pool.cc

Public Member Functions

Static Public Member Functions

Member Function Documentation

◆ filter()