std1.0.0-beta-7 SafeC docs

std :: ml::device

Macro DEV_CONV_FWD

#include <std/ml/device.h> source

#define DEV_CONV_FWD 0

A 2D convolution (transpose 0) or transposed convolution (1) in NCHW, in the layer's terms: x the input [batch, inC, inH, inW], y the output [batch, outC, outH, outW], w [outC, inC, kH, kW] (a transposed one's [inC, outC, kH, kW]). op 0: y = conv(x, w); 1: x = the input gradient from y's; 2: w = the weight gradient from y's and x. The result is overwritten.