Conv2D - Analog Devices VISUALDSP++ 3.5 Manual

conv2d

2-D convolution
Synopsis
#include <filter.h>
void conv2d_fr16(min1, mrow1, mcol1, min2, mrow2, mcol2, mout )
const fract16 *min1;
int mrow1;
int mcol1;
const fract16 *min2;
fract16 *mrow2;
int mcol2;
fract16 *mout;
Description
This function computes the two-dimensional convolution of input matrix
of size
min1 mrow1
result in matrix
Algorithm
Two dimensional input matrix
placing the result in a matrix pointed to by
−
mcol 2
[ ]
∑ ∑
=
mout c , r
=
i 0
for c = 0 to
mcol1
Domain
-1.0 to +1.0
VisualDSP++ 3.5 C/C++ Compiler and Library Manual
for ADSP-219x DSPs
x and
mcol1 min2
of dimension (
mout
min1
−
1 mrow 2 1
+ +
min1 [ c i , r j
=
j 0
+ and r = 0 to
mcol2-1
DSP Run-Time Library
/* pointer to input matrix 1
/* number of rows in matrix 1
/* number of columns in matrix 1
/* pointer to input matrix 2
/* number of rows in matrix 1
/* number of columns in matrix 2
/* pointer to output matrix
of size x
mrow2
+
mrow1 mrow2-1
is convolved with input matrix
.
mout
( )
• − −
] min2 [ mcol 2 1
mrow2-1
and store the
mcol2 s
) x ( +
mcol1 mcol2-1
min2
( )
− −
i , mrow 2 1 j ]
4-49
*/
*/
*/
*/
*/
*/
*/
).
,