fsfw/globalfunctions/math/MatrixOperations.h
Ulrich Mohr 575f70ba03 updating code from Flying Laptop
This is the framework of Flying Laptop OBSW version A.13.0.
2018-07-12 16:29:32 +02:00

116 lines
3.5 KiB
C++

#ifndef MATRIXOPERATIONS_H_
#define MATRIXOPERATIONS_H_
#include <cmath>
#include <stdint.h>
template<typename T1, typename T2=T1, typename T3=T2>
class MatrixOperations {
public:
//do not use with result == matrix1 or matrix2
static void multiply(const T1 *matrix1, const T2 *matrix2, T3 *result,
uint8_t rows1, uint8_t columns1, uint8_t columns2) {
if ((matrix1 == (T1*)result) || (matrix2 == (T2*)result)){
//SHOULDDO find an implementation that is tolerant to this
return;
}
for (uint8_t resultColumn = 0; resultColumn < columns2;
resultColumn++) {
for (uint8_t resultRow = 0; resultRow < rows1; resultRow++) {
result[resultColumn + columns2 * resultRow] = 0;
for (uint8_t i = 0; i < columns1; i++) {
result[resultColumn + columns2 * resultRow] += matrix1[i
+ resultRow * columns1]
* matrix2[resultColumn + i * columns2];
}
}
}
}
static void transpose(const T1 *matrix, T2 *transposed, uint8_t size) {
uint8_t row, column;
transposed[0] = matrix[0];
for (column = 1; column < size; column++) {
transposed[column + size * column] = matrix[column + size * column];
for (row = 0; row < column; row++) {
T1 temp = matrix[column + size * row];
transposed[column + size * row] = matrix[row + size * column];
transposed[row + size * column] = temp;
}
}
}
// Overload transpose to support non symmetrical matrices
//do not use with transposed == matrix && columns != rows
static void transpose(const T1 *matrix, T2 *transposed, uint8_t rows, uint8_t columns) {
uint8_t row, column;
transposed[0] = matrix[0];
if (matrix == transposed && columns == rows)
{
transpose(matrix, transposed, rows);
}
else if (matrix == transposed && columns != rows)
{
// not permitted
return;
}
for (column = 0; column < columns; column++) {
for (row = 0; row < rows; row++) {
transposed[row + column * rows] = matrix[column + row * columns];
}
}
}
static void add(const T1 *matrix1, const T2 *matrix2, T3 *result,
uint8_t rows, uint8_t columns)
{
for (uint8_t resultColumn = 0; resultColumn < columns; resultColumn++)
{
for (uint8_t resultRow = 0; resultRow < rows; resultRow++)
{
result[resultColumn + columns * resultRow] = matrix1[resultColumn + columns * resultRow]+
matrix2[resultColumn + columns * resultRow];
}
}
}
static void subtract(const T1 *matrix1, const T2 *matrix2, T3 *result,
uint8_t rows, uint8_t columns)
{
for (uint8_t resultColumn = 0; resultColumn < columns; resultColumn++)
{
for (uint8_t resultRow = 0; resultRow < rows; resultRow++)
{
result[resultColumn + columns * resultRow] = matrix1[resultColumn + columns * resultRow]-
matrix2[resultColumn + columns * resultRow];
}
}
}
static void addScalar(const T1 *matrix1, const T2 scalar, T3 *result,
uint8_t rows, uint8_t columns)
{
for (uint8_t resultColumn = 0; resultColumn < columns; resultColumn++)
{
for (uint8_t resultRow = 0; resultRow < rows; resultRow++)
{
result[resultColumn + columns * resultRow] = matrix1[resultColumn + columns * resultRow]+scalar;
}
}
}
static void multiplyScalar(const T1 *matrix1, const T2 scalar, T3 *result,
uint8_t rows, uint8_t columns)
{
for (uint8_t resultColumn = 0; resultColumn < columns; resultColumn++)
{
for (uint8_t resultRow = 0; resultRow < rows; resultRow++)
{
result[resultColumn + columns * resultRow] = matrix1[resultColumn + columns * resultRow]*scalar;
}
}
}
};
#endif /* MATRIXOPERATIONS_H_ */