int main(int argc, char** argv, char** envp) {
float x=4.5f;
float y=3.2f;
double dx=4.5f;
double dy=3.2f;
printf("%f+%f=%f\n", x, y, addfloat(x, y));
printf("%lf+%lf=%lf\n", dx, dy, adddouble(dx, dy));
return EXIT_SUCCESS;
}
/**
* @brief Perform inverse DWT (periodic boundaries) on 3D data.
*
* @param wc Wavelet presentation of 3D data.
* @param img Reconstructed signal.
*/
void
idpwt3 (const Matrix <T> & wc, Matrix <T> & img)
{
// assign dwt to result image
img = wc;
# pragma omp parallel default (shared) num_threads (_num_threads)
{
T * wcplo, * wcphi, * templo, * temphi, * temptop, * tmp;
size_t stride;
int sl1 = _sl1_scale,
sl2 = _sl2_scale,
sl3 = _sl3_scale;
const int t_num = omp_get_thread_num ();
// loop over levels of backwards DWT
for (int j = _min_level; j < _max_level; j++)
{
// update stride
stride = 6 * sl3 * t_num;
tmp = & _temp [stride];
templo = & _temp [2 * sl3 + stride];
temphi = & _temp [3 * sl3 + stride];
temptop = & _temp [4 * sl3 + stride];
# pragma omp for schedule (OMP_SCHEDULE)
// loop over lines along third dimension ('third') of result image
for (int c1_loc = 0; c1_loc < 2 * sl1 * 2 * sl2; c1_loc++)
{
int c1_glob = (c1_loc % (2 * sl1)) + (c1_loc / (2 * sl1)) * _sl1;
// copy lowpass part of current line to temporary memory
unpackdouble (& img [c1_glob], sl3, _ld12, 0, templo);
// copy highpass part of current line to temporary memory
unpackdouble (& img [c1_glob + sl3 * _ld12], sl3, _ld12, 0, temphi);
// perform lowpass reconstruction
uplo (templo, sl3, tmp);
// perform highpass reconstruction
uphi (temphi, sl3, temptop);
// fusion of reconstruction parts
adddouble (tmp, temptop, sl3 * 2, tmp);
// write back reconstructed line
packdouble (tmp, sl3 * 2, _ld12, 0, & img [c1_glob]);
} // loop over lines along third dimension of result image
// update stride
stride = 6 * sl2 * t_num;
tmp = & _temp [stride];
templo = & _temp [2 * sl2 + stride];
temphi = & _temp [3 * sl2 + stride];
temptop = & _temp [4 * sl2 + stride];
# pragma omp for schedule (OMP_SCHEDULE)
// loop over lines along second dimension ('rows') of result image
for (int c1_loc = 0; c1_loc < 2 * sl1 * 2 * sl3; c1_loc++)
{
int c1_glob = (c1_loc / (2 * sl1)) * _sl1 * _sl2;
// copy lowpass part of current line to temporary memory
unpackdouble (& img [c1_glob], sl2, _sl1, c1_loc % (2 * sl1), templo);
// copy highpass part of current line to temporary memory
unpackdouble (& img [c1_glob + sl2 * _sl1], sl2, _sl1, c1_loc % (2 * sl1), temphi);
// perform lowpass reconstruction
uplo (templo, sl2, tmp);
// perform highpass reconstruction
uphi (temphi, sl2, temptop);
// fusion of reconstruction parts
adddouble (tmp, temptop, sl2 * 2, tmp);
// write back reconstructed line
packdouble (tmp, sl2 * 2, _sl1, c1_loc % (2 * sl1), & img [c1_glob]);
} // loop over lines along second dimension of result image
// update stride
stride = 5 * sl1 * t_num;
tmp = & _temp [stride];
templo = & _temp [ sl1 + stride];
temphi = & _temp [3 * sl1 + stride];
# pragma omp for schedule (OMP_SCHEDULE)
// loop over lines along first dimension ('columns') of result image
for (int c2_loc = 0; c2_loc < 2 * sl2 * 2 * sl3; c2_loc++)
{
int c2_glob = (c2_loc / (2 * sl2)) * _sl2 * _sl1 + (c2_loc % (2 * sl2)) * _sl1;
// assign address of current line's lowpass part
wcplo = & img [c2_glob];
// assign address of current line's highpass part
wcphi = & img [c2_glob + sl1];
// copy lowpass part to temporary memory
copydouble (wcplo, tmp, sl1);
// perform lowpass reconstruction
uplo (wcplo, sl1, templo);
// perform highpass reconstruction
uphi (wcphi, sl1, temphi);
// combine reconstructed parts and write back to current line
adddouble (templo, temphi, sl1 * 2, wcplo);
} // loop over lines along first dimension ('columns') of result image
// update current row / column size
sl2 *= 2;
sl1 *= 2;
sl3 *= 2;
} // loop over levels of backwards DWT
} // omp parallel
}