///////////////////////////////////////////////transforms/////////////////////////////////////////////////////////////////////
void mBior53::transrows(char** dest, char** sour, unsigned int w, unsigned int h) const
{
char srck0[8] = {0, 0, 0, 0, 0, 0, 0, 0};
//int n;
float s, d;
unsigned int w2 = w / 2;
const vec1D& tH = gettH();
const vec1D& tG = gettG();
for (unsigned int y = 0; y < h; y++) {
//k=0
unsigned int k = 0;
srck0[0] = sour[y][2];
srck0[1] = sour[y][1];
srck0[2] = sour[y][0];
srck0[3] = sour[y][1];
s = conv(tH.data(), srck0);
s += tH[2] * (float)sour[y][2*k + 2];
d = conv(tG.data(), &sour[y][2*k]);
dest[y][k] = mmxround(s);
dest[y][k+w2] = mmxroundTH(d);
//k=1, k<w2-1
for (k = 1; k < w2 - 1; k++) {
s = conv(tH.data(), &sour[y][2*k - 2]);
s += tH[2] * (float)sour[y][2*k + 2];
d = conv(tG.data(), &sour[y][2*k]);
dest[y][k] = mmxround(s);
dest[y][k+w2] = mmxroundTH(d);
}
//k=w2-1
k = w2 - 1;
srck0[0] = sour[y][2*k];
srck0[1] = sour[y][2*k+1];
srck0[2] = sour[y][2*k];
srck0[3] = sour[y][2*k-1];
s = conv(tH.data(), &sour[y][2*k - 2]);
s += tH[2] * (float)sour[y][2*k];
d = conv(tG.data(), srck0);
dest[y][k] = mmxround(s);
dest[y][k+w2] = mmxroundTH(d);
}
}
void FWT2D::transcols(char** dest, char** sour, unsigned int w, unsigned int h) const
{
int n;
float s, d;
unsigned int h2 = h / 2;
const vec1D& tH = gettH();
const vec1D& tG = gettG();
for (unsigned int x = 0; x < w; x++)
{
for (unsigned int k = 0; k < h2; k++)
{
s = 0.0f;
d = 0.0f;
for (int m = tH.first(); m <= tH.last(); m++)
{
n = 2 * k + m;
if (n < 0) n = 0 - n;
if (n >= (int)h) n -= 2 * (1 + n - h);
s += tH[m] * float(sour[n][x]);
}
for (int m = tG.first(); m <= tG.last(); m++)
{
n = 2 * k + m;
if (n < 0) n = 0 - n;
if (n >= (int)h) n -= 2 * (1 + n - h);
d += tG[m] * float(sour[n][x]);
}
if (x < w / 2)
dest[k][x] = mmxround(s);
else
dest[k][x] = mmxroundTH(s); //is this needed? hi band were TH'ed on transrows
dest[k+h2][x] = mmxroundTH(d); //is this needed? hi band were TH'ed on transrows on x>w/2
}
}
}
///////////////////////////////////////////////transforms/////////////////////////////////////////////////////////////////////
void FWT2D::transrows(char** dest, char** sour, unsigned int w, unsigned int h) const
{
int n;
float s, d;
unsigned int w2 = w / 2;
const vec1D& tH = gettH();
const vec1D& tG = gettG();
for (unsigned int y = 0; y < h; y++)
{
for (unsigned int k = 0; k < w2; k++)
{
s = 0.0f;
d = 0.0f;
for (int m = tH.first(); m <= tH.last(); m++)
{
n = 2 * k + m;
if (n < 0) n = 0 - n;
if (n >= (int)w) n -= 2 * (1 + n - w);
s += tH[m] * float(sour[y][n]);
}
for (int m = tG.first(); m <= tG.last(); m++)
{
n = 2 * k + m;
if (n < 0) n = 0 - n;
if (n >= (int)w) n -= 2 * (1 + n - w);
d += tG[m] * float(sour[y][n]);
}
dest[y][k] = mmxround(s);
dest[y][k+w2] = mmxroundTH(d);
}
}
}
void mBior53::transcols(char** dest, char** sour, unsigned int w, unsigned int h) const
{
float fz = 0.0f;
int n;
float s, d;
__m128 ms, md;
unsigned int h2 = h / 2;
const vec1D& tH = gettH();
const vec1D& tG = gettG();
for (unsigned int x = 0; x < w / 4; x++) { //x<w/4 x = 4*x
for (unsigned int k = 0; k < h2; k++) {
ms = _mm_load_ss(&fz);
md = ms;
for (int m = -2; m <= 2; m++) {
n = 2 * k + m;
if (n < 0) n = 0 - n;
if (n >= (int)h) n -= 2 * (1 + n - h);
ms = _mm_add_ps(ms, _mm_mul_ps(_mm_load_ps1(tH.data(m)), _mm_cvtpi8_ps(*(__m64 *)(&sour[n][4*x]))));
}
for (int m = 0; m <= 2; m++) {
n = 2 * k + m;
if (n < 0) n = 0 - n;
if (n >= (int)h) n -= 2 * (1 + n - h);
md = _mm_add_ps(md, _mm_mul_ps(_mm_load_ps1(tG.data(m)), _mm_cvtpi8_ps(*(__m64 *)(&sour[n][4*x]))));
}
if (4*x < w / 2) {
if ((w / 2) - (4*x) >= 4)
mmxround4(&dest[k][4*x], ms);
else
mmxround4TH(&dest[k][4*x], ms, (w / 2) - (4*x)); //skip first from LL part 10/2-4=1 [lo] o o o o * | * * * o o [hi]
} else
mmxround4TH(&dest[k][4*x], ms);
mmxround4TH(&dest[k+h2][4*x], md);
}
}
_mm_empty();
//odd remainder
for (unsigned int x = w - (w % 4); x < w; x++) {
for (unsigned int k = 0; k < h2; k++) {
s = 0;
d = 0;
for (int m = -2; m <= 2; m++) {
n = 2 * k + m;
if (n < 0) n = 0 - n;
if (n >= (int)h) n -= 2 * (1 + n - h);
s += tH[m] * float(sour[n][x]);
}
for (int m = 0; m <= 2; m++) {
n = 2 * k + m;
if (n < 0) n = 0 - n;
if (n >= (int)h) n -= 2 * (1 + n - h);
d += tG[m] * float(sour[n][x]);
}
if (x < w / 2)
dest[k][x] = mmxround(s);
else
dest[k][x] = mmxroundTH(s); //is this needed? hi band were TH'ed on transrows
dest[k+h2][x] = mmxroundTH(d); //is this needed? hi band were TH'ed on transrows on x>w/2
}
}
}