double f22(double k,void * params) /* No integral before the deltaM*/
{
double alpha = *(double *) params;
double f1 = pow(k,2.0) / (2 * pow(M_PI,2.0));
double f2 = Pk( k) * pow(Windowfunc( k,Radius(alpha)),2.0) ;
double f = f1 * f2;
return sqrt(f);
}
void P_graded(vector<double> &P, const vector<double> &par,
const NumericMatrix &Theta, const NumericVector &ot, const int &N,
const int &nfact, const int &nint, const int &itemexp, const int &israting)
{
const int parsize = par.size();
vector<double> a(nfact);
for(int i = 0; i < nfact; ++i) a[i] = par[i];
vector<double> d(nint, 0.0);
if(israting){
const double t = par[parsize-1];
for(int i = nfact; i < parsize - 1; ++i)
d[i - nfact] = par[i] + t;
} else {
for(int i = nfact; i < parsize; ++i)
d[i - nfact] = par[i];
}
int notordered = 0;
for(int i = 1; i < nint; ++i)
notordered += d[i-1] <= d[i];
if(notordered){
for(int i = 0; i < P.size(); ++i)
P[i] = 0.0;
} else {
const double nullzero = 0.0, nullone = 1.0;
NumericMatrix Pk(N, nint + 2);
for(int i = 0; i < N; ++i)
Pk(i,0) = 1.0;
for(int i = 0; i < nint; ++i){
vector<double> tmp1(N), tmp2(N);
itemTrace(tmp1, tmp2, a, &d[i], Theta, &nullzero, &nullone, ot);
for(int j = 0; j < N; ++j)
Pk(j,i+1) = tmp2[j];
}
if(itemexp){
int which = N * (nint + 1) - 1;
for(int i = (Pk.ncol()-2); i >= 0; --i){
for(int j = (N-1); j >= 0; --j){
P[which] = Pk(j,i) - Pk(j,i+1);
if(P[which] < 1e-20) P[which] = 1e-20;
else if((1.0 - P[which]) < 1e-20) P[which] = 1.0 - 1e-20;
--which;
}
}
} else {
int which = 0;
for(int i = 0; i < Pk.ncol(); ++i){
for(int j = 0; j < Pk.nrow(); ++j){
P[which] = Pk(j,i);
++which;
}
}
}
}
}
void getScale(short *mirror, short *result, int height, int width, int delay, float ix, float iy) {
int i, j;
// int n, m;
float a, b;
//float pn, pm;
__m128 pm,
pn_1, pn0, pn1, pn2;
__m128 sum_1, sum0, sum1, sum2;
int size = height*iy;
int depth = width+2*delay;
float sum;
#pragma omp parallel private(i, j, sum, a, pn_1, pn0, pn1, pn2, pm, sum_1, sum0, sum1, sum2)
{
#pragma omp for
for (i = 0; i < size; i++) {
for (j = 0; j < width*ix; j++) {
// sum = 0.0f;
a = ((float) i)/ix - ((int) i/ix);
b = ((float) j)/iy - ((int) j/iy);
//Get all pn
pn_1 = _mm_set1_ps(Pk(-1 - a));
pn0 = _mm_set1_ps(Pk(- a));
pn1 = _mm_set1_ps(Pk(1 - a));
pn2 = _mm_set1_ps(Pk(2 - a));
//get all pm
pm = _mm_set_ps(Pk(b-2),Pk(b-1),Pk(b),Pk(b+1));
//tmp mul pn*pm
pn_1 = _mm_mul_ps(pm,pn_1);
pn0 = _mm_mul_ps(pm,pn0);
pn1 = _mm_mul_ps(pm,pn1);
pn2 = _mm_mul_ps(pm,pn2);
//get all mirror pos
sum_1 = _mm_cvtepi32_ps(_mm_setr_epi32(
mirror INDEX ((int) (i/ix+delay-1), (int) (j/iy+delay-1), depth),
mirror INDEX ((int) (i/ix+delay-1), (int) (j/iy+delay), depth),
mirror INDEX ((int) (i/ix+delay-1), (int) (j/iy+delay+1), depth),
mirror INDEX ((int) (i/ix+delay-1), (int) (j/iy+delay+2), depth)));
sum0 = _mm_cvtepi32_ps(_mm_setr_epi32(
mirror INDEX ((int) (i/ix+delay), (int) (j/iy+delay-1), depth),
mirror INDEX ((int) (i/ix+delay), (int) (j/iy+delay), depth),
mirror INDEX ((int) (i/ix+delay), (int) (j/iy+delay+1), depth),
mirror INDEX ((int) (i/ix+delay), (int) (j/iy+delay+2), depth)));
sum1 = _mm_cvtepi32_ps(_mm_setr_epi32(
mirror INDEX ((int) (i/ix+delay+1), (int) (j/iy+delay-1), depth),
mirror INDEX ((int) (i/ix+delay+1), (int) (j/iy+delay), depth),
mirror INDEX ((int) (i/ix+delay+1), (int) (j/iy+delay+1), depth),
mirror INDEX ((int) (i/ix+delay+1), (int) (j/iy+delay+2), depth)));
sum2 = _mm_cvtepi32_ps(_mm_setr_epi32(
mirror INDEX ((int) (i/ix+delay+2), (int) (j/iy+delay-1), depth),
mirror INDEX ((int) (i/ix+delay+2), (int) (j/iy+delay), depth),
mirror INDEX ((int) (i/ix+delay+2), (int) (j/iy+delay+1), depth),
mirror INDEX ((int) (i/ix+delay+2), (int) (j/iy+delay+2), depth)));
//get sum for all mirror pos
sum_1 = _mm_mul_ps(sum_1, pn_1);
sum0 = _mm_mul_ps(sum0, pn0);
sum1 = _mm_mul_ps(sum1, pn1);
sum2 = _mm_mul_ps(sum2, pn2);
//sum all record sse for mirror pos *pn*pm
sum_1 = _mm_add_ps(sum_1, sum0);
sum1 = _mm_add_ps(sum1, sum2);
sum_1 = _mm_add_ps(sum_1, sum1);
sum_1 = _mm_hadd_ps(sum_1, sum_1);
sum_1 = _mm_hadd_ps(sum_1, sum_1);
_mm_store_ss(&sum, sum_1);
result INDEX(i, j, (int) (width*ix)) = (int) sum;
}
}
}
}
__m128 Pm(float b){
return _mm_set_ps(Pk(b-2),Pk(b-1),Pk(b),Pk(b+1));
}
