struct ath_3d_fft_plan *ath_3d_fft_create_plan(DomainS *pD, int gnx3, int gnx2,
int gnx1, int gks, int gke, int gjs, int gje,
int gis, int gie, ath_fft_data *data, int al,
ath_fft_direction dir)
{
int nbuf, tmp;
struct ath_3d_fft_plan *ath_plan;
if ((dir != ATH_FFT_FORWARD) && (dir != ATH_FFT_BACKWARD)) {
ath_error("Invalid Athena FFT direction.\n");
}
/* Allocate memory for the plan */
ath_plan = (struct ath_3d_fft_plan *)malloc(sizeof(struct ath_3d_fft_plan));
if (ath_plan == NULL) {
ath_error("[ath_3d_fft_plan] Couldn't malloc for FFT plan.\n");
}
/* Set forward/backward FFT */
ath_plan->dir = dir;
/* Set element count (for easy malloc and memset) */
ath_plan->cnt = (gke-gks+1)*(gje-gjs+1)*(gie-gis+1);
ath_plan->gcnt = gnx3*gnx2*gnx1;
tmp = (al==0 ? 1 : 0);
if (data != NULL) tmp = 0;
/* If data == NULL, then allocate something (temporarily if tmp=1) */
if (data == NULL)
data = (ath_fft_data *)ath_3d_fft_malloc(ath_plan);
if (data == NULL)
ath_error("[ath_3d_fft_plan] Couln't malloc for FFT plan data.\n");
/* Create the plan */
#ifdef FFT_BLOCK_DECOMP
/* Block decomp library plans don't care if forward or backward */
ath_plan->plan = fft_3d_create_plan(pD->Comm_Domain, gnx3, gnx2, gnx1,
gks, gke, gjs, gje, gis, gie,
gks, gke, gjs, gje, gis, gie,
0, 0, &nbuf);
#else /* FFT_BLOCK_DECOMP */
if (dir == ATH_FFT_FORWARD) {
ath_plan->plan = fftw_plan_dft_3d(gnx1, gnx2, gnx3, data, data,
FFTW_FORWARD, FFTW_MEASURE);
} else {
ath_plan->plan = fftw_plan_dft_3d(gnx1, gnx2, gnx3, data, data,
FFTW_BACKWARD, FFTW_MEASURE);
}
#endif /* FFT_BLOCK_DECOMP */
if (tmp) ath_3d_fft_free(data);
return ath_plan;
}
int main(void)
{
fftw_complex in[N0][N1][N2], out[N0][N1][N2], out2[N0][N1][N2]; /* double [2] */
fftw_plan p;
int i0, i1, i2;
p = fftw_plan_dft_3d(N0, N1, N2, &in[0][0][0], &out[0][0][0], FFTW_FORWARD, FFTW_ESTIMATE);
for (i0 = 0; i0 < N0; i0++)
for (i1 = 0; i1 < N1; i1++)
for (i2 = 0; i2 < N2; i2++) {
in[i0][i1][i2][0] = sin(2*M_PI*i0/N0) + 3*cos(6*M_PI*i1/N1);
in[i0][i1][i2][1] = 5*cos(4*M_PI*i2/N2);
}
fftw_execute(p);
ft3d(N0, N1, N2, &in[0][0][0], &out2[0][0][0]);
for (i0 = 0; i0 < N0; i0++)
for (i1 = 0; i1 < N1; i1++)
for (i2 = 0; i2 < N2; i2++)
if ( fabs(out[i0][i1][i2][0]) > 1e-3 || fabs(out[i0][i1][i2][1]) > 1e-3
|| fabs(out2[i0][i1][i2][0]) > 1e-3 || fabs(out2[i0][i1][i2][1]) > 1e-3 )
printf("%6d %6d %6d: %20.10f %20.10f | %20.10f %20.10f\n",
i0, i1, i2, out[i0][i1][i2][0], out[i0][i1][i2][1],
out2[i0][i1][i2][0], out2[i0][i1][i2][1]);
fftw_destroy_plan(p);
fftw_cleanup();
return 0;
}
PetscErrorCode MatApply_USFFT_Private(Mat A, fftw_plan *plan, int direction, Vec x,Vec y)
{
#if 0
PetscErrorCode ierr;
PetscScalar *r_array, *y_array;
Mat_USFFT* = (Mat_USFFT*)(A->data);
#endif
PetscFunctionBegin;
#if 0
/* resample x to usfft->resample */
ierr = MatResample_USFFT_Private(A, x);CHKERRQ(ierr);
/* NB: for now we use outdim for both x and y; this will change once a full USFFT is implemented */
ierr = VecGetArray(usfft->resample,&r_array);CHKERRQ(ierr);
ierr = VecGetArray(y,&y_array);CHKERRQ(ierr);
if (!*plan) { /* create a plan then execute it*/
if (usfft->dof == 1) {
#if defined(PETSC_DEBUG_USFFT)
ierr = PetscPrintf(PetscObjectComm((PetscObject)A), "direction = %d, usfft->ndim = %d\n", direction, usfft->ndim);CHKERRQ(ierr);
for (int ii = 0; ii < usfft->ndim; ++ii) {
ierr = PetscPrintf(PetscObjectComm((PetscObject)A), "usfft->outdim[%d] = %d\n", ii, usfft->outdim[ii]);CHKERRQ(ierr);
}
#endif
switch (usfft->dim) {
case 1:
*plan = fftw_plan_dft_1d(usfft->outdim[0],(fftw_complex*)x_array,(fftw_complex*)y_array,direction,usfft->p_flag);
break;
case 2:
*plan = fftw_plan_dft_2d(usfft->outdim[0],usfft->outdim[1],(fftw_complex*)x_array,(fftw_complex*)y_array,direction,usfft->p_flag);
break;
case 3:
*plan = fftw_plan_dft_3d(usfft->outdim[0],usfft->outdim[1],usfft->outdim[2],(fftw_complex*)x_array,(fftw_complex*)y_array,direction,usfft->p_flag);
break;
default:
*plan = fftw_plan_dft(usfft->ndim,usfft->outdim,(fftw_complex*)x_array,(fftw_complex*)y_array,direction,usfft->p_flag);
break;
}
fftw_execute(*plan);
} /* if (dof == 1) */
else { /* if (dof > 1) */
*plan = fftw_plan_many_dft(/*rank*/usfft->ndim, /*n*/usfft->outdim, /*howmany*/usfft->dof,
(fftw_complex*)x_array, /*nembed*/usfft->outdim, /*stride*/usfft->dof, /*dist*/1,
(fftw_complex*)y_array, /*nembed*/usfft->outdim, /*stride*/usfft->dof, /*dist*/1,
/*sign*/direction, /*flags*/usfft->p_flag);
fftw_execute(*plan);
} /* if (dof > 1) */
} /* if (!*plan) */
else { /* if (*plan) */
/* use existing plan */
fftw_execute_dft(*plan,(fftw_complex*)x_array,(fftw_complex*)y_array);
}
ierr = VecRestoreArray(y,&y_array);CHKERRQ(ierr);
ierr = VecRestoreArray(x,&x_array);CHKERRQ(ierr);
#endif
PetscFunctionReturn(0);
} /* MatApply_USFFT_Private() */
void
MPC::compute_g_G(double &g_0, vector<double> &g_G, int N)
{
double L = PtclRef->Lattice.WignerSeitzRadius;
double Linv = 1.0/L;
double Linv3 = Linv*Linv*Linv;
// create an FFTW plan
Array<complex<double>,3> rBox(N,N,N);
Array<complex<double>,3> GBox(N,N,N);
// app_log() << "Doing " << N << " x " << N << " x " << N << " FFT.\n";
//create BC handler
DTD_BConds<double,3,SUPERCELL_BULK> mybc(PtclRef->Lattice);
// Fill the real-space array with f(r)
double Ninv = 1.0/(double)N;
TinyVector<double,3> u, r;
for (int ix=0; ix<N; ix++)
{
u[0] = Ninv*ix;
for (int iy=0; iy<N; iy++)
{
u[1] = Ninv*iy;
for (int iz=0; iz<N; iz++)
{
u[2] = Ninv*iz;
r = PtclRef->Lattice.toCart (u);
//DTD_BConds<double,3,SUPERCELL_BULK>::apply (PtclRef->Lattice, r);
//double rmag = std::sqrt(dot(r,r));
double rmag = std::sqrt(mybc.apply_bc(r));
if (rmag < L)
rBox(ix,iy,iz) = -0.5*rmag*rmag*Linv3 + 1.5*Linv;
else
rBox(ix,iy,iz) = 1.0/rmag;
}
}
}
fftw_plan fft = fftw_plan_dft_3d
(N, N, N, (fftw_complex*)rBox.data(),
(fftw_complex*) GBox.data(), 1, FFTW_ESTIMATE);
fftw_execute (fft);
fftw_destroy_plan (fft);
// Now, copy data into output, and add on analytic part
double norm = Ninv*Ninv*Ninv;
int numG = Gints.size();
for (int iG=0; iG < numG; iG++)
{
TinyVector<int,OHMMS_DIM> gint = Gints[iG];
for (int j=0; j<OHMMS_DIM; j++)
gint[j] = (gint[j] + N)%N;
g_G[iG] = norm * real(GBox(gint[0], gint[1], gint[2]));
}
g_0 = norm * real(GBox(0,0,0));
}
void fft3dCPU(T1* d_data, int nx, int ny, int nz)
{
cout << "Running forward xform 3d" << endl;
fftw_plan plan;
plan = fftw_plan_dft_3d(nz,
ny, nx, (fftw_complex*) d_data,
(fftw_complex*) d_data, FFTW_FORWARD, FFTW_ESTIMATE);
// Inverse transform 'gridData_d' in place.
fftw_execute(plan);
fftw_destroy_plan(plan);
}
convolution_plan::convolution_plan(int width, int height, int depth, int kw, int mode, int threadMaxCount) {
switch (mode) {
case 0:
this->width = width;
this->height = height;
this->depth = depth;
break;
case 1:
this->width = width + kw - 1;
this->height = height + kw - 1;
this->depth = depth + kw - 1;
break;
default:
std::cout << mode << std::endl;
throw std::invalid_argument("Warning: 3d convolution plan: Invalid mode");
}
if (threadMaxCount > 1) {
fftw_init_threads(); // This MUST come before all other fftw calls
fftw_plan_with_nthreads(threadMaxCount);
}
this->dim = 3;
this->kw = kw;
this->threadMaxCount = threadMaxCount;
fftw_complex* benchmarkArray1 = (fftw_complex*) fftw_malloc(sizeof(fftw_complex) * this->width * this->height * this->depth);
fftw_complex* benchmarkArray2 = (fftw_complex*) fftw_malloc(sizeof(fftw_complex) * this->width * this->height * this->depth);
this->forwardPlan = fftw_plan_dft_3d(this->depth, this->height, this->width, benchmarkArray1, benchmarkArray2, FFTW_FORWARD, FFTW_MEASURE);
this->backwardPlan = fftw_plan_dft_3d(this->depth, this->height, this->width, benchmarkArray1, benchmarkArray2, FFTW_BACKWARD, FFTW_MEASURE);
fftw_free(benchmarkArray1);
fftw_free(benchmarkArray2);
this->staticKernel = NULL;
}
void fft3d(fftw_complex * out, double * k, fftw_complex * in, int * n, double * delta)
{
fftw_plan plan;
plan = fftw_plan_dft_3d(n[0], n[1], n[2], in, out, FFTW_FORWARD, FFTW_ESTIMATE);
k[0] = 2 * M_PI / (n[0] * delta[0]);
k[1] = 2 * M_PI / (n[1] * delta[1]);
k[2] = 2 * M_PI / (n[2] * delta[2]);
fftw_execute(plan);
fftw_destroy_plan(plan);
for (int i = 0; i < n[0]*n[1]*n[2]; i++)
{
out[i] /= (n[0]*n[1]*n[2]);
}
}
WGSLIB_DECL bool fft_varmap_3d(
VarOut& Yout,
const std::vector<double>& data,
const std::vector<int>& has_point,
int N, int M, int K)
{
omp_set_num_threads(fft_get_num_threads());
fftw_plan_with_nthreads(fft_get_num_threads());
fftw_complex *z, *Z, *ZI;
fftw_complex *z2, *Z2;
fftw_complex *ni, *NI, *INI;
fftw_plan p;
int ZS = (2 * N + 1) * (2 * M + 1) * (2 * K + 1);
fft_lock();
z = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
Z = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
ZI = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
z2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
Z2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
ni = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
NI = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
INI = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
fft_unlock();
#pragma omp parallel for
for (int i = 0; i < ZS; ++i) {
z[i][REAL] = z[i][IMAG] = 0;
z2[i][REAL] = z2[i][IMAG] = 0;
ni[i][REAL] = ni[i][IMAG] = 0;
}
#pragma omp parallel for
for (int i = 0; i < N; ++i) {
for (int j = 0; j < M; ++j) {
for (int k = 0; k < K; ++k) {
double v = data _p(i, j, k);
int hp = has_point _p(i, j, k);
ni _p3(i, j, k)[REAL] = hp;
if (hp) {
z _p3(i, j, k)[REAL] = v;
z2 _p3(i, j, k)[REAL] = v * v;
}
}
}
}
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, z, Z, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, z2, Z2, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, ni, NI, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
#pragma omp parallel for
for (int h = 0; h < ZS; ++h) {
fftw_complex Z2_I;
fftw_complex Z_Z;
fftw_complex I_Z2;
fftw_complex I_I;
mul_conj(Z2_I, Z2[h], NI[h]);
mul_conj(Z_Z, Z[h], Z[h]);
mul_conj(I_Z2, NI[h], Z2[h]);
mul_conj(I_I, NI[h], NI[h]);
Z[h][REAL] = Z2_I[REAL] - 2 * Z_Z[REAL] + I_Z2[REAL];
Z[h][IMAG] = Z2_I[IMAG] - 2 * Z_Z[IMAG] + I_Z2[IMAG];
NI[h][REAL] = I_I[REAL];
NI[h][IMAG] = I_I[IMAG];
}
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, Z, ZI, FFTW_BACKWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, NI, INI, FFTW_BACKWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
#pragma omp parallel for
for (int hx = 0; hx < N; ++hx) {
for (int hy = 0; hy < M; ++hy) {
for (int hz = 0; hz < K; ++hz) {
double Y = ZI _p3(hx, hy, hz)[REAL] / ZS;
double N = INI _p3(hx, hy, hz)[REAL] / ZS;
if (N > 0.01) {
Yout.varmap _p(hx, hy, hz) = Y / (2 * N);
} else {
Yout.varmap _p(hx, hy, hz) = 0;
}
Yout.ni _p(hx, hy, hz) = N;
}
}
}
fft_lock();
fftw_destroy_plan(p);
fftw_free(z);
fftw_free(Z);
fftw_free(ZI);
fftw_free(z2);
fftw_free(Z2);
fftw_free(ni);
fftw_free(NI);
fftw_free(INI);
fft_unlock();
return true;
}
WGSLIB_DECL bool fft_crossvarmap_3d_declus(
VarOut& Yout,
const std::vector<double>& weigth1,
const std::vector<double>& data1,
const std::vector<int>& has_point1,
const std::vector<double>& data2,
const std::vector<int>& has_point2,
int N, int M, int K)
{
omp_set_num_threads(fft_get_num_threads());
fftw_plan_with_nthreads(fft_get_num_threads());
fftw_complex *i1i2, *I1I2;
fftw_complex *z1i2, *Z1I2;
fftw_complex *i1z2, *I1Z2;
fftw_complex *z1z2, *Z1Z2;
fftw_complex *w, *W;
fftw_complex *wz1, *WZ1;
fftw_complex *wz2, *WZ2;
fftw_complex *wz1z2, *WZ1Z2;
fftw_plan p;
int ZS = (2 * N + 1) * (2 * M + 1) * (2 * K + 1);
fft_lock();
i1i2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
I1I2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
z1i2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
Z1I2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
i1z2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
I1Z2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
z1z2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
Z1Z2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
w = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
W = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
wz1 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
WZ1 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
wz2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
WZ2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
wz1z2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
WZ1Z2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
fft_unlock();
#pragma omp parallel for
for (int i = 0; i < ZS; ++i) {
i1i2[i][REAL] = i1i2[i][IMAG] = 0;
I1I2[i][REAL] = I1I2[i][IMAG] = 0;
z1i2[i][REAL] = z1i2[i][IMAG] = 0;
Z1I2[i][REAL] = Z1I2[i][IMAG] = 0;
i1z2[i][REAL] = i1z2[i][IMAG] = 0;
I1Z2[i][REAL] = I1Z2[i][IMAG] = 0;
z1z2[i][REAL] = z1z2[i][IMAG] = 0;
Z1Z2[i][REAL] = Z1Z2[i][IMAG] = 0;
w[i][REAL] = w[i][IMAG] = 0;
W[i][REAL] = W[i][IMAG] = 0;
wz1[i][REAL] = wz1[i][IMAG] = 0;
WZ1[i][REAL] = WZ1[i][IMAG] = 0;
wz2[i][REAL] = wz2[i][IMAG] = 0;
WZ2[i][REAL] = WZ2[i][IMAG] = 0;
wz1z2[i][REAL] = wz1z2[i][IMAG] = 0;
WZ1Z2[i][REAL] = WZ1Z2[i][IMAG] = 0;
}
#pragma omp parallel for
for (int i = 0; i < N; ++i) {
for (int j = 0; j < M; ++j) {
for (int k = 0; k < K; ++k) {
double v1 = data1 _p(i, j, k);
double w1_ = weigth1 _p(i, j, k);
int i1 = has_point1 _p(i, j, k);
double v2 = data2 _p(i, j, k);
int i2 = has_point2 _p(i, j, k);
i1i2 _p3(i, j, k)[REAL] = i1 * i2;
z1i2 _p3(i, j, k)[REAL] = v1 * i2;
i1z2 _p3(i, j, k)[REAL] = i1 * v2;
z1z2 _p3(i, j, k)[REAL] = v1 * v2;
w _p3(i, j, k)[REAL] = w1_;
wz1 _p3(i, j, k)[REAL] = w1_ * v1;
wz2 _p3(i, j, k)[REAL] = w1_ * v2;
wz1z2 _p3(i, j, k)[REAL] = w1_ * v1 * v2;
}
}
}
///////////////////////////////////////////////////////////////////
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, i1i2, I1I2, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, z1i2, Z1I2, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, i1z2, I1Z2, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, z1z2, Z1Z2, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
////////////////////////////////////////////////////////////////////////////////////
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, w, W, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, wz1, WZ1, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, wz2, WZ2, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, wz1z2, WZ1Z2, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
#pragma omp parallel for
for (int h = 0; h < ZS; ++h) {
fftw_complex W_I1I2;
fftw_complex I1I2_W;
fftw_complex WZ1Z2_I1I2;
fftw_complex I1I2_WZ1Z2;
fftw_complex WZ2_Z1I2;
fftw_complex Z1I2_WZ2;
fftw_complex WZ1_I1Z2;
fftw_complex I1Z2_WZ1;
fftw_complex W_Z1Z2;
fftw_complex Z1Z2_W;
mul_conj(W_I1I2, W[h], I1I2[h]);
mul_conj(I1I2_W, I1I2[h], W[h]);
mul_conj(WZ1Z2_I1I2, WZ1Z2[h], I1I2[h]);
mul_conj(I1I2_WZ1Z2, I1I2[h], WZ1Z2[h]);
mul_conj(WZ2_Z1I2, WZ2[h], Z1I2[h]);
mul_conj(Z1I2_WZ2, Z1I2[h], WZ2[h]);
mul_conj(WZ1_I1Z2, WZ1[h], I1Z2[h]);
mul_conj(I1Z2_WZ1, I1Z2[h], WZ1[h]);
mul_conj(W_Z1Z2, W[h], Z1Z2[h]);
mul_conj(Z1Z2_W, Z1Z2[h], W[h]);
Z1Z2[h][REAL] =
WZ1Z2_I1I2[REAL] -
WZ2_Z1I2[REAL] -
WZ1_I1Z2[REAL] +
W_Z1Z2[REAL] +
Z1Z2_W[REAL] -
I1Z2_WZ1[REAL] -
Z1I2_WZ2[REAL] +
I1I2_WZ1Z2[REAL];
Z1Z2[h][IMAG] =
WZ1Z2_I1I2[IMAG] -
WZ2_Z1I2[IMAG] -
WZ1_I1Z2[IMAG] +
W_Z1Z2[IMAG] +
Z1Z2_W[IMAG] -
I1Z2_WZ1[IMAG] -
Z1I2_WZ2[IMAG] +
I1I2_WZ1Z2[IMAG];
W[h][REAL] = 2 * (W_I1I2[REAL] + I1I2_W[REAL]);
W[h][IMAG] = 2 * (W_I1I2[IMAG] + I1I2_W[IMAG]);
}
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, Z1Z2, z1z2, FFTW_BACKWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, W, w, FFTW_BACKWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
#pragma omp parallel for
for (int hx = 0; hx < N; ++hx) {
for (int hy = 0; hy < M; ++hy) {
for (int hz = 0; hz < K; ++hz) {
double Y = z1z2 _p3(hx, hy, hz)[REAL] / ZS;
double N = w _p3(hx, hy, hz)[REAL] / ZS;
if (N > 0.01) {
Yout.varmap _p(hx, hy, hz) = Y / (N);
}
else {
Yout.varmap _p(hx, hy, hz) = 0;
}
Yout.ni _p(hx, hy, hz) = N;
}
}
}
fft_lock();
fftw_destroy_plan(p);
fftw_free(i1i2);
fftw_free(I1I2);
fftw_free(z1i2);
fftw_free(Z1I2);
fftw_free(i1z2);
fftw_free(I1Z2);
fftw_free(z1z2);
fftw_free(Z1Z2);
fftw_free(w);
fftw_free(W);
fftw_free(wz1);
fftw_free(WZ1);
fftw_free(wz2);
fftw_free(WZ2);
fftw_free(wz1z2);
fftw_free(WZ1Z2);
fft_unlock();
return true;
}
WGSLIB_DECL bool fft_crossvarmap_3d(
VarOut& Yout,
const std::vector<double>& data1,
const std::vector<int>& has_point1,
const std::vector<double>& data2,
const std::vector<int>& has_point2,
int N, int M, int K)
{
omp_set_num_threads(fft_get_num_threads());
fftw_plan_with_nthreads(fft_get_num_threads());
fftw_complex *Z1I;
fftw_complex *I1I;
fftw_complex *z1z2, *Z1Z2;
fftw_complex *i2z1, *I2Z1;
fftw_complex *i1z2, *I1Z2;
fftw_complex *i1i2, *I1I2;
fftw_plan p;
int ZS = (2 * N + 1) * (2 * M + 1) * (2 * K + 1);
fft_lock();
Z1I = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
I1I = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
z1z2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
Z1Z2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
i2z1 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
I2Z1 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
i1z2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
I1Z2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
i1i2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
I1I2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
fft_unlock();
#pragma omp parallel for
for (int i = 0; i < ZS; ++i) {
z1z2[i][REAL] = z1z2[i][IMAG] = 0;
i1z2[i][REAL] = i1z2[i][IMAG] = 0;
i2z1[i][REAL] = i2z1[i][IMAG] = 0;
i1i2[i][REAL] = i1i2[i][IMAG] = 0;
}
#pragma omp parallel for
for (int i = 0; i < N; ++i) {
for (int j = 0; j < M; ++j) {
for (int k = 0; k < K; ++k) {
double v1 = data1 _p(i, j, k);
double v2 = data2 _p(i, j, k);
int hp1 = has_point1 _p(i, j, k);
int hp2 = has_point2 _p(i, j, k);
z1z2 _p3(i, j, k)[REAL] = v1 * v2;
i1z2 _p3(i, j, k)[REAL] = hp1 * v2;
i2z1 _p3(i, j, k)[REAL] = hp2 * v1;
i1i2 _p3(i, j, k)[REAL] = hp1 * hp2;
}
}
}
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, z1z2, Z1Z2, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, i1z2, I1Z2, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, i2z1, I2Z1, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, i1i2, I1I2, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
#pragma omp parallel for
for (int h = 0; h < ZS; ++h) {
fftw_complex A;
fftw_complex B;
fftw_complex C;
fftw_complex D;
fftw_complex I1_I2;
mul_conj(A, Z1Z2[h], I1I2[h]);
mul_conj(D, I1I2[h], Z1Z2[h]);
mul_conj(B, I2Z1[h], I1Z2[h]);
mul_conj(C, I1Z2[h], I2Z1[h]);
mul_conj(I1_I2, I1I2[h], I1I2[h]);
Z1Z2[h][REAL] = A[REAL] - B[REAL] - C[REAL] + D[REAL];
Z1Z2[h][IMAG] = A[IMAG] - B[IMAG] - C[IMAG] + D[IMAG];
I1I2[h][REAL] = I1_I2[REAL];
I1I2[h][IMAG] = I1_I2[IMAG];
}
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, Z1Z2, Z1I, FFTW_BACKWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, I1I2, I1I, FFTW_BACKWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
#pragma omp parallel for
for (int hx = 0; hx < N; ++hx) {
for (int hy = 0; hy < M; ++hy) {
for (int hz = 0; hz < K; ++hz) {
double Y = Z1I _p3(hx, hy, hz)[REAL] / ZS;
double N = I1I _p3(hx, hy, hz)[REAL] / ZS;
if (N > 0.01) {
Yout.varmap _p(hx, hy, hz) = Y / (2 * N);
} else {
Yout.varmap _p(hx, hy, hz) = 0;
}
Yout.ni _p(hx, hy, hz) = N;
}
}
}
fft_lock();
fftw_destroy_plan(p);
fftw_free(Z1I);
fftw_free(I1I);
fftw_free(z1z2);
fftw_free(Z1Z2);
fftw_free(i2z1);
fftw_free(I2Z1);
fftw_free(i1z2);
fftw_free(I1Z2);
fftw_free(i1i2);
fftw_free(I1I2);
fft_unlock();
return true;
}
WGSLIB_DECL bool fft_varmap_3d_declus(
VarOut& Yout,
const std::vector<double>& data,
const std::vector<double>& weigth,
const std::vector<int>& has_point,
int N, int M, int K)
{
omp_set_num_threads(fft_get_num_threads());
fftw_plan_with_nthreads(fft_get_num_threads());
fftw_complex *z, *Z, *ZI;
fftw_complex *z2, *Z2;
fftw_complex *ni, *NI, *INI;
fftw_complex *w, *W;
fftw_complex *z2w, *Z2W;
fftw_complex *zw, *ZW;
fftw_plan p;
int ZS = (2 * N + 1) * (2 * M + 1) * (2 * K + 1);
fft_lock();
z = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
Z = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
ZI = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
z2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
Z2 = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
ni = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
NI = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
INI = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
w = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
W = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
z2w = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
Z2W = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
zw = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
ZW = (fftw_complex*)fftw_malloc(sizeof(fftw_complex)* ZS);
fft_unlock();
#pragma omp parallel for
for (int i = 0; i < ZS; ++i) {
z[i][REAL] = z[i][IMAG] = 0;
z2[i][REAL] = z2[i][IMAG] = 0;
ni[i][REAL] = ni[i][IMAG] = 0;
w[i][REAL] = w[i][IMAG] = 0;
z2w[i][REAL] = z2w[i][IMAG] = 0;
zw[i][REAL] = zw[i][IMAG] = 0;
}
#pragma omp parallel for
for (int i = 0; i < N; ++i) {
for (int j = 0; j < M; ++j) {
for (int k = 0; k < K; ++k) {
double v = data _p(i, j, k);
double w_ = weigth _p(i, j, k);
int hp = has_point _p(i, j, k);
ni _p3(i, j, k)[REAL] = hp;
if (hp) {
z _p3(i, j, k)[REAL] = v;
z2 _p3(i, j, k)[REAL] = v * v;
zw _p3(i, j, k)[REAL] = v * w_;
z2w _p3(i, j, k)[REAL] = v * v * w_;
w _p3(i, j, k)[REAL] = w_;
}
}
}
}
//LEVA z para o dominio da frequencia
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, z, Z, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, w, W, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, zw, ZW, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, z2w, Z2W, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, z2, Z2, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, ni, NI, FFTW_FORWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
//Realiza convolução
#pragma omp parallel for
for (int h = 0; h < ZS; ++h) {
fftw_complex Z2W_I;
fftw_complex ZW_Z;
fftw_complex W_Z2;
fftw_complex Z2_W;
fftw_complex Z_ZW;
fftw_complex I_Z2W;
fftw_complex W_I;
fftw_complex I_W;
mul_conj(Z2W_I, Z2W[h], NI[h]);
mul_conj(ZW_Z, ZW[h], Z[h]);
mul_conj(W_Z2, W[h], Z2[h]);
mul_conj(Z2_W, Z2[h], W[h]);
mul_conj(Z_ZW, Z[h], ZW[h]);
mul_conj(I_Z2W, NI[h], Z2W[h]);
mul_conj(I_W, NI[h], W[h]);
mul_conj(W_I, W[h], NI[h]);
Z[h][REAL] = Z2W_I[REAL] - 2 * ZW_Z[REAL] + W_Z2[REAL] +
Z2_W[REAL] - 2 * Z_ZW[REAL] + I_Z2W[REAL];
Z[h][IMAG] = Z2W_I[IMAG] - 2 * ZW_Z[IMAG] + W_Z2[IMAG] +
Z2_W[IMAG] - 2 * Z_ZW[IMAG] + I_Z2W[IMAG];
NI[h][REAL] = 2 * (W_I[REAL] + I_W[REAL]);
NI[h][IMAG] = 2 * (W_I[IMAG] + I_W[IMAG]);
}
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, Z, ZI, FFTW_BACKWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
fft_lock();
p = fftw_plan_dft_3d(2 * N + 1, 2 * M + 1, 2 * K + 1, NI, INI, FFTW_BACKWARD, FFTW_ESTIMATE);
fft_unlock();
fftw_execute(p);
#pragma omp parallel for
for (int hx = 0; hx < N; ++hx) {
for (int hy = 0; hy < M; ++hy) {
for (int hz = 0; hz < K; ++hz) {
double Y = ZI _p3(hx, hy, hz)[REAL] / ZS;
double N = INI _p3(hx, hy, hz)[REAL] / ZS;
if (N > 0.01) {
Yout.varmap _p(hx, hy, hz) = Y / N;
}
else {
Yout.varmap _p(hx, hy, hz) = 0;
}
Yout.ni _p(hx, hy, hz) = N;
}
}
}
fft_lock();
fftw_destroy_plan(p);
fftw_free(z);
fftw_free(Z);
fftw_free(ZI);
fftw_free(z2);
fftw_free(Z2);
fftw_free(zw);
fftw_free(ZW);
fftw_free(z2w);
fftw_free(Z2W);
fftw_free(w);
fftw_free(W);
fftw_free(ni);
fftw_free(NI);
fftw_free(INI);
fft_unlock();
return true;
}
void
MPC::init_spline()
{
Array<complex<double>,3> rBox(SplineDim[0], SplineDim[1], SplineDim[2]),
GBox(SplineDim[0], SplineDim[1], SplineDim[2]);
Array<double,3> splineData(SplineDim[0], SplineDim[1], SplineDim[2]);
GBox = complex<double>();
Vconst = 0.0;
// Now fill in elements of GBox
double vol = PtclRef->Lattice.Volume;
double volInv = 1.0/vol;
for (int iG=0; iG < Gvecs.size(); iG++)
{
TinyVector<int,OHMMS_DIM> gint = Gints[iG];
PosType G = Gvecs[iG];
double G2 = dot(G,G);
TinyVector<int,OHMMS_DIM> index;
for (int j=0; j<OHMMS_DIM; j++)
index[j] = (gint[j] + SplineDim[j]) % SplineDim[j];
if (!(index[0]==0 && index[1]==0 && index[2]==0))
{
GBox(index[0], index[1], index[2]) = vol *
Rho_G[iG] * (4.0*M_PI*volInv/G2 - f_G[iG]);
Vconst -= 0.5 * vol * vol * norm(Rho_G[iG])
* (4.0*M_PI*volInv/G2 - f_G[iG]);
}
}
// G=0 component calculated seperately
GBox(0,0,0) = -vol * f_0 * Rho_G[0];
Vconst += 0.5 * vol * vol * f_0 * norm(Rho_G[0]);
app_log() << " Constant potential = " << Vconst << endl;
fftw_plan fft = fftw_plan_dft_3d
(SplineDim[0], SplineDim[1], SplineDim[2], (fftw_complex*)GBox.data(),
(fftw_complex*) rBox.data(), -1, FFTW_ESTIMATE);
fftw_execute (fft);
fftw_destroy_plan (fft);
for (int i0=0; i0<SplineDim[0]; i0++)
for (int i1=0; i1<SplineDim[1]; i1++)
for (int i2=0; i2<SplineDim[2]; i2++)
splineData(i0, i1, i2) = real(rBox(i0,i1,i2));
BCtype_d bc0, bc1, bc2;
Ugrid grid0, grid1, grid2;
grid0.start=0.0;
grid0.end=1.0;
grid0.num = SplineDim[0];
grid1.start=0.0;
grid1.end=1.0;
grid1.num = SplineDim[1];
grid2.start=0.0;
grid2.end=1.0;
grid2.num = SplineDim[2];
bc0.lCode = bc0.rCode = PERIODIC;
bc1.lCode = bc1.rCode = PERIODIC;
bc2.lCode = bc2.rCode = PERIODIC;
VlongSpline = create_UBspline_3d_d (grid0, grid1, grid2, bc0, bc1, bc2,
splineData.data());
// grid0.num = PtclRef->Density_r.size(0);
// grid1.num = PtclRef->Density_r.size(1);
// grid2.num = PtclRef->Density_r.size(2);
// DensitySpline = create_UBspline_3d_d (grid0, grid1, grid2, bc0, bc1, bc2,
// PtclRef->Density_r.data());
}
/*
*
* To use FFTW in parallel, the lattice needs to be redistributed;
* this Fourier-transform function takes a number (n) of 3D blocks;
* the n blocks shall be destributed over as many prosesses as possible;
* then the parallel FFTW is called, destributing the z-axis over remaining processes
*
*/
void
qpb_ft(qpb_complex **out, qpb_complex **in, int n, int mom[][4], int nmom)
{
int rank = problem_params.proc_id;
int nprocs = problem_params.nprocs;
int Lz = problem_params.g_dim[1];
int Ly = problem_params.g_dim[2];
int Lx = problem_params.g_dim[3];
int lz = problem_params.l_dim[1];
int ly = problem_params.l_dim[2];
int lx = problem_params.l_dim[3];
int vol3d = Lx*Ly*Lz;
int lvol3d = lx*ly*lz;
MPI_Comm comm_cart = problem_params.mpi_comm_cart;
int nprocs_n = 0;
for(int i=1; i<nprocs+1; i++)
if((n % i) == 0)
nprocs_n = i;
int n_loc = n / nprocs_n;
fftw_complex *corr[n_loc];
qpb_complex *swap = NULL;
if(rank < nprocs_n)
for(int i=0; i<n_loc; i++)
corr[i] = fftw_malloc(sizeof(fftw_complex)*vol3d);
if(rank < nprocs_n)
swap = qpb_alloc(sizeof(qpb_complex)*vol3d);
for(int i=0; i<n_loc; i++)
for(int j=0; j<nprocs_n; j++)
{
MPI_Gather(in[i+j*n_loc], lvol3d*sizeof(qpb_complex), MPI_BYTE,
swap, lvol3d*sizeof(qpb_complex), MPI_BYTE, j, MPI_COMM_WORLD);
if(rank == j)
{
for(int p=0; p<nprocs; p++)
{
int coords[ND-1];
qpb_complex *ptr = swap + p*lvol3d;
MPI_Cart_coords(comm_cart, p, ND-1, coords);
int zoff = coords[0]*lz;
int yoff = coords[1]*ly;
int xoff = coords[2]*lx;
for(int z=zoff; z<lz+zoff; z++)
for(int y=yoff; y<ly+yoff; y++)
for(int x=xoff; x<lx+xoff; x++)
{
corr[i][x + y*Lx + z*Lx*Ly][0] = ptr->re;
corr[i][x + y*Lx + z*Lx*Ly][1] = ptr->im;
ptr++;
}
}
}
MPI_Barrier(MPI_COMM_WORLD);
}
if(rank < nprocs_n)
for(int i=0; i<n_loc; i++)
{
fftw_plan plan = fftw_plan_dft_3d(Lz, Ly, Lx, corr[i], corr[i],
FFTW_FORWARD, FFTW_ESTIMATE);
fftw_execute(plan);
fftw_destroy_plan(plan);
}
if(rank < nprocs_n)
for(int i=0; i<n_loc; i++)
for(int p=0; p<nmom; p++)
{
int kx = (Lx + mom[p][3]) % Lx;
int ky = (Ly + mom[p][2]) % Ly;
int kz = (Lz + mom[p][1]) % Lz;
out[i][p] = (qpb_complex){corr[i][kx + ky*Lx + kz*Lx*Ly][0],
corr[i][kx + ky*Lx + kz*Lx*Ly][1]};
}
for(int p=1; p<nprocs_n; p++)
for(int i=0; i<n_loc; i++)
{
if(rank == 0)
MPI_Recv(out[p*n_loc+i], nmom*sizeof(qpb_complex), MPI_BYTE, p, p, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
if(rank == p)
MPI_Send(out[i], nmom*sizeof(qpb_complex), MPI_BYTE, 0, p, MPI_COMM_WORLD);
}
if(rank < nprocs_n)
{
for(int i=0; i<n_loc; i++)
fftw_free(corr[i]);
free(swap);
}
return;
}
bool c_FourierTransfrom::ifftw_complex_3d(const Mat_<Vec6d> &_input,
Mat_<Vec6d> &_output)
{
size_t height = _input.rows;
size_t width = _input.cols;
size_t n_channels = _input.channels() / 2;
size_t n_pixels = height * width;
size_t n_data = n_pixels * n_channels;
fftw_complex *in, *out;
fftw_plan p;
in = (fftw_complex *)fftw_malloc(sizeof(fftw_complex) * n_data);
out = (fftw_complex *)fftw_malloc(sizeof(fftw_complex) * n_data);
p = fftw_plan_dft_3d(height, width, n_channels, in, out, FFTW_BACKWARD,
FFTW_ESTIMATE);
/*!< prepare the data */
for (size_t i_row = 0; i_row < height; ++i_row)
{
const Vec3d *p = _input.ptr<Vec3d>(i_row);
for (size_t i_col = 0; i_col < width; ++i_col)
{
size_t index = i_row * width + i_col;
for (size_t k = 0; k < n_channels; ++k)
{
in[n_pixels * k + index][0] = p[i_col][k];
in[n_pixels * k + index][1] = p[i_col][k + n_channels];
}
#if 0
in[index][0] = p[i_col][4];
in[index][1] = p[i_col][5];
in[n_pixels + index][0] = p[i_col][2];
in[n_pixels + index][1] = p[i_col][3];
in[n_pixels * 2 + index][0] = p[i_col][0];
in[n_pixels * 2 + index][1] = p[i_col][1];
#endif
}
}
fftw_execute(p);
/*!< write back data */
_output = Mat_<Vec6d>::zeros(_input.size());
for (size_t i_row = 0; i_row < height; ++i_row)
{
Vec6d *p = _output.ptr<Vec6d>(i_row);
for (size_t i_col = 0; i_col < width; ++i_col)
{
size_t index = i_row * width + i_col;
for (size_t k = 0; k < n_channels; ++k)
{
p[i_col][k] = out[n_pixels * k + index][0];
p[i_col][k + n_channels] = out[n_pixels * k + index][1];
}
#if 0
p[i_col][0] = out[n_pixels * 2 + index][0];
p[i_col][1] = out[n_pixels + index][0];
p[i_col][2] = out[index][0];
p[i_col][3] = out[n_pixels * 2 + index][1];
p[i_col][4] = out[n_pixels + index][1];
p[i_col][5] = out[index][1];
#endif
}
}
_output /= n_data;
fftw_destroy_plan(p);
fftw_free(in);
fftw_free(out);
return true;
}
void EinsplineSetBuilder::ReadBands_ESHDF(int spin, EinsplineSetExtended<double>* orbitalSet)
{
update_token(__FILE__,__LINE__,"ReadBands_ESHDF:double");
ReportEngine PRE("EinsplineSetBuilder","ReadBands_ESHDF(EinsplineSetExtended<double>*");
vector<AtomicOrbital<double> > realOrbs(AtomicOrbitals.size());
for (int iat=0; iat<realOrbs.size(); iat++)
{
AtomicOrbital<complex<double> > &corb (AtomicOrbitals[iat]);
realOrbs[iat].set_pos (corb.Pos);
realOrbs[iat].set_lmax (corb.lMax);
realOrbs[iat].set_cutoff (corb.CutoffRadius);
realOrbs[iat].set_spline (corb.SplineRadius, corb.SplinePoints);
realOrbs[iat].set_polynomial (corb.PolyRadius, corb.PolyOrder);
realOrbs[iat].Lattice = corb.Lattice;
}
bool root = myComm->rank()==0;
// bcast other stuff
myComm->bcast (NumDistinctOrbitals);
myComm->bcast (NumValenceOrbs);
myComm->bcast (NumCoreOrbs);
int N = NumDistinctOrbitals;
orbitalSet->kPoints.resize(N);
orbitalSet->MakeTwoCopies.resize(N);
orbitalSet->StorageValueVector.resize(N);
orbitalSet->BlendValueVector.resize(N);
orbitalSet->StorageLaplVector.resize(N);
orbitalSet->BlendLaplVector.resize(N);
orbitalSet->StorageGradVector.resize(N);
orbitalSet->BlendGradVector.resize(N);
orbitalSet->StorageHessVector.resize(N);
orbitalSet->StorageGradHessVector.resize(N);
orbitalSet->phase.resize(N);
orbitalSet->eikr.resize(N);
orbitalSet->NumValenceOrbs = NumValenceOrbs;
orbitalSet->NumCoreOrbs = NumCoreOrbs;
orbitalSet->FirstOrderSplines.resize(IonPos.size());
// Read in k-points
int numOrbs = orbitalSet->getOrbitalSetSize();
int num = 0;
vector<BandInfo>& SortBands(*FullBands[spin]);
if (root)
{
for (int iorb=0; iorb<N; iorb++)
{
int ti = SortBands[iorb].TwistIndex;
PosType twist = TwistAngles[ti];
orbitalSet->kPoints[iorb] = orbitalSet->PrimLattice.k_cart(twist);
orbitalSet->MakeTwoCopies[iorb] =
(num < (numOrbs-1)) && SortBands[iorb].MakeTwoCopies;
num += orbitalSet->MakeTwoCopies[iorb] ? 2 : 1;
}
PosType twist0 = TwistAngles[SortBands[0].TwistIndex];
for (int i=0; i<OHMMS_DIM; i++)
if (std::fabs(std::fabs(twist0[i]) - 0.5) < 1.0e-8)
orbitalSet->HalfG[i] = 1;
else
orbitalSet->HalfG[i] = 0;
EinsplineSetBuilder::RotateBands_ESHDF(spin, orbitalSet);
}
myComm->bcast(orbitalSet->kPoints);
myComm->bcast(orbitalSet->MakeTwoCopies);
myComm->bcast(orbitalSet->HalfG);
// First, check to see if we have already read this in
H5OrbSet set(H5FileName, spin, N);
bool havePsir=!ReadGvectors_ESHDF();
app_log() << "MeshSize = (" << MeshSize[0] << ", "
<< MeshSize[1] << ", " << MeshSize[2] << ")\n";
//int nx, ny, nz, bi, ti;
int nx, ny, nz;
nx=MeshSize[0];
ny=MeshSize[1];
nz=MeshSize[2];
Ugrid x_grid, y_grid, z_grid;
BCtype_d xBC, yBC, zBC;
if (orbitalSet->HalfG[0])
{
xBC.lCode = ANTIPERIODIC;
xBC.rCode = ANTIPERIODIC;
}
else
{
xBC.lCode = PERIODIC;
xBC.rCode = PERIODIC;
}
if (orbitalSet->HalfG[1])
{
yBC.lCode = ANTIPERIODIC;
yBC.rCode = ANTIPERIODIC;
}
else
{
yBC.lCode = PERIODIC;
yBC.rCode = PERIODIC;
}
if (orbitalSet->HalfG[2])
{
zBC.lCode = ANTIPERIODIC;
zBC.rCode = ANTIPERIODIC;
}
else
{
zBC.lCode = PERIODIC;
zBC.rCode = PERIODIC;
}
x_grid.start = 0.0;
x_grid.end = 1.0;
x_grid.num = nx;
y_grid.start = 0.0;
y_grid.end = 1.0;
y_grid.num = ny;
z_grid.start = 0.0;
z_grid.end = 1.0;
z_grid.num = nz;
// Create the multiUBspline object
orbitalSet->MultiSpline =
create_multi_UBspline_3d_d (x_grid, y_grid, z_grid, xBC, yBC, zBC, NumValenceOrbs);
if (HaveOrbDerivs)
{
orbitalSet->FirstOrderSplines.resize(IonPos.size());
for (int ion=0; ion<IonPos.size(); ion++)
for (int dir=0; dir<OHMMS_DIM; dir++)
orbitalSet->FirstOrderSplines[ion][dir] =
create_multi_UBspline_3d_d (x_grid, y_grid, z_grid, xBC, yBC, zBC, NumValenceOrbs);
}
//////////////////////////////////////
// Create the MuffinTin APW splines //
//////////////////////////////////////
orbitalSet->MuffinTins.resize(NumMuffinTins);
for (int tin=0; tin<NumMuffinTins; tin++)
{
orbitalSet->MuffinTins[tin].Atom = tin;
orbitalSet->MuffinTins[tin].set_center (MT_centers[tin]);
orbitalSet->MuffinTins[tin].set_lattice(Lattice);
orbitalSet->MuffinTins[tin].init_APW
(MT_APW_rgrids[tin], MT_APW_lmax[tin],
NumValenceOrbs);
}
for (int iat=0; iat<realOrbs.size(); iat++)
{
realOrbs[iat].set_num_bands(NumValenceOrbs);
realOrbs[iat].allocate();
}
int isComplex;
if (root)
{
HDFAttribIO<int> h_isComplex(isComplex);
h_isComplex.read(H5FileID, "/electrons/psi_r_is_complex");
}
myComm->bcast(isComplex);
bool isCore = bcastSortBands(spin,N,root);
if(isCore)
{
APP_ABORT("Core states not supported by ES-HDF yet.");
}
//this is common
Array<double,3> splineData(nx,ny,nz);
if(havePsir)
{
if(isComplex)
{
app_log() << " Reading complex psi_r and convert to real" << endl;
Array<complex<double>,3> rawData;
for(int iorb=0,ival=0; iorb<N; ++iorb, ++ival)
{
int ti=SortBands[iorb].TwistIndex;
if(root)
{
ostringstream path;
path << "/electrons/kpoint_" << SortBands[iorb].TwistIndex
<< "/spin_" << spin << "/state_" << SortBands[iorb].BandIndex << "/psi_r";
HDFAttribIO<Array<complex<double>,3> > h_splineData(rawData);
h_splineData.read(H5FileID, path.str().c_str());
}
myComm->bcast(rawData);
//multiply twist factor and project on the real
fix_phase_c2r(rawData,splineData,TwistAngles[ti]);
set_multi_UBspline_3d_d (orbitalSet->MultiSpline, ival, splineData.data());
}
}
else
{
app_log() << " Reading real psi_r" << endl;
for(int iorb=0,ival=0; iorb<N; ++iorb, ++ival)
{
if(root)
{
ostringstream path;
path << "/electrons/kpoint_" << SortBands[iorb].TwistIndex
<< "/spin_" << spin << "/state_" << SortBands[iorb].BandIndex << "/psi_r";
HDFAttribIO<Array<double,3> > h_splineData(splineData);
h_splineData.read(H5FileID, path.str().c_str());
}
myComm->bcast(splineData);
set_multi_UBspline_3d_d (orbitalSet->MultiSpline, ival, splineData.data());
}
}
}
else
{
Array<ComplexType,3> FFTbox;
FFTbox.resize(MeshSize[0], MeshSize[1], MeshSize[2]);
fftw_plan FFTplan = fftw_plan_dft_3d
(MeshSize[0], MeshSize[1], MeshSize[2],
reinterpret_cast<fftw_complex*>(FFTbox.data()),
reinterpret_cast<fftw_complex*>(FFTbox.data()),
+1, FFTW_ESTIMATE);
for(int iorb=0,ival=0; iorb<N; ++iorb, ++ival)
{
Vector<complex<double> > cG;
int ncg=0;
int ti=SortBands[iorb].TwistIndex;
if(root)
{
ostringstream path;
path << "/electrons/kpoint_" << SortBands[iorb].TwistIndex
<< "/spin_" << spin << "/state_" << SortBands[iorb].BandIndex << "/psi_g";
HDFAttribIO<Vector<complex<double> > > h_cG(cG);
h_cG.read (H5FileID, path.str().c_str());
ncg=cG.size();
}
myComm->bcast(ncg);
if(ncg != Gvecs[0].size())
{
APP_ABORT("Failed : ncg != Gvecs[0].size()");
}
if(!root)
cG.resize(ncg);
myComm->bcast(cG);
unpack4fftw(cG,Gvecs[0],MeshSize,FFTbox);
fftw_execute (FFTplan);
fix_phase_rotate_c2r(FFTbox,splineData,TwistAngles[ti]);
set_multi_UBspline_3d_d (orbitalSet->MultiSpline, ival, splineData.data());
}
fftw_destroy_plan(FFTplan);
}
for(int iorb=0,ival=0; iorb<N; ++iorb, ++ival)
{
// Read atomic orbital information
for (int iat=0; iat<realOrbs.size(); iat++)
{
app_log() << "Reading orbital " << iat << " for band " << ival << endl;
AtomicOrbital<double> &orb = realOrbs[iat];
//AtomicOrbital<complex<double> > &orb = realOrbs[iat];
Array<complex<double>,2> radial_spline(orb.SplinePoints,orb.Numlm),
poly_coefs(orb.PolyOrder+1,orb.Numlm);
int ti = SortBands[iorb].TwistIndex;
if (root)
{
int bi = SortBands[iorb].BandIndex;
ostringstream path;
path << "/electrons/kpoint_" << ti << "/spin_" << spin << "/state_" << bi << "/";
ostringstream spline_path, poly_path;
spline_path << path.str() << "radial_spline_" << iat;
poly_path << path.str() << "poly_coefs_" << iat;
HDFAttribIO<Array<complex<double>,2> > h_radial_spline(radial_spline);
HDFAttribIO<Array<complex<double>,2> > h_poly_coefs(poly_coefs);
h_radial_spline.read(H5FileID, spline_path.str().c_str());
h_poly_coefs.read (H5FileID, poly_path.str().c_str());
}
myComm->bcast(radial_spline);
myComm->bcast(poly_coefs);
realOrbs[iat].set_band (ival, radial_spline, poly_coefs, TwistAngles[ti]);
}
}
for(int iorb=0,ival=0; iorb<N; ++iorb, ++ival)
{
// Now read muffin tin data
for (int tin=0; tin<NumMuffinTins; tin++)
{
// app_log() << "Reading data for muffin tin " << tin << endl;
PosType twist, k;
int lmax = MT_APW_lmax[tin];
int numYlm = (lmax+1)*(lmax+1);
Array<complex<double>,2>
u_lm_r(numYlm, MT_APW_num_radial_points[tin]);
Array<complex<double>,1> du_lm_dr (numYlm);
int ti = SortBands[iorb].TwistIndex;
if (root)
{
int bi = SortBands[iorb].BandIndex;
twist = TwistAngles[ti];
k = orbitalSet->PrimLattice.k_cart(twist);
string uName = MuffinTinPath (ti, bi,tin) + "u_lm_r";
string duName = MuffinTinPath (ti, bi,tin) + "du_lm_dr";
HDFAttribIO<Array<complex<double>,2> > h_u_lm_r(u_lm_r);
HDFAttribIO<Array<complex<double>,1> > h_du_lm_dr(du_lm_dr);
h_u_lm_r.read(H5FileID, uName.c_str());
h_du_lm_dr.read(H5FileID, duName.c_str());
}
myComm->bcast(u_lm_r);
myComm->bcast(du_lm_dr);
myComm->bcast(k);
double Z = (double)IonTypes(tin);
OrbitalSet->MuffinTins[tin].set_APW (ival, k, u_lm_r, du_lm_dr, Z);
}
}
//FIX HaveOrbDerivs after debugging
// // Now read orbital derivatives if we have them
// if (HaveOrbDerivs) {
// for (int ion=0; ion<IonPos.size(); ion++)
// for (int dim=0; dim<OHMMS_DIM; dim++) {
// if (root) {
// int ti = SortBands[iorb].TwistIndex;
// int bi = SortBands[iorb].BandIndex;
//
// app_log() << "Reading orbital derivative for ion " << ion
// << " dim " << dim << " spin " << spin << " band "
// << bi << " kpoint " << ti << endl;
// ostringstream path;
// path << "/electrons/kpoint_" << ti << "/spin_" << spin << "/state_" << bi << "/"
// << "dpsi_" << ion << "_" << dim << "_r";
// string psirName = path.str();
// if (isComplex) {
// HDFAttribIO<Array<complex<double>,3> > h_rawData(rawData);
// h_rawData.read(H5FileID, psirName.c_str());
// if ((rawData.size(0) != nx) ||
// (rawData.size(1) != ny) ||
// (rawData.size(2) != nz)) {
// fprintf (stderr, "Error in EinsplineSetBuilder::ReadBands.\n");
// fprintf (stderr, "Extended orbitals should all have the same dimensions\n");
// abort();
// }
//#pragma omp parallel for
// for (int ix=0; ix<nx; ix++) {
// PosType ru;
// ru[0] = (RealType)ix / (RealType)nx;
// for (int iy=0; iy<ny; iy++) {
// ru[1] = (RealType)iy / (RealType)ny;
// for (int iz=0; iz<nz; iz++) {
// ru[2] = (RealType)iz / (RealType)nz;
// double phi = -2.0*M_PI*dot (ru, TwistAngles[ti]);
// double s, c;
// sincos(phi, &s, &c);
// complex<double> phase(c,s);
// complex<double> z = phase*rawData(ix,iy,iz);
// splineData(ix,iy,iz) = z.real();
// }
// }
// }
// }
// else {
// HDFAttribIO<Array<double,3> > h_splineData(splineData);
// h_splineData.read(H5FileID, psirName.c_str());
// if ((splineData.size(0) != nx) ||
// (splineData.size(1) != ny) ||
// (splineData.size(2) != nz)) {
// fprintf (stderr, "Error in EinsplineSetBuilder::ReadBands.\n");
// fprintf (stderr, "Extended orbitals should all have the same dimensions\n");
// abort();
// }
// }
// }
// myComm->bcast(splineData);
// set_multi_UBspline_3d_d
// (orbitalSet->FirstOrderSplines[ion][dim], ival, splineData.data());
// }
// }
//
//
//
orbitalSet->AtomicOrbitals = realOrbs;
for (int i=0; i<orbitalSet->AtomicOrbitals.size(); i++)
orbitalSet->AtomicOrbitals[i].registerTimers();
//ExtendedMap_d[set] = orbitalSet->MultiSpline;
}
void EinsplineSetBuilder::ReadBands_ESHDF(int spin, EinsplineSetExtended<complex<double > >* orbitalSet)
{
update_token(__FILE__,__LINE__,"ReadBands_ESHDF:complex");
ReportEngine PRE("EinsplineSetBuilder","ReadBands_ESHDF(EinsplineSetExtended<complex<double > >*");
Timer c_prep, c_unpack,c_fft, c_phase, c_spline, c_newphase, c_h5, c_init;
double t_prep=0.0, t_unpack=0.0, t_fft=0.0, t_phase=0.0, t_spline=0.0, t_newphase=0.0, t_h5=0.0, t_init=0.0;
c_prep.restart();
bool root = myComm->rank()==0;
vector<BandInfo>& SortBands(*FullBands[spin]);
// bcast other stuff
myComm->bcast (NumDistinctOrbitals);
myComm->bcast (NumValenceOrbs);
myComm->bcast (NumCoreOrbs);
int N = NumDistinctOrbitals;
orbitalSet->kPoints.resize(N);
orbitalSet->MakeTwoCopies.resize(N);
orbitalSet->StorageValueVector.resize(N);
orbitalSet->BlendValueVector.resize(N);
orbitalSet->StorageLaplVector.resize(N);
orbitalSet->BlendLaplVector.resize(N);
orbitalSet->StorageGradVector.resize(N);
orbitalSet->BlendGradVector.resize(N);
orbitalSet->StorageHessVector.resize(N);
orbitalSet->StorageGradHessVector.resize(N);
orbitalSet->phase.resize(N);
orbitalSet->eikr.resize(N);
orbitalSet->NumValenceOrbs = NumValenceOrbs;
orbitalSet->NumCoreOrbs = NumCoreOrbs;
// Read in k-points
int numOrbs = orbitalSet->getOrbitalSetSize();
int num = 0;
if (root)
{
for (int iorb=0; iorb<N; iorb++)
{
int ti = SortBands[iorb].TwistIndex;
PosType twist = TwistAngles[ti];
orbitalSet->kPoints[iorb] = orbitalSet->PrimLattice.k_cart(twist);
orbitalSet->MakeTwoCopies[iorb] =
(num < (numOrbs-1)) && SortBands[iorb].MakeTwoCopies;
num += orbitalSet->MakeTwoCopies[iorb] ? 2 : 1;
}
}
myComm->bcast(orbitalSet->kPoints);
myComm->bcast(orbitalSet->MakeTwoCopies);
// First, check to see if we have already read this in
H5OrbSet set(H5FileName, spin, N);
///check mesh or ready for FFT grid
bool havePsig=ReadGvectors_ESHDF();
app_log() << "MeshSize = (" << MeshSize[0] << ", " << MeshSize[1] << ", " << MeshSize[2] << ")\n";
int nx, ny, nz, bi, ti;
nx=MeshSize[0];
ny=MeshSize[1];
nz=MeshSize[2];
Ugrid x_grid, y_grid, z_grid;
BCtype_z xBC, yBC, zBC;
xBC.lCode = PERIODIC;
xBC.rCode = PERIODIC;
yBC.lCode = PERIODIC;
yBC.rCode = PERIODIC;
zBC.lCode = PERIODIC;
zBC.rCode = PERIODIC;
x_grid.start = 0.0;
x_grid.end = 1.0;
x_grid.num = nx;
y_grid.start = 0.0;
y_grid.end = 1.0;
y_grid.num = ny;
z_grid.start = 0.0;
z_grid.end = 1.0;
z_grid.num = nz;
// Create the multiUBspline object
orbitalSet->MultiSpline =
create_multi_UBspline_3d_z (x_grid, y_grid, z_grid, xBC, yBC, zBC, NumValenceOrbs);
//////////////////////////////////////
// Create the MuffinTin APW splines //
//////////////////////////////////////
orbitalSet->MuffinTins.resize(NumMuffinTins);
for (int tin=0; tin<NumMuffinTins; tin++)
{
orbitalSet->MuffinTins[tin].Atom = tin;
orbitalSet->MuffinTins[tin].set_center (MT_centers[tin]);
orbitalSet->MuffinTins[tin].set_lattice(Lattice);
orbitalSet->MuffinTins[tin].init_APW
(MT_APW_rgrids[tin], MT_APW_lmax[tin],
NumValenceOrbs);
}
for (int iat=0; iat<AtomicOrbitals.size(); iat++)
{
AtomicOrbitals[iat].set_num_bands(NumValenceOrbs);
AtomicOrbitals[iat].allocate();
}
int isComplex=1;
if (root)
{
HDFAttribIO<int> h_isComplex(isComplex);
h_isComplex.read(H5FileID, "/electrons/psi_r_is_complex");
}
myComm->bcast(isComplex);
if (!isComplex)
{
APP_ABORT("Expected complex orbitals in ES-HDF file, but found real ones.");
}
EinsplineSetBuilder::RotateBands_ESHDF(spin, orbitalSet);
bool isCore = bcastSortBands(spin,N,root);
if(isCore)
{
APP_ABORT("Core states not supported by ES-HDF yet.");
}
t_prep += c_prep.elapsed();
/** For valence orbitals,
* - extended orbitals either in G or in R
* - localized orbitals
*/
//this can potentially break
Array<ComplexType,3> splineData(nx,ny,nz);
if(havePsig)//perform FFT using FFTW
{
c_init.restart();
Array<ComplexType,3> FFTbox;
FFTbox.resize(MeshSize[0], MeshSize[1], MeshSize[2]);
fftw_plan FFTplan = fftw_plan_dft_3d
(MeshSize[0], MeshSize[1], MeshSize[2],
reinterpret_cast<fftw_complex*>(FFTbox.data()),
reinterpret_cast<fftw_complex*>(FFTbox.data()),
+1, FFTW_ESTIMATE);
Vector<complex<double> > cG(MaxNumGvecs);
//this will be parallelized with OpenMP
for(int iorb=0,ival=0; iorb<N; ++iorb, ++ival)
{
//Vector<complex<double> > cG;
int ncg=0;
int ti=SortBands[iorb].TwistIndex;
c_h5.restart();
if(root)
{
ostringstream path;
path << "/electrons/kpoint_" << SortBands[iorb].TwistIndex
<< "/spin_" << spin << "/state_" << SortBands[iorb].BandIndex << "/psi_g";
HDFAttribIO<Vector<complex<double> > > h_cG(cG);
h_cG.read (H5FileID, path.str().c_str());
ncg=cG.size();
}
myComm->bcast(ncg);
if(ncg != Gvecs[0].size())
{
APP_ABORT("Failed : ncg != Gvecs[0].size()");
}
if(!root)
cG.resize(ncg);
myComm->bcast(cG);
t_h5 += c_h5.elapsed();
c_unpack.restart();
unpack4fftw(cG,Gvecs[0],MeshSize,FFTbox);
t_unpack+= c_unpack.elapsed();
c_fft.restart();
fftw_execute (FFTplan);
t_fft+= c_fft.elapsed();
c_phase.restart();
fix_phase_rotate_c2c(FFTbox,splineData,TwistAngles[ti]);
t_phase+= c_phase.elapsed();
c_spline.restart();
set_multi_UBspline_3d_z(orbitalSet->MultiSpline, ival, splineData.data());
t_spline+= c_spline.elapsed();
}
fftw_destroy_plan(FFTplan);
t_init+=c_init.elapsed();
}
else
{
//this will be parallelized with OpenMP
for(int iorb=0,ival=0; iorb<N; ++iorb, ++ival)
{
//check dimension
if(root)
{
ostringstream path;
path << "/electrons/kpoint_" << SortBands[iorb].TwistIndex
<< "/spin_" << spin << "/state_" << SortBands[iorb].BandIndex << "/psi_r";
HDFAttribIO<Array<complex<double>,3> > h_splineData(splineData);
h_splineData.read(H5FileID, path.str().c_str());
}
myComm->bcast(splineData);
set_multi_UBspline_3d_z(orbitalSet->MultiSpline, ival, splineData.data());
}
//return true;
}
app_log() << " READBANDS::PREP = " << t_prep << endl;
app_log() << " READBANDS::H5 = " << t_h5 << endl;
app_log() << " READBANDS::UNPACK = " << t_unpack << endl;
app_log() << " READBANDS::FFT = " << t_fft << endl;
app_log() << " READBANDS::PHASE = " << t_phase << endl;
app_log() << " READBANDS::SPLINE = " << t_spline << endl;
app_log() << " READBANDS::SUM = " << t_init << endl;
//now localized orbitals
for(int iorb=0,ival=0; iorb<N; ++iorb, ++ival)
{
PosType twist=TwistAngles[SortBands[iorb].TwistIndex];
// Read atomic orbital information
for (int iat=0; iat<AtomicOrbitals.size(); iat++)
{
app_log() << "Reading orbital " << iat << " for band " << ival << endl;
AtomicOrbital<complex<double> > &orb = AtomicOrbitals[iat];
Array<complex<double>,2> radial_spline(orb.SplinePoints,orb.Numlm),
poly_coefs(orb.PolyOrder+1,orb.Numlm);
if (root)
{
int ti = SortBands[iorb].TwistIndex;
int bi = SortBands[iorb].BandIndex;
ostringstream path;
path << "/electrons/kpoint_" << ti << "/spin_" << spin << "/state_" << bi << "/";
AtomicOrbital<complex<double> > &orb = AtomicOrbitals[iat];
ostringstream spline_path, poly_path;
spline_path << path.str() << "radial_spline_" << iat;
poly_path << path.str() << "poly_coefs_" << iat;
HDFAttribIO<Array<complex<double>,2> > h_radial_spline(radial_spline);
HDFAttribIO<Array<complex<double>,2> > h_poly_coefs(poly_coefs);
h_radial_spline.read(H5FileID, spline_path.str().c_str());
h_poly_coefs.read (H5FileID, poly_path.str().c_str());
// cerr << "radial_spline.size = (" << radial_spline.size(0)
// << ", " << radial_spline.size(1) << ")\n";
// cerr << "poly_coefs.size = (" << poly_coefs.size(0)
// << ", " << poly_coefs.size(1) << ")\n";
}
myComm->bcast(radial_spline);
myComm->bcast(poly_coefs);
AtomicOrbitals[iat].set_band (ival, radial_spline, poly_coefs, twist);
}
// Now read muffin tin data
for (int tin=0; tin<NumMuffinTins; tin++)
{
// app_log() << "Reading data for muffin tin " << tin << endl;
PosType twist, k;
int lmax = MT_APW_lmax[tin];
int numYlm = (lmax+1)*(lmax+1);
Array<complex<double>,2>
u_lm_r(numYlm, MT_APW_num_radial_points[tin]);
Array<complex<double>,1> du_lm_dr (numYlm);
if (root)
{
int ti = SortBands[iorb].TwistIndex;
int bi = SortBands[iorb].BandIndex;
twist = TwistAngles[ti];
k = orbitalSet->PrimLattice.k_cart(twist);
string uName = MuffinTinPath (ti, bi,tin) + "u_lm_r";
string duName = MuffinTinPath (ti, bi,tin) + "du_lm_dr";
HDFAttribIO<Array<complex<double>,2> > h_u_lm_r(u_lm_r);
HDFAttribIO<Array<complex<double>,1> > h_du_lm_dr(du_lm_dr);
h_u_lm_r.read(H5FileID, uName.c_str());
h_du_lm_dr.read(H5FileID, duName.c_str());
}
myComm->bcast(u_lm_r);
myComm->bcast(du_lm_dr);
myComm->bcast(k);
double Z = (double)IonTypes(tin);
OrbitalSet->MuffinTins[tin].set_APW (ival, k, u_lm_r, du_lm_dr, Z);
}
}
orbitalSet->AtomicOrbitals = AtomicOrbitals;
for (int i=0; i<orbitalSet->AtomicOrbitals.size(); i++)
orbitalSet->AtomicOrbitals[i].registerTimers();
//ExtendedMap_z[set] = orbitalSet->MultiSpline;
}
