void cSystem::startNthreadsFFTW(void)
{
require( fftw_init_threads() != 0, "void cSystem::startNthreadsFFTW(void)");
require(fftwf_init_threads() != 0, "void cSystem::startNthreadsFFTW(void)");
fftw_plan_with_nthreads(getNumProcessors());
fftwf_plan_with_nthreads(getNumProcessors());
std::cout << "FFTW multithreading is turned on: " << getNumProcessors() << " threads\n\n";
}
int main(){
int nthreads = 4;
omp_set_num_threads(nthreads);
#pragma omp parallel
fprintf(stderr,"nthreads %d \n", omp_get_num_threads());
int n3 = 128;
int n2 = 128;
int n1 = 128;
// float ***array = sf_floatalloc3(n1,n2,n3);
float *array = fftwf_alloc_real(n3*n2*n1);
fftwf_complex* cout = fftwf_alloc_complex(n3*n2*n1);
int err = fftwf_init_threads();
if (err == 0) {
fprintf(stderr,"something went wrong with fftw\n");
}
fprintf(stderr,"Got here\n");
double start,end;
start = omp_get_wtime()*omp_get_wtick();
fftwf_plan_with_nthreads(nthreads);
fftwf_plan plan = fftwf_plan_dft_r2c_3d(
n1,n2,n3,
array,cout,
FFTW_MEASURE);
end = omp_get_wtime()*omp_get_wtick();
fprintf(stderr,"elapsed time: %f %f %f\n",end,start,end-start);
for(int i = 0; i < n3*n2*n1; ++i)
array[i] = rand()/RAND_MAX;
//float start = clock()/CLOCKS_PER_SEC;
start = omp_get_wtime();
for(int i=0; i < 1001; ++i)
fftwf_execute(plan);
//float end = clock()/CLOCKS_PER_SEC;
end = omp_get_wtime();
fprintf(stderr,"elapsed time: %f time/calc %f\n",
end-start,(end-start)/100.0);
fftwf_cleanup_threads();
fftwf_cleanup();
fftwf_destroy_plan(plan);
fftwf_free(cout);
fftwf_free(array);
//free(**array); free(*array); free(array);
return 0;
}
int mcfft3_init(int pad1 /* padding on the first axis */,
int nx, int ny, int nz /* input data size */,
int *nx2, int *ny2, int *nz2 /* padded data size */,
int *n_local, int *o_local /* local size & start */)
/*< initialize >*/
{
int cpuid;
MPI_Comm_rank(MPI_COMM_WORLD, &cpuid);
if (threads_ok) threads_ok = fftwf_init_threads();
fftwf_mpi_init();
if (false)
sf_warning("Using threaded FFTW3! \n");
if (threads_ok)
fftwf_plan_with_nthreads(omp_get_max_threads());
/* axis 1 */
nk = n1 = kiss_fft_next_fast_size(nx*pad1);
/* axis 2 */
n2 = kiss_fft_next_fast_size(ny);
/* axis 3 */
n3 = kiss_fft_next_fast_size(nz);
alloc_local = fftwf_mpi_local_size_3d(n3, n2, n1, MPI_COMM_WORLD, &local_n0, &local_0_start);
//cc = sf_complexalloc3(n1,n2,n3);
cc = sf_complexalloc(alloc_local);
cfg = fftwf_mpi_plan_dft_3d(n3,n2,n1,
(fftwf_complex *) cc,
(fftwf_complex *) cc,
MPI_COMM_WORLD,
FFTW_FORWARD, FFTW_MEASURE);
icfg = fftwf_mpi_plan_dft_3d(n3,n2,n1,
(fftwf_complex *) cc,
(fftwf_complex *) cc,
MPI_COMM_WORLD,
FFTW_BACKWARD, FFTW_MEASURE);
if (NULL == cfg || NULL == icfg) sf_error("FFTW failure.");
*nx2 = n1;
*ny2 = n2;
*nz2 = n3;
*n_local = (int) local_n0;
*o_local = (int) local_0_start;
wt = 1.0/(n3*n2*n1);
return (nk*n2*n3);
}
int cfft2_init(int pad1 /* padding on the first axis */,
int nx, int ny /* input data size */,
int *nx2, int *ny2 /* padded data size */)
/*< initialize >*/
{
#ifdef SF_HAS_FFTW
#ifdef _OPENMP
fftwf_init_threads();
sf_warning("Using threaded FFTW3! \n");
fftwf_plan_with_nthreads(omp_get_max_threads());
#endif
#endif
#ifndef SF_HAS_FFTW
int i2;
#endif
nk = n1 = kiss_fft_next_fast_size(nx*pad1);
#ifndef SF_HAS_FFTW
cfg1 = kiss_fft_alloc(n1,0,NULL,NULL);
icfg1 = kiss_fft_alloc(n1,1,NULL,NULL);
#endif
n2 = kiss_fft_next_fast_size(ny);
cc = sf_complexalloc2(n1,n2);
dd = sf_complexalloc2(nk,n2);
#ifndef SF_HAS_FFTW
cfg2 = kiss_fft_alloc(n2,0,NULL,NULL);
icfg2 = kiss_fft_alloc(n2,1,NULL,NULL);
tmp = (kiss_fft_cpx **) sf_alloc(n2,sizeof(*tmp));
tmp[0] = (kiss_fft_cpx *) sf_alloc(nk*n2,sizeof(kiss_fft_cpx));
for (i2=0; i2 < n2; i2++) {
tmp[i2] = tmp[0]+i2*nk;
}
trace2 = sf_complexalloc(n2);
ctrace2 = (kiss_fft_cpx *) trace2;
#endif
*nx2 = n1;
*ny2 = n2;
wt = 1.0/(n1*n2);
return (nk*n2);
}
int cfft2_init(int pad1 /* padding on the first axis */,
int nx, int ny /* input data size */,
int *nx2, int *ny2 /* padded data size */,
int *n_local, int *o_local /* local size & start */,
MPI_Comm comm)
/*< initialize >*/
{
if (threads_ok) threads_ok = fftwf_init_threads();
fftwf_mpi_init();
if (false)
sf_warning("Using threaded FFTW3! \n");
if (threads_ok)
fftwf_plan_with_nthreads(omp_get_max_threads());
nk = n1 = kiss_fft_next_fast_size(nx*pad1);
n2 = kiss_fft_next_fast_size(ny);
alloc_local = fftwf_mpi_local_size_2d(n2, n1, comm, &local_n0, &local_0_start);
//cc = sf_complexalloc2(n1,n2);
//dd = sf_complexalloc2(nk,n2);
cc = sf_complexalloc(alloc_local);
dd = sf_complexalloc(alloc_local);
cfg = fftwf_mpi_plan_dft_2d(n2,n1,
(fftwf_complex *) cc,
(fftwf_complex *) dd,
comm,
FFTW_FORWARD, FFTW_MEASURE);
icfg = fftwf_mpi_plan_dft_2d(n2,n1,
(fftwf_complex *) dd,
(fftwf_complex *) cc,
comm,
FFTW_BACKWARD, FFTW_MEASURE);
if (NULL == cfg || NULL == icfg) sf_error("FFTW failure.");
*nx2 = n1;
*ny2 = n2;
*n_local = (int) local_n0;
*o_local = (int) local_0_start;
wt = 1.0/(n1*n2);
return (nk*n2);
}
void fft_set_num_threads(unsigned int n)
{
#ifdef FFTWTHREADS
#pragma omp critical
if (!fft_threads_init) {
fft_threads_init = true;
fftwf_init_threads();
}
#pragma omp critical
fftwf_plan_with_nthreads(n);
#else
UNUSED(n);
#endif
}
void Fft::prepareFft(){
fftwf_init_threads();
fftwf_plan_with_nthreads(FFT_THREADS);
}
int fft2_init(bool cmplx1 /* if complex transform */,
int pad1 /* padding on the first axis */,
int nx, int ny /* input data size */,
int *nx2, int *ny2 /* padded data size */)
/*< initialize >*/
{
#ifdef SF_HAS_FFTW
#ifdef _OPENMP
fftwf_init_threads();
sf_warning("Using threaded FFTW3!\n");
fftwf_plan_with_nthreads(omp_get_max_threads());
#endif
#else
int i2;
#endif
cmplx = cmplx1;
if (cmplx) {
nk = n1 = kiss_fft_next_fast_size(nx*pad1);
#ifndef SF_HAS_FFTW
cfg1 = kiss_fft_alloc(n1,0,NULL,NULL);
icfg1 = kiss_fft_alloc(n1,1,NULL,NULL);
#endif
} else {
nk = kiss_fft_next_fast_size(pad1*(nx+1)/2)+1;
n1 = 2*(nk-1);
#ifndef SF_HAS_FFTW
cfg = kiss_fftr_alloc(n1,0,NULL,NULL);
icfg = kiss_fftr_alloc(n1,1,NULL,NULL);
#endif
}
n2 = kiss_fft_next_fast_size(ny);
if (cmplx) {
cc = sf_complexalloc2(n1,n2);
} else {
ff = sf_floatalloc2(n1,n2);
}
dd = sf_complexalloc(nk*n2);
#ifndef SF_HAS_FFTW
cfg2 = kiss_fft_alloc(n2,0,NULL,NULL);
icfg2 = kiss_fft_alloc(n2,1,NULL,NULL);
tmp = (kiss_fft_cpx **) sf_alloc(n2,sizeof(*tmp));
tmp[0] = (kiss_fft_cpx *) sf_alloc(nk*n2,sizeof(kiss_fft_cpx));
for (i2=0; i2 < n2; i2++) {
tmp[i2] = tmp[0]+i2*nk;
}
trace2 = sf_complexalloc(n2);
ctrace2 = (kiss_fft_cpx *) trace2;
#endif
*nx2 = n1;
*ny2 = n2;
wt = 1.0/(n1*n2);
return (nk*n2);
}
int main (int argc, char *argv[])
{
bool verb, snap;
bool abc, adj;
int nz, nx, nt, ns, nr;
float dz, dx, dt, oz, ox;
int nz0, nx0, nb;
float oz0, ox0;
int nkz, nkx;
int nzpad, nxpad;
float **u1, **u0;
float *ws, *wr;
sf_file file_src = NULL, file_rec = NULL;
sf_file file_inp = NULL, file_out = NULL;
sf_file file_mdl = NULL;
sf_axis az = NULL, ax = NULL, at = NULL, as = NULL, ar = NULL;
pt2d *src2d = NULL;
pt2d *rec2d = NULL;
scoef2d cssinc = NULL;
scoef2d crsinc = NULL;
float *wi = NULL, *wo = NULL;
sf_axis ai = NULL, ao = NULL;
scoef2d cisinc = NULL, cosinc = NULL;
bool spt = false, rpt = false;
bool ipt = false, opt = false;
sf_init(argc, argv);
if (!sf_getbool("verb", &verb)) verb = false;
if (!sf_getbool("snap", &snap)) snap = false;
if (!sf_getbool("adj", &adj)) adj = false;
if (!sf_getint("nb", &nb)) nb = 4;
if (sf_getstring("sou") != NULL) {
spt = true;
if (adj) opt = true;
else ipt = true;
}
if (sf_getstring("rec") != NULL) {
rpt = true;
if (adj) ipt = true;
else opt = true;
}
file_inp = sf_input("in");
file_mdl = sf_input("model");
if (spt) file_src = sf_input("sou");
if (rpt) file_rec = sf_input("rec");
file_out = sf_output("out");
if (ipt) at = sf_iaxa(file_inp, 2);
else at = sf_iaxa(file_inp, 3);
if (spt) as = sf_iaxa(file_src, 2);
if (rpt) ar = sf_iaxa(file_rec, 2);
az = sf_iaxa(file_mdl, 1);
ax = sf_iaxa(file_mdl, 2);
nt = sf_n(at); dt = sf_d(at); //ot = sf_o(at);
nz0 = sf_n(az); dz = sf_d(az); oz0 = sf_o(az);
nx0 = sf_n(ax); dx = sf_d(ax); ox0 = sf_o(ax);
if (spt) ns = sf_n(as);
if (rpt) nr = sf_n(ar);
nz = nz0 + 2 * nb;
nx = nx0 + 2 * nb;
oz = oz0 - nb * dz;
ox = ox0 - nb * dx;
abc = nb ? true : false;
// sf_error("ox=%f ox0=%f oz=%f oz0=%f",ox,ox0,oz,oz0);
nzpad = kiss_fft_next_fast_size( ((nz+1)>>1)<<1 );
nkx = nxpad = kiss_fft_next_fast_size(nx);
nkz = nzpad / 2 + 1;
/* float okx = - 0.5f / dx; */
float okx = 0.f;
float okz = 0.f;
float dkx = 1.f / (nxpad * dx);
float dkz = 1.f / (nzpad * dz);
float **vp, **eps, **del;
vp = sf_floatalloc2(nz, nx);
eps = sf_floatalloc2(nz, nx);
del = sf_floatalloc2(nz, nx);
float **tmparray = sf_floatalloc2(nz0, nx0);
sf_floatread(tmparray[0], nz0*nx0, file_mdl); expand2d(vp[0], tmparray[0], nz, nx, nz0, nx0);
sf_floatread(tmparray[0], nz0*nx0, file_mdl); expand2d(eps[0], tmparray[0], nz, nx, nz0, nx0);
sf_floatread(tmparray[0], nz0*nx0, file_mdl); expand2d(del[0], tmparray[0], nz, nx, nz0, nx0);
float **vn, **vh;
float **eta, **lin_eta;
lin_eta = NULL, vh = NULL;
vn = sf_floatalloc2(nz, nx);
vh = sf_floatalloc2(nz, nx);
eta = sf_floatalloc2(nz, nx);
lin_eta = sf_floatalloc2(nz, nx);
for (int ix=0; ix<nx; ix++) {
for (int iz=0; iz<nz; iz++){
vp[ix][iz] *= vp[ix][iz];
vn[ix][iz] = vp[ix][iz] * (1.f + 2.f * del[ix][iz]);
vh[ix][iz] = vp[ix][iz] * (1.f + 2.f * eps[ix][iz]);
eta[ix][iz] = (eps[ix][iz] - del[ix][iz]) / (1.f + 2.f * del[ix][iz]);
lin_eta[ix][iz] = eta[ix][iz] * (1.f + 2.f * del[ix][iz]);
}
}
float *kx = sf_floatalloc(nkx);
float *kz = sf_floatalloc(nkz);
for (int ikx=0; ikx<nkx; ++ikx) {
kx[ikx] = okx + ikx * dkx;
/* if (ikx >= nkx/2) kx[ikx] = (nkx - ikx) * dkx; */
if (ikx >= nkx/2) kx[ikx] = (ikx - nkx) * dkx;
kx[ikx] *= 2 * SF_PI;
kx[ikx] *= kx[ikx];
}
for (int ikz=0; ikz<nkz; ++ikz) {
kz[ikz] = okz + ikz * dkz;
kz[ikz] *= 2 * SF_PI;
kz[ikz] *= kz[ikz];
}
if (adj) {
ai = ar; ao = as;
} else {
ai = as; ao = ar;
}
if (opt) {
sf_oaxa(file_out, ao, 1);
sf_oaxa(file_out, at, 2);
} else {
sf_oaxa(file_out, az, 1);
sf_oaxa(file_out, ax, 2);
sf_oaxa(file_out, at, 3);
}
sf_fileflush(file_out, NULL);
if (spt) {
src2d = pt2dalloc1(ns);
pt2dread1(file_src, src2d, ns, 2);
cssinc = sinc2d_make(ns, src2d, nz, nx, dz, dx, oz, ox);
ws = sf_floatalloc(ns);
if (adj) { cosinc = cssinc; wo = ws; }
else { cisinc = cssinc; wi = ws; }
}
if (rpt) {
rec2d = pt2dalloc1(nr);
pt2dread1(file_rec, rec2d, nr, 2);
crsinc = sinc2d_make(nr, rec2d, nz, nx, dz, dx, oz, ox);
wr = sf_floatalloc(nr);
if (adj) { cisinc = crsinc; wi = wr; }
else { cosinc = crsinc; wo = wr; }
}
u0 = sf_floatalloc2(nz, nx);
u1 = sf_floatalloc2(nz, nx);
float *rwave = (float *) fftwf_malloc(nzpad*nxpad*sizeof(float));
float *rwavem = (float *) fftwf_malloc(nzpad*nxpad*sizeof(float));
fftwf_complex *cwave = (fftwf_complex *) fftwf_malloc(nkz*nkx*sizeof(fftwf_complex));
fftwf_complex *cwavem = (fftwf_complex *) fftwf_malloc(nkz*nkx*sizeof(fftwf_complex));
/* float *rwavem = (float *) fftwf_malloc(nzpad*nxpad*sizeof(float));
fftwf_complex *cwave = (fftwf_complex *) fftwf_malloc(nkz*nkx*sizeof(fftwf_complex));
fftwf_complex *cwavem = (fftwf_complex *) fftwf_malloc(nkz*nkx*sizeof(fftwf_complex)); */
/* boundary conditions */
float **ucut = NULL;
float *damp = NULL;
if (!(ipt &&opt)) ucut = sf_floatalloc2(nz0, nx0);
damp = damp_make(nb);
float wt = 1./(nxpad * nzpad);
wt *= dt * dt;
fftwf_plan forward_plan;
fftwf_plan inverse_plan;
#ifdef _OPENMP
#ifdef SF_HAS_FFTW_OMP
fftwf_init_threads();
fftwf_plan_with_nthreads(omp_get_max_threads());
#endif
#endif
forward_plan = fftwf_plan_dft_r2c_2d(nxpad, nzpad,
rwave, cwave, FFTW_MEASURE);
#ifdef _OPENMP
#ifdef SF_HAS_FFTW_OMP
fftwf_plan_with_nthreads(omp_get_max_threads());
#endif
#endif
inverse_plan = fftwf_plan_dft_c2r_2d(nxpad, nzpad,
cwavem, rwavem, FFTW_MEASURE);
int itb, ite, itc;
if (adj) {
itb = nt -1; ite = -1; itc = -1;
} else {
itb = 0; ite = nt; itc = 1;
}
if (adj) {
for (int it=0; it<nt; it++) {
if (opt) sf_floatwrite(wo, sf_n(ao), file_out);
else sf_floatwrite(ucut[0], nz0*nx0, file_out);
}
sf_seek(file_out, 0, SEEK_SET);
}
float **ptrtmp = NULL;
memset(u0[0], 0, sizeof(float)*nz*nx);
memset(u1[0], 0, sizeof(float)*nz*nx);
memset(rwave, 0, sizeof(float)*nzpad*nxpad);
memset(rwavem, 0, sizeof(float)*nzpad*nxpad);
memset(cwave, 0, sizeof(float)*nkz*nkx*2);
memset(cwavem, 0, sizeof(float)*nkz*nkx*2);
for (int it=itb; it!=ite; it+=itc) { if (verb) sf_warning("it = %d;",it);
#ifdef _OPENMP
double tic = omp_get_wtime();
#endif
if (ipt) {
if (adj) sf_seek(file_inp, (off_t)(it)*sizeof(float)*sf_n(ai), SEEK_SET);
sf_floatread(wi, sf_n(ai), file_inp);
for (int i=0; i<sf_n(ai); i++)
wi[i] *= dt* dt;
} else {
if (adj) sf_seek(file_inp, (off_t)(it)*sizeof(float)*nz0*nx0, SEEK_SET);
sf_floatread(ucut[0], nz0*nx0, file_inp);
for (int j=0; j<nx0; j++)
for (int i=0; i<nz0; i++)
ucut[j][i] *= dt * dt;
}
/* apply absorbing boundary condition: E \times u@n-1 */
damp2d_apply(u0, damp, nz, nx, nb);
fft_stepforward(u0, u1, rwave, rwavem, cwave, cwavem,
vp, vn, eta, vh, eps, lin_eta, kz, kx,
forward_plan, inverse_plan,
nz, nx, nzpad, nxpad, nkz, nkx, wt, adj);
// sinc2d_inject1(u0, ws[it][s_idx], cssinc[s_idx]);
if (ipt) sinc2d_inject(u0, wi, cisinc);
else wfld2d_inject(u0, ucut, nz0, nx0, nb);
/* apply absorbing boundary condition: E \times u@n+1 */
damp2d_apply(u0, damp, nz, nx, nb);
/* loop over pointers */
ptrtmp = u0; u0 = u1; u1 = ptrtmp;
if (opt) {
if (adj) sf_seek(file_out, (off_t)(it)*sizeof(float)*sf_n(ao),SEEK_SET);
sinc2d_extract(u0, wo, cosinc);
sf_floatwrite(wo, sf_n(ao), file_out);
} else {
if (adj) sf_seek(file_out, (off_t)(it)*sizeof(float)*nz0*nx0,SEEK_SET);
wwin2d(ucut, u0, nz0, nx0, nb);
sf_floatwrite(ucut[0], nz0*nx0, file_out);
}
#ifdef _OPENMP
double toc = omp_get_wtime();
if (verb) fprintf(stderr," clock = %lf;", toc-tic);
#endif
} /* END OF TIME LOOP */
return 0;
}
int cfft3_init(int pad1 /* padding on the first axis */,
int nx, int ny, int nz /* input data size */,
int *nx2, int *ny2, int *nz2 /* padded data size */)
/*< initialize >*/
{
#ifdef SF_HAS_FFTW
#ifdef _OPENMP
fftwf_init_threads();
sf_warning("Using threaded FFTW3! %d\n",omp_get_max_threads());
fftwf_plan_with_nthreads(omp_get_max_threads());
#endif
#else
int i2, i3;
#endif
/* axis 1 */
nk = n1 = kiss_fft_next_fast_size(nx*pad1);
#ifndef SF_HAS_FFTW
cfg1 = kiss_fft_alloc(n1,0,NULL,NULL);
icfg1 = kiss_fft_alloc(n1,1,NULL,NULL);
#endif
/* axis 2 */
n2 = kiss_fft_next_fast_size(ny);
#ifndef SF_HAS_FFTW
cfg2 = kiss_fft_alloc(n2,0,NULL,NULL);
icfg2 = kiss_fft_alloc(n2,1,NULL,NULL);
trace2 = sf_complexalloc(n2);
ctrace2 = (kiss_fft_cpx *) trace2;
#endif
/* axis 3 */
n3 = kiss_fft_next_fast_size(nz);
#ifndef SF_HAS_FFTW
cfg3 = kiss_fft_alloc(n3,0,NULL,NULL);
icfg3 = kiss_fft_alloc(n3,1,NULL,NULL);
trace3 = sf_complexalloc(n3);
ctrace3 = (kiss_fft_cpx *) trace3;
/* --- */
tmp = (kiss_fft_cpx***) sf_alloc (n3,sizeof(kiss_fft_cpx**));
tmp[0] = (kiss_fft_cpx**) sf_alloc (n2*n3,sizeof(kiss_fft_cpx*));
tmp[0][0] = (kiss_fft_cpx*) sf_alloc (nk*n2*n3,sizeof(kiss_fft_cpx));
for (i2=1; i2 < n2*n3; i2++) {
tmp[0][i2] = tmp[0][0]+i2*nk;
}
for (i3=1; i3 < n3; i3++) {
tmp[i3] = tmp[0]+i3*n2;
}
#endif
cc = sf_complexalloc3(n1,n2,n3);
*nx2 = n1;
*ny2 = n2;
*nz2 = n3;
wt = 1.0/(n3*n2*n1);
return (nk*n2*n3);
}
int main(int argc, char** argv){
float tr[6];
const float ZAP=32;
const uint64_t TSIZE=18;
const uint64_t zapE=64;
fftwf_init_threads();
fftwf_plan_with_nthreads(omp_get_max_threads());
logmsg("Open file '%s'",argv[1]);
FILE* f = fopen(argv[1],"r");
int hdr_bytes = read_header(f);
const uint64_t nskip = hdr_bytes;
const uint64_t nchan = nchans;
logmsg("Nchan=%"PRIu64", tsamp=%f",nchan,tsamp);
mjk_rand_t *random = mjk_rand_init(12345);
rewind(f);
FILE* of = fopen("clean.fil","w");
uint8_t hdr[nskip];
fread(hdr,1,nskip,f);
fwrite(hdr,1,nskip,of);
const uint64_t nsamp_per_block=round(pow(2,TSIZE));
logmsg("Tblock = %f",nsamp_per_block*tsamp);
mjk_clock_t *t_all = init_clock();
start_clock(t_all);
mjk_clock_t *t_read = init_clock();
mjk_clock_t *t_trns= init_clock();
mjk_clock_t *t_rms = init_clock();
mjk_clock_t *t_fft = init_clock();
mjk_clock_t *t_spec = init_clock();
const uint64_t bytes_per_block = nchan*nsamp_per_block;
uint8_t *buffer = calloc(bytes_per_block,1);
float **data = malloc_2df(nchan,nsamp_per_block);
float **clean = malloc_2df(nchan,nsamp_per_block);
float *bpass = calloc(nchan,sizeof(float));
float *ch_var=NULL;
float *ch_mean=NULL;
float *ch_fft_n=NULL;
float *ch_fft_p=NULL;
logmsg("Planning FFT - this will take a long time the first time it is run!");
start_clock(t_fft);
FILE * wisfile;
if(wisfile=fopen("wisdom.txt","r")){
fftwf_import_wisdom_from_file(wisfile);
fclose(wisfile);
}
const int fftX=nsamp_per_block;
const int fftY=nchan;
const int fftXo=nsamp_per_block/2+1;
float *X = fftwf_malloc(sizeof(float)*fftX);
for (uint64_t i = 0; i < nsamp_per_block ; i++){
X[i]=i;
}
float *tseries = fftwf_malloc(sizeof(float)*fftX);
float complex *fseries = fftwf_malloc(sizeof(float complex)*fftXo);
float *pseries = fftwf_malloc(sizeof(float)*fftXo);
uint8_t *mask = malloc(sizeof(uint8_t)*fftXo);
fftwf_plan fft_1d = fftwf_plan_dft_r2c_1d(fftX,tseries,fseries,FFTW_MEASURE|FFTW_DESTROY_INPUT);
complex float * fftd = fftwf_malloc(sizeof(complex float)*(fftXo*fftY));
fftwf_plan fft_plan = fftwf_plan_many_dft_r2c(
1,&fftX,fftY,
data[0] ,&fftX,1,fftX,
fftd ,&fftXo,1,fftXo,
FFTW_MEASURE|FFTW_PRESERVE_INPUT);
logmsg("Planning iFFT - this will take a long time the first time it is run!");
fftwf_plan ifft_plan = fftwf_plan_many_dft_c2r(
1,&fftX,fftY,
fftd ,&fftXo,1,fftXo,
clean[0] ,&fftX,1,fftX,
FFTW_MEASURE|FFTW_PRESERVE_INPUT);
if(!fft_plan){
logmsg("Error - could not do FFT plan");
exit(2);
}
wisfile=fopen("wisdom.txt","w");
fftwf_export_wisdom_to_file(wisfile);
fclose(wisfile);
stop_clock(t_fft);
logmsg("T(planFFT)= %.2lfs",read_clock(t_fft));
reset_clock(t_fft);
float min_var=1e9;
float max_var=0;
float min_fft_n=1e9;
float max_fft_n=0;
float min_fft_p=1e9;
float max_fft_p=0;
float min_mean=1e9;
float max_mean=0;
uint64_t nblocks=0;
uint64_t totread=0;
while(!feof(f)){
nblocks++;
ch_var = realloc(ch_var,nchan*nblocks*sizeof(float));
ch_mean = realloc(ch_mean,nchan*nblocks*sizeof(float));
ch_fft_n = realloc(ch_fft_n,nchan*nblocks*sizeof(float));
ch_fft_p = realloc(ch_fft_p,nchan*nblocks*sizeof(float));
start_clock(t_read);
uint64_t read = fread(buffer,1,bytes_per_block,f);
stop_clock(t_read);
if (read!=bytes_per_block){
nblocks--;
break;
}
totread+=read;
logmsg("read=%"PRIu64" bytes. T=%fs",read,totread*tsamp/(float)nchan);
uint64_t offset = (nblocks-1)*nchan;
start_clock(t_trns);
// transpose with small blocks in order to increase cache efficiency.
#define BLK 8
#pragma omp parallel for schedule(static,2) shared(buffer,data)
for (uint64_t j = 0; j < nchan ; j+=BLK){
for (uint64_t i = 0; i < nsamp_per_block ; i++){
for (uint64_t k = 0; k < BLK ; k++){
data[j+k][i] = buffer[i*nchan+j+k];
}
}
}
#pragma omp parallel for shared(data)
for (uint64_t j = 0; j < nchan ; j++){
if(j<zapE || (nchan-j) < zapE){
for (uint64_t i = 0; i < nsamp_per_block ; i++){
data[j][i]=ZAP;
}
}
}
if(nblocks==1){
#pragma omp parallel for shared(data,bpass)
for (uint64_t j = 0; j < nchan ; j++){
for (uint64_t i = 0; i < nsamp_per_block ; i++){
bpass[j]+=data[j][i];
}
bpass[j]/=(float)nsamp_per_block;
bpass[j]-=ZAP;
}
}
#pragma omp parallel for shared(data,bpass)
for (uint64_t j = 0; j < nchan ; j++){
for (uint64_t i = 0; i < nsamp_per_block ; i++){
data[j][i]-=bpass[j];
}
}
stop_clock(t_trns);
start_clock(t_rms);
#pragma omp parallel for shared(data,ch_mean,ch_var)
for (uint64_t j = 0; j < nchan ; j++){
float mean=0;
for (uint64_t i = 0; i < nsamp_per_block ; i++){
mean+=data[j][i];
}
mean/=(float)nsamp_per_block;
if(mean > ZAP+5 || mean < ZAP-5){
logmsg("ZAP ch=%"PRIu64,j);
for (uint64_t i = 0; i < nsamp_per_block ; i++){
data[j][i]=ZAP;
}
}
float ss=0;
float x=0;
for (uint64_t i = 0; i < nsamp_per_block ; i++){
x = data[j][i]-mean;
ss+=x*x;
}
float var=ss/(float)nsamp_per_block;
if (var > 0){
for (uint64_t i = 0; i < nsamp_per_block ; i++){
float v = (data[j][i]-mean)/sqrt(var);
if(v > 3 || v < -3){
data[j][i]=mjk_rand_gauss(random)*sqrt(var)+mean;
}
}
}
ch_var[offset+j] = var;
ch_mean[offset+j] = mean;
}
stop_clock(t_rms);
for (uint64_t i = 0; i < nsamp_per_block ; i++){
tseries[i]=0;
}
float tmean=0;
float tvar=0;
float max=0;
float min=1e99;
//#pragma omp parallel for shared(data,tseries)
// NOT THREAD SAFE
for (uint64_t j = 0; j < nchan ; j++){
tmean+=ch_mean[offset+j];
tvar+=ch_var[offset+j];
for (uint64_t i = 0; i < nsamp_per_block ; i++){
tseries[i]+=data[j][i];
if(data[j][i]>max)max=data[j][i];
if(data[j][i]<min)min=data[j][i];
}
}
float ss=0;
float mm=0;
for (uint64_t i = 0; i < nsamp_per_block ; i++){
float x=tseries[i]-tmean;
mm+=tseries[i];
ss+=x*x;
}
float rvar=ss/(float)nsamp_per_block;
logmsg("var=%g tvar=%g",ss/(float)nsamp_per_block,tvar);
logmsg("mean=%g tmean=%g",mm/(float)nsamp_per_block,tmean);
cpgopen("3/xs");
cpgsvp(0.1,0.9,0.1,0.9);
cpgswin(0,fftX,tmean-sqrt(tvar)*30,tmean+sqrt(tvar)*30);
cpgbox("ABN",0,0,"ABN",0,0);
cpgline(fftX,X,tseries);
cpgsci(2);
cpgclos();
tr[0] = 0.0 ;
tr[1] = 1;
tr[2] = 0;
tr[3] = 0.5;
tr[4] = 0;
tr[5] = 1;
logmsg("max=%g min=%g",max,min);
cpgopen("4/xs");
cpgsvp(0.1,0.9,0.1,0.9);
cpgswin(0,nsamp_per_block,0,nchan);
cpgbox("ABN",0,0,"ABN",0,0);
cpggray(*data,nsamp_per_block,nchan,1,nsamp_per_block,1,nchan,tmean/(float)nchan+sqrt(rvar/(float)nchan),tmean/(float)nchan-sqrt(rvar/(float)nchan),tr);
cpgclos();
start_clock(t_fft);
fftwf_execute(fft_1d);
fftwf_execute(fft_plan);
stop_clock(t_fft);
{
float T = sqrt(fftXo*tvar)*12;
logmsg("Zap T=%.2e",T);
float fx[fftXo];
float fT[fftXo];
#pragma omp parallel for shared(fseries,pseries,mask)
for (uint64_t i = 0; i < fftXo ; i++){
mask[i]=1;
}
#pragma omp parallel for shared(fseries,pseries,mask)
for (uint64_t i = 0; i < fftXo ; i++){
pseries[i]=camp(fseries[i]);
fx[i]=i;
float TT = T;
if (i>512)TT=T/2.0;
if(i>32){
fT[i]=TT;
if (pseries[i] > TT) {
mask[i]=0;
}
} else fT[i]=0;
}
uint64_t nmask=0;
for (uint64_t i = 0; i < fftXo ; i++){
if (mask[i]==0){
nmask++;
}
}
logmsg("masked=%d (%.2f%%)",nmask,100*nmask/(float)fftXo);
cpgopen("1/xs");
cpgsvp(0.1,0.9,0.1,0.9);
cpgswin(0,fftXo,0,T*10);
cpgbox("ABN",0,0,"ABN",0,0);
cpgline(fftXo,fx,pseries);
cpgsci(2);
cpgline(fftXo,fx,fT);
cpgclos();
}
// exit(1);
start_clock(t_spec);
//FILE* ff=fopen("plot","w");
#pragma omp parallel for shared(fftd,ch_mean,ch_fft_n,ch_fft_p)
for (uint64_t j = 0; j < nchan ; j++){
float var = ch_var[offset+j];
float m=sqrt(var*fftXo/2.0);
float T = sqrt(var*fftXo)*3;
uint64_t n=0;
float p=0;
float complex *fftch = fftd + fftXo*j;
for(uint64_t i = 1; i < fftXo; i++){
if (camp(fftch[i]) > T) {
n++;
p+=camp(fftch[i]);
}
// if(j==512)fprintf(ff,"%f ",camp(fftch[i]));
if(mask[i]==0){
fftch[i]=m*(mjk_rand_gauss(random) + I*mjk_rand_gauss(random));
}
// if(j==512)fprintf(ff,"%f\n",camp(fftch[i]));
}
ch_fft_n[offset+j]=n;
ch_fft_p[offset+j]=p;
}
// fclose(ff);
logmsg("iFFT");
fftwf_execute(ifft_plan);
#pragma omp parallel for schedule(static,2) shared(buffer,clean)
for (uint64_t j = 0; j < nchan ; j+=BLK){
for (uint64_t i = 0; i < nsamp_per_block ; i++){
for (uint64_t k = 0; k < BLK ; k++){
clean[j+k][i]/=(float)fftX;
buffer[i*nchan+j+k] = round(clean[j+k][i]);
}
}
if(j==512){
cpgopen("2/xs");
cpgsvp(0.1,0.9,0.1,0.9);
cpgswin(0,fftX,ch_mean[j]-sqrt(ch_var[j])*10,ch_mean[j]+sqrt(ch_var[j])*10);
cpgbox("ABN",0,0,"ABN",0,0);
cpgline(fftX,X,data[j]);
cpgsci(2);
cpgline(fftX,X,clean[j]);
cpgclos();
}
}
fwrite(buffer,1,bytes_per_block,of);
for (uint64_t i = 0; i < nsamp_per_block ; i++){
tseries[i]=0;
}
tmean=0;
tvar=0;
max=0;
min=1e99;
//#pragma omp parallel for shared(clean,tseries)
// NOT THREAD SAFE
for (uint64_t j = 0; j < nchan ; j++){
tmean+=ch_mean[offset+j];
tvar+=ch_var[offset+j];
for (uint64_t i = 0; i < nsamp_per_block ; i++){
tseries[i]+=clean[j][i];
if(clean[j][i]>max)max=clean[j][i];
if(clean[j][i]<min)min=clean[j][i];
}
}
ss=0;
mm=0;
for (uint64_t i = 0; i < nsamp_per_block ; i++){
float x=tseries[i]-tmean;
mm+=tseries[i];
ss+=x*x;
}
rvar=ss/(float)nsamp_per_block;
logmsg("var=%g tvar=%g",ss/(float)nsamp_per_block,tvar);
logmsg("mean=%g tmean=%g",mm/(float)nsamp_per_block,tmean);
cpgopen("5/xs");
cpgsvp(0.1,0.9,0.1,0.9);
cpgswin(0,fftX,tmean-sqrt(tvar)*30,tmean+sqrt(tvar)*30);
cpgbox("ABN",0,0,"ABN",0,0);
cpgline(fftX,X,tseries);
cpgsci(2);
cpgclos();
tr[0] = 0.0 ;
tr[1] = 1;
tr[2] = 0;
tr[3] = 0.5;
tr[4] = 0;
tr[5] = 1;
logmsg("max=%g min=%g",max,min);
cpgopen("6/xs");
cpgsvp(0.1,0.9,0.1,0.9);
cpgswin(0,nsamp_per_block,0,nchan);
cpgbox("ABN",0,0,"ABN",0,0);
cpggray(*clean,nsamp_per_block,nchan,1,nsamp_per_block,1,nchan,tmean/(float)nchan+sqrt(rvar/(float)nchan),tmean/(float)nchan-sqrt(rvar/(float)nchan),tr);
cpgclos();
stop_clock(t_spec);
for (uint64_t j = 0; j < nchan ; j++){
float mean=ch_mean[offset+j];
if (mean > max_mean)max_mean=mean;
if (mean < min_mean)min_mean=mean;
float var=ch_var[offset+j];
if (var > max_var)max_var=var;
if (var < min_var)min_var=var;
float fft_n=ch_fft_n[offset+j];
if (fft_n > max_fft_n)max_fft_n=fft_n;
if (fft_n < min_fft_n)min_fft_n=fft_n;
float fft_p=ch_fft_p[offset+j];
if (fft_p > max_fft_p)max_fft_p=fft_p;
if (fft_p < min_fft_p)min_fft_p=fft_p;
}
}
stop_clock(t_all);
fclose(of);
logmsg("T(all) = %.2lfs",read_clock(t_all));
logmsg("T(read) = %.2lfs",read_clock(t_read));
logmsg("T(trans)= %.2lfs",read_clock(t_trns));
logmsg("T(fft) = %.2lfs",read_clock(t_fft));
logmsg("T(fan) = %.2lfs",read_clock(t_spec));
logmsg("T(rms) = %.2lfs",read_clock(t_rms));
logmsg("T(rest) = %.2lfs",read_clock(t_all)-read_clock(t_read)-read_clock(t_trns)-read_clock(t_rms)-read_clock(t_fft)-read_clock(t_spec));
tr[0] = -tsamp*nsamp_per_block*0.5;
tr[2] = tsamp*nsamp_per_block;
tr[1] = 0;
tr[3] = 0.5;
tr[5] = 0;
tr[4] = 1;
cpgopen("1/xs");
cpgsvp(0.1,0.9,0.1,0.9);
cpgswin(0,nblocks*tsamp*nsamp_per_block,0,nchan);
cpgbox("ABN",600,10,"ABN",100,1);
cpggray(ch_mean,nchan,nblocks,1,nchan,1,nblocks,max_mean,min_mean,tr);
cpgclos();
cpgopen("2/xs");
cpgsvp(0.1,0.9,0.1,0.9);
cpgswin(0,nblocks*tsamp*nsamp_per_block,0,nchan);
cpgbox("ABN",600,10,"ABN",100,1);
cpggray(ch_var,nchan,nblocks,1,nchan,1,nblocks,max_var,min_var,tr);
cpgclos();
cpgopen("3/xs");
cpgsvp(0.1,0.9,0.1,0.9);
cpgswin(0,nblocks*tsamp*nsamp_per_block,0,nchan);
cpgbox("ABN",600,10,"ABN",100,1);
cpggray(ch_fft_n,nchan,nblocks,1,nchan,1,nblocks,max_fft_n,min_fft_n,tr);
cpgclos();
cpgopen("4/xs");
cpgsvp(0.1,0.9,0.1,0.9);
cpgswin(0,nblocks*tsamp*nsamp_per_block,0,nchan);
cpgbox("ABN",600,10,"ABN",100,1);
cpggray(ch_fft_p,nchan,nblocks,1,nchan,1,nblocks,max_fft_p,min_fft_p,tr);
cpgclos();
cpgopen("mean.ps/vcps");
cpgsvp(0.1,0.9,0.1,0.9);
cpgswin(0,nblocks*tsamp*nsamp_per_block,0,nchan);
cpgbox("ABN",600,10,"ABN",100,1);
cpggray(ch_mean,nchan,nblocks,1,nchan,1,nblocks,max_mean,min_mean,tr);
cpgclos();
cpgopen("var.ps/vcps");
cpgsvp(0.1,0.9,0.1,0.9);
cpgswin(0,nblocks*tsamp*nsamp_per_block,0,nchan);
cpgbox("ABN",600,10,"ABN",100,1);
cpggray(ch_var,nchan,nblocks,1,nchan,1,nblocks,max_var,min_var,tr);
cpgclos();
cpgopen("fft_n.ps/vcps");
cpgsvp(0.1,0.9,0.1,0.9);
cpgswin(0,nblocks*tsamp*nsamp_per_block,0,nchan);
cpgbox("ABN",600,10,"ABN",100,1);
cpggray(ch_fft_n,nchan,nblocks,1,nchan,1,nblocks,max_fft_n,min_fft_n,tr);
cpgclos();
cpgopen("fft_p.ps/vcps");
cpgsvp(0.1,0.9,0.1,0.9);
cpgswin(0,nblocks*tsamp*nsamp_per_block,0,nchan);
cpgbox("ABN",600,10,"ABN",100,1);
cpggray(ch_fft_p,nchan,nblocks,1,nchan,1,nblocks,max_fft_p,min_fft_p,tr);
cpgclos();
fclose(f);
free(buffer);
free_2df(data);
return 0;
}
int main(int argc, char *argv[]) {
int opt=0, verb=0;
int max_harm = 64, max_lag=0;
int isub = 1;
while ((opt=getopt(argc,argv,"hvI:H:L:"))!=-1) {
switch (opt) {
case 'v':
verb++;
break;
case 'I':
isub = atoi(optarg);
break;
case 'H':
max_harm = atoi(optarg);
break;
case 'L':
max_lag = atoi(optarg);
break;
case 'h':
usage();
exit(0);
break;
}
}
if (optind==argc) {
usage();
exit(1);
}
int i, rv;
/* Open file */
fitsfile *f;
int status;
fits_open_file(&f, argv[optind], READONLY, &status);
fits_error_check_fatal();
/* Get basic dims */
struct cyclic_work w;
cyclic_load_params(f, &w, &status);
fits_error_check_fatal();
if (verb) {
printf("Read nphase=%d npol=%d nchan=%d\n",
w.nphase, w.npol, w.nchan);
fflush(stdout);
}
int orig_npol = w.npol;
w.npol = 1;
/* Init FFTs */
fftwf_init_threads();
fftwf_plan_with_nthreads(4);
if (verb) { printf("Planning FFTs\n"); fflush(stdout); }
#define WF "/home/pdemores/share/cyclic_wisdom.dat"
FILE *wf = fopen(WF,"r");
if (wf!=NULL) { fftwf_import_wisdom_from_file(wf); fclose(wf); }
rv = cyclic_init_ffts(&w);
if (rv) {
fprintf(stderr, "Error planning ffts (rv=%d)\n", rv);
exit(1);
}
wf = fopen(WF,"w");
if (wf!=NULL) { fftwf_export_wisdom_to_file(wf); fclose(wf); }
/* Alloc some stuff */
struct periodic_spectrum raw;
struct cyclic_spectrum cs, model_cs;
struct filter_time ht;
struct filter_freq hf;
struct profile_phase pp;
struct profile_harm ph;
raw.nphase = pp.nphase = w.nphase;
raw.nchan = cs.nchan = hf.nchan = w.nchan;
cs.nharm = ph.nharm = w.nharm;
ht.nlag = w.nlag;
raw.npol = orig_npol;
cs.npol = 1;
model_cs.nchan = cs.nchan;
model_cs.nharm = cs.nharm;
model_cs.npol = cs.npol;
cyclic_alloc_ps(&raw);
cyclic_alloc_cs(&cs);
cyclic_alloc_cs(&model_cs);
filter_alloc_time(&ht);
filter_alloc_freq(&hf);
profile_alloc_phase(&pp);
profile_alloc_harm(&ph);
#if 0 // XXX not implemented yet
/* Check bounds */
if (max_harm > w.nharm) { max_harm = w.nharm; }
if (max_lag > w.nlag/2) { max_lag = w.nlag/2; }
if (verb) {
printf("Using max of %d harmonics and %d lags\n", max_harm, max_lag);
}
#endif
/* Load data */
cyclic_load_ps(f, &raw, isub, &status);
fits_error_check_fatal();
/* Add polns w/o calibration */
cyclic_pscrunch_ps(&raw, 1.0, 1.0);
/* Initialize H, profile guesses */
cyclic_fscrunch_ps(&pp, &raw);
profile_phase2harm(&pp, &ph, &w);
ht.data[0] = 1.0;
for (i=1; i<ht.nlag; i++) { ht.data[i] = 0.0; }
filter_profile_norm(&ht, &ph, w.nharm);
profile_harm2phase(&ph, &pp, &w);
/* convert input data to cyclic spectrum */
cyclic_ps2cs(&raw, &cs, &w);
/* could output initial profile */
/* Fill in data struct for nlopt */
struct cyclic_data cdata;
cdata.cs = &cs;
cdata.s0 = &ph;
cdata.ht = &ht;
cdata.model_cs = &model_cs;
cdata.w = &w;
/* Set up minimizer */
const int dim = 2*(w.nharm-1) + 2*w.nlag; /* number of free params */
printf("number of fit params = %d\n", dim);
nlopt_opt op;
op = nlopt_create(NLOPT_LN_COBYLA, dim);
nlopt_set_min_objective(op, cyclic_ms_difference_nlopt, &cdata);
nlopt_set_xtol_rel(op, 1e-4);
/* Set up initial params */
double *x = (double *)malloc(sizeof(double) * dim);
double *xtmp = x;
for (i=1; i<ph.nharm; i++) {
xtmp[0] = creal(ph.data[i]);
xtmp[1] = cimag(ph.data[i]);
xtmp += 2;
}
for (i=0; i<ht.nlag; i++) {
xtmp[0] = creal(ht.data[i]);
xtmp[1] = cimag(ht.data[i]);
xtmp += 2;
}
/* Run optimization */
double min;
if (nlopt_optimize(op, x, &min)) {
fprintf(stderr, "nlopt_optimize failed\n");
exit(1);
}
/* TODO: some kind of output */
/* All done :) */
nlopt_destroy(op);
exit(0);
}
int
main(int argc, char *argv[])
{
uint16_t *addr_in;
float *addr_out;
int fd_in;
int fd_out;
struct stat sb;
size_t length;
uint64_t fftlen;
size_t num_ffts;
size_t index;
if (argc < 4|| argc > 5) {
fprintf(stderr, "%s filein fileout fftlength [nthreads]\n", argv[0]);
exit(EXIT_FAILURE);
}
fd_in = open(argv[1], O_RDONLY|O_LARGEFILE);
if (fd_in == -1)
handle_error("in open");
fd_out = open(argv[2], O_LARGEFILE|O_CREAT|O_TRUNC|O_RDWR, 0666);
if (fd_out == -1)
handle_error("out open");
fftlen = atoll(argv[3]);
int nthreads = (argc == 5) ? 4 : atoi(argv[4]);
posix_fallocate(fd_out, 0, fftlen* sizeof(float));
if (fstat(fd_in, &sb) == -1) /* To obtain file size */
handle_error("fstat");
length = sb.st_size; //two bytes per short, two shorts per complex value
num_ffts = length/fftlen;
//map the input into memory.
addr_in = mmap(NULL, length, PROT_READ,
MAP_PRIVATE, fd_in, 0);
if (addr_in == MAP_FAILED)
handle_error("input mmap");
//map the output into memory.
addr_out = mmap(NULL, fftlen*sizeof(float), PROT_WRITE,
MAP_SHARED, fd_out, 0);
if (addr_out == MAP_FAILED)
handle_error("output mmap");
//fftwf stuff
fftwf_init_threads();
fftwf_plan_with_nthreads(nthreads);
fftwf_complex *in, *out;
fftwf_plan my_plan;
in = (fftwf_complex*) fftwf_malloc(sizeof(fftwf_complex)*fftlen);
out = (fftwf_complex*) fftwf_malloc(sizeof(fftwf_complex)*fftlen);
my_plan = fftwf_plan_dft_1d(fftlen, in, out, FFTW_FORWARD, FFTW_ESTIMATE);
float *fin = (float*)in;
float *fout = (float*)out;
for(index = 0; index < fftlen*2; index++) {
float tmp = (float) addr_in[index];
fin[index] =tmp;
}
addr_in += 2 * fftlen;
//run that FFT
fftwf_execute(my_plan);
//calculate Power
size_t currindex=0;
float real, imag;
for(index = 0; index < fftlen; index++) {
real = fout[currindex++];
imag = fout[currindex++];
addr_out[index] = 10.0f*log10f( real * real + imag * imag );
}
fftwf_destroy_plan(my_plan);
fftwf_free(in);
fftwf_free(out);
msync(addr_out, fftlen * sizeof(float), MS_SYNC);
exit(EXIT_SUCCESS);
}
int main(int argc, char *argv[]) {
int opt=0, verb=0;
int max_harm = 64, max_lag=0;
int causal_filter = 0;
while ((opt=getopt(argc,argv,"hvH:L:C"))!=-1) {
switch (opt) {
case 'v':
verb++;
break;
case 'H':
max_harm = atoi(optarg);
break;
case 'L':
max_lag = atoi(optarg);
break;
case 'C':
causal_filter = 1;
break;
case 'h':
usage();
exit(0);
break;
}
}
if (optind==argc) {
usage();
exit(1);
}
int i, rv;
/* Open file */
fitsfile *f;
int status;
fits_open_file(&f, argv[optind], READONLY, &status);
fits_error_check_fatal();
/* Get basic dims */
struct cyclic_work w;
cyclic_load_params(f, &w, &status);
fits_error_check_fatal();
if (verb) {
printf("Read nphase=%d npol=%d nchan=%d\n",
w.nphase, w.npol, w.nchan);
fflush(stdout);
}
int orig_npol = w.npol;
w.npol = 1;
/* Init FFTs */
fftwf_init_threads();
fftwf_plan_with_nthreads(4);
if (verb) { printf("Planning FFTs\n"); fflush(stdout); }
#define WF "/home/pdemores/share/cyclic_wisdom.dat"
FILE *wf = fopen(WF,"r");
if (wf!=NULL) { fftwf_import_wisdom_from_file(wf); fclose(wf); }
rv = cyclic_init_ffts(&w);
if (rv) {
fprintf(stderr, "Error planning ffts (rv=%d)\n", rv);
exit(1);
}
wf = fopen(WF,"w");
if (wf!=NULL) { fftwf_export_wisdom_to_file(wf); fclose(wf); }
/* Alloc some stuff */
struct periodic_spectrum raw;
struct cyclic_spectrum cs, cs_neg;
struct filter_time ht, ht_new;
struct filter_freq hf, hf_new;
struct filter_freq *hf_shift_pos, *hf_shift_neg;
hf_shift_pos = (struct filter_freq *)malloc(
sizeof(struct filter_freq)*w.nharm);
hf_shift_neg = (struct filter_freq *)malloc(
sizeof(struct filter_freq)*w.nharm);
struct profile_phase pp, pp_new;
struct profile_harm ph, ph_new;
raw.nphase = pp.nphase = pp_new.nphase = w.nphase;
raw.nchan = cs.nchan = hf.nchan = hf_new.nchan = w.nchan;
cs.nharm = ph.nharm = ph_new.nharm = w.nharm;
ht.nlag = ht_new.nlag = w.nlag;
for (i=0; i<w.nharm; i++) { hf_shift_pos[i].nchan = w.nchan; }
for (i=0; i<w.nharm; i++) { hf_shift_neg[i].nchan = w.nchan; }
raw.npol = orig_npol;
cs.npol = 1;
cs_neg.nchan = cs.nchan;
cs_neg.nharm = cs.nharm;
cs_neg.npol = cs.npol;
cyclic_alloc_ps(&raw);
cyclic_alloc_cs(&cs);
cyclic_alloc_cs(&cs_neg);
filter_alloc_time(&ht);
filter_alloc_time(&ht_new);
filter_alloc_freq(&hf);
filter_alloc_freq(&hf_new);
for (i=0; i<w.nharm; i++) {
filter_alloc_freq(&hf_shift_pos[i]);
filter_alloc_freq(&hf_shift_neg[i]);
}
profile_alloc_phase(&pp);
profile_alloc_phase(&pp_new);
profile_alloc_harm(&ph);
profile_alloc_harm(&ph_new);
/* Check bounds */
if (max_harm > w.nharm) { max_harm = w.nharm; }
if (max_lag > w.nlag/2) { max_lag = w.nlag/2; }
if (verb) {
printf("Using max of %d harmonics and %d lags\n", max_harm, max_lag);
}
/* Run procedure on subint 0 */
int isub = 1;
/* Load data */
cyclic_load_ps(f, &raw, isub, &status);
fits_error_check_fatal();
/* Add polns w/o calibration */
cyclic_pscrunch_ps(&raw, 1.0, 1.0);
/* Initialize H, profile guesses */
cyclic_fscrunch_ps(&pp, &raw);
profile_phase2harm(&pp, &ph, &w);
ht.data[0] = 1.0;
for (i=1; i<ht.nlag; i++) { ht.data[i] = 0.0; }
filter_profile_norm(&ht, &ph, max_harm);
profile_harm2phase(&ph, &pp, &w);
/* convert to CS, produce shifted version */
cyclic_ps2cs(&raw, &cs, &w);
cyclic_ps2cs(&raw, &cs_neg, &w);
cyclic_shift_cs(&cs, +1, &w);
cyclic_shift_cs(&cs_neg, -1, &w);
/* TODO output initial profile */
/* Remove old files */
#define FILT "filters.dat"
#define TFILT "tfilters.dat"
#define PROF "profs.dat"
#define FPROF "fprofs.dat"
unlink(FILT);
unlink(TFILT);
unlink(PROF);
unlink(FPROF);
FILE *it = fopen("iter.dat", "w");
/* iterate */
int nit=0;
double mse=0.0, last_mse=0.0;
signal(SIGINT, cc);
do {
if (verb) {
printf("iter %d\n", nit);
fflush(stdout);
}
/* Make freq domain filter */
filter_time2freq(&ht, &hf, &w);
write_filter(TFILT, &ht);
write_filter_freq(FILT, &hf);
/* Make shifted filter array */
filter_shift(hf_shift_pos, &ht, w.nharm,
raw.ref_freq/(raw.bw*1e6), &w);
filter_shift(hf_shift_neg, &ht, w.nharm,
-1.0*raw.ref_freq/(raw.bw*1e6), &w);
mse = cyclic_mse(&cs, &cs_neg, &ph, hf_shift_pos, hf_shift_neg,
max_harm);
/* Update filter, prof */
cyclic_update_filter(&hf_new, &cs, &cs_neg, &ph,
hf_shift_pos, hf_shift_neg, max_harm);
cyclic_update_profile(&ph_new, &cs, &cs_neg,
hf_shift_pos, hf_shift_neg);
/* Back to time domain filter */
filter_freq2time(&hf_new, &ht_new, &w);
/* Fix filter normalization */
for (i=0; i<ht_new.nlag; i++)
ht_new.data[i] /= (float)ht_new.nlag;
/* Zero out negative lags */
if (causal_filter) {
for (i=ht_new.nlag/2; i<ht_new.nlag; i++)
ht_new.data[i] = 0.0;
}
/* Zero out large lags */
if (max_lag>0) {
for (i=max_lag; i<ht_new.nlag-max_lag; i++)
ht_new.data[i] = 0.0;
}
/* Kill nyquist point?? */
ht_new.data[ht_new.nlag/2] = 0.0;
/* Normalize prof and filter */
filter_profile_norm(&ht_new, &ph_new, max_harm);
/* TODO some kind of convergence test */
double prof_diff = profile_ms_difference(&ph, &ph_new, max_harm);
double filt_diff = filter_ms_difference(&ht, &ht_new);
/* TODO zero out high harmonics ?? */
/* Step halfway to new versions, except first time */
if (nit==0) {
for (i=0; i<w.nharm; i++)
ph.data[i] = ph_new.data[i];
for (i=0; i<w.nlag; i++)
ht.data[i] = ht_new.data[i];
} else {
//double fac = (mse<last_mse) ? 1.0 : 0.5*sqrt(mse/last_mse);
double fac=0.25;
for (i=0; i<w.nharm; i++)
ph.data[i] = (1.0-fac)*ph.data[i] + fac*ph_new.data[i];
for (i=0; i<w.nlag; i++)
ht.data[i] = (1.0-fac)*ht.data[i] + fac*ht_new.data[i];
}
/* Back to phase domain profile */
ph.data[0] = 0.0;
profile_harm2phase(&ph, &pp_new, &w);
/* Write out current profiles */
write_profile(PROF, &pp_new);
write_fprofile(FPROF, &ph);
/* Print convergence params */
if (verb) {
fprintf(it,"%.3e %.3e %.8e %.8e\n", prof_diff, filt_diff, mse,
mse - last_mse);
}
last_mse = mse;
/* Update iter count */
nit++;
} while (run);
fclose(it);
exit(0);
}
int cfft2_init(int pad1 /* padding on the first axis */,
int nx, int ny /* input data size */,
int *nx2, int *ny2 /* padded data size */)
/*< initialize >*/
{
#ifdef SF_HAS_FFTW
#ifdef _OPENMP
fftwf_init_threads();
if (false)
sf_warning("Using threaded FFTW3! \n");
fftwf_plan_with_nthreads(omp_get_max_threads());
#else
if (false)
sf_warning("Using FFTW3! \n");
#endif
#else
if (false)
sf_warning("Using KissFFT! \n");
#endif
#ifndef SF_HAS_FFTW
int i2;
#endif
nk = n1 = kiss_fft_next_fast_size(nx*pad1);
n2 = kiss_fft_next_fast_size(ny);
cc = sf_complexalloc2(n1,n2);
#ifdef SF_HAS_FFTW
dd = sf_complexalloc2(nk,n2);
cfg = fftwf_plan_dft_2d(n2,n1,
(fftwf_complex *) cc[0],
(fftwf_complex *) dd[0],
FFTW_FORWARD, FFTW_MEASURE);
icfg = fftwf_plan_dft_2d(n2,n1,
(fftwf_complex *) dd[0],
(fftwf_complex *) cc[0],
FFTW_BACKWARD, FFTW_MEASURE);
if (NULL == cfg || NULL == icfg) sf_error("FFTW failure.");
#else
cfg1 = kiss_fft_alloc(n1,0,NULL,NULL);
icfg1 = kiss_fft_alloc(n1,1,NULL,NULL);
cfg2 = kiss_fft_alloc(n2,0,NULL,NULL);
icfg2 = kiss_fft_alloc(n2,1,NULL,NULL);
tmp = (kiss_fft_cpx **) sf_alloc(n2,sizeof(*tmp));
tmp[0] = (kiss_fft_cpx *) sf_alloc(nk*n2,sizeof(kiss_fft_cpx));
#ifdef _OPENMP
#pragma omp parallel for private(i2) default(shared)
#endif
for (i2=0; i2 < n2; i2++) {
tmp[i2] = tmp[0]+i2*nk;
}
trace2 = sf_complexalloc(n2);
ctrace2 = (kiss_fft_cpx *) trace2;
#endif
*nx2 = n1;
*ny2 = n2;
wt = 1.0/(n1*n2);
return (nk*n2);
}
static int init_threads() {
int res = fftwf_init_threads();
return res;
}
float* CalcFFT(GDALDataset *srcDS1, GDALDataset *srcDS2, GDALDataset *dstDS)
{
fftwf_plan plan1, plan2, planI;
fftwf_complex *img1, *img2;
unsigned char *out;
int band;
const size_t px_count = dstDS->GetRasterXSize() * dstDS->GetRasterYSize();
const size_t buffer_len = sizeof(fftwf_complex) * px_count;
img1 = (fftwf_complex*) fftwf_malloc(buffer_len);
img2 = (fftwf_complex*) fftwf_malloc(buffer_len);
out = (unsigned char*) fftwf_malloc(sizeof(unsigned char) * px_count);
/* ^ not used in fft, but aligned is good anyway */
if(img1 == NULL || img2 == NULL || out == NULL)
error("Could not allocate memory\n");
if(fftwf_init_threads())
fftwf_plan_with_nthreads(CORES);
plan1 = fftwf_plan_dft_2d(dstDS->GetRasterYSize(), dstDS->GetRasterXSize(),
img1, img1, FFTW_FORWARD, FFTW_ESTIMATE);
plan2 = fftwf_plan_dft_2d(dstDS->GetRasterYSize(), dstDS->GetRasterXSize(),
img2, img2, FFTW_FORWARD, FFTW_ESTIMATE);
planI = fftwf_plan_dft_2d(dstDS->GetRasterYSize(), dstDS->GetRasterXSize(),
img2, img2, FFTW_BACKWARD, FFTW_ESTIMATE);
if(plan1 == NULL || plan2 == NULL || planI == NULL)
error("Could not plan FFT\n");
for(band = 1; band <= dstDS->GetRasterCount(); band++) {
printf("FFT 1 band %d\n", band);
runFFT( plan1, srcDS1, img1, band, dstDS );
printf("FFT 2 band %d\n", band);
runFFT( plan2, srcDS2, img2, band, dstDS );
printf("Complex Conj band %d\n", band);
/* mult img1 and conj of img2 */
for(int px = 0; px < px_count; px++) {
img2[px] = img1[px] * conj(img2[px]);
}
/* IFFT of result */
printf("IFFT band %d\n", band);
fftwf_execute(planI);
printf("normalize band %d\n", band);
complex float norm = csqrt(px_count + 0I);
float max = cabs(img2[0] / norm);
float min = cabs(img2[0] / norm);
for(int i = 0; i < px_count; i++) {
img2[i] = img2[i] / norm;
if(cabs(img2[i]) < min)
min = cabs(img2[i]);
if(cabs(img2[i]) > max)
max = cabs(img2[i]);
}
/* img2 should now be real - normalize 0-255 and -- write output */
printf("Save band %d; min = %f max = %f\n", band, min, max);
for(int i = 0; i < px_count; i++) {
out[i] = floor( ((cabs(img2[i]) - min) / (max-min) ) * 255.0 );
}
fft2shift(out, dstDS->GetRasterYSize(), dstDS->GetRasterXSize());
dstDS->GetRasterBand(band)->RasterIO( GF_Write, 0, 0,
dstDS->GetRasterXSize(), dstDS->GetRasterYSize(),
out, dstDS->GetRasterXSize(), dstDS->GetRasterYSize(),
GDT_Byte, 0, 0);
}
fftwf_destroy_plan(plan1);
fftwf_destroy_plan(plan2);
fftwf_destroy_plan(planI);
fftwf_free(img1);
fftwf_free(img2);
fftwf_free(out);
}
int main(int argc, char *argv[]) {
char fname[300];
FILE *fid;
DIR* dir;
size_t elem;
long int ii,ij,ik, ii_c, ij_c, ik_c, a, b, c;
long int ncells_1D;
long int i,j,p,indi,indj,ind;
int flag_bub,iz;
double redshift,tmp;
double kk;
double bfactor; /* value by which to divide bubble size R */
double neutral,*xHI;
float *halo_map, *top_hat_r, *density_map,*bubblef, *bubble;
fftwf_complex *halo_map_c, *top_hat_c, *collapsed_mass_c, *density_map_c, *total_mass_c, *bubble_c;
fftwf_plan pr2c1,pr2c2,pr2c3,pr2c4,pc2r1,pc2r2,pc2r3;
double zmin,zmax,dz;
double R;
if(argc != 2) {
printf("Generates boxes with ionization fraction for a range of redshifts\n");
printf("usage: get_HIIbubbles base_dir\n");
printf("base_dir contains simfast21.ini and directory structure\n");
exit(1);
}
get_Simfast21_params(argv[1]);
zmin=global_Zminsim;
zmax=global_Zmaxsim;
dz=global_Dzsim;
bfactor=pow(10.0,log10(global_bubble_Rmax/global_dx_smooth)/global_bubble_Nbins);
printf("Bubble radius ratio (bfactor): %f\n", bfactor); fflush(0);
#ifdef _OMPTHREAD_
omp_set_num_threads(global_nthreads);
fftwf_init_threads();
fftwf_plan_with_nthreads(global_nthreads);
printf("Using %d threads\n",global_nthreads);fflush(0);
#endif
/* Create directory Ionization */
sprintf(fname,"%s/Ionization",argv[1]);
if((dir=opendir(fname))==NULL) {
printf("Creating Ionization directory\n");
if(mkdir(fname,(S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH))!=0) {
printf("Error creating directory!\n");
exit(1);
}
}
sprintf(fname,"%s/Output_text_files",argv[1]);
if((dir=opendir(fname))==NULL) {
printf("Creating Output_text_files directory\n");
if(mkdir(fname,(S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH))!=0) {
printf("Error creating directory!\n");
exit(1);
}
}
/* Memory allocation - we could do some of the FFTs inline... */
/*************************************************************/
/* density_map mass */
if(!(density_map=(float *) fftwf_malloc(global_N3_smooth*sizeof(float)))) {
printf("Problem1...\n");
exit(1);
}
if(!(density_map_c=(fftwf_complex *) fftwf_malloc(global_N_smooth*global_N_smooth*(global_N_smooth/2+1)*sizeof(fftwf_complex)))) {
printf("Problem2...\n");
exit(1);
}
if(!(pr2c1=fftwf_plan_dft_r2c_3d(global_N_smooth, global_N_smooth, global_N_smooth, density_map, density_map_c, FFTWflag))) {
printf("Problem3...\n");
exit(1);
}
/* halo_map mass */
if(!(halo_map=(float *) fftwf_malloc(global_N3_smooth*sizeof(float)))) {
printf("Problem4...\n");
exit(1);
}
if(!(halo_map_c=(fftwf_complex *) fftwf_malloc(global_N_smooth*global_N_smooth*(global_N_smooth/2+1)*sizeof(fftwf_complex)))) {
printf("Problem5...\n");
exit(1);
}
if(!(pr2c2=fftwf_plan_dft_r2c_3d(global_N_smooth, global_N_smooth, global_N_smooth, halo_map, halo_map_c, FFTWflag))) {
printf("Problem6...\n");
exit(1);
}
/* total mass */
if(!(total_mass_c=(fftwf_complex *) fftwf_malloc(global_N_smooth*global_N_smooth*(global_N_smooth/2+1)*sizeof(fftwf_complex)))) {
printf("Problem7...\n");
exit(1);
}
if(!(pc2r1=fftwf_plan_dft_c2r_3d(global_N_smooth, global_N_smooth, global_N_smooth, total_mass_c, density_map, FFTWflag))) {
printf("Problem8...\n");
exit(1);
}
/* collapsed mass */
if(!(collapsed_mass_c=(fftwf_complex *) fftwf_malloc(global_N_smooth*global_N_smooth*(global_N_smooth/2+1)*sizeof(fftwf_complex)))) {
printf("Problem9...\n");
exit(1);
}
if(!(pc2r2=fftwf_plan_dft_c2r_3d(global_N_smooth, global_N_smooth, global_N_smooth, collapsed_mass_c, halo_map, FFTWflag))) {
printf("Problem10...\n");
exit(1);
}
/* top hat window */
if(!(top_hat_r=(float *) fftwf_malloc(global_N3_smooth*sizeof(float)))) {
printf("Problem11...\n");
exit(1);
}
if(!(top_hat_c=(fftwf_complex *) fftwf_malloc(global_N_smooth*global_N_smooth*(global_N_smooth/2+1)*sizeof(fftwf_complex)))) {
printf("Problem12...\n");
exit(1);
}
if(!(pr2c3=fftwf_plan_dft_r2c_3d(global_N_smooth, global_N_smooth, global_N_smooth, top_hat_r, top_hat_c, FFTWflag))) {
printf("Problem13...\n");
exit(1);
}
/* bubble boxes */
if(!(bubble=(float *) fftwf_malloc(global_N3_smooth*sizeof(float)))) {
printf("Problem14...\n");
exit(1);
}
if(!(bubble_c=(fftwf_complex *) fftwf_malloc(global_N_smooth*global_N_smooth*(global_N_smooth/2+1)*sizeof(fftwf_complex)))) {
printf("Problem15...\n");
exit(1);
}
if(!(bubblef=(float *) malloc(global_N3_smooth*sizeof(float)))) {
printf("Problem16...\n");
exit(1);
}
if(!(pr2c4=fftwf_plan_dft_r2c_3d(global_N_smooth, global_N_smooth, global_N_smooth, bubble, bubble_c, FFTWflag))) {
printf("Problem17...\n");
exit(1);
}
if(!(pc2r3=fftwf_plan_dft_c2r_3d(global_N_smooth, global_N_smooth, global_N_smooth, bubble_c, bubble, FFTWflag))) {
printf("Problem18...\n");
exit(1);
}
if(!(xHI=(double *) malloc((int)((zmax-zmin)/dz+2)*sizeof(double)))) {
printf("Problem19...\n");
exit(1);
}
/****************************************************/
/***************** Redshift cycle *******************/
printf("Number of bubble sizes: %d\n",(int)((log(global_bubble_Rmax)-log(2.*global_dx_smooth))/log(bfactor)));
printf("Redshift cycle...\n");fflush(0);
iz=0;
neutral=0.;
for(redshift=zmin;redshift<(zmax+dz/10) && (neutral < xHlim);redshift+=dz){
printf("z = %f\n",redshift);fflush(0);
sprintf(fname, "%s/delta/deltanl_z%.3f_N%ld_L%.1f.dat",argv[1],redshift,global_N_smooth,global_L/global_hubble);
fid=fopen(fname,"rb");
if (fid==NULL) {printf("\nError reading deltanl file... Check path or if the file exists..."); exit (1);}
elem=fread(density_map,sizeof(float),global_N3_smooth,fid);
fclose(fid);
#ifdef _OMPTHREAD_
#pragma omp parallel for shared(global_N3_smooth, density_map,global_rho_m, global_dx_smooth,bubblef) private(i)
#endif
for(i=0;i<(global_N3_smooth);i++){
density_map[i]=(1.0+density_map[i])*global_rho_m*global_dx_smooth*global_dx_smooth*global_dx_smooth; /* total mass in 1 cell */
bubblef[i]=0.0;
}
sprintf(fname, "%s/Halos/masscoll_z%.3f_N%ld_L%.1f.dat",argv[1],redshift,global_N_smooth,global_L/global_hubble);
fid=fopen(fname,"rb");
if (fid==NULL) {printf("\nError reading %s file... Check path or if the file exists...",fname); exit (1);}
elem=fread(halo_map,sizeof(float),global_N3_smooth,fid);
fclose(fid);
/* Quick fill of single cells before going to bubble cycle */
#ifdef _OMPTHREAD_
#pragma omp parallel for shared(global_N3_smooth,halo_map,density_map,global_eff,bubblef) private(i,tmp)
#endif
for(i=0;i<global_N3_smooth;i++) {
if(halo_map[i]>0.) {
if(density_map[i]>0.)
tmp=(double)halo_map[i]*global_eff/density_map[i];
else tmp=1.0;
}else tmp=0.;
if(tmp>=1.0) bubblef[i]=1.0; else bubblef[i]=tmp;
}
/* FFT density and halos */
fftwf_execute(pr2c1);
fftwf_execute(pr2c2);
/************** going over the bubble sizes ****************/
R=global_bubble_Rmax; /* Maximum bubble size...*/
while(R>=2*global_dx_smooth){
printf("bubble radius R= %lf\n", R);fflush(0);
// printf("Filtering halo and density boxes...\n");fflush(0);
#ifdef _OMPTHREAD_
#pragma omp parallel for shared(collapsed_mass_c,halo_map_c,total_mass_c,density_map_c,global_N_smooth,global_dk,R) private(i,j,p,indi,indj,kk)
#endif
for(i=0;i<global_N_smooth;i++) {
if(i>global_N_smooth/2) {
indi=-(global_N_smooth-i);
}else indi=i;
for(j=0;j<global_N_smooth;j++) {
if(j>global_N_smooth/2) {
indj=-(global_N_smooth-j);
}else indj=j;
for(p=0;p<=global_N_smooth/2;p++) {
kk=global_dk*sqrt(indi*indi+indj*indj+p*p);
total_mass_c[i*global_N_smooth*(global_N_smooth/2+1)+j*(global_N_smooth/2+1)+p]=density_map_c[i*global_N_smooth*(global_N_smooth/2+1)+j*(global_N_smooth/2+1)+p]*W_filter(kk*R);
collapsed_mass_c[i*global_N_smooth*(global_N_smooth/2+1)+j*(global_N_smooth/2+1)+p]=halo_map_c[i*global_N_smooth*(global_N_smooth/2+1)+j*(global_N_smooth/2+1)+p]*W_filter(kk*R);
}
}
}
fftwf_execute(pc2r1);
fftwf_execute(pc2r2);
flag_bub=0;
// printf("Starting to find and fill bubbles...\n");fflush(0);
/* signal center of bubbles */
#ifdef _OMPTHREAD_
#pragma omp parallel for shared(halo_map,density_map,bubble,global_N_smooth,global_eff,flag_bub) private(ii,ij,ik,ind)
#endif
for(ii=0;ii<global_N_smooth;ii++){
for(ij=0;ij<global_N_smooth;ij++){
for(ik=0;ik<global_N_smooth;ik++){
ind=ii*global_N_smooth*global_N_smooth+ij*global_N_smooth+ik;
if(halo_map[ind]>0.) {
if(density_map[ind]>0.) {
if((double)halo_map[ind]/density_map[ind]>=1.0/global_eff) {
flag_bub=1;
bubble[ind]=1.0;
}else bubble[ind]=0;
}else {
flag_bub=1;
bubble[ind]=1.0;
}
}else bubble[ind]=0;
}
}
}
/* generate spherical window in real space for a given R */
if(flag_bub>0){
printf("Found bubble...\n");fflush(0);
#ifdef _OMPTHREAD_
#pragma omp parallel for shared(top_hat_r,R,global_dx_smooth,global_N_smooth) private(i,j,p)
#endif
for(i=0;i<global_N_smooth;i++){
for(j=0;j<global_N_smooth;j++){
for(p=0;p<global_N_smooth;p++){
if(sqrt(i*i+j*j+p*p)*global_dx_smooth<=R || sqrt(i*i+(j-global_N_smooth)*(j-global_N_smooth)+p*p)*global_dx_smooth<=R || sqrt(i*i+(j-global_N_smooth)*(j-global_N_smooth)+(p-global_N_smooth)*(p-global_N_smooth))*global_dx_smooth <=R || sqrt(i*i+(p-global_N_smooth)*(p-global_N_smooth)+j*j)*global_dx_smooth<=R || sqrt((i-global_N_smooth)*(i-global_N_smooth)+j*j+p*p)*global_dx_smooth<=R || sqrt((i-global_N_smooth)*(i-global_N_smooth)+(j-global_N_smooth)*(j-global_N_smooth)+p*p)*global_dx_smooth<=R || sqrt((i-global_N_smooth)*(i-global_N_smooth)+(j-global_N_smooth)*(j-global_N_smooth)+(p-global_N_smooth)*(p-global_N_smooth))*global_dx_smooth<=R || sqrt((i-global_N_smooth)*(i-global_N_smooth)+j*j+(p-global_N_smooth)*(p-global_N_smooth))*global_dx_smooth<=R ) {
top_hat_r[i*global_N_smooth*global_N_smooth+j*global_N_smooth+p]=1.0;
}else top_hat_r[i*global_N_smooth*global_N_smooth+j*global_N_smooth+p]=0.0;
}
}
}
/* FFT bubble centers and window */
fftwf_execute(pr2c3);
fftwf_execute(pr2c4);
/* Make convolution */
#ifdef _OMPTHREAD_
#pragma omp parallel for shared(bubble_c,top_hat_c,global_N_smooth) private(i)
#endif
for(i=0;i<global_N_smooth*global_N_smooth*(global_N_smooth/2+1);i++) {
bubble_c[i]*=top_hat_c[i];
}
fftwf_execute(pc2r3);
/* after dividing by global_N3_smooth, values in bubble are between 0 (neutral)and global_N3_smooth */
#ifdef _OMPTHREAD_
#pragma omp parallel for shared(bubble,bubblef,global_N3_smooth) private(i)
#endif
for (i=0; i<global_N3_smooth; i++){
bubble[i]/=global_N3_smooth;
if (bubble[i]>0.2) bubblef[i]=1.0; /* neutral should be zero */
}
} /* ends filling out bubbles in box for R */
R/=bfactor;
} /* ends R cycle */
/* just to check smallest bubbles through older method */
printf("Going to smaller R cycle...\n"); fflush(0);
while(R>=global_dx_smooth){
printf("bubble radius R= %lf\n", R);fflush(0);
flag_bub=0;
#ifdef _OMPTHREAD_
#pragma omp parallel for shared(collapsed_mass_c,halo_map_c,total_mass_c,density_map_c,global_N_smooth,global_dx_smooth,global_dk,R) private(i,j,p,indi,indj,kk)
#endif
for(i=0;i<global_N_smooth;i++) {
if(i>global_N_smooth/2) {
indi=-(global_N_smooth-i);
}else indi=i;
for(j=0;j<global_N_smooth;j++) {
if(j>global_N_smooth/2) {
indj=-(global_N_smooth-j);
}else indj=j;
for(p=0;p<=global_N_smooth/2;p++) {
kk=global_dk*sqrt(indi*indi+indj*indj+p*p);
total_mass_c[i*global_N_smooth*(global_N_smooth/2+1)+j*(global_N_smooth/2+1)+p]=density_map_c[i*global_N_smooth*(global_N_smooth/2+1)+j*(global_N_smooth/2+1)+p]*W_filter(kk*R);
collapsed_mass_c[i*global_N_smooth*(global_N_smooth/2+1)+j*(global_N_smooth/2+1)+p]=halo_map_c[i*global_N_smooth*(global_N_smooth/2+1)+j*(global_N_smooth/2+1)+p]*W_filter(kk*R);
}
}
}
fftwf_execute(pc2r1);
fftwf_execute(pc2r2);
/* fill smaller bubbles in box */
ncells_1D=(long int)(R/global_dx_smooth);
// printf("Starting to find and fill bubbles...\n");fflush(0);
#ifdef _OMPTHREAD_
#pragma omp parallel for shared(halo_map,density_map,global_eff,global_N_smooth,global_dx_smooth,R,ncells_1D,bubblef,flag_bub) private(ii_c,ij_c,ik_c,ii,ij,ik,a,b,c,ind)
#endif
for(ii_c=0;ii_c<global_N_smooth;ii_c++){
for(ij_c=0;ij_c<global_N_smooth;ij_c++){
for(ik_c=0;ik_c<global_N_smooth;ik_c++){
ind=ii_c*global_N_smooth*global_N_smooth+ij_c*global_N_smooth+ik_c;
if(halo_map[ind]>0.) {
if(!(density_map[ind]>0.) || ((double)halo_map[ind]/density_map[ind]>=1.0/global_eff)) {
flag_bub=1;
for(ii=-(ncells_1D+1);ii<=ncells_1D+1;ii++){
a=check_borders(ii_c+ii,global_N_smooth);
for(ij=-(ncells_1D+1);ij<=ncells_1D+1;ij++){
if(sqrt(ii*ii+ij*ij)*global_dx_smooth <= R){
b=check_borders(ij_c+ij,global_N_smooth);
for(ik=-(ncells_1D+1);ik<=ncells_1D+1;ik++){
c=check_borders(ik_c+ik,global_N_smooth);
if(sqrt(ii*ii+ij*ij+ik*ik)*global_dx_smooth <= R){
bubblef[a*global_N_smooth*global_N_smooth+b*global_N_smooth+c]=1.0;
}
}
}
}
}
}
}
}
}
}
if(flag_bub>0){printf("Found bubble...\n");fflush(0);}
R/=bfactor;
} /* ends small bubbles R cycle */
neutral=0.;
for (i=0; i<global_N3_smooth; i++){
neutral+=1.0-bubblef[i];
}
neutral/=global_N3_smooth;
printf("neutral fraction=%lf\n",neutral);fflush(0);
xHI[iz]=neutral;
sprintf(fname, "%s/Ionization/xHII_z%.3f_eff%.2lf_N%ld_L%.1f.dat",argv[1],redshift,global_eff,global_N_smooth,global_L/global_hubble);
if((fid = fopen(fname,"wb"))==NULL) {
printf("Error opening file:%s\n",fname);
exit(1);
}
elem=fwrite(bubblef,sizeof(float),global_N3_smooth,fid);
fclose(fid);
iz++;
} /* ends redshift cycle */
/* z cycle for neutral>=xHlim */
while(redshift<(zmax+dz/10)) {
printf("z(>%f) = %f\n",xHlim,redshift);fflush(0);
xHI[iz]=1.0;
sprintf(fname, "%s/Ionization/xHII_z%.3f_eff%.2lf_N%ld_L%.1f.dat",argv[1],redshift,global_eff,global_N_smooth,global_L/global_hubble);
if((fid = fopen(fname,"wb"))==NULL) {
printf("Error opening file:%s\n",fname);
exit(1);
}
#ifdef _OMPTHREAD_
#pragma omp parallel for shared(bubblef,global_N3_smooth) private(i)
#endif
for(i=0;i<global_N3_smooth;i++) bubblef[i]=0.0;
elem=fwrite(bubblef,sizeof(float),global_N3_smooth,fid);
fclose(fid);
iz++;
redshift+=dz;
}
sprintf(fname, "%s/Output_text_files/zsim.txt",argv[1]);
if((fid = fopen(fname,"a"))==NULL) {
printf("Error opening file:%s\n",fname);
exit(1);
}
for(redshift=zmax;redshift>(zmin-dz/10);redshift-=dz) fprintf(fid,"%f\n",redshift); /* first line should be highest redshift */
fclose(fid);
sprintf(fname, "%s/Output_text_files/x_HI_eff%.2lf_N%ld_L%.1f.dat",argv[1],global_eff,global_N_smooth,global_L/global_hubble);
if((fid = fopen(fname,"a"))==NULL) {
printf("Error opening file:%s\n",fname);
exit(1);
}
for(i=iz-1;i>=0;i--) fprintf(fid,"%lf\n",xHI[i]); /* first line should be highest redshift */
fclose(fid);
free(xHI);
free(bubblef);
fftwf_free(top_hat_r);
fftwf_free(top_hat_c);
fftwf_free(collapsed_mass_c);
fftwf_free(density_map);
fftwf_free(density_map_c);
fftwf_free(halo_map);
fftwf_free(halo_map_c);
fftwf_free(total_mass_c);
fftwf_free(bubble);
fftwf_free(bubble_c);
exit(0);
}
GLFFTWater::GLFFTWater(GLFFTWaterParams ¶ms) {
#ifdef _WIN32
m_h = (float *)__mingw_aligned_malloc((sizeof(float)*(params.N+2)*(params.N)), 4);
m_dx = (float *)__mingw_aligned_malloc((sizeof(float)*(params.N+2)*(params.N)), 4);
m_dz = (float *)__mingw_aligned_malloc((sizeof(float)*(params.N+2)*(params.N)), 4);
m_w = (float *)__mingw_aligned_malloc((sizeof(float)*(params.N)*(params.N)), 4);
#else
posix_memalign((void **)&m_h,4,sizeof(float)*(params.N+2)*(params.N));
posix_memalign((void **)&m_dx,4,sizeof(float)*(params.N+2)*(params.N));
posix_memalign((void **)&m_dz,4,sizeof(float)*(params.N+2)*(params.N));
posix_memalign((void **)&m_w,4,sizeof(float)*(params.N)*(params.N));
#endif
m_htilde0 = (fftwf_complex *)fftwf_malloc(sizeof(fftwf_complex)*(params.N)*(params.N));
m_heightmap = new float3[(params.N)*(params.N)];
m_params = params;
std::tr1::mt19937 prng(1337);
std::tr1::normal_distribution<float> normal;
std::tr1::uniform_real<float> uniform;
std::tr1::variate_generator<std::tr1::mt19937, std::tr1::normal_distribution<float> > randn(prng,normal);
std::tr1::variate_generator<std::tr1::mt19937, std::tr1::uniform_real<float> > randu(prng,uniform);
for(int i=0, k=0; i<params.N; i++) {
float k_x = (-(params.N-1)*0.5f+i)*(2.f*3.141592654f / params.L);
for(int j=0; j<params.N; j++, k++) {
float k_y = (-(params.N-1)*0.5f+j)*(2.f*3.141592654f / params.L);
float A = randn();
float theta = randu()*2.f*3.141592654f;
float P = (k_x==0.f && k_y==0.0f) ? 0.f : sqrtf(phillips(k_x,k_y,m_w[k]));
m_htilde0[k][0] = m_htilde0[k][1] = P*A*sinf(theta);
}
}
m_kz = new float[params.N*(params.N / 2 + 1)];
m_kx = new float[params.N*(params.N / 2 + 1)];
const int hN = m_params.N / 2;
for(int y=0; y<m_params.N; y++) {
float kz = (float) (y - hN);
for(int x=0; x<=hN; x++) {
float kx = (float) (x - hN);
float k = 1.f/sqrtf(kx*kx+kz*kz);
m_kz[y*(hN+1)+x] = kz*k;
m_kx[y*(hN+1)+x] = kx*k;
}
}
if(!fftwf_init_threads()) {
cerr << "Error initializing multithreaded fft." << endl;
} else {
fftwf_plan_with_nthreads(2);
}
m_fftplan = fftwf_plan_dft_c2r_2d(m_params.N, m_params.N, (fftwf_complex *)m_h, m_h,
FFTW_ESTIMATE);
glGenTextures(1, &m_texId);
glBindTexture(GL_TEXTURE_2D, m_texId);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
glTexImage2D(GL_TEXTURE_2D, 0, GL_RGB16F, params.N, params.N, 0, GL_RGB, GL_FLOAT, 0);
glBindTexture(GL_TEXTURE_2D, 0);
}
int main(int argc, char **argv){
FILE *fid;
size_t elem;
long int i,j,p;
long int indi, indj;
double kk;
fftwf_complex *map_vel_c;
float *map;
fftwf_complex *map_in_c;
fftwf_plan pc2r;
fftwf_plan pr2c;
char fname[256];
DIR* dir;
if(argc != 2) {
printf("Usage: get_velocityfield work_dir\n");
printf("work_dir - directory containing simfast21.ini \n");
exit(1);
}
get_Simfast21_params(argv[1]);
#ifdef _OMPTHREAD_
omp_set_num_threads(global_nthreads);
fftwf_init_threads();
fftwf_plan_with_nthreads(global_nthreads);
printf("Using %d threads\n",global_nthreads);
#endif
if(!(map=(float *) fftwf_malloc(global_N_halo*global_N_halo*global_N_halo*sizeof(float)))) {
printf("Problem...\n");
exit(1);
}
if(!(map_in_c = (fftwf_complex*) fftwf_malloc(sizeof(fftwf_complex) * global_N_halo*global_N_halo*(global_N_halo/2+1)))) {
printf(" Out of memory...\n");
exit(1);
}
if(!(map_vel_c = (fftwf_complex*) fftwf_malloc(sizeof(fftwf_complex) * global_N_halo*global_N_halo*(global_N_halo/2+1)))) {
printf("Problem allocating memory for x velocity field in k-space...\n");
exit(1);
}
/* Tansformacoes de fourier para calcular as caixas vx(x) vx(y) e vx(z) */
if(!(pc2r=fftwf_plan_dft_c2r_3d(global_N_halo, global_N_halo, global_N_halo, map_vel_c, map, FFTW_ESTIMATE))) {
printf("Problem...\n");
exit(1);
}
/* FFT para map */
if(!(pr2c=fftwf_plan_dft_r2c_3d(global_N_halo, global_N_halo, global_N_halo, map , map_in_c, FFTW_ESTIMATE))) {
printf("Problem...\n");
exit(1);
}
sprintf(fname, "%s/delta/delta_z0_N%ld_L%d.dat", argv[1],global_N_halo,(int)(global_L));
/*Leitura do campo de densidades no espaco real*/
fid=fopen(fname,"rb"); /* second argument contains name of input file */
if (fid==NULL) {
printf("\n Density file path is not correct or the file does not exit...\n");
exit (1);
}
elem=fread(map,sizeof(float),global_N_halo*global_N_halo*global_N_halo,fid);
fclose(fid);
/***********************************************************************************/
// Conversao do mapa de densidades de real para complexo
fftwf_execute(pr2c);
/********************************************************************/
/********************************************************************/
/********************************************************************/
/* Computing velocity fields */
/* Create directory Velocity */
sprintf(fname,"%s/Velocity",argv[1]);
if((dir=opendir(fname))==NULL) {
printf("Creating Velocity directory\n");
if(mkdir(fname,(S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH))!=0) {
printf("Error creating directory!\n");
exit(1);
}
}
/********************************************************************/
printf("\nComputing v_x field...\n");fflush(0);
#ifdef _OMPTHREAD_
#pragma omp parallel for shared(global_N_halo,global_dk,map_vel_c,map_in_c,global_dx_halo) private(i,indi,j,indj,p,kk)
#endif
for(i=0;i<global_N_halo;i++) {
if(i>global_N_halo/2) { /* Large frequencies are equivalent to smaller negative ones */
indi=-(global_N_halo-i);
}else indi=i;
for(j=0;j<global_N_halo;j++) {
if(j>global_N_halo/2) {
indj=-(global_N_halo-j);
}else indj=j;
for(p=0;p<=global_N_halo/2;p++) {
kk=global_dk*sqrt(indi*indi+indj*indj+p*p);
if(kk>0){
//Normalizacao pois a biblioteca fftw3 não tem dx nem global_dk nos integrais
map_in_c[i*global_N_halo*(global_N_halo/2+1)+j*(global_N_halo/2+1)+p]=I*(global_dk)*(1/(kk*kk))*map_in_c[i*global_N_halo*(global_N_halo/2+1)+j*(global_N_halo/2+1)+p]*global_dx_halo*global_dx_halo*global_dx_halo/global_L3;
map_vel_c[i*global_N_halo*(global_N_halo/2+1)+j*(global_N_halo/2+1)+p]=indi*map_in_c[i*global_N_halo*(global_N_halo/2+1)+j*(global_N_halo/2+1)+p];
}else{
map_vel_c[i*global_N_halo*(global_N_halo/2+1)+j*(global_N_halo/2+1)+p]=0;
}
}
}
}
box_symmetriesf(map_vel_c,global_N_halo);
/* Executes FFT */
fftwf_execute(pc2r);
printf("\nWriting v_x field to file...\n");fflush(0);
sprintf(fname, "%s/Velocity/vel_x_z0_N%ld_L%d.dat", argv[1],global_N_halo,(int)(global_L));
if((fid=fopen(fname,"wb"))==NULL){
printf("\nThe file cannot be open\n");
return 0;
}
elem=fwrite(map,sizeof(float),global_N_halo*global_N_halo*global_N_halo,fid);
fclose(fid);
/********************************************************************/
printf("\nComputing v_y field...\n");fflush(0);
#ifdef _OMPTHREAD_
#pragma omp parallel for shared(global_N_halo,global_dk,map_vel_c,map_in_c,global_dx_halo) private(i,indi,j,indj,p,kk)
#endif
for(i=0;i<global_N_halo;i++) {
if(i>global_N_halo/2) { /* Large frequencies are equivalent to smaller negative ones */
indi=-(global_N_halo-i);
}else indi=i;
for(j=0;j<global_N_halo;j++) {
if(j>global_N_halo/2) {
indj=-(global_N_halo-j);
}else indj=j;
for(p=0;p<=global_N_halo/2;p++) {
kk=global_dk*sqrt(indi*indi+indj*indj+p*p);
if(kk>0){
//Normalizacao pois a biblioteca fftw3 não tem dx nem global_dk nos integrais
map_vel_c[i*global_N_halo*(global_N_halo/2+1)+j*(global_N_halo/2+1)+p]=indj*map_in_c[i*global_N_halo*(global_N_halo/2+1)+j*(global_N_halo/2+1)+p];
}else{
map_vel_c[i*global_N_halo*(global_N_halo/2+1)+j*(global_N_halo/2+1)+p]=0;
}
}
}
}
box_symmetriesf(map_vel_c,global_N_halo);
/* Executes FFT */
fftwf_execute(pc2r);
printf("\nWriting v_y field to file...\n");fflush(0);
sprintf(fname, "%s/Velocity/vel_y_z0_N%ld_L%d.dat", argv[1],global_N_halo,(int)(global_L));
if((fid=fopen(fname,"wb"))==NULL){
printf("\nThe file cannot be open\n");
return 0;
}
elem=fwrite(map,sizeof(float),global_N_halo*global_N_halo*global_N_halo,fid);
fclose(fid);
/********************************************************************/
printf("\nComputing v_z field...\n");fflush(0);
#ifdef _OMPTHREAD_
#pragma omp parallel for shared(global_N_halo,global_dk,map_vel_c,map_in_c,global_dx_halo) private(i,indi,j,indj,p,kk)
#endif
for(i=0;i<global_N_halo;i++) {
if(i>global_N_halo/2) { /* Large frequencies are equivalent to smaller negative ones */
indi=-(global_N_halo-i);
}else indi=i;
for(j=0;j<global_N_halo;j++) {
if(j>global_N_halo/2) {
indj=-(global_N_halo-j);
}else indj=j;
for(p=0;p<=global_N_halo/2;p++) {
kk=global_dk*sqrt(indi*indi+indj*indj+p*p);
if(kk>0){
//Normalizacao pois a biblioteca fftw3 não tem dx nem global_dk nos integrais
map_vel_c[i*global_N_halo*(global_N_halo/2+1)+j*(global_N_halo/2+1)+p]=p*map_in_c[i*global_N_halo*(global_N_halo/2+1)+j*(global_N_halo/2+1)+p];
}else{
map_vel_c[i*global_N_halo*(global_N_halo/2+1)+j*(global_N_halo/2+1)+p]=0;
}
}
}
}
box_symmetriesf(map_vel_c,global_N_halo);
/* Executes FFT */
fftwf_execute(pc2r);
printf("\nWriting v_z field to file...\n");fflush(0);
sprintf(fname, "%s/Velocity/vel_z_z0_N%ld_L%d.dat", argv[1],global_N_halo,(int)(global_L));
if((fid=fopen(fname,"wb"))==NULL){
printf("\nThe file cannot be open\n");
return 0;
}
elem=fwrite(map,sizeof(float),global_N_halo*global_N_halo*global_N_halo,fid);
fclose(fid);
fftwf_free(map);
fftwf_free(map_in_c);
fftwf_free(map_vel_c);
fftwf_destroy_plan(pc2r);
fftwf_destroy_plan(pr2c);
exit(0);
}