void sparse_solve(int npml, int pad1, int pad2, sf_complex ***f, bool hermite, int ns, int uts)
/*< sparse solver>*/
{
SuiteSparse_long status;
int is, its;
/* loop over shots */
#ifdef _OPENMP
#pragma omp parallel for num_threads(uts) private(its)
#endif
for ( is=0; is < ns; is++ ) {
#ifdef _OPENMP
its = omp_get_thread_num();
#else
its = 0;
#endif
fdpad(npml, pad1, pad2, f[is], Bx[its], Bz[its]);
status = umfpack_zl_solve(hermite?UMFPACK_At: UMFPACK_A,
NULL, NULL, NULL, NULL,
Xx[its], Xz[its], Bx[its], Bz[its],
Numeric[its], Control, NULL);
fdcut(npml, pad1, pad2, f[is], Xx[its], Xz[its]);
}
}
int main(int argc, char* argv[])
{
bool verb, save, load;
int npml, pad1, pad2, n1, n2;
int ih, nh, is, ns, iw, nw, i, j;
SuiteSparse_long n, nz, *Ti, *Tj;
float d1, d2, **vel, ****image, ****timage, dw, ow;
double omega, *Tx, *Tz;
SuiteSparse_long *Ap, *Ai, *Map;
double *Ax, *Az, **Xx, **Xz, **Bx, **Bz;
void *Symbolic, **Numeric;
double Control[UMFPACK_CONTROL];
sf_complex ***srce, ***recv;
char *datapath, *insert, *append;
size_t srclen, inslen;
sf_file in, out, source, data, us, ur, timg;
int uts, its, mts;
sf_timer timer;
char *order;
sf_init(argc,argv);
in = sf_input("in");
out = sf_output("out");
if (!sf_getbool("verb",&verb)) verb=false;
/* verbosity flag */
if (verb)
timer = sf_timer_init();
else
timer = NULL;
if (!sf_getbool("save",&save)) save=false;
/* save LU */
if (!sf_getbool("load",&load)) load=false;
/* load LU */
if (save || load) {
datapath = sf_histstring(in,"in");
srclen = strlen(datapath);
insert = sf_charalloc(6);
} else {
datapath = NULL;
srclen = 0;
insert = NULL;
append = NULL;
}
if (!sf_getint("uts",&uts)) uts=0;
/* number of OMP threads */
#ifdef _OPENMP
mts = omp_get_max_threads();
#else
mts = 1;
#endif
uts = (uts < 1)? mts: uts;
if (verb) sf_warning("Using %d out of %d threads.",uts,mts);
if (!sf_getint("nh",&nh)) nh=0;
/* horizontal space-lag */
if (!sf_getint("npml",&npml)) npml=10;
/* PML width */
if (NULL == (order = sf_getstring("order"))) order="j";
/* discretization scheme (default optimal 9-point) */
fdprep_order(order);
/* read model */
if (!sf_histint(in,"n1",&n1)) sf_error("No n1= in input.");
if (!sf_histint(in,"n2",&n2)) sf_error("No n2= in input.");
if (!sf_histfloat(in,"d1",&d1)) sf_error("No d1= in input.");
if (!sf_histfloat(in,"d2",&d2)) sf_error("No d2= in input.");
vel = sf_floatalloc2(n1,n2);
sf_floatread(vel[0],n1*n2,in);
/* read source */
if (NULL == sf_getstring("source"))
sf_error("Need source=");
source = sf_input("source");
if (!sf_histint(source,"n3",&ns)) sf_error("No ns=.");
if (!sf_histint(source,"n4",&nw)) sf_error("No nw=.");
if (!sf_histfloat(source,"d4",&dw)) sf_error("No dw=.");
if (!sf_histfloat(source,"o4",&ow)) sf_error("No ow=.");
srce = sf_complexalloc3(n1,n2,ns);
/* read receiver */
if (NULL == sf_getstring("data"))
sf_error("Need data=");
data = sf_input("data");
recv = sf_complexalloc3(n1,n2,ns);
/* write output header */
sf_putint(out,"n3",2*nh+1);
sf_putfloat(out,"d3",d2);
sf_putfloat(out,"o3",(float) -nh*d2);
/* output source wavefield */
if (NULL != sf_getstring("us")) {
us = sf_output("us");
sf_settype(us,SF_COMPLEX);
sf_putint(us,"n3",ns);
sf_putstring(us,"label3","Shot");
sf_putstring(us,"unit3","");
sf_putint(us,"n4",nw);
sf_putfloat(us,"d4",dw);
sf_putfloat(us,"o4",ow);
sf_putstring(us,"label4","Frequency");
sf_putstring(us,"unit4","Hz");
} else {
us = NULL;
}
/* output receiver wavefield */
if (NULL != sf_getstring("ur")) {
ur = sf_output("ur");
sf_settype(ur,SF_COMPLEX);
sf_putint(ur,"n3",ns);
sf_putstring(ur,"label3","Shot");
sf_putstring(ur,"unit3","");
sf_putint(ur,"n4",nw);
sf_putfloat(ur,"d4",dw);
sf_putfloat(ur,"o4",ow);
sf_putstring(ur,"label4","Frequency");
sf_putstring(ur,"unit4","Hz");
} else {
ur = NULL;
}
/* output time-shift image derivative */
if (NULL != sf_getstring("timg")) {
timg = sf_output("timg");
sf_putint(timg,"n3",2*nh+1);
sf_putfloat(timg,"d3",d2);
sf_putfloat(timg,"o3",(float) -nh*d2);
timage = (float****) sf_alloc(uts,sizeof(float***));
for (its=0; its < uts; its++) {
timage[its] = sf_floatalloc3(n1,n2,2*nh+1);
}
} else {
timg = NULL;
timage = NULL;
}
/* allocate temporary memory */
if (load) {
Ti = NULL; Tj = NULL; Tx = NULL; Tz = NULL;
Ap = NULL; Ai = NULL; Map = NULL; Ax = NULL; Az = NULL;
}
Bx = (double**) sf_alloc(uts,sizeof(double*));
Bz = (double**) sf_alloc(uts,sizeof(double*));
Xx = (double**) sf_alloc(uts,sizeof(double*));
Xz = (double**) sf_alloc(uts,sizeof(double*));
image = (float****) sf_alloc(uts,sizeof(float***));
for (its=0; its < uts; its++) {
image[its] = sf_floatalloc3(n1,n2,2*nh+1);
}
Numeric = (void**) sf_alloc(uts,sizeof(void*));
/* LU control */
umfpack_zl_defaults (Control);
Control [UMFPACK_IRSTEP] = 0;
/* loop over frequency */
for (iw=0; iw < nw; iw++) {
omega = (double) 2.*SF_PI*(ow+iw*dw);
/* PML padding */
pad1 = n1+2*npml;
pad2 = n2+2*npml;
n = fdprep_n (pad1,pad2);
nz = fdprep_nz(pad1,pad2);
if (!load) {
Ti = (SuiteSparse_long*) sf_alloc(nz,sizeof(SuiteSparse_long));
Tj = (SuiteSparse_long*) sf_alloc(nz,sizeof(SuiteSparse_long));
Tx = (double*) sf_alloc(nz,sizeof(double));
Tz = (double*) sf_alloc(nz,sizeof(double));
Ap = (SuiteSparse_long*) sf_alloc(n+1,sizeof(SuiteSparse_long));
Ai = (SuiteSparse_long*) sf_alloc(nz,sizeof(SuiteSparse_long));
Map = (SuiteSparse_long*) sf_alloc(nz,sizeof(SuiteSparse_long));
Ax = (double*) sf_alloc(nz,sizeof(double));
Az = (double*) sf_alloc(nz,sizeof(double));
}
for (its=0; its < uts; its++) {
Bx[its] = (double*) sf_alloc(n,sizeof(double));
Bz[its] = (double*) sf_alloc(n,sizeof(double));
Xx[its] = (double*) sf_alloc(n,sizeof(double));
Xz[its] = (double*) sf_alloc(n,sizeof(double));
}
if (verb) {
sf_warning("Frequency %d of %d.",iw+1,nw);
sf_timer_start(timer);
}
/* LU file (append _lu* after velocity file) */
if (save || load) {
sprintf(insert,"_lu%d",iw);
inslen = strlen(insert);
append = malloc(srclen+inslen+1);
memcpy(append,datapath,srclen-5);
memcpy(append+srclen-5,insert,inslen);
memcpy(append+srclen-5+inslen,datapath+srclen-5,5+1);
}
if (!load) {
/* assemble matrix */
fdprep(omega,
n1, n2, d1, d2, vel,
npml, pad1, pad2,
Ti, Tj, Tx, Tz);
(void) umfpack_zl_triplet_to_col (n, n, nz,
Ti, Tj, Tx, Tz,
Ap, Ai, Ax, Az, Map);
/* LU */
(void) umfpack_zl_symbolic (n, n,
Ap, Ai, Ax, Az,
&Symbolic, Control, NULL);
(void) umfpack_zl_numeric (Ap, Ai, Ax, Az,
Symbolic, &Numeric[0],
Control, NULL);
/* save Numeric */
#ifdef _OPENMP
(void) umfpack_zl_save_numeric (Numeric[0], append);
for (its=1; its < uts; its++) {
(void) umfpack_zl_load_numeric (&Numeric[its], append);
}
if (!save) {
(void) remove (append);
(void) remove ("numeric.umf");
}
#else
if (save) (void) umfpack_zl_save_numeric (Numeric[0], append);
#endif
} else {
/* load Numeric */
for (its=0; its < uts; its++) {
(void) umfpack_zl_load_numeric (&Numeric[its], append);
}
}
if (save || load) free(append);
/* read source and data */
sf_complexread(srce[0][0],n1*n2*ns,source);
sf_complexread(recv[0][0],n1*n2*ns,data);
/* loop over shots */
#ifdef _OPENMP
#pragma omp parallel for num_threads(uts) private(its,ih,j,i)
#endif
for (is=0; is < ns; is++) {
#ifdef _OPENMP
its = omp_get_thread_num();
#else
its = 0;
#endif
/* source wavefield */
fdpad(npml,pad1,pad2, srce[is],Bx[its],Bz[its]);
(void) umfpack_zl_solve (UMFPACK_A,
NULL, NULL, NULL, NULL,
Xx[its], Xz[its], Bx[its], Bz[its],
Numeric[its], Control, NULL);
fdcut(npml,pad1,pad2, srce[is],Xx[its],Xz[its]);
/* receiver wavefield */
fdpad(npml,pad1,pad2, recv[is],Bx[its],Bz[its]);
(void) umfpack_zl_solve (UMFPACK_At,
NULL, NULL, NULL, NULL,
Xx[its], Xz[its], Bx[its], Bz[its],
Numeric[its], Control, NULL);
fdcut(npml,pad1,pad2, recv[is],Xx[its],Xz[its]);
/* imaging condition */
for (ih=-nh; ih < nh+1; ih++) {
for (j=0; j < n2; j++) {
for (i=0; i < n1; i++) {
if (j-abs(ih) >= 0 && j+abs(ih) < n2) {
image[its][ih+nh][j][i] += crealf(conjf(srce[is][j-ih][i])*recv[is][j+ih][i]);
if (timg != NULL) timage[its][ih+nh][j][i]
+= crealf(2.*I*omega*conjf(srce[is][j-ih][i])*recv[is][j+ih][i]);
}
}
}
}
}
if (verb) {
sf_timer_stop (timer);
sf_warning("Finished in %g seconds.",sf_timer_get_diff_time(timer)/1.e3);
}
if (!load) (void) umfpack_zl_free_symbolic (&Symbolic);
for (its=0; its < uts; its++) {
(void) umfpack_zl_free_numeric (&Numeric[its]);
}
if (!load) {
free(Ti); free(Tj); free(Tx); free(Tz);
free(Ap); free(Ai); free(Map);
free(Ax); free(Az);
}
for (its=0; its < uts; its++) {
free(Bx[its]); free(Bz[its]); free(Xx[its]); free(Xz[its]);
}
if (us != NULL) sf_complexwrite(srce[0][0],n1*n2*ns,us);
if (ur != NULL) sf_complexwrite(recv[0][0],n1*n2*ns,ur);
}
#ifdef _OPENMP
#pragma omp parallel for num_threads(uts) private(j,i,its)
for (ih=-nh; ih < nh+1; ih++) {
for (j=0; j < n2; j++) {
for (i=0; i < n1; i++) {
for (its=1; its < uts; its++) {
image[0][ih+nh][j][i] += image[its][ih+nh][j][i];
if (timg != NULL) timage[0][ih+nh][j][i]
+= timage[its][ih+nh][j][i];
}
}
}
}
#endif
sf_floatwrite(image[0][0][0],n1*n2*(2*nh+1),out);
if (timg != NULL) sf_floatwrite(timage[0][0][0],n1*n2*(2*nh+1),timg);
exit(0);
}
void iwigrad_oper(bool adj, bool add, int nx, int nr, float *x, float *r)
/*< linear operator >*/
{
int iw, is, its, i;
int pad1, pad2;
SuiteSparse_long n, nz;
double omega;
sf_adjnull(adj,add,nx,nr,x,r);
/* PML padding */
pad1 = n1+2*npml;
pad2 = n2+2*npml;
n = fdprep_n (pad1,pad2);
nz = fdprep_nz(pad1,pad2);
/* loop over frequency */
for (iw=0; iw < nw; iw++) {
omega = (double) 2.*SF_PI*(ow+iw*dw);
if (!load) {
Ti = (SuiteSparse_long*) sf_alloc(nz,sizeof(SuiteSparse_long));
Tj = (SuiteSparse_long*) sf_alloc(nz,sizeof(SuiteSparse_long));
Tx = (double*) sf_alloc(nz,sizeof(double));
Tz = (double*) sf_alloc(nz,sizeof(double));
Ap = (SuiteSparse_long*) sf_alloc(n+1,sizeof(SuiteSparse_long));
Ai = (SuiteSparse_long*) sf_alloc(nz,sizeof(SuiteSparse_long));
Map = (SuiteSparse_long*) sf_alloc(nz,sizeof(SuiteSparse_long));
Ax = (double*) sf_alloc(nz,sizeof(double));
Az = (double*) sf_alloc(nz,sizeof(double));
}
for (its=0; its < uts; its++) {
Bx[its] = (double*) sf_alloc(n,sizeof(double));
Bz[its] = (double*) sf_alloc(n,sizeof(double));
Xx[its] = (double*) sf_alloc(n,sizeof(double));
Xz[its] = (double*) sf_alloc(n,sizeof(double));
}
/* LU file (append _inv* after velocity file) */
if (load) {
sprintf(insert,"_inv%d",iw);
inslen = strlen(insert);
append = malloc(srclen+inslen+1);
memcpy(append,datapath,srclen-5);
memcpy(append+srclen-5,insert,inslen);
memcpy(append+srclen-5+inslen,datapath+srclen-5,5+1);
}
if (!load) {
/* assemble matrix */
fdprep(omega,
n1, n2, d1, d2, vel,
npml, pad1, pad2,
Ti, Tj, Tx, Tz);
(void) umfpack_zl_triplet_to_col (n, n, nz,
Ti, Tj, Tx, Tz,
Ap, Ai, Ax, Az, Map);
/* LU */
(void) umfpack_zl_symbolic (n, n,
Ap, Ai, Ax, Az,
&Symbolic, Control, NULL);
(void) umfpack_zl_numeric (Ap, Ai, Ax, Az,
Symbolic, &Numeric[0],
Control, NULL);
#ifdef _OPENMP
(void) umfpack_zl_save_numeric (Numeric[0], append);
for (its=1; its < uts; its++) {
(void) umfpack_zl_load_numeric (&Numeric[its], append);
}
(void) remove (append);
(void) remove ("numeric.umf");
#endif
} else {
/* load Numeric */
for (its=0; its < uts; its++) {
(void) umfpack_zl_load_numeric (&Numeric[its], append);
}
}
if (load) free(append);
/* read wavefields from temporary file */
sf_fslice_get(sfile,iw,us[0][0]);
sf_fslice_get(rfile,iw,ur[0][0]);
/* loop over shots */
#ifdef _OPENMP
#pragma omp parallel for num_threads(uts) private(its)
#endif
for (is=0; is < ns; is++) {
#ifdef _OPENMP
its = omp_get_thread_num();
#else
its = 0;
#endif
/* adjoint source */
adjsrce(ur[is],as[is], x,r,adj);
fdpad(npml,pad1,pad2, as[is],Bx[its],Bz[its]);
(void) umfpack_zl_solve (UMFPACK_At,
NULL, NULL, NULL, NULL,
Xx[its], Xz[its], Bx[its], Bz[its],
Numeric[its], Control, NULL);
fdcut(npml,pad1,pad2, as[is],Xx[its],Xz[its]);
/* adjoint receiver */
adjrecv(us[is],ar[is], x,r,adj);
fdpad(npml,pad1,pad2, ar[is],Bx[its],Bz[its]);
(void) umfpack_zl_solve (UMFPACK_A,
NULL, NULL, NULL, NULL,
Xx[its], Xz[its], Bx[its], Bz[its],
Numeric[its], Control, NULL);
fdcut(npml,pad1,pad2, ar[is],Xx[its],Xz[its]);
/* assemble */
iwiadd(omega, us[is],ur[is],as[is],ar[is], tempx[its],tempr[its],adj);
/* clean up */
if (adj) adjclean(as[is],ar[is]);
}
if (!load) (void) umfpack_zl_free_symbolic (&Symbolic);
for (its=0; its < uts; its++) {
(void) umfpack_zl_free_numeric (&Numeric[its]);
}
if (!load) {
free(Ti);
free(Tj);
free(Tx);
free(Tz);
free(Ap);
free(Ai);
free(Map);
free(Ax);
free(Az);
}
for (its=0; its < uts; its++) {
free(Bx[its]);
free(Bz[its]);
free(Xx[its]);
free(Xz[its]);
}
}
#ifdef _OPENMP
if (adj) {
#pragma omp parallel for num_threads(uts) private(its)
for (i=0; i < n1*n2; i++) {
for (its=0; its < uts; its++) {
x[i] += tempx[its][i];
tempx[its][i] = 0.;
}
}
} else {
#pragma omp parallel for num_threads(uts) private(its)
for (i=0; i < n1*n2*(2*nh+1); i++) {
for (its=0; its < uts; its++) {
r[i] += tempr[its][i];
tempr[its][i] = 0.;
}
}
}
#else
if (adj) {
for (i=0; i < n1*n2; i++) {
x[i] = tempx[0][i];
tempx[0][i] = 0.;
}
} else {
for (i=0; i < n1*n2*(2*nh+1); i++) {
r[i] = tempr[0][i];
tempr[0][i] = 0.;
}
}
#endif
}
