void F77_FUNC_(rfftwnd_f77_real_to_complex,RFFTWND_F77_REAL_TO_COMPLEX)
(fftwnd_plan *p, int *howmany, fftw_real *in, int *istride, int *idist,
fftw_complex *out, int *ostride, int *odist)
{
rfftwnd_real_to_complex(*p,*howmany,in,*istride,*idist,
out,*ostride,*odist);
}
/*
* Class: jfftw_real_nd_Plan
* Method: transform
* Signature: (I[DII[DII)V
*/
JNIEXPORT void JNICALL Java_jfftw_real_nd_Plan_transform__I_3DII_3DII( JNIEnv *env, jobject obj, jint howmany, jdoubleArray in, jint istride, jint idist, jdoubleArray out, jint ostride, jint odist )
{
jdouble *cin, *cout;
int i;
jclass clazz = (*env)->GetObjectClass( env, obj );
jfieldID id = (*env)->GetFieldID( env, clazz, "plan", "[B" );
jbyteArray arr = (jbyteArray)(*env)->GetObjectField( env, obj, id );
unsigned char* carr = (*env)->GetByteArrayElements( env, arr, 0 );
rfftwnd_plan plan = *(rfftwnd_plan*)carr;
int length = 1;
int clength = 2;
for( i = 0; i < plan->rank; ++i ) length *= plan->plans[i]->n;
for( i = 0; i < plan->rank; ++i ) clength *= plan->n[i];
if( (howmany - 1) * idist + (plan->dir==FFTW_REAL_TO_COMPLEX?length:clength) != (*env)->GetArrayLength( env, in ) )
if( (plan->dir==FFTW_REAL_TO_COMPLEX?length:clength) != (*env)->GetArrayLength( env, in ) )
{
(*env)->ThrowNew( env, (*env)->FindClass( env, "java/lang/IndexOutOfBoundsException" ), "the Plan was created for a different length (in)" );
(*env)->ReleaseByteArrayElements( env, arr, carr, 0 );
return;
}
if( (howmany - 1) * odist + (plan->dir==FFTW_REAL_TO_COMPLEX?clength:length) != (*env)->GetArrayLength( env, out ) )
{
(*env)->ThrowNew( env, (*env)->FindClass( env, "java/lang/IndexOutOfBoundsException" ), "the Plan was created for a different length (out)" );
(*env)->ReleaseByteArrayElements( env, arr, carr, 0 );
return;
}
cin = (*env)->GetDoubleArrayElements( env, in, 0 );
cout = (*env)->GetDoubleArrayElements( env, out, 0 );
if( plan->rank > 0 && ! plan->plans[0]->flags & FFTW_THREADSAFE )
{
// synchronization
(*env)->MonitorEnter( env, obj );
}
if( plan->dir == FFTW_REAL_TO_COMPLEX )
{
rfftwnd_real_to_complex( plan, howmany, cin, istride, idist, (fftw_complex*)cout, ostride, odist );
}
else
{
rfftwnd_complex_to_real( plan, howmany, (fftw_complex*)cin, istride, idist, cout, ostride, odist );
}
if( plan->rank > 0 && ! plan->plans[0]->flags & FFTW_THREADSAFE )
{
// synchronization
(*env)->MonitorExit( env, obj );
}
(*env)->ReleaseByteArrayElements( env, arr, carr, 0 );
(*env)->ReleaseDoubleArrayElements( env, in, cin, 0 );
(*env)->ReleaseDoubleArrayElements( env, out, cout, 0 );
}
static void other_dims_aux(rfftwnd_mpi_plan p,
int n_fields, fftw_real *local_data)
{
int local_nx = p->p_transpose->local_nx;
int n_after_x = p->p_fft->n[0] * p->p_fft->n_after[0];
if (n_fields > 1) {
rfftwnd_plan p_fft = p->p_fft;
int fft_iter;
if (p_fft->dir == FFTW_REAL_TO_COMPLEX)
for (fft_iter = 0; fft_iter < local_nx; ++fft_iter)
rfftwnd_real_to_complex(p_fft, n_fields,
local_data
+ (2 * n_after_x * n_fields) * fft_iter,
n_fields, 1,
NULL, 0, 0);
else
for (fft_iter = 0; fft_iter < local_nx; ++fft_iter)
rfftwnd_complex_to_real(p_fft, n_fields,
((fftw_complex *) local_data)
+ (n_after_x * n_fields) * fft_iter,
n_fields, 1,
NULL, 0, 0);
}
else {
if (p->p_fft->dir == FFTW_REAL_TO_COMPLEX)
rfftwnd_real_to_complex(p->p_fft, local_nx,
local_data, 1, 2*n_after_x,
NULL, 0, 0);
else
rfftwnd_complex_to_real(p->p_fft, local_nx,
(fftw_complex *) local_data,
1, n_after_x,
NULL, 0, 0);
}
}
int increBoundary(void)
{
/* External Variables */
extern int Nx, Nz;
extern fftw_complex ***CT; /* 6-by-(3Nz/2)-by-(3*Nx/4+1) */
extern mcomplex **Uxb, **Uzb;
extern fftw_plan pf1, pf2;
extern rfftwnd_plan pr1, pr2;
extern double *Kx, *Kz;
int x, i, z, idx;
double norm, tmp1, tmp2, tmp3;
fftw_real *RT; /* real to complex transform */
fftw_complex *fout = NULL;
fftw_real *rout = NULL;
idx = (3 * Nz / 2) * (3 * Nx / 2 + 2);
RT = (fftw_real *) CT[0][0];
norm = 1.0 / ((3. * Nx / 2.) * (3. * Nz / 2.));
memset(CT[0][0], 0,
MAXT * (3 * Nz / 2) * (3 * Nx / 4 + 1) * sizeof(fftw_complex));
/* store Uxb hat and Uzb hat and w hat on CT for inverse FFT */
for (z = 0; z < Nz / 2; ++z) {
/* CT[0] store the data of Uxb, CT[1] storedata for Uzb */
memcpy(CT[0][z], Uxb[z], (Nx / 2) * sizeof(fftw_complex));
memcpy(CT[1][z], Uzb[z], (Nx / 2) * sizeof(fftw_complex));
/* for(x=0; x<Nx/2; ++x)
{
Re(CT[2][z][x])=1.0;
Im(CT[2][z][x])=0.;
} */
}
for (z = Nz / 2 + 1; z < Nz; ++z) {
memcpy(CT[0][z + Nz / 2], Uxb[z], (Nx / 2) * sizeof(fftw_complex));
memcpy(CT[1][z + Nz / 2], Uzb[z], (Nx / 2) * sizeof(fftw_complex));
/*for(x=0; x<Nx/2; ++x)
{
Re(CT[2][z+Nz/2][x])=1.0;
Im(CT[2][z+Nz/2][x])=0.;
}
*/
}
Re(CT[2][1][1]) = 1.;
//Re(CT[2][3*Nz/2-1][0])=1.;
//Re(CT[2][0][0])=1.;
// Re(CT[2][0][0])=1.;
/* inverse Fourier transform */
for (i = 0; i < 3; ++i) {
/* Each column of CT[i] */
fftw(pf1, Nx / 2, CT[i][0], 3 * Nx / 4 + 1, 1, fout, -1, -1);
/* Each row of CT[i] */
rfftwnd_complex_to_real(pr1, 3 * Nz / 2, CT[i][0], 1,
3 * Nx / 4 + 1, rout, -1, -1);
}
/* compute (dux)*(w.n) and (duz)*(w.n) */
for (z = 0; z < (3 * Nz / 2); ++z) {
for (x = 0; x < 3 * Nx / 2; ++x) {
RT[(z * (3 * Nx / 2 + 2) + x)] =
RT[(z * (3 * Nx / 2 + 2) + x)] * RT[2 * idx +
(z * (3 * Nx / 2 + 2) +
x)];
RT[idx + (z * (3 * Nx / 2 + 2) + x)] =
RT[idx + (z * (3 * Nx / 2 + 2) + x)] * RT[2 * idx +
(z *
(3 * Nx / 2 +
2) + x)];
}
}
/* Fourier transform to get Uxb hats and Uzb hats. */
for (i = 0; i < 3; ++i) {
/* Each row of RT[i] */
rfftwnd_real_to_complex(pr2, 3 * Nz / 2, RT + (i * idx), 1,
3 * Nx / 2 + 2, fout, -1, -1);
/* Each column of CT[i] */
fftw(pf2, Nx / 2, CT[i][0], 3 * Nx / 4 + 1, 1, fout, -1, -1);
/* constant of FFT */
for (z = 0; z < Nz / 2; ++z) {
for (x = 0; x < Nx / 2; ++x) {
Re(CT[i][z][x]) = norm * Re(CT[i][z][x]);
Im(CT[i][z][x]) = norm * Im(CT[i][z][x]);
}
}
for (z = Nz + 1; z < 3 * Nz / 2; ++z) {
for (x = 0; x < Nx / 2; ++x) {
Re(CT[i][z][x]) = norm * Re(CT[i][z][x]);
Im(CT[i][z][x]) = norm * Im(CT[i][z][x]);
}
}
}
/*put date back in array Uxb and Uzb */
memset(Uxb[0], 0, Nz * (Nx / 2) * sizeof(mcomplex));
memset(Uzb[0], 0, Nz * (Nx / 2) * sizeof(mcomplex));
for (z = 0; z < Nz / 2; ++z) {
memcpy(Uxb[z], CT[0][z], Nx / 2 * sizeof(fftw_complex));
memcpy(Uzb[z], CT[1][z], Nx / 2 * sizeof(fftw_complex));
}
for (z = Nz + 1; z < 3 * Nz / 2; ++z) {
memcpy(Uxb[z - Nz / 2], CT[0][z], Nx / 2 * sizeof(fftw_complex));
memcpy(Uzb[z - Nz / 2], CT[1][z], Nx / 2 * sizeof(fftw_complex));
}
/* further computation to get c1, c2, c3, c4 as in the note and results are rewritten in
Uxb and Uzb:
c1=g hat=iKz*Uxb-iKx*Uzb;-------rewrittin in Uxb
c2=du_y=-iKx*Uxb-iKz*Uzb;-------rewrittin in Uzb
c3=U=Uxb(0,0); -------rewritten in Uxb[0][0]
c4=Uzb(0,0) -------rewritten in Uzb[0][0] */
for (z = 0; z < Nz; ++z) {
for (x = 0; x < Nx / 2; ++x) {
if (z * z + x * x > 0) {
tmp1 = -Kz[z] * Im(Uxb[z][x]) + Kx[x] * Im(Uzb[z][x]);
tmp2 = Kz[z] * Re(Uxb[z][x]) - Kx[x] * Re(Uzb[z][x]);
tmp3 = Kx[x] * Im(Uxb[z][x]) + Kz[z] * Im(Uzb[z][x]);
Im(Uzb[z][x]) =
-Kx[x] * Re(Uxb[z][x]) - Kz[z] * Re(Uzb[z][x]);
Re(Uzb[z][x]) = tmp3;
Re(Uxb[z][x]) = tmp1;
Im(Uxb[z][x]) = tmp2;
}
}
}
return (NO_ERR);
}
void rfftwnd_one_real_to_complex(fftwnd_plan p,
fftw_real *in, fftw_complex *out)
{
rfftwnd_real_to_complex(p, 1, in, 1, 1, out, 1, 1);
}
void test_speed_nd_aux(struct size sz,
fftw_direction dir, int flags, int specific)
{
fftw_real *in;
fftwnd_plan plan;
double t;
fftw_time begin, end;
int i, N;
/* only bench in-place multi-dim transforms */
flags |= FFTW_IN_PLACE;
N = 1;
for (i = 0; i < sz.rank - 1; ++i)
N *= sz.narray[i];
N *= (sz.narray[i] + 2);
in = (fftw_real *) fftw_malloc(N * howmany_fields * sizeof(fftw_real));
if (specific) {
begin = fftw_get_time();
plan = rfftwnd_create_plan_specific(sz.rank, sz.narray, dir,
speed_flag | flags
| wisdom_flag | no_vector_flag,
in, howmany_fields, 0, 1);
} else {
begin = fftw_get_time();
plan = rfftwnd_create_plan(sz.rank, sz.narray,
dir, speed_flag | flags
| wisdom_flag | no_vector_flag);
}
end = fftw_get_time();
CHECK(plan != NULL, "can't create plan");
t = fftw_time_to_sec(fftw_time_diff(end, begin));
WHEN_VERBOSE(2, printf("time for planner: %f s\n", t));
WHEN_VERBOSE(2, printf("\n"));
WHEN_VERBOSE(2, (rfftwnd_print_plan(plan)));
WHEN_VERBOSE(2, printf("\n"));
if (dir == FFTW_REAL_TO_COMPLEX) {
FFTW_TIME_FFT(rfftwnd_real_to_complex(plan, howmany_fields,
in, howmany_fields, 1,
0, 0, 0),
in, N * howmany_fields, t);
} else {
FFTW_TIME_FFT(rfftwnd_complex_to_real(plan, howmany_fields,
(fftw_complex *) in,
howmany_fields, 1,
0, 0, 0),
in, N * howmany_fields, t);
}
rfftwnd_destroy_plan(plan);
WHEN_VERBOSE(1, printf("time for one fft: %s", smart_sprint_time(t)));
WHEN_VERBOSE(1, printf(" (%s/point)\n", smart_sprint_time(t / N)));
WHEN_VERBOSE(1, printf("\"mflops\" = 5/2 (N log2 N) / (t in microseconds)"
" = %f\n", 0.5 * howmany_fields * mflops(t, N)));
fftw_free(in);
WHEN_VERBOSE(1, printf("\n"));
}
void testnd_in_place(int rank, int *n, fftwnd_plan validated_plan,
int alternate_api, int specific)
{
int istride, ostride, howmany;
int N, dim, i, j, k;
int nc, nhc, nr;
fftw_real *in1, *out3;
fftw_complex *in2, *out1, *out2;
fftwnd_plan p, ip;
int flags = measure_flag | wisdom_flag | FFTW_IN_PLACE;
if (coinflip())
flags |= FFTW_THREADSAFE;
N = nc = nr = nhc = 1;
for (dim = 0; dim < rank; ++dim)
N *= n[dim];
if (rank > 0) {
nr = n[rank - 1];
nc = N / nr;
nhc = nr / 2 + 1;
}
in1 = (fftw_real *) fftw_malloc(2 * nhc * nc * MAX_STRIDE * sizeof(fftw_real));
out3 = in1;
out1 = (fftw_complex *) in1;
in2 = (fftw_complex *) fftw_malloc(N * sizeof(fftw_complex));
out2 = (fftw_complex *) fftw_malloc(N * sizeof(fftw_complex));
if (alternate_api && specific && (rank == 2 || rank == 3)) {
if (rank == 2) {
p = rfftw2d_create_plan_specific(n[0], n[1],
FFTW_REAL_TO_COMPLEX, flags,
in1, MAX_STRIDE, 0, 0);
ip = rfftw2d_create_plan_specific(n[0], n[1],
FFTW_COMPLEX_TO_REAL, flags,
in1, MAX_STRIDE, 0, 0);
} else {
p = rfftw3d_create_plan_specific(n[0], n[1], n[2],
FFTW_REAL_TO_COMPLEX, flags,
in1, MAX_STRIDE, 0, 0);
ip = rfftw3d_create_plan_specific(n[0], n[1], n[2],
FFTW_COMPLEX_TO_REAL, flags,
in1, MAX_STRIDE, 0, 0);
}
} else if (specific) {
p = rfftwnd_create_plan_specific(rank, n, FFTW_REAL_TO_COMPLEX,
flags,
in1, MAX_STRIDE, in1, MAX_STRIDE);
ip = rfftwnd_create_plan_specific(rank, n, FFTW_COMPLEX_TO_REAL,
flags,
in1, MAX_STRIDE, in1, MAX_STRIDE);
} else if (alternate_api && (rank == 2 || rank == 3)) {
if (rank == 2) {
p = rfftw2d_create_plan(n[0], n[1], FFTW_REAL_TO_COMPLEX,
flags);
ip = rfftw2d_create_plan(n[0], n[1], FFTW_COMPLEX_TO_REAL,
flags);
} else {
p = rfftw3d_create_plan(n[0], n[1], n[2], FFTW_REAL_TO_COMPLEX,
flags);
ip = rfftw3d_create_plan(n[0], n[1], n[2], FFTW_COMPLEX_TO_REAL,
flags);
}
} else {
p = rfftwnd_create_plan(rank, n, FFTW_REAL_TO_COMPLEX, flags);
ip = rfftwnd_create_plan(rank, n, FFTW_COMPLEX_TO_REAL, flags);
}
CHECK(p != NULL && ip != NULL, "can't create plan");
for (i = 0; i < nc * nhc * 2 * MAX_STRIDE; ++i)
out3[i] = 0;
for (istride = 1; istride <= MAX_STRIDE; ++istride) {
/* generate random inputs */
for (i = 0; i < nc; ++i)
for (j = 0; j < nr; ++j) {
c_re(in2[i * nr + j]) = DRAND();
c_im(in2[i * nr + j]) = 0.0;
for (k = 0; k < istride; ++k)
in1[(i * nhc * 2 + j) * istride + k]
= c_re(in2[i * nr + j]);
}
fftwnd(validated_plan, 1, in2, 1, 1, out2, 1, 1);
howmany = ostride = istride;
WHEN_VERBOSE(2, printf("\n testing in-place stride %d...",
istride));
if (howmany != 1 || istride != 1 || ostride != 1 || coinflip())
rfftwnd_real_to_complex(p, howmany, in1, istride, 1,
out1, ostride, 1);
else
rfftwnd_one_real_to_complex(p, in1, NULL);
for (i = 0; i < nc; ++i)
for (k = 0; k < howmany; ++k)
CHECK(compute_error_complex(out1 + i * nhc * ostride + k,
ostride,
out2 + i * nr, 1,
nhc) < TOLERANCE,
"in-place (r2c): wrong answer");
if (howmany != 1 || istride != 1 || ostride != 1 || coinflip())
rfftwnd_complex_to_real(ip, howmany, out1, ostride, 1,
out3, istride, 1);
else
rfftwnd_one_complex_to_real(ip, out1, NULL);
for (i = 0; i < nc * nhc * 2 * istride; ++i)
out3[i] *= 1.0 / N;
for (i = 0; i < nc; ++i)
for (k = 0; k < howmany; ++k)
CHECK(compute_error(out3 + i * nhc * 2 * istride + k,
istride,
(fftw_real *) (in2 + i * nr), 2,
nr) < TOLERANCE,
"in-place (c2r): wrong answer (check 2)");
}
rfftwnd_destroy_plan(p);
rfftwnd_destroy_plan(ip);
fftw_free(out2);
fftw_free(in2);
fftw_free(in1);
}
void testnd_out_of_place(int rank, int *n, fftwnd_plan validated_plan)
{
int istride, ostride;
int N, dim, i, j, k;
int nc, nhc, nr;
fftw_real *in1, *out3;
fftw_complex *in2, *out1, *out2;
fftwnd_plan p, ip;
int flags = measure_flag | wisdom_flag;
if (coinflip())
flags |= FFTW_THREADSAFE;
N = nc = nr = nhc = 1;
for (dim = 0; dim < rank; ++dim)
N *= n[dim];
if (rank > 0) {
nr = n[rank - 1];
nc = N / nr;
nhc = nr / 2 + 1;
}
in1 = (fftw_real *) fftw_malloc(N * MAX_STRIDE * sizeof(fftw_real));
out3 = (fftw_real *) fftw_malloc(N * MAX_STRIDE * sizeof(fftw_real));
out1 = (fftw_complex *) fftw_malloc(nhc * nc * MAX_STRIDE
* sizeof(fftw_complex));
in2 = (fftw_complex *) fftw_malloc(N * sizeof(fftw_complex));
out2 = (fftw_complex *) fftw_malloc(N * sizeof(fftw_complex));
p = rfftwnd_create_plan(rank, n, FFTW_REAL_TO_COMPLEX, flags);
ip = rfftwnd_create_plan(rank, n, FFTW_COMPLEX_TO_REAL, flags);
CHECK(p != NULL && ip != NULL, "can't create plan");
for (istride = 1; istride <= MAX_STRIDE; ++istride) {
/* generate random inputs */
for (i = 0; i < nc; ++i)
for (j = 0; j < nr; ++j) {
c_re(in2[i * nr + j]) = DRAND();
c_im(in2[i * nr + j]) = 0.0;
for (k = 0; k < istride; ++k)
in1[(i * nr + j) * istride + k]
= c_re(in2[i * nr + j]);
}
for (i = 0; i < N * istride; ++i)
out3[i] = 0.0;
fftwnd(validated_plan, 1, in2, 1, 1, out2, 1, 1);
for (ostride = 1; ostride <= MAX_STRIDE; ++ostride) {
int howmany = (istride < ostride) ? istride : ostride;
WHEN_VERBOSE(2, printf("\n testing stride %d/%d...",
istride, ostride));
if (howmany != 1 || istride != 1 || ostride != 1 || coinflip())
rfftwnd_real_to_complex(p, howmany, in1, istride, 1,
out1, ostride, 1);
else
rfftwnd_one_real_to_complex(p, in1, out1);
for (i = 0; i < nc; ++i)
for (k = 0; k < howmany; ++k)
CHECK(compute_error_complex(out1 + i * nhc * ostride + k,
ostride,
out2 + i * nr, 1,
nhc) < TOLERANCE,
"out-of-place (r2c): wrong answer");
if (howmany != 1 || istride != 1 || ostride != 1 || coinflip())
rfftwnd_complex_to_real(ip, howmany, out1, ostride, 1,
out3, istride, 1);
else
rfftwnd_one_complex_to_real(ip, out1, out3);
for (i = 0; i < N * istride; ++i)
out3[i] *= 1.0 / N;
if (istride == howmany)
CHECK(compute_error(out3, 1, in1, 1, N * istride)
< TOLERANCE, "out-of-place (c2r): wrong answer");
for (i = 0; i < nc; ++i)
for (k = 0; k < howmany; ++k)
CHECK(compute_error(out3 + i * nr * istride + k,
istride,
(fftw_real *) (in2 + i * nr), 2,
nr) < TOLERANCE,
"out-of-place (c2r): wrong answer (check 2)");
}
}
rfftwnd_destroy_plan(p);
rfftwnd_destroy_plan(ip);
fftw_free(out3);
fftw_free(out2);
fftw_free(in2);
fftw_free(out1);
fftw_free(in1);
}
