diff options
Diffstat (limited to 'src')
| -rw-r--r-- | src/Makefile.am | 27 | ||||
| -rw-r--r-- | src/fft.h | 30 | ||||
| -rw-r--r-- | src/fftsg.c | 3492 | ||||
| -rw-r--r-- | src/fftsg.h | 21 | ||||
| -rw-r--r-- | src/helper.c | 2 | ||||
| -rw-r--r-- | src/init.c | 321 | ||||
| -rw-r--r-- | src/scalar.c | 2 | ||||
| -rw-r--r-- | src/vector.c | 136 | ||||
| -rw-r--r-- | src/window.c | 16 | ||||
| -rw-r--r-- | src/xtract_globals_private.h | 14 |
10 files changed, 405 insertions, 3656 deletions
diff --git a/src/Makefile.am b/src/Makefile.am index d768d2d..93b9eac 100644 --- a/src/Makefile.am +++ b/src/Makefile.am @@ -1,11 +1,28 @@ MAINTAINERCLEANFILES = Makefile.in -SOURCES = libxtract.c descriptors.c scalar.c vector.c delta.c init.c\ - window.c fini.c helper.c fftsg.c +if BUILD_OOURA +OOURA = ooura/fftsg.c +else +OOURA = +ACCELERATE_FLAGS = -framework Accelerate +endif + +SOURCES = libxtract.c \ + descriptors.c \ + scalar.c \ + vector.c \ + delta.c \ + init.c \ + window.c \ + fini.c \ + helper.c \ + $(OOURA) lib_LTLIBRARIES = libxtract.la -libxtract_la_CFLAGS = $(FFT_DEFINE) +libxtract_la_CFLAGS = libxtract_la_SOURCES = $(SOURCES) -libxtract_la_LDFLAGS = -export-dynamic +libxtract_la_LDFLAGS = -export-dynamic $(ACCELERATE_FLAGS) -EXTRA_DIST = xtract_globals_private.h xtract_macros_private.h xtract_window_private.h +EXTRA_DIST = xtract_globals_private.h \ + xtract_macros_private.h \ + xtract_window_private.h @@ -21,4 +21,34 @@ * */ +#ifndef FFT_H +#define FFT_H + +#ifdef HAVE_CONFIG_H +#include <config.h> +#endif + +#ifdef USE_OOURA #include "ooura/fftsg.h" +#else +#include <Accelerate/Accelerate.h> +#endif + +#ifdef USE_OOURA +typedef struct xtract_ooura_data_ +{ + int *ooura_ip; + double *ooura_w; + bool initialised; +} xtract_ooura_data; +#else +typedef struct xtract_vdsp_data_ +{ + FFTSetupD setup; + DSPDoubleSplitComplex fft; + vDSP_Length log2N; + bool initialised; +} xtract_vdsp_data; +#endif + +#endif /* Header guard */ diff --git a/src/fftsg.c b/src/fftsg.c deleted file mode 100644 index 0f3092b..0000000 --- a/src/fftsg.c +++ /dev/null @@ -1,3492 +0,0 @@ -/* -Fast Fourier/Cosine/Sine Transform - dimension :one - data length :power of 2 - decimation :frequency - radix :split-radix - data :inplace - table :use -functions - cdft: Complex Discrete Fourier Transform - rdft: Real Discrete Fourier Transform - ddct: Discrete Cosine Transform - ddst: Discrete Sine Transform - dfct: Cosine Transform of RDFT (Real Symmetric DFT) - dfst: Sine Transform of RDFT (Real Anti-symmetric DFT) -function prototypes - void cdft(int, int, double *, int *, double *); - void rdft(int, int, double *, int *, double *); - void ddct(int, int, double *, int *, double *); - void ddst(int, int, double *, int *, double *); - void dfct(int, double *, double *, int *, double *); - void dfst(int, double *, double *, int *, double *); -macro definitions - USE_CDFT_PTHREADS : default=not defined - CDFT_THREADS_BEGIN_N : must be >= 512, default=8192 - CDFT_4THREADS_BEGIN_N : must be >= 512, default=65536 - USE_CDFT_WINTHREADS : default=not defined - CDFT_THREADS_BEGIN_N : must be >= 512, default=32768 - CDFT_4THREADS_BEGIN_N : must be >= 512, default=524288 - - --------- Complex DFT (Discrete Fourier Transform) -------- - [definition] - <case1> - X[k] = sum_j=0^n-1 x[j]*exp(2*pi*i*j*k/n), 0<=k<n - <case2> - X[k] = sum_j=0^n-1 x[j]*exp(-2*pi*i*j*k/n), 0<=k<n - (notes: sum_j=0^n-1 is a summation from j=0 to n-1) - [usage] - <case1> - ip[0] = 0; // first time only - cdft(2*n, 1, a, ip, w); - <case2> - ip[0] = 0; // first time only - cdft(2*n, -1, a, ip, w); - [parameters] - 2*n :data length (int) - n >= 1, n = power of 2 - a[0...2*n-1] :input/output data (double *) - input data - a[2*j] = Re(x[j]), - a[2*j+1] = Im(x[j]), 0<=j<n - output data - a[2*k] = Re(X[k]), - a[2*k+1] = Im(X[k]), 0<=k<n - ip[0...*] :work area for bit reversal (int *) - length of ip >= 2+sqrt(n) - strictly, - length of ip >= - 2+(1<<(int)(log(n+0.5)/log(2))/2). - ip[0],ip[1] are pointers of the cos/sin table. - w[0...n/2-1] :cos/sin table (double *) - w[],ip[] are initialized if ip[0] == 0. - [remark] - Inverse of - cdft(2*n, -1, a, ip, w); - is - cdft(2*n, 1, a, ip, w); - for (j = 0; j <= 2 * n - 1; j++) { - a[j] *= 1.0 / n; - } - . - - --------- Real DFT / Inverse of Real DFT -------- - [definition] - <case1> RDFT - R[k] = sum_j=0^n-1 a[j]*cos(2*pi*j*k/n), 0<=k<=n/2 - I[k] = sum_j=0^n-1 a[j]*sin(2*pi*j*k/n), 0<k<n/2 - <case2> IRDFT (excluding scale) - a[k] = (R[0] + R[n/2]*cos(pi*k))/2 + - sum_j=1^n/2-1 R[j]*cos(2*pi*j*k/n) + - sum_j=1^n/2-1 I[j]*sin(2*pi*j*k/n), 0<=k<n - [usage] - <case1> - ip[0] = 0; // first time only - rdft(n, 1, a, ip, w); - <case2> - ip[0] = 0; // first time only - rdft(n, -1, a, ip, w); - [parameters] - n :data length (int) - n >= 2, n = power of 2 - a[0...n-1] :input/output data (double *) - <case1> - output data - a[2*k] = R[k], 0<=k<n/2 - a[2*k+1] = I[k], 0<k<n/2 - a[1] = R[n/2] - <case2> - input data - a[2*j] = R[j], 0<=j<n/2 - a[2*j+1] = I[j], 0<j<n/2 - a[1] = R[n/2] - ip[0...*] :work area for bit reversal (int *) - length of ip >= 2+sqrt(n/2) - strictly, - length of ip >= - 2+(1<<(int)(log(n/2+0.5)/log(2))/2). - ip[0],ip[1] are pointers of the cos/sin table. - w[0...n/2-1] :cos/sin table (double *) - w[],ip[] are initialized if ip[0] == 0. - [remark] - Inverse of - rdft(n, 1, a, ip, w); - is - rdft(n, -1, a, ip, w); - for (j = 0; j <= n - 1; j++) { - a[j] *= 2.0 / n; - } - . - - --------- DCT (Discrete Cosine Transform) / Inverse of DCT -------- - [definition] - <case1> IDCT (excluding scale) - C[k] = sum_j=0^n-1 a[j]*cos(pi*j*(k+1/2)/n), 0<=k<n - <case2> DCT - C[k] = sum_j=0^n-1 a[j]*cos(pi*(j+1/2)*k/n), 0<=k<n - [usage] - <case1> - ip[0] = 0; // first time only - ddct(n, 1, a, ip, w); - <case2> - ip[0] = 0; // first time only - ddct(n, -1, a, ip, w); - [parameters] - n :data length (int) - n >= 2, n = power of 2 - a[0...n-1] :input/output data (double *) - output data - a[k] = C[k], 0<=k<n - ip[0...*] :work area for bit reversal (int *) - length of ip >= 2+sqrt(n/2) - strictly, - length of ip >= - 2+(1<<(int)(log(n/2+0.5)/log(2))/2). - ip[0],ip[1] are pointers of the cos/sin table. - w[0...n*5/4-1] :cos/sin table (double *) - w[],ip[] are initialized if ip[0] == 0. - [remark] - Inverse of - ddct(n, -1, a, ip, w); - is - a[0] *= 0.5; - ddct(n, 1, a, ip, w); - for (j = 0; j <= n - 1; j++) { - a[j] *= 2.0 / n; - } - . - - --------- DST (Discrete Sine Transform) / Inverse of DST -------- - [definition] - <case1> IDST (excluding scale) - S[k] = sum_j=1^n A[j]*sin(pi*j*(k+1/2)/n), 0<=k<n - <case2> DST - S[k] = sum_j=0^n-1 a[j]*sin(pi*(j+1/2)*k/n), 0<k<=n - [usage] - <case1> - ip[0] = 0; // first time only - ddst(n, 1, a, ip, w); - <case2> - ip[0] = 0; // first time only - ddst(n, -1, a, ip, w); - [parameters] - n :data length (int) - n >= 2, n = power of 2 - a[0...n-1] :input/output data (double *) - <case1> - input data - a[j] = A[j], 0<j<n - a[0] = A[n] - output data - a[k] = S[k], 0<=k<n - <case2> - output data - a[k] = S[k], 0<k<n - a[0] = S[n] - ip[0...*] :work area for bit reversal (int *) - length of ip >= 2+sqrt(n/2) - strictly, - length of ip >= - 2+(1<<(int)(log(n/2+0.5)/log(2))/2). - ip[0],ip[1] are pointers of the cos/sin table. - w[0...n*5/4-1] :cos/sin table (double *) - w[],ip[] are initialized if ip[0] == 0. - [remark] - Inverse of - ddst(n, -1, a, ip, w); - is - a[0] *= 0.5; - ddst(n, 1, a, ip, w); - for (j = 0; j <= n - 1; j++) { - a[j] *= 2.0 / n; - } - . - - --------- Cosine Transform of RDFT (Real Symmetric DFT) -------- - [definition] - C[k] = sum_j=0^n a[j]*cos(pi*j*k/n), 0<=k<=n - [usage] - ip[0] = 0; // first time only - dfct(n, a, t, ip, w); - [parameters] - n :data length - 1 (int) - n >= 2, n = power of 2 - a[0...n] :input/output data (double *) - output data - a[k] = C[k], 0<=k<=n - t[0...n/2] :work area (double *) - ip[0...*] :work area for bit reversal (int *) - length of ip >= 2+sqrt(n/4) - strictly, - length of ip >= - 2+(1<<(int)(log(n/4+0.5)/log(2))/2). - ip[0],ip[1] are pointers of the cos/sin table. - w[0...n*5/8-1] :cos/sin table (double *) - w[],ip[] are initialized if ip[0] == 0. - [remark] - Inverse of - a[0] *= 0.5; - a[n] *= 0.5; - dfct(n, a, t, ip, w); - is - a[0] *= 0.5; - a[n] *= 0.5; - dfct(n, a, t, ip, w); - for (j = 0; j <= n; j++) { - a[j] *= 2.0 / n; - } - . - - --------- Sine Transform of RDFT (Real Anti-symmetric DFT) -------- - [definition] - S[k] = sum_j=1^n-1 a[j]*sin(pi*j*k/n), 0<k<n - [usage] - ip[0] = 0; // first time only - dfst(n, a, t, ip, w); - [parameters] - n :data length + 1 (int) - n >= 2, n = power of 2 - a[0...n-1] :input/output data (double *) - output data - a[k] = S[k], 0<k<n - (a[0] is used for work area) - t[0...n/2-1] :work area (double *) - ip[0...*] :work area for bit reversal (int *) - length of ip >= 2+sqrt(n/4) - strictly, - length of ip >= - 2+(1<<(int)(log(n/4+0.5)/log(2))/2). - ip[0],ip[1] are pointers of the cos/sin table. - w[0...n*5/8-1] :cos/sin table (double *) - w[],ip[] are initialized if ip[0] == 0. - [remark] - Inverse of - dfst(n, a, t, ip, w); - is - dfst(n, a, t, ip, w); - for (j = 1; j <= n - 1; j++) { - a[j] *= 2.0 / n; - } - . - - -Appendix : - The cos/sin table is recalculated when the larger table required. - w[] and ip[] are compatible with all routines. -*/ - - -void cdft(int n, int isgn, double *a, int *ip, double *w) -{ - void makewt(int nw, int *ip, double *w); - void cftfsub(int n, double *a, int *ip, int nw, double *w); - void cftbsub(int n, double *a, int *ip, int nw, double *w); - int nw; - - nw = ip[0]; - if (n > (nw << 2)) - { - nw = n >> 2; - makewt(nw, ip, w); - } - if (isgn >= 0) - { - cftfsub(n, a, ip, nw, w); - } - else - { - cftbsub(n, a, ip, nw, w); - } -} - - -void rdft(int n, int isgn, double *a, int *ip, double *w) -{ - void makewt(int nw, int *ip, double *w); - void makect(int nc, int *ip, double *c); - void cftfsub(int n, double *a, int *ip, int nw, double *w); - void cftbsub(int n, double *a, int *ip, int nw, double *w); - void rftfsub(int n, double *a, int nc, double *c); - void rftbsub(int n, double *a, int nc, double *c); - int nw, nc; - double xi; - - nw = ip[0]; - if (n > (nw << 2)) - { - nw = n >> 2; - makewt(nw, ip, w); - } - nc = ip[1]; - if (n > (nc << 2)) - { - nc = n >> 2; - makect(nc, ip, w + nw); - } - if (isgn >= 0) - { - if (n > 4) - { - cftfsub(n, a, ip, nw, w); - rftfsub(n, a, nc, w + nw); - } - else if (n == 4) - { - cftfsub(n, a, ip, nw, w); - } - xi = a[0] - a[1]; - a[0] += a[1]; - a[1] = xi; - } - else - { - a[1] = 0.5 * (a[0] - a[1]); - a[0] -= a[1]; - if (n > 4) - { - rftbsub(n, a, nc, w + nw); - cftbsub(n, a, ip, nw, w); - } - else if (n == 4) - { - cftbsub(n, a, ip, nw, w); - } - } -} - - -void ddct(int n, int isgn, double *a, int *ip, double *w) -{ - void makewt(int nw, int *ip, double *w); - void makect(int nc, int *ip, double *c); - void cftfsub(int n, double *a, int *ip, int nw, double *w); - void cftbsub(int n, double *a, int *ip, int nw, double *w); - void rftfsub(int n, double *a, int nc, double *c); - void rftbsub(int n, double *a, int nc, double *c); - void dctsub(int n, double *a, int nc, double *c); - int j, nw, nc; - double xr; - - nw = ip[0]; - if (n > (nw << 2)) - { - nw = n >> 2; - makewt(nw, ip, w); - } - nc = ip[1]; - if (n > nc) - { - nc = n; - makect(nc, ip, w + nw); - } - if (isgn < 0) - { - xr = a[n - 1]; - for (j = n - 2; j >= 2; j -= 2) - { - a[j + 1] = a[j] - a[j - 1]; - a[j] += a[j - 1]; - } - a[1] = a[0] - xr; - a[0] += xr; - if (n > 4) - { - rftbsub(n, a, nc, w + nw); - cftbsub(n, a, ip, nw, w); - } - else if (n == 4) - { - cftbsub(n, a, ip, nw, w); - } - } - dctsub(n, a, nc, w + nw); - if (isgn >= 0) - { - if (n > 4) - { - cftfsub(n, a, ip, nw, w); - rftfsub(n, a, nc, w + nw); - } - else if (n == 4) - { - cftfsub(n, a, ip, nw, w); - } - xr = a[0] - a[1]; - a[0] += a[1]; - for (j = 2; j < n; j += 2) - { - a[j - 1] = a[j] - a[j + 1]; - a[j] += a[j + 1]; - } - a[n - 1] = xr; - } -} - - -void ddst(int n, int isgn, double *a, int *ip, double *w) -{ - void makewt(int nw, int *ip, double *w); - void makect(int nc, int *ip, double *c); - void cftfsub(int n, double *a, int *ip, int nw, double *w); - void cftbsub(int n, double *a, int *ip, int nw, double *w); - void rftfsub(int n, double *a, int nc, double *c); - void rftbsub(int n, double *a, int nc, double *c); - void dstsub(int n, double *a, int nc, double *c); - int j, nw, nc; - double xr; - - nw = ip[0]; - if (n > (nw << 2)) - { - nw = n >> 2; - makewt(nw, ip, w); - } - nc = ip[1]; - if (n > nc) - { - nc = n; - makect(nc, ip, w + nw); - } - if (isgn < 0) - { - xr = a[n - 1]; - for (j = n - 2; j >= 2; j -= 2) - { - a[j + 1] = -a[j] - a[j - 1]; - a[j] -= a[j - 1]; - } - a[1] = a[0] + xr; - a[0] -= xr; - if (n > 4) - { - rftbsub(n, a, nc, w + nw); - cftbsub(n, a, ip, nw, w); - } - else if (n == 4) - { - cftbsub(n, a, ip, nw, w); - } - } - dstsub(n, a, nc, w + nw); - if (isgn >= 0) - { - if (n > 4) - { - cftfsub(n, a, ip, nw, w); - rftfsub(n, a, nc, w + nw); - } - else if (n == 4) - { - cftfsub(n, a, ip, nw, w); - } - xr = a[0] - a[1]; - a[0] += a[1]; - for (j = 2; j < n; j += 2) - { - a[j - 1] = -a[j] - a[j + 1]; - a[j] -= a[j + 1]; - } - a[n - 1] = -xr; - } -} - - -void dfct(int n, double *a, double *t, int *ip, double *w) -{ - void makewt(int nw, int *ip, double *w); - void makect(int nc, int *ip, double *c); - void cftfsub(int n, double *a, int *ip, int nw, double *w); - void rftfsub(int n, double *a, int nc, double *c); - void dctsub(int n, double *a, int nc, double *c); - int j, k, l, m, mh, nw, nc; - double xr, xi, yr, yi; - - nw = ip[0]; - if (n > (nw << 3)) - { - nw = n >> 3; - makewt(nw, ip, w); - } - nc = ip[1]; - if (n > (nc << 1)) - { - nc = n >> 1; - makect(nc, ip, w + nw); - } - m = n >> 1; - yi = a[m]; - xi = a[0] + a[n]; - a[0] -= a[n]; - t[0] = xi - yi; - t[m] = xi + yi; - if (n > 2) - { - mh = m >> 1; - for (j = 1; j < mh; j++) - { - k = m - j; - xr = a[j] - a[n - j]; - xi = a[j] + a[n - j]; - yr = a[k] - a[n - k]; - yi = a[k] + a[n - k]; - a[j] = xr; - a[k] = yr; - t[j] = xi - yi; - t[k] = xi + yi; - } - t[mh] = a[mh] + a[n - mh]; - a[mh] -= a[n - mh]; - dctsub(m, a, nc, w + nw); - if (m > 4) - { - cftfsub(m, a, ip, nw, w); - rftfsub(m, a, nc, w + nw); - } - else if (m == 4) - { - cftfsub(m, a, ip, nw, w); - } - a[n - 1] = a[0] - a[1]; - a[1] = a[0] + a[1]; - for (j = m - 2; j >= 2; j -= 2) - { - a[2 * j + 1] = a[j] + a[j + 1]; - a[2 * j - 1] = a[j] - a[j + 1]; - } - l = 2; - m = mh; - while (m >= 2) - { - dctsub(m, t, nc, w + nw); - if (m > 4) - { - cftfsub(m, t, ip, nw, w); - rftfsub(m, t, nc, w + nw); - } - else if (m == 4) - { - cftfsub(m, t, ip, nw, w); - } - a[n - l] = t[0] - t[1]; - a[l] = t[0] + t[1]; - k = 0; - for (j = 2; j < m; j += 2) - { - k += l << 2; - a[k - l] = t[j] - t[j + 1]; - a[k + l] = t[j] + t[j + 1]; - } - l <<= 1; - mh = m >> 1; - for (j = 0; j < mh; j++) - { - k = m - j; - t[j] = t[m + k] - t[m + j]; - t[k] = t[m + k] + t[m + j]; - } - t[mh] = t[m + mh]; - m = mh; - } - a[l] = t[0]; - a[n] = t[2] - t[1]; - a[0] = t[2] + t[1]; - } - else - { - a[1] = a[0]; - a[2] = t[0]; - a[0] = t[1]; - } -} - - -void dfst(int n, double *a, double *t, int *ip, double *w) -{ - void makewt(int nw, int *ip, double *w); - void makect(int nc, int *ip, double *c); - void cftfsub(int n, double *a, int *ip, int nw, double *w); - void rftfsub(int n, double *a, int nc, double *c); - void dstsub(int n, double *a, int nc, double *c); - int j, k, l, m, mh, nw, nc; - double xr, xi, yr, yi; - - nw = ip[0]; - if (n > (nw << 3)) - { - nw = n >> 3; - makewt(nw, ip, w); - } - nc = ip[1]; - if (n > (nc << 1)) - { - nc = n >> 1; - makect(nc, ip, w + nw); - } - if (n > 2) - { - m = n >> 1; - mh = m >> 1; - for (j = 1; j < mh; j++) - { - k = m - j; - xr = a[j] + a[n - j]; - xi = a[j] - a[n - j]; - yr = a[k] + a[n - k]; - yi = a[k] - a[n - k]; - a[j] = xr; - a[k] = yr; - t[j] = xi + yi; - t[k] = xi - yi; - } - t[0] = a[mh] - a[n - mh]; - a[mh] += a[n - mh]; - a[0] = a[m]; - dstsub(m, a, nc, w + nw); - if (m > 4) - { - cftfsub(m, a, ip, nw, w); - rftfsub(m, a, nc, w + nw); - } - else if (m == 4) - { - cftfsub(m, a, ip, nw, w); - } - a[n - 1] = a[1] - a[0]; - a[1] = a[0] + a[1]; - for (j = m - 2; j >= 2; j -= 2) - { - a[2 * j + 1] = a[j] - a[j + 1]; - a[2 * j - 1] = -a[j] - a[j + 1]; - } - l = 2; - m = mh; - while (m >= 2) - { - dstsub(m, t, nc, w + nw); - if (m > 4) - { - cftfsub(m, t, ip, nw, w); - rftfsub(m, t, nc, w + nw); - } - else if (m == 4) - { - cftfsub(m, t, ip, nw, w); - } - a[n - l] = t[1] - t[0]; - a[l] = t[0] + t[1]; - k = 0; - for (j = 2; j < m; j += 2) - { - k += l << 2; - a[k - l] = -t[j] - t[j + 1]; - a[k + l] = t[j] - t[j + 1]; - } - l <<= 1; - mh = m >> 1; - for (j = 1; j < mh; j++) - { - k = m - j; - t[j] = t[m + k] + t[m + j]; - t[k] = t[m + k] - t[m + j]; - } - t[0] = t[m + mh]; - m = mh; - } - a[l] = t[0]; - } - a[0] = 0; -} - - -/* -------- initializing routines -------- */ - - -#include <math.h> - -void makewt(int nw, int *ip, double *w) -{ - void makeipt(int nw, int *ip); - int j, nwh, nw0, nw1; - double delta, wn4r, wk1r, wk1i, wk3r, wk3i; - - ip[0] = nw; - ip[1] = 1; - if (nw > 2) - { - nwh = nw >> 1; - delta = atan(1.0) / nwh; - wn4r = cos(delta * nwh); - w[0] = 1; - w[1] = wn4r; - if (nwh == 4) - { - w[2] = cos(delta * 2); - w[3] = sin(delta * 2); - } - else if (nwh > 4) - { - makeipt(nw, ip); - w[2] = 0.5 / cos(delta * 2); - w[3] = 0.5 / cos(delta * 6); - for (j = 4; j < nwh; j += 4) - { - w[j] = cos(delta * j); - w[j + 1] = sin(delta * j); - w[j + 2] = cos(3 * delta * j); - w[j + 3] = -sin(3 * delta * j); - } - } - nw0 = 0; - while (nwh > 2) - { - nw1 = nw0 + nwh; - nwh >>= 1; - w[nw1] = 1; - w[nw1 + 1] = wn4r; - if (nwh == 4) - { - wk1r = w[nw0 + 4]; - wk1i = w[nw0 + 5]; - w[nw1 + 2] = wk1r; - w[nw1 + 3] = wk1i; - } - else if (nwh > 4) - { - wk1r = w[nw0 + 4]; - wk3r = w[nw0 + 6]; - w[nw1 + 2] = 0.5 / wk1r; - w[nw1 + 3] = 0.5 / wk3r; - for (j = 4; j < nwh; j += 4) - { - wk1r = w[nw0 + 2 * j]; - wk1i = w[nw0 + 2 * j + 1]; - wk3r = w[nw0 + 2 * j + 2]; - wk3i = w[nw0 + 2 * j + 3]; - w[nw1 + j] = wk1r; - w[nw1 + j + 1] = wk1i; - w[nw1 + j + 2] = wk3r; - w[nw1 + j + 3] = wk3i; - } - } - nw0 = nw1; - } - } -} - - -void makeipt(int nw, int *ip) -{ - int j, l, m, m2, p, q; - - ip[2] = 0; - ip[3] = 16; - m = 2; - for (l = nw; l > 32; l >>= 2) - { - m2 = m << 1; - q = m2 << 3; - for (j = m; j < m2; j++) - { - p = ip[j] << 2; - ip[m + j] = p; - ip[m2 + j] = p + q; - } - m = m2; - } -} - - -void makect(int nc, int *ip, double *c) -{ - int j, nch; - double delta; - - ip[1] = nc; - if (nc > 1) - { - nch = nc >> 1; - delta = atan(1.0) / nch; - c[0] = cos(delta * nch); - c[nch] = 0.5 * c[0]; - for (j = 1; j < nch; j++) - { - c[j] = 0.5 * cos(delta * j); - c[nc - j] = 0.5 * sin(delta * j); - } - } -} - - -/* -------- child routines -------- */ - - -#ifdef USE_CDFT_PTHREADS -#define USE_CDFT_THREADS -#ifndef CDFT_THREADS_BEGIN_N -#define CDFT_THREADS_BEGIN_N 8192 -#endif -#ifndef CDFT_4THREADS_BEGIN_N -#define CDFT_4THREADS_BEGIN_N 65536 -#endif -#include <pthread.h> -#include <stdio.h> -#include <stdlib.h> -#define cdft_thread_t pthread_t -#define cdft_thread_create(thp,func,argp) { \ - if (pthread_create(thp, NULL, func, (void *) argp) != 0) { \ - fprintf(stderr, "cdft thread error\n"); \ - exit(1); \ - } \ -} -#define cdft_thread_wait(th) { \ - if (pthread_join(th, NULL) != 0) { \ - fprintf(stderr, "cdft thread error\n"); \ - exit(1); \ - } \ -} -#endif /* USE_CDFT_PTHREADS */ - - -#ifdef USE_CDFT_WINTHREADS -#define USE_CDFT_THREADS -#ifndef CDFT_THREADS_BEGIN_N -#define CDFT_THREADS_BEGIN_N 32768 -#endif -#ifndef CDFT_4THREADS_BEGIN_N -#define CDFT_4THREADS_BEGIN_N 524288 -#endif -#include <windows.h> -#include <stdio.h> -#include <stdlib.h> -#define cdft_thread_t HANDLE -#define cdft_thread_create(thp,func,argp) { \ - DWORD thid; \ - *(thp) = CreateThread(NULL, 0, (LPTHREAD_START_ROUTINE) func, (LPVOID) argp, 0, &thid); \ - if (*(thp) == 0) { \ - fprintf(stderr, "cdft thread error\n"); \ - exit(1); \ - } \ -} -#define cdft_thread_wait(th) { \ - WaitForSingleObject(th, INFINITE); \ - CloseHandle(th); \ -} -#endif /* USE_CDFT_WINTHREADS */ - - -void cftfsub(int n, double *a, int *ip, int nw, double *w) -{ - void bitrv2(int n, int *ip, double *a); - void bitrv216(double *a); - void bitrv208(double *a); - void cftf1st(int n, double *a, double *w); - void cftrec4(int n, double *a, int nw, double *w); - void cftleaf(int n, int isplt, double *a, int nw, double *w); - void cftfx41(int n, double *a, int nw, double *w); - void cftf161(double *a, double *w); - void cftf081(double *a, double *w); - void cftf040(double *a); - void cftx020(double *a); -#ifdef USE_CDFT_THREADS - void cftrec4_th(int n, double *a, int nw, double *w); -#endif /* USE_CDFT_THREADS */ - - if (n > 8) - { - if (n > 32) - { - cftf1st(n, a, &w[nw - (n >> 2)]); -#ifdef USE_CDFT_THREADS - if (n > CDFT_THREADS_BEGIN_N) - { - cftrec4_th(n, a, nw, w); - } - else -#endif /* USE_CDFT_THREADS */ - if (n > 512) - { - cftrec4(n, a, nw, w); - } - else if (n > 128) - { - cftleaf(n, 1, a, nw, w); - } - else - { - cftfx41(n, a, nw, w); - } - bitrv2(n, ip, a); - } - else if (n == 32) - { - cftf161(a, &w[nw - 8]); - bitrv216(a); - } - else - { - cftf081(a, w); - bitrv208(a); - } - } - else if (n == 8) - { - cftf040(a); - } - else if (n == 4) - { - cftx020(a); - } -} - - -void cftbsub(int n, double *a, int *ip, int nw, double *w) -{ - void bitrv2conj(int n, int *ip, double *a); - void bitrv216neg(double *a); - void bitrv208neg(double *a); - void cftb1st(int n, double *a, double *w); - void cftrec4(int n, double *a, int nw, double *w); - void cftleaf(int n, int isplt, double *a, int nw, double *w); - void cftfx41(int n, double *a, int nw, double *w); - void cftf161(double *a, double *w); - void cftf081(double *a, double *w); - void cftb040(double *a); - void cftx020(double *a); -#ifdef USE_CDFT_THREADS - void cftrec4_th(int n, double *a, int nw, double *w); -#endif /* USE_CDFT_THREADS */ - - if (n > 8) - { - if (n > 32) - { - cftb1st(n, a, &w[nw - (n >> 2)]); -#ifdef USE_CDFT_THREADS - if (n > CDFT_THREADS_BEGIN_N) - { - cftrec4_th(n, a, nw, w); - } - else -#endif /* USE_CDFT_THREADS */ - if (n > 512) - { - cftrec4(n, a, nw, w); - } - else if (n > 128) - { - cftleaf(n, 1, a, nw, w); - } - else - { - cftfx41(n, a, nw, w); - } - bitrv2conj(n, ip, a); - } - else if (n == 32) - { - cftf161(a, &w[nw - 8]); - bitrv216neg(a); - } - else - { - cftf081(a, w); - bitrv208neg(a); - } - } - else if (n == 8) - { - cftb040(a); - } - else if (n == 4) - { - cftx020(a); - } -} - - -void bitrv2(int n, int *ip, double *a) -{ - int j, j1, k, k1, l, m, nh, nm; - double xr, xi, yr, yi; - - m = 1; - for (l = n >> 2; l > 8; l >>= 2) - { - m <<= 1; - } - nh = n >> 1; - nm = 4 * m; - if (l == 8) - { - for (k = 0; k < m; k++) - { - for (j = 0; j < k; j++) - { - j1 = 4 * j + 2 * ip[m + k]; - k1 = 4 * k + 2 * ip[m + j]; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 += 2 * nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 -= nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 += 2 * nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nh; - k1 += 2; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nm; - k1 -= 2 * nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nm; - k1 += nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nm; - k1 -= 2 * nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += 2; - k1 += nh; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 += 2 * nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 -= nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 += 2 * nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nh; - k1 -= 2; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nm; - k1 -= 2 * nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nm; - k1 += nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nm; - k1 -= 2 * nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - } - k1 = 4 * k + 2 * ip[m + k]; - j1 = k1 + 2; - k1 += nh; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 += 2 * nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 -= nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= 2; - k1 -= nh; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nh + 2; - k1 += nh + 2; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nh - nm; - k1 += 2 * nm - 2; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - } - } - else - { - for (k = 0; k < m; k++) - { - for (j = 0; j < k; j++) - { - j1 = 4 * j + ip[m + k]; - k1 = 4 * k + ip[m + j]; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 += nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nh; - k1 += 2; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nm; - k1 -= nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += 2; - k1 += nh; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 += nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nh; - k1 -= 2; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nm; - k1 -= nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - } - k1 = 4 * k + ip[m + k]; - j1 = k1 + 2; - k1 += nh; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 += nm; - xr = a[j1]; - xi = a[j1 + 1]; - yr = a[k1]; - yi = a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - } - } -} - - -void bitrv2conj(int n, int *ip, double *a) -{ - int j, j1, k, k1, l, m, nh, nm; - double xr, xi, yr, yi; - - m = 1; - for (l = n >> 2; l > 8; l >>= 2) - { - m <<= 1; - } - nh = n >> 1; - nm = 4 * m; - if (l == 8) - { - for (k = 0; k < m; k++) - { - for (j = 0; j < k; j++) - { - j1 = 4 * j + 2 * ip[m + k]; - k1 = 4 * k + 2 * ip[m + j]; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 += 2 * nm; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 -= nm; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 += 2 * nm; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nh; - k1 += 2; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nm; - k1 -= 2 * nm; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nm; - k1 += nm; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nm; - k1 -= 2 * nm; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += 2; - k1 += nh; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 += 2 * nm; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 -= nm; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 += 2 * nm; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nh; - k1 -= 2; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nm; - k1 -= 2 * nm; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nm; - k1 += nm; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nm; - k1 -= 2 * nm; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - } - k1 = 4 * k + 2 * ip[m + k]; - j1 = k1 + 2; - k1 += nh; - a[j1 - 1] = -a[j1 - 1]; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - a[k1 + 3] = -a[k1 + 3]; - j1 += nm; - k1 += 2 * nm; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 -= nm; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= 2; - k1 -= nh; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nh + 2; - k1 += nh + 2; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nh - nm; - k1 += 2 * nm - 2; - a[j1 - 1] = -a[j1 - 1]; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - a[k1 + 3] = -a[k1 + 3]; - } - } - else - { - for (k = 0; k < m; k++) - { - for (j = 0; j < k; j++) - { - j1 = 4 * j + ip[m + k]; - k1 = 4 * k + ip[m + j]; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 += nm; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nh; - k1 += 2; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nm; - k1 -= nm; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += 2; - k1 += nh; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 += nm; - k1 += nm; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nh; - k1 -= 2; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - j1 -= nm; - k1 -= nm; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - } - k1 = 4 * k + ip[m + k]; - j1 = k1 + 2; - k1 += nh; - a[j1 - 1] = -a[j1 - 1]; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - a[k1 + 3] = -a[k1 + 3]; - j1 += nm; - k1 += nm; - a[j1 - 1] = -a[j1 - 1]; - xr = a[j1]; - xi = -a[j1 + 1]; - yr = a[k1]; - yi = -a[k1 + 1]; - a[j1] = yr; - a[j1 + 1] = yi; - a[k1] = xr; - a[k1 + 1] = xi; - a[k1 + 3] = -a[k1 + 3]; - } - } -} - - -void bitrv216(double *a) -{ - double x1r, x1i, x2r, x2i, x3r, x3i, x4r, x4i, - x5r, x5i, x7r, x7i, x8r, x8i, x10r, x10i, - x11r, x11i, x12r, x12i, x13r, x13i, x14r, x14i; - - x1r = a[2]; - x1i = a[3]; - x2r = a[4]; - x2i = a[5]; - x3r = a[6]; - x3i = a[7]; - x4r = a[8]; - x4i = a[9]; - x5r = a[10]; - x5i = a[11]; - x7r = a[14]; - x7i = a[15]; - x8r = a[16]; - x8i = a[17]; - x10r = a[20]; - x10i = a[21]; - x11r = a[22]; - x11i = a[23]; - x12r = a[24]; - x12i = a[25]; - x13r = a[26]; - x13i = a[27]; - x14r = a[28]; - x14i = a[29]; - a[2] = x8r; - a[3] = x8i; - a[4] = x4r; - a[5] = x4i; - a[6] = x12r; - a[7] = x12i; - a[8] = x2r; - a[9] = x2i; - a[10] = x10r; - a[11] = x10i; - a[14] = x14r; - a[15] = x14i; - a[16] = x1r; - a[17] = x1i; - a[20] = x5r; - a[21] = x5i; - a[22] = x13r; - a[23] = x13i; - a[24] = x3r; - a[25] = x3i; - a[26] = x11r; - a[27] = x11i; - a[28] = x7r; - a[29] = x7i; -} - - -void bitrv216neg(double *a) -{ - double x1r, x1i, x2r, x2i, x3r, x3i, x4r, x4i, - x5r, x5i, x6r, x6i, x7r, x7i, x8r, x8i, - x9r, x9i, x10r, x10i, x11r, x11i, x12r, x12i, - x13r, x13i, x14r, x14i, x15r, x15i; - - x1r = a[2]; - x1i = a[3]; - x2r = a[4]; - x2i = a[5]; - x3r = a[6]; - x3i = a[7]; - x4r = a[8]; - x4i = a[9]; - x5r = a[10]; - x5i = a[11]; - x6r = a[12]; - x6i = a[13]; - x7r = a[14]; - x7i = a[15]; - x8r = a[16]; - x8i = a[17]; - x9r = a[18]; - x9i = a[19]; - x10r = a[20]; - x10i = a[21]; - x11r = a[22]; - x11i = a[23]; - x12r = a[24]; - x12i = a[25]; - x13r = a[26]; - x13i = a[27]; - x14r = a[28]; - x14i = a[29]; - x15r = a[30]; - x15i = a[31]; - a[2] = x15r; - a[3] = x15i; - a[4] = x7r; - a[5] = x7i; - a[6] = x11r; - a[7] = x11i; - a[8] = x3r; - a[9] = x3i; - a[10] = x13r; - a[11] = x13i; - a[12] = x5r; - a[13] = x5i; - a[14] = x9r; - a[15] = x9i; - a[16] = x1r; - a[17] = x1i; - a[18] = x14r; - a[19] = x14i; - a[20] = x6r; - a[21] = x6i; - a[22] = x10r; - a[23] = x10i; - a[24] = x2r; - a[25] = x2i; - a[26] = x12r; - a[27] = x12i; - a[28] = x4r; - a[29] = x4i; - a[30] = x8r; - a[31] = x8i; -} - - -void bitrv208(double *a) -{ - double x1r, x1i, x3r, x3i, x4r, x4i, x6r, x6i; - - x1r = a[2]; - x1i = a[3]; - x3r = a[6]; - x3i = a[7]; - x4r = a[8]; - x4i = a[9]; - x6r = a[12]; - x6i = a[13]; - a[2] = x4r; - a[3] = x4i; - a[6] = x6r; - a[7] = x6i; - a[8] = x1r; - a[9] = x1i; - a[12] = x3r; - a[13] = x3i; -} - - -void bitrv208neg(double *a) -{ - double x1r, x1i, x2r, x2i, x3r, x3i, x4r, x4i, - x5r, x5i, x6r, x6i, x7r, x7i; - - x1r = a[2]; - x1i = a[3]; - x2r = a[4]; - x2i = a[5]; - x3r = a[6]; - x3i = a[7]; - x4r = a[8]; - x4i = a[9]; - x5r = a[10]; - x5i = a[11]; - x6r = a[12]; - x6i = a[13]; - x7r = a[14]; - x7i = a[15]; - a[2] = x7r; - a[3] = x7i; - a[4] = x3r; - a[5] = x3i; - a[6] = x5r; - a[7] = x5i; - a[8] = x1r; - a[9] = x1i; - a[10] = x6r; - a[11] = x6i; - a[12] = x2r; - a[13] = x2i; - a[14] = x4r; - a[15] = x4i; -} - - -void cftf1st(int n, double *a, double *w) -{ - int j, j0, j1, j2, j3, k, m, mh; - double wn4r, csc1, csc3, wk1r, wk1i, wk3r, wk3i, - wd1r, wd1i, wd3r, wd3i; - double x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i, - y0r, y0i, y1r, y1i, y2r, y2i, y3r, y3i; - - mh = n >> 3; - m = 2 * mh; - j1 = m; - j2 = j1 + m; - j3 = j2 + m; - x0r = a[0] + a[j2]; - x0i = a[1] + a[j2 + 1]; - x1r = a[0] - a[j2]; - x1i = a[1] - a[j2 + 1]; - x2r = a[j1] + a[j3]; - x2i = a[j1 + 1] + a[j3 + 1]; - x3r = a[j1] - a[j3]; - x3i = a[j1 + 1] - a[j3 + 1]; - a[0] = x0r + x2r; - a[1] = x0i + x2i; - a[j1] = x0r - x2r; - a[j1 + 1] = x0i - x2i; - a[j2] = x1r - x3i; - a[j2 + 1] = x1i + x3r; - a[j3] = x1r + x3i; - a[j3 + 1] = x1i - x3r; - wn4r = w[1]; - csc1 = w[2]; - csc3 = w[3]; - wd1r = 1; - wd1i = 0; - wd3r = 1; - wd3i = 0; - k = 0; - for (j = 2; j < mh - 2; j += 4) - { - k += 4; - wk1r = csc1 * (wd1r + w[k]); - wk1i = csc1 * (wd1i + w[k + 1]); - wk3r = csc3 * (wd3r + w[k + 2]); - wk3i = csc3 * (wd3i + w[k + 3]); - wd1r = w[k]; - wd1i = w[k + 1]; - wd3r = w[k + 2]; - wd3i = w[k + 3]; - j1 = j + m; - j2 = j1 + m; - j3 = j2 + m; - x0r = a[j] + a[j2]; - x0i = a[j + 1] + a[j2 + 1]; - x1r = a[j] - a[j2]; - x1i = a[j + 1] - a[j2 + 1]; - y0r = a[j + 2] + a[j2 + 2]; - y0i = a[j + 3] + a[j2 + 3]; - y1r = a[j + 2] - a[j2 + 2]; - y1i = a[j + 3] - a[j2 + 3]; - x2r = a[j1] + a[j3]; - x2i = a[j1 + 1] + a[j3 + 1]; - x3r = a[j1] - a[j3]; - x3i = a[j1 + 1] - a[j3 + 1]; - y2r = a[j1 + 2] + a[j3 + 2]; - y2i = a[j1 + 3] + a[j3 + 3]; - y3r = a[j1 + 2] - a[j3 + 2]; - y3i = a[j1 + 3] - a[j3 + 3]; - a[j] = x0r + x2r; - a[j + 1] = x0i + x2i; - a[j + 2] = y0r + y2r; - a[j + 3] = y0i + y2i; - a[j1] = x0r - x2r; - a[j1 + 1] = x0i - x2i; - a[j1 + 2] = y0r - y2r; - a[j1 + 3] = y0i - y2i; - x0r = x1r - x3i; - x0i = x1i + x3r; - a[j2] = wk1r * x0r - wk1i * x0i; - a[j2 + 1] = wk1r * x0i + wk1i * x0r; - x0r = y1r - y3i; - x0i = y1i + y3r; - a[j2 + 2] = wd1r * x0r - wd1i * x0i; - a[j2 + 3] = wd1r * x0i + wd1i * x0r; - x0r = x1r + x3i; - x0i = x1i - x3r; - a[j3] = wk3r * x0r + wk3i * x0i; - a[j3 + 1] = wk3r * x0i - wk3i * x0r; - x0r = y1r + y3i; - x0i = y1i - y3r; - a[j3 + 2] = wd3r * x0r + wd3i * x0i; - a[j3 + 3] = wd3r * x0i - wd3i * x0r; - j0 = m - j; - j1 = j0 + m; - j2 = j1 + m; - j3 = j2 + m; - x0r = a[j0] + a[j2]; - x0i = a[j0 + 1] + a[j2 + 1]; - x1r = a[j0] - a[j2]; - x1i = a[j0 + 1] - a[j2 + 1]; - y0r = a[j0 - 2] + a[j2 - 2]; - y0i = a[j0 - 1] + a[j2 - 1]; - y1r = a[j0 - 2] - a[j2 - 2]; - y1i = a[j0 - 1] - a[j2 - 1]; - x2r = a[j1] + a[j3]; - x2i = a[j1 + 1] + a[j3 + 1]; - x3r = a[j1] - a[j3]; - x3i = a[j1 + 1] - a[j3 + 1]; - y2r = a[j1 - 2] + a[j3 - 2]; - y2i = a[j1 - 1] + a[j3 - 1]; - y3r = a[j1 - 2] - a[j3 - 2]; - y3i = a[j1 - 1] - a[j3 - 1]; - a[j0] = x0r + x2r; - a[j0 + 1] = x0i + x2i; - a[j0 - 2] = y0r + y2r; - a[j0 - 1] = y0i + y2i; - a[j1] = x0r - x2r; - a[j1 + 1] = x0i - x2i; - a[j1 - 2] = y0r - y2r; - a[j1 - 1] = y0i - y2i; - x0r = x1r - x3i; - x0i = x1i + x3r; - a[j2] = wk1i * x0r - wk1r * x0i; - a[j2 + 1] = wk1i * x0i + wk1r * x0r; - x0r = y1r - y3i; - x0i = y1i + y3r; - a[j2 - 2] = wd1i * x0r - wd1r * x0i; - a[j2 - 1] = wd1i * x0i + wd1r * x0r; - x0r = x1r + x3i; - x0i = x1i - x3r; - a[j3] = wk3i * x0r + wk3r * x0i; - a[j3 + 1] = wk3i * x0i - wk3r * x0r; - x0r = y1r + y3i; - x0i = y1i - y3r; - a[j3 - 2] = wd3i * x0r + wd3r * x0i; - a[j3 - 1] = wd3i * x0i - wd3r * x0r; - } - wk1r = csc1 * (wd1r + wn4r); - wk1i = csc1 * (wd1i + wn4r); - wk3r = csc3 * (wd3r - wn4r); - wk3i = csc3 * (wd3i - wn4r); - j0 = mh; - j1 = j0 + m; - j2 = j1 + m; - j3 = j2 + m; - x0r = a[j0 - 2] + a[j2 - 2]; - x0i = a[j0 - 1] + a[j2 - 1]; - x1r = a[j0 - 2] - a[j2 - 2]; - x1i = a[j0 - 1] - a[j2 - 1]; - x2r = a[j1 - 2] + a[j3 - 2]; - x2i = a[j1 - 1] + a[j3 - 1]; - x3r = a[j1 - 2] - a[j3 - 2]; - x3i = a[j1 - 1] - a[j3 - 1]; - a[j0 - 2] = x0r + x2r; - a[j0 - 1] = x0i + x2i; - a[j1 - 2] = x0r - x2r; - a[j1 - 1] = x0i - x2i; - x0r = x1r - x3i; - x0i = x1i + x3r; - a[j2 - 2] = wk1r * x0r - wk1i * x0i; - a[j2 - 1] = wk1r * x0i + wk1i * x0r; - x0r = x1r + x3i; - x0i = x1i - x3r; - a[j3 - 2] = wk3r * x0r + wk3i * x0i; - a[j3 - 1] = wk3r * x0i - wk3i * x0r; - x0r = a[j0] + a[j2]; - x0i = a[j0 + 1] + a[j2 + 1]; - x1r = a[j0] - a[j2]; - x1i = a[j0 + 1] - a[j2 + 1]; - x2r = a[j1] + a[j3]; - x2i = a[j1 + 1] + a[j3 + 1]; - x3r = a[j1] - a[j3]; - x3i = a[j1 + 1] - a[j3 + 1]; - a[j0] = x0r + x2r; - a[j0 + 1] = x0i + x2i; - a[j1] = x0r - x2r; - a[j1 + 1] = x0i - x2i; - x0r = x1r - x3i; - x0i = x1i + x3r; - a[j2] = wn4r * (x0r - x0i); - a[j2 + 1] = wn4r * (x0i + x0r); - x0r = x1r + x3i; - x0i = x1i - x3r; - a[j3] = -wn4r * (x0r + x0i); - a[j3 + 1] = -wn4r * (x0i - x0r); - x0r = a[j0 + 2] + a[j2 + 2]; - x0i = a[j0 + 3] + a[j2 + 3]; - x1r = a[j0 + 2] - a[j2 + 2]; - x1i = a[j0 + 3] - a[j2 + 3]; - x2r = a[j1 + 2] + a[j3 + 2]; - x2i = a[j1 + 3] + a[j3 + 3]; - x3r = a[j1 + 2] - a[j3 + 2]; - x3i = a[j1 + 3] - a[j3 + 3]; - a[j0 + 2] = x0r + x2r; - a[j0 + 3] = x0i + x2i; - a[j1 + 2] = x0r - x2r; - a[j1 + 3] = x0i - x2i; - x0r = x1r - x3i; - x0i = x1i + x3r; - a[j2 + 2] = wk1i * x0r - wk1r * x0i; - a[j2 + 3] = wk1i * x0i + wk1r * x0r; - x0r = x1r + x3i; - x0i = x1i - x3r; - a[j3 + 2] = wk3i * x0r + wk3r * x0i; - a[j3 + 3] = wk3i * x0i - wk3r * x0r; -} - - -void cftb1st(int n, double *a, double *w) -{ - int j, j0, j1, j2, j3, k, m, mh; - double wn4r, csc1, csc3, wk1r, wk1i, wk3r, wk3i, - wd1r, wd1i, wd3r, wd3i; - double x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i, - y0r, y0i, y1r, y1i, y2r, y2i, y3r, y3i; - - mh = n >> 3; - m = 2 * mh; - j1 = m; - j2 = j1 + m; - j3 = j2 + m; - x0r = a[0] + a[j2]; - x0i = -a[1] - a[j2 + 1]; - x1r = a[0] - a[j2]; - x1i = -a[1] + a[j2 + 1]; - x2r = a[j1] + a[j3]; - x2i = a[j1 + 1] + a[j3 + 1]; - x3r = a[j1] - a[j3]; - x3i = a[j1 + 1] - a[j3 + 1]; - a[0] = x0r + x2r; - a[1] = x0i - x2i; - a[j1] = x0r - x2r; - a[j1 + 1] = x0i + x2i; - a[j2] = x1r + x3i; - a[j2 + 1] = x1i + x3r; - a[j3] = x1r - x3i; - a[j3 + 1] = x1i - x3r; - wn4r = w[1]; - csc1 = w[2]; - csc3 = w[3]; - wd1r = 1; - wd1i = 0; - wd3r = 1; - wd3i = 0; - k = 0; - for (j = 2; j < mh - 2; j += 4) - { - k += 4; - wk1r = csc1 * (wd1r + w[k]); - wk1i = csc1 * (wd1i + w[k + 1]); - wk3r = csc3 * (wd3r + w[k + 2]); - wk3i = csc3 * (wd3i + w[k + 3]); - wd1r = w[k]; - wd1i = w[k + 1]; - wd3r = w[k + 2]; - wd3i = w[k + 3]; - j1 = j + m; - j2 = j1 + m; - j3 = j2 + m; - x0r = a[j] + a[j2]; - x0i = -a[j + 1] - a[j2 + 1]; - x1r = a[j] - a[j2]; - x1i = -a[j + 1] + a[j2 + 1]; - y0r = a[j + 2] + a[j2 + 2]; - y0i = -a[j + 3] - a[j2 + 3]; - y1r = a[j + 2] - a[j2 + 2]; - y1i = -a[j + 3] + a[j2 + 3]; - x2r = a[j1] + a[j3]; - x2i = a[j1 + 1] + a[j3 + 1]; - x3r = a[j1] - a[j3]; - x3i = a[j1 + 1] - a[j3 + 1]; - y2r = a[j1 + 2] + a[j3 + 2]; - y2i = a[j1 + 3] + a[j3 + 3]; - y3r = a[j1 + 2] - a[j3 + 2]; - y3i = a[j1 + 3] - a[j3 + 3]; - a[j] = x0r + x2r; - a[j + 1] = x0i - x2i; - a[j + 2] = y0r + y2r; - a[j + 3] = y0i - y2i; - a[j1] = x0r - x2r; - a[j1 + 1] = x0i + x2i; - a[j1 + 2] = y0r - y2r; - a[j1 + 3] = y0i + y2i; - x0r = x1r + x3i; - x0i = x1i + x3r; - a[j2] = wk1r * x0r - wk1i * x0i; - a[j2 + 1] = wk1r * x0i + wk1i * x0r; - x0r = y1r + y3i; - x0i = y1i + y3r; - a[j2 + 2] = wd1r * x0r - wd1i * x0i; - a[j2 + 3] = wd1r * x0i + wd1i * x0r; - x0r = x1r - x3i; - x0i = x1i - x3r; - a[j3] = wk3r * x0r + wk3i * x0i; - a[j3 + 1] = wk3r * x0i - wk3i * x0r; - x0r = y1r - y3i; - x0i = y1i - y3r; - a[j3 + 2] = wd3r * x0r + wd3i * x0i; - a[j3 + 3] = wd3r * x0i - wd3i * x0r; - j0 = m - j; - j1 = j0 + m; - j2 = j1 + m; - j3 = j2 + m; - x0r = a[j0] + a[j2]; - x0i = -a[j0 + 1] - a[j2 + 1]; - x1r = a[j0] - a[j2]; - x1i = -a[j0 + 1] + a[j2 + 1]; - y0r = a[j0 - 2] + a[j2 - 2]; - y0i = -a[j0 - 1] - a[j2 - 1]; - y1r = a[j0 - 2] - a[j2 - 2]; - y1i = -a[j0 - 1] + a[j2 - 1]; - x2r = a[j1] + a[j3]; - x2i = a[j1 + 1] + a[j3 + 1]; - x3r = a[j1] - a[j3]; - x3i = a[j1 + 1] - a[j3 + 1]; - y2r = a[j1 - 2] + a[j3 - 2]; - y2i = a[j1 - 1] + a[j3 - 1]; - y3r = a[j1 - 2] - a[j3 - 2]; - y3i = a[j1 - 1] - a[j3 - 1]; - a[j0] = x0r + x2r; - a[j0 + 1] = x0i - x2i; - a[j0 - 2] = y0r + y2r; - a[j0 - 1] = y0i - y2i; - a[j1] = x0r - x2r; - a[j1 + 1] = x0i + x2i; - a[j1 - 2] = y0r - y2r; - a[j1 - 1] = y0i + y2i; - x0r = x1r + x3i; - x0i = x1i + x3r; - a[j2] = wk1i * x0r - wk1r * x0i; - a[j2 + 1] = wk1i * x0i + wk1r * x0r; - x0r = y1r + y3i; - x0i = y1i + y3r; - a[j2 - 2] = wd1i * x0r - wd1r * x0i; - a[j2 - 1] = wd1i * x0i + wd1r * x0r; - x0r = x1r - x3i; - x0i = x1i - x3r; - a[j3] = wk3i * x0r + wk3r * x0i; - a[j3 + 1] = wk3i * x0i - wk3r * x0r; - x0r = y1r - y3i; - x0i = y1i - y3r; - a[j3 - 2] = wd3i * x0r + wd3r * x0i; - a[j3 - 1] = wd3i * x0i - wd3r * x0r; - } - wk1r = csc1 * (wd1r + wn4r); - wk1i = csc1 * (wd1i + wn4r); - wk3r = csc3 * (wd3r - wn4r); - wk3i = csc3 * (wd3i - wn4r); - j0 = mh; - j1 = j0 + m; - j2 = j1 + m; - j3 = j2 + m; - x0r = a[j0 - 2] + a[j2 - 2]; - x0i = -a[j0 - 1] - a[j2 - 1]; - x1r = a[j0 - 2] - a[j2 - 2]; - x1i = -a[j0 - 1] + a[j2 - 1]; - x2r = a[j1 - 2] + a[j3 - 2]; - x2i = a[j1 - 1] + a[j3 - 1]; - x3r = a[j1 - 2] - a[j3 - 2]; - x3i = a[j1 - 1] - a[j3 - 1]; - a[j0 - 2] = x0r + x2r; - a[j0 - 1] = x0i - x2i; - a[j1 - 2] = x0r - x2r; - a[j1 - 1] = x0i + x2i; - x0r = x1r + x3i; - x0i = x1i + x3r; - a[j2 - 2] = wk1r * x0r - wk1i * x0i; - a[j2 - 1] = wk1r * x0i + wk1i * x0r; - x0r = x1r - x3i; - x0i = x1i - x3r; - a[j3 - 2] = wk3r * x0r + wk3i * x0i; - a[j3 - 1] = wk3r * x0i - wk3i * x0r; - x0r = a[j0] + a[j2]; - x0i = -a[j0 + 1] - a[j2 + 1]; - x1r = a[j0] - a[j2]; - x1i = -a[j0 + 1] + a[j2 + 1]; - x2r = a[j1] + a[j3]; - x2i = a[j1 + 1] + a[j3 + 1]; - x3r = a[j1] - a[j3]; - x3i = a[j1 + 1] - a[j3 + 1]; - a[j0] = x0r + x2r; - a[j0 + 1] = x0i - x2i; - a[j1] = x0r - x2r; - a[j1 + 1] = x0i + x2i; - x0r = x1r + x3i; - x0i = x1i + x3r; - a[j2] = wn4r * (x0r - x0i); - a[j2 + 1] = wn4r * (x0i + x0r); - x0r = x1r - x3i; - x0i = x1i - x3r; - a[j3] = -wn4r * (x0r + x0i); - a[j3 + 1] = -wn4r * (x0i - x0r); - x0r = a[j0 + 2] + a[j2 + 2]; - x0i = -a[j0 + 3] - a[j2 + 3]; - x1r = a[j0 + 2] - a[j2 + 2]; - x1i = -a[j0 + 3] + a[j2 + 3]; - x2r = a[j1 + 2] + a[j3 + 2]; - x2i = a[j1 + 3] + a[j3 + 3]; - x3r = a[j1 + 2] - a[j3 + 2]; - x3i = a[j1 + 3] - a[j3 + 3]; - a[j0 + 2] = x0r + x2r; - a[j0 + 3] = x0i - x2i; - a[j1 + 2] = x0r - x2r; - a[j1 + 3] = x0i + x2i; - x0r = x1r + x3i; - x0i = x1i + x3r; - a[j2 + 2] = wk1i * x0r - wk1r * x0i; - a[j2 + 3] = wk1i * x0i + wk1r * x0r; - x0r = x1r - x3i; - x0i = x1i - x3r; - a[j3 + 2] = wk3i * x0r + wk3r * x0i; - a[j3 + 3] = wk3i * x0i - wk3r * x0r; -} - - -#ifdef USE_CDFT_THREADS -struct cdft_arg_st -{ - int n0; - int n; - double *a; - int nw; - double *w; -}; -typedef struct cdft_arg_st cdft_arg_t; - - -void cftrec4_th(int n, double *a, int nw, double *w) -{ - void *cftrec1_th(void *p); - void *cftrec2_th(void *p); - int i, idiv4, m, nthread; - cdft_thread_t th[4]; - cdft_arg_t ag[4]; - - nthread = 2; - idiv4 = 0; - m = n >> 1; - if (n > CDFT_4THREADS_BEGIN_N) - { - nthread = 4; - idiv4 = 1; - m >>= 1; - } - for (i = 0; i < nthread; i++) - { - ag[i].n0 = n; - ag[i].n = m; - ag[i].a = &a[i * m]; - ag[i].nw = nw; - ag[i].w = w; - if (i != idiv4) - { - cdft_thread_create(&th[i], cftrec1_th, &ag[i]); - } - else - { - cdft_thread_create(&th[i], cftrec2_th, &ag[i]); - } - } - for (i = 0; i < nthread; i++) - { - cdft_thread_wait(th[i]); - } -} - - -void *cftrec1_th(void *p) -{ - int cfttree(int n, int j, int k, double *a, int nw, double *w); - void cftleaf(int n, int isplt, double *a, int nw, double *w); - void cftmdl1(int n, double *a, double *w); - int isplt, j, k, m, n, n0, nw; - double *a, *w; - - n0 = ((cdft_arg_t *) p)->n0; - n = ((cdft_arg_t *) p)->n; - a = ((cdft_arg_t *) p)->a; - nw = ((cdft_arg_t *) p)->nw; - w = ((cdft_arg_t *) p)->w; - m = n0; - while (m > 512) - { - m >>= 2; - cftmdl1(m, &a[n - m], &w[nw - (m >> 1)]); - } - cftleaf(m, 1, &a[n - m], nw, w); - k = 0; - for (j = n - m; j > 0; j -= m) - { - k++; - isplt = cfttree(m, j, k, a, nw, w); - cftleaf(m, isplt, &a[j - m], nw, w); - } - return (void *) 0; -} - - -void *cftrec2_th(void *p) -{ - int cfttree(int n, int j, int k, double *a, int nw, double *w); - void cftleaf(int n, int isplt, double *a, int nw, double *w); - void cftmdl2(int n, double *a, double *w); - int isplt, j, k, m, n, n0, nw; - double *a, *w; - - n0 = ((cdft_arg_t *) p)->n0; - n = ((cdft_arg_t *) p)->n; - a = ((cdft_arg_t *) p)->a; - nw = ((cdft_arg_t *) p)->nw; - w = ((cdft_arg_t *) p)->w; - k = 1; - m = n0; - while (m > 512) - { - m >>= 2; - k <<= 2; - cftmdl2(m, &a[n - m], &w[nw - m]); - } - cftleaf(m, 0, &a[n - m], nw, w); - k >>= 1; - for (j = n - m; j > 0; j -= m) - { - k++; - isplt = cfttree(m, j, k, a, nw, w); - cftleaf(m, isplt, &a[j - m], nw, w); - } - return (void *) 0; -} -#endif /* USE_CDFT_THREADS */ - - -void cftrec4(int n, double *a, int nw, double *w) -{ - int cfttree(int n, int j, int k, double *a, int nw, double *w); - void cftleaf(int n, int isplt, double *a, int nw, double *w); - void cftmdl1(int n, double *a, double *w); - int isplt, j, k, m; - - m = n; - while (m > 512) - { - m >>= 2; - cftmdl1(m, &a[n - m], &w[nw - (m >> 1)]); - } - cftleaf(m, 1, &a[n - m], nw, w); - k = 0; - for (j = n - m; j > 0; j -= m) - { - k++; - isplt = cfttree(m, j, k, a, nw, w); - cftleaf(m, isplt, &a[j - m], nw, w); - } -} - - -int cfttree(int n, int j, int k, double *a, int nw, double *w) -{ - void cftmdl1(int n, double *a, double *w); - void cftmdl2(int n, double *a, double *w); - int i, isplt, m; - - if ((k & 3) != 0) - { - isplt = k & 1; - if (isplt != 0) - { - cftmdl1(n, &a[j - n], &w[nw - (n >> 1)]); - } - else - { - cftmdl2(n, &a[j - n], &w[nw - n]); - } - } - else - { - m = n; - for (i = k; (i & 3) == 0; i >>= 2) - { - m <<= 2; - } - isplt = i & 1; - if (isplt != 0) - { - while (m > 128) - { - cftmdl1(m, &a[j - m], &w[nw - (m >> 1)]); - m >>= 2; - } - } - else - { - while (m > 128) - { - cftmdl2(m, &a[j - m], &w[nw - m]); - m >>= 2; - } - } - } - return isplt; -} - - -void cftleaf(int n, int isplt, double *a, int nw, double *w) -{ - void cftmdl1(int n, double *a, double *w); - void cftmdl2(int n, double *a, double *w); - void cftf161(double *a, double *w); - void cftf162(double *a, double *w); - void cftf081(double *a, double *w); - void cftf082(double *a, double *w); - - if (n == 512) - { - cftmdl1(128, a, &w[nw - 64]); - cftf161(a, &w[nw - 8]); - cftf162(&a[32], &w[nw - 32]); - cftf161(&a[64], &w[nw - 8]); - cftf161(&a[96], &w[nw - 8]); - cftmdl2(128, &a[128], &w[nw - 128]); - cftf161(&a[128], &w[nw - 8]); - cftf162(&a[160], &w[nw - 32]); - cftf161(&a[192], &w[nw - 8]); - cftf162(&a[224], &w[nw - 32]); - cftmdl1(128, &a[256], &w[nw - 64]); - cftf161(&a[256], &w[nw - 8]); - cftf162(&a[288], &w[nw - 32]); - cftf161(&a[320], &w[nw - 8]); - cftf161(&a[352], &w[nw - 8]); - if (isplt != 0) - { - cftmdl1(128, &a[384], &w[nw - 64]); - cftf161(&a[480], &w[nw - 8]); - } - else - { - cftmdl2(128, &a[384], &w[nw - 128]); - cftf162(&a[480], &w[nw - 32]); - } - cftf161(&a[384], &w[nw - 8]); - cftf162(&a[416], &w[nw - 32]); - cftf161(&a[448], &w[nw - 8]); - } - else - { - cftmdl1(64, a, &w[nw - 32]); - cftf081(a, &w[nw - 8]); - cftf082(&a[16], &w[nw - 8]); - cftf081(&a[32], &w[nw - 8]); - cftf081(&a[48], &w[nw - 8]); - cftmdl2(64, &a[64], &w[nw - 64]); - cftf081(&a[64], &w[nw - 8]); - cftf082(&a[80], &w[nw - 8]); - cftf081(&a[96], &w[nw - 8]); - cftf082(&a[112], &w[nw - 8]); - cftmdl1(64, &a[128], &w[nw - 32]); - cftf081(&a[128], &w[nw - 8]); - cftf082(&a[144], &w[nw - 8]); - cftf081(&a[160], &w[nw - 8]); - cftf081(&a[176], &w[nw - 8]); - if (isplt != 0) - { - cftmdl1(64, &a[192], &w[nw - 32]); - cftf081(&a[240], &w[nw - 8]); - } - else - { - cftmdl2(64, &a[192], &w[nw - 64]); - cftf082(&a[240], &w[nw - 8]); - } - cftf081(&a[192], &w[nw - 8]); - cftf082(&a[208], &w[nw - 8]); - cftf081(&a[224], &w[nw - 8]); - } -} - - -void cftmdl1(int n, double *a, double *w) -{ - int j, j0, j1, j2, j3, k, m, mh; - double wn4r, wk1r, wk1i, wk3r, wk3i; - double x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i; - - mh = n >> 3; - m = 2 * mh; - j1 = m; - j2 = j1 + m; - j3 = j2 + m; - x0r = a[0] + a[j2]; - x0i = a[1] + a[j2 + 1]; - x1r = a[0] - a[j2]; - x1i = a[1] - a[j2 + 1]; - x2r = a[j1] + a[j3]; - x2i = a[j1 + 1] + a[j3 + 1]; - x3r = a[j1] - a[j3]; - x3i = a[j1 + 1] - a[j3 + 1]; - a[0] = x0r + x2r; - a[1] = x0i + x2i; - a[j1] = x0r - x2r; - a[j1 + 1] = x0i - x2i; - a[j2] = x1r - x3i; - a[j2 + 1] = x1i + x3r; - a[j3] = x1r + x3i; - a[j3 + 1] = x1i - x3r; - wn4r = w[1]; - k = 0; - for (j = 2; j < mh; j += 2) - { - k += 4; - wk1r = w[k]; - wk1i = w[k + 1]; - wk3r = w[k + 2]; - wk3i = w[k + 3]; - j1 = j + m; - j2 = j1 + m; - j3 = j2 + m; - x0r = a[j] + a[j2]; - x0i = a[j + 1] + a[j2 + 1]; - x1r = a[j] - a[j2]; - x1i = a[j + 1] - a[j2 + 1]; - x2r = a[j1] + a[j3]; - x2i = a[j1 + 1] + a[j3 + 1]; - x3r = a[j1] - a[j3]; - x3i = a[j1 + 1] - a[j3 + 1]; - a[j] = x0r + x2r; - a[j + 1] = x0i + x2i; - a[j1] = x0r - x2r; - a[j1 + 1] = x0i - x2i; - x0r = x1r - x3i; - x0i = x1i + x3r; - a[j2] = wk1r * x0r - wk1i * x0i; - a[j2 + 1] = wk1r * x0i + wk1i * x0r; - x0r = x1r + x3i; - x0i = x1i - x3r; - a[j3] = wk3r * x0r + wk3i * x0i; - a[j3 + 1] = wk3r * x0i - wk3i * x0r; - j0 = m - j; - j1 = j0 + m; - j2 = j1 + m; - j3 = j2 + m; - x0r = a[j0] + a[j2]; - x0i = a[j0 + 1] + a[j2 + 1]; - x1r = a[j0] - a[j2]; - x1i = a[j0 + 1] - a[j2 + 1]; - x2r = a[j1] + a[j3]; - x2i = a[j1 + 1] + a[j3 + 1]; - x3r = a[j1] - a[j3]; - x3i = a[j1 + 1] - a[j3 + 1]; - a[j0] = x0r + x2r; - a[j0 + 1] = x0i + x2i; - a[j1] = x0r - x2r; - a[j1 + 1] = x0i - x2i; - x0r = x1r - x3i; - x0i = x1i + x3r; - a[j2] = wk1i * x0r - wk1r * x0i; - a[j2 + 1] = wk1i * x0i + wk1r * x0r; - x0r = x1r + x3i; - x0i = x1i - x3r; - a[j3] = wk3i * x0r + wk3r * x0i; - a[j3 + 1] = wk3i * x0i - wk3r * x0r; - } - j0 = mh; - j1 = j0 + m; - j2 = j1 + m; - j3 = j2 + m; - x0r = a[j0] + a[j2]; - x0i = a[j0 + 1] + a[j2 + 1]; - x1r = a[j0] - a[j2]; - x1i = a[j0 + 1] - a[j2 + 1]; - x2r = a[j1] + a[j3]; - x2i = a[j1 + 1] + a[j3 + 1]; - x3r = a[j1] - a[j3]; - x3i = a[j1 + 1] - a[j3 + 1]; - a[j0] = x0r + x2r; - a[j0 + 1] = x0i + x2i; - a[j1] = x0r - x2r; - a[j1 + 1] = x0i - x2i; - x0r = x1r - x3i; - x0i = x1i + x3r; - a[j2] = wn4r * (x0r - x0i); - a[j2 + 1] = wn4r * (x0i + x0r); - x0r = x1r + x3i; - x0i = x1i - x3r; - a[j3] = -wn4r * (x0r + x0i); - a[j3 + 1] = -wn4r * (x0i - x0r); -} - - -void cftmdl2(int n, double *a, double *w) -{ - int j, j0, j1, j2, j3, k, kr, m, mh; - double wn4r, wk1r, wk1i, wk3r, wk3i, wd1r, wd1i, wd3r, wd3i; - double x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i, y0r, y0i, y2r, y2i; - - mh = n >> 3; - m = 2 * mh; - wn4r = w[1]; - j1 = m; - j2 = j1 + m; - j3 = j2 + m; - x0r = a[0] - a[j2 + 1]; - x0i = a[1] + a[j2]; - x1r = a[0] + a[j2 + 1]; - x1i = a[1] - a[j2]; - x2r = a[j1] - a[j3 + 1]; - x2i = a[j1 + 1] + a[j3]; - x3r = a[j1] + a[j3 + 1]; - x3i = a[j1 + 1] - a[j3]; - y0r = wn4r * (x2r - x2i); - y0i = wn4r * (x2i + x2r); - a[0] = x0r + y0r; - a[1] = x0i + y0i; - a[j1] = x0r - y0r; - a[j1 + 1] = x0i - y0i; - y0r = wn4r * (x3r - x3i); - y0i = wn4r * (x3i + x3r); - a[j2] = x1r - y0i; - a[j2 + 1] = x1i + y0r; - a[j3] = x1r + y0i; - a[j3 + 1] = x1i - y0r; - k = 0; - kr = 2 * m; - for (j = 2; j < mh; j += 2) - { - k += 4; - wk1r = w[k]; - wk1i = w[k + 1]; - wk3r = w[k + 2]; - wk3i = w[k + 3]; - kr -= 4; - wd1i = w[kr]; - wd1r = w[kr + 1]; - wd3i = w[kr + 2]; - wd3r = w[kr + 3]; - j1 = j + m; - j2 = j1 + m; - j3 = j2 + m; - x0r = a[j] - a[j2 + 1]; - x0i = a[j + 1] + a[j2]; - x1r = a[j] + a[j2 + 1]; - x1i = a[j + 1] - a[j2]; - x2r = a[j1] - a[j3 + 1]; - x2i = a[j1 + 1] + a[j3]; - x3r = a[j1] + a[j3 + 1]; - x3i = a[j1 + 1] - a[j3]; - y0r = wk1r * x0r - wk1i * x0i; - y0i = wk1r * x0i + wk1i * x0r; - y2r = wd1r * x2r - wd1i * x2i; - y2i = wd1r * x2i + wd1i * x2r; - a[j] = y0r + y2r; - a[j + 1] = y0i + y2i; - a[j1] = y0r - y2r; - a[j1 + 1] = y0i - y2i; - y0r = wk3r * x1r + wk3i * x1i; - y0i = wk3r * x1i - wk3i * x1r; - y2r = wd3r * x3r + wd3i * x3i; - y2i = wd3r * x3i - wd3i * x3r; - a[j2] = y0r + y2r; - a[j2 + 1] = y0i + y2i; - a[j3] = y0r - y2r; - a[j3 + 1] = y0i - y2i; - j0 = m - j; - j1 = j0 + m; - j2 = j1 + m; - j3 = j2 + m; - x0r = a[j0] - a[j2 + 1]; - x0i = a[j0 + 1] + a[j2]; - x1r = a[j0] + a[j2 + 1]; - x1i = a[j0 + 1] - a[j2]; - x2r = a[j1] - a[j3 + 1]; - x2i = a[j1 + 1] + a[j3]; - x3r = a[j1] + a[j3 + 1]; - x3i = a[j1 + 1] - a[j3]; - y0r = wd1i * x0r - wd1r * x0i; - y0i = wd1i * x0i + wd1r * x0r; - y2r = wk1i * x2r - wk1r * x2i; - y2i = wk1i * x2i + wk1r * x2r; - a[j0] = y0r + y2r; - a[j0 + 1] = y0i + y2i; - a[j1] = y0r - y2r; - a[j1 + 1] = y0i - y2i; - y0r = wd3i * x1r + wd3r * x1i; - y0i = wd3i * x1i - wd3r * x1r; - y2r = wk3i * x3r + wk3r * x3i; - y2i = wk3i * x3i - wk3r * x3r; - a[j2] = y0r + y2r; - a[j2 + 1] = y0i + y2i; - a[j3] = y0r - y2r; - a[j3 + 1] = y0i - y2i; - } - wk1r = w[m]; - wk1i = w[m + 1]; - j0 = mh; - j1 = j0 + m; - j2 = j1 + m; - j3 = j2 + m; - x0r = a[j0] - a[j2 + 1]; - x0i = a[j0 + 1] + a[j2]; - x1r = a[j0] + a[j2 + 1]; - x1i = a[j0 + 1] - a[j2]; - x2r = a[j1] - a[j3 + 1]; - x2i = a[j1 + 1] + a[j3]; - x3r = a[j1] + a[j3 + 1]; - x3i = a[j1 + 1] - a[j3]; - y0r = wk1r * x0r - wk1i * x0i; - y0i = wk1r * x0i + wk1i * x0r; - y2r = wk1i * x2r - wk1r * x2i; - y2i = wk1i * x2i + wk1r * x2r; - a[j0] = y0r + y2r; - a[j0 + 1] = y0i + y2i; - a[j1] = y0r - y2r; - a[j1 + 1] = y0i - y2i; - y0r = wk1i * x1r - wk1r * x1i; - y0i = wk1i * x1i + wk1r * x1r; - y2r = wk1r * x3r - wk1i * x3i; - y2i = wk1r * x3i + wk1i * x3r; - a[j2] = y0r - y2r; - a[j2 + 1] = y0i - y2i; - a[j3] = y0r + y2r; - a[j3 + 1] = y0i + y2i; -} - - -void cftfx41(int n, double *a, int nw, double *w) -{ - void cftf161(double *a, double *w); - void cftf162(double *a, double *w); - void cftf081(double *a, double *w); - void cftf082(double *a, double *w); - - if (n == 128) - { - cftf161(a, &w[nw - 8]); - cftf162(&a[32], &w[nw - 32]); - cftf161(&a[64], &w[nw - 8]); - cftf161(&a[96], &w[nw - 8]); - } - else - { - cftf081(a, &w[nw - 8]); - cftf082(&a[16], &w[nw - 8]); - cftf081(&a[32], &w[nw - 8]); - cftf081(&a[48], &w[nw - 8]); - } -} - - -void cftf161(double *a, double *w) -{ - double wn4r, wk1r, wk1i, - x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i, - y0r, y0i, y1r, y1i, y2r, y2i, y3r, y3i, - y4r, y4i, y5r, y5i, y6r, y6i, y7r, y7i, - y8r, y8i, y9r, y9i, y10r, y10i, y11r, y11i, - y12r, y12i, y13r, y13i, y14r, y14i, y15r, y15i; - - wn4r = w[1]; - wk1r = w[2]; - wk1i = w[3]; - x0r = a[0] + a[16]; - x0i = a[1] + a[17]; - x1r = a[0] - a[16]; - x1i = a[1] - a[17]; - x2r = a[8] + a[24]; - x2i = a[9] + a[25]; - x3r = a[8] - a[24]; - x3i = a[9] - a[25]; - y0r = x0r + x2r; - y0i = x0i + x2i; - y4r = x0r - x2r; - y4i = x0i - x2i; - y8r = x1r - x3i; - y8i = x1i + x3r; - y12r = x1r + x3i; - y12i = x1i - x3r; - x0r = a[2] + a[18]; - x0i = a[3] + a[19]; - x1r = a[2] - a[18]; - x1i = a[3] - a[19]; - x2r = a[10] + a[26]; - x2i = a[11] + a[27]; - x3r = a[10] - a[26]; - x3i = a[11] - a[27]; - y1r = x0r + x2r; - y1i = x0i + x2i; - y5r = x0r - x2r; - y5i = x0i - x2i; - x0r = x1r - x3i; - x0i = x1i + x3r; - y9r = wk1r * x0r - wk1i * x0i; - y9i = wk1r * x0i + wk1i * x0r; - x0r = x1r + x3i; - x0i = x1i - x3r; - y13r = wk1i * x0r - wk1r * x0i; - y13i = wk1i * x0i + wk1r * x0r; - x0r = a[4] + a[20]; - x0i = a[5] + a[21]; - x1r = a[4] - a[20]; - x1i = a[5] - a[21]; - x2r = a[12] + a[28]; - x2i = a[13] + a[29]; - x3r = a[12] - a[28]; - x3i = a[13] - a[29]; - y2r = x0r + x2r; - y2i = x0i + x2i; - y6r = x0r - x2r; - y6i = x0i - x2i; - x0r = x1r - x3i; - x0i = x1i + x3r; - y10r = wn4r * (x0r - x0i); - y10i = wn4r * (x0i + x0r); - x0r = x1r + x3i; - x0i = x1i - x3r; - y14r = wn4r * (x0r + x0i); - y14i = wn4r * (x0i - x0r); - x0r = a[6] + a[22]; - x0i = a[7] + a[23]; - x1r = a[6] - a[22]; - x1i = a[7] - a[23]; - x2r = a[14] + a[30]; - x2i = a[15] + a[31]; - x3r = a[14] - a[30]; - x3i = a[15] - a[31]; - y3r = x0r + x2r; - y3i = x0i + x2i; - y7r = x0r - x2r; - y7i = x0i - x2i; - x0r = x1r - x3i; - x0i = x1i + x3r; - y11r = wk1i * x0r - wk1r * x0i; - y11i = wk1i * x0i + wk1r * x0r; - x0r = x1r + x3i; - x0i = x1i - x3r; - y15r = wk1r * x0r - wk1i * x0i; - y15i = wk1r * x0i + wk1i * x0r; - x0r = y12r - y14r; - x0i = y12i - y14i; - x1r = y12r + y14r; - x1i = y12i + y14i; - x2r = y13r - y15r; - x2i = y13i - y15i; - x3r = y13r + y15r; - x3i = y13i + y15i; - a[24] = x0r + x2r; - a[25] = x0i + x2i; - a[26] = x0r - x2r; - a[27] = x0i - x2i; - a[28] = x1r - x3i; - a[29] = x1i + x3r; - a[30] = x1r + x3i; - a[31] = x1i - x3r; - x0r = y8r + y10r; - x0i = y8i + y10i; - x1r = y8r - y10r; - x1i = y8i - y10i; - x2r = y9r + y11r; - x2i = y9i + y11i; - x3r = y9r - y11r; - x3i = y9i - y11i; - a[16] = x0r + x2r; - a[17] = x0i + x2i; - a[18] = x0r - x2r; - a[19] = x0i - x2i; - a[20] = x1r - x3i; - a[21] = x1i + x3r; - a[22] = x1r + x3i; - a[23] = x1i - x3r; - x0r = y5r - y7i; - x0i = y5i + y7r; - x2r = wn4r * (x0r - x0i); - x2i = wn4r * (x0i + x0r); - x0r = y5r + y7i; - x0i = y5i - y7r; - x3r = wn4r * (x0r - x0i); - x3i = wn4r * (x0i + x0r); - x0r = y4r - y6i; - x0i = y4i + y6r; - x1r = y4r + y6i; - x1i = y4i - y6r; - a[8] = x0r + x2r; - a[9] = x0i + x2i; - a[10] = x0r - x2r; - a[11] = x0i - x2i; - a[12] = x1r - x3i; - a[13] = x1i + x3r; - a[14] = x1r + x3i; - a[15] = x1i - x3r; - x0r = y0r + y2r; - x0i = y0i + y2i; - x1r = y0r - y2r; - x1i = y0i - y2i; - x2r = y1r + y3r; - x2i = y1i + y3i; - x3r = y1r - y3r; - x3i = y1i - y3i; - a[0] = x0r + x2r; - a[1] = x0i + x2i; - a[2] = x0r - x2r; - a[3] = x0i - x2i; - a[4] = x1r - x3i; - a[5] = x1i + x3r; - a[6] = x1r + x3i; - a[7] = x1i - x3r; -} - - -void cftf162(double *a, double *w) -{ - double wn4r, wk1r, wk1i, wk2r, wk2i, wk3r, wk3i, - x0r, x0i, x1r, x1i, x2r, x2i, - y0r, y0i, y1r, y1i, y2r, y2i, y3r, y3i, - y4r, y4i, y5r, y5i, y6r, y6i, y7r, y7i, - y8r, y8i, y9r, y9i, y10r, y10i, y11r, y11i, - y12r, y12i, y13r, y13i, y14r, y14i, y15r, y15i; - - wn4r = w[1]; - wk1r = w[4]; - wk1i = w[5]; - wk3r = w[6]; - wk3i = -w[7]; - wk2r = w[8]; - wk2i = w[9]; - x1r = a[0] - a[17]; - x1i = a[1] + a[16]; - x0r = a[8] - a[25]; - x0i = a[9] + a[24]; - x2r = wn4r * (x0r - x0i); - x2i = wn4r * (x0i + x0r); - y0r = x1r + x2r; - y0i = x1i + x2i; - y4r = x1r - x2r; - y4i = x1i - x2i; - x1r = a[0] + a[17]; - x1i = a[1] - a[16]; - x0r = a[8] + a[25]; - x0i = a[9] - a[24]; - x2r = wn4r * (x0r - x0i); - x2i = wn4r * (x0i + x0r); - y8r = x1r - x2i; - y8i = x1i + x2r; - y12r = x1r + x2i; - y12i = x1i - x2r; - x0r = a[2] - a[19]; - x0i = a[3] + a[18]; - x1r = wk1r * x0r - wk1i * x0i; - x1i = wk1r * x0i + wk1i * x0r; - x0r = a[10] - a[27]; - x0i = a[11] + a[26]; - x2r = wk3i * x0r - wk3r * x0i; - x2i = wk3i * x0i + wk3r * x0r; - y1r = x1r + x2r; - y1i = x1i + x2i; - y5r = x1r - x2r; - y5i = x1i - x2i; - x0r = a[2] + a[19]; - x0i = a[3] - a[18]; - x1r = wk3r * x0r - wk3i * x0i; - x1i = wk3r * x0i + wk3i * x0r; - x0r = a[10] + a[27]; - x0i = a[11] - a[26]; - x2r = wk1r * x0r + wk1i * x0i; - x2i = wk1r * x0i - wk1i * x0r; - y9r = x1r - x2r; - y9i = x1i - x2i; - y13r = x1r + x2r; - y13i = x1i + x2i; - x0r = a[4] - a[21]; - x0i = a[5] + a[20]; - x1r = wk2r * x0r - wk2i * x0i; - x1i = wk2r * x0i + wk2i * x0r; - x0r = a[12] - a[29]; - x0i = a[13] + a[28]; - x2r = wk2i * x0r - wk2r * x0i; - x2i = wk2i * x0i + wk2r * x0r; - y2r = x1r + x2r; - y2i = x1i + x2i; - y6r = x1r - x2r; - y6i = x1i - x2i; - x0r = a[4] + a[21]; - x0i = a[5] - a[20]; - x1r = wk2i * x0r - wk2r * x0i; - x1i = wk2i * x0i + wk2r * x0r; - x0r = a[12] + a[29]; - x0i = a[13] - a[28]; - x2r = wk2r * x0r - wk2i * x0i; - x2i = wk2r * x0i + wk2i * x0r; - y10r = x1r - x2r; - y10i = x1i - x2i; - y14r = x1r + x2r; - y14i = x1i + x2i; - x0r = a[6] - a[23]; - x0i = a[7] + a[22]; - x1r = wk3r * x0r - wk3i * x0i; - x1i = wk3r * x0i + wk3i * x0r; - x0r = a[14] - a[31]; - x0i = a[15] + a[30]; - x2r = wk1i * x0r - wk1r * x0i; - x2i = wk1i * x0i + wk1r * x0r; - y3r = x1r + x2r; - y3i = x1i + x2i; - y7r = x1r - x2r; - y7i = x1i - x2i; - x0r = a[6] + a[23]; - x0i = a[7] - a[22]; - x1r = wk1i * x0r + wk1r * x0i; - x1i = wk1i * x0i - wk1r * x0r; - x0r = a[14] + a[31]; - x0i = a[15] - a[30]; - x2r = wk3i * x0r - wk3r * x0i; - x2i = wk3i * x0i + wk3r * x0r; - y11r = x1r + x2r; - y11i = x1i + x2i; - y15r = x1r - x2r; - y15i = x1i - x2i; - x1r = y0r + y2r; - x1i = y0i + y2i; - x2r = y1r + y3r; - x2i = y1i + y3i; - a[0] = x1r + x2r; - a[1] = x1i + x2i; - a[2] = x1r - x2r; - a[3] = x1i - x2i; - x1r = y0r - y2r; - x1i = y0i - y2i; - x2r = y1r - y3r; - x2i = y1i - y3i; - a[4] = x1r - x2i; - a[5] = x1i + x2r; - a[6] = x1r + x2i; - a[7] = x1i - x2r; - x1r = y4r - y6i; - x1i = y4i + y6r; - x0r = y5r - y7i; - x0i = y5i + y7r; - x2r = wn4r * (x0r - x0i); - x2i = wn4r * (x0i + x0r); - a[8] = x1r + x2r; - a[9] = x1i + x2i; - a[10] = x1r - x2r; - a[11] = x1i - x2i; - x1r = y4r + y6i; - x1i = y4i - y6r; - x0r = y5r + y7i; - x0i = y5i - y7r; - x2r = wn4r * (x0r - x0i); - x2i = wn4r * (x0i + x0r); - a[12] = x1r - x2i; - a[13] = x1i + x2r; - a[14] = x1r + x2i; - a[15] = x1i - x2r; - x1r = y8r + y10r; - x1i = y8i + y10i; - x2r = y9r - y11r; - x2i = y9i - y11i; - a[16] = x1r + x2r; - a[17] = x1i + x2i; - a[18] = x1r - x2r; - a[19] = x1i - x2i; - x1r = y8r - y10r; - x1i = y8i - y10i; - x2r = y9r + y11r; - x2i = y9i + y11i; - a[20] = x1r - x2i; - a[21] = x1i + x2r; - a[22] = x1r + x2i; - a[23] = x1i - x2r; - x1r = y12r - y14i; - x1i = y12i + y14r; - x0r = y13r + y15i; - x0i = y13i - y15r; - x2r = wn4r * (x0r - x0i); - x2i = wn4r * (x0i + x0r); - a[24] = x1r + x2r; - a[25] = x1i + x2i; - a[26] = x1r - x2r; - a[27] = x1i - x2i; - x1r = y12r + y14i; - x1i = y12i - y14r; - x0r = y13r - y15i; - x0i = y13i + y15r; - x2r = wn4r * (x0r - x0i); - x2i = wn4r * (x0i + x0r); - a[28] = x1r - x2i; - a[29] = x1i + x2r; - a[30] = x1r + x2i; - a[31] = x1i - x2r; -} - - -void cftf081(double *a, double *w) -{ - double wn4r, x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i, - y0r, y0i, y1r, y1i, y2r, y2i, y3r, y3i, - y4r, y4i, y5r, y5i, y6r, y6i, y7r, y7i; - - wn4r = w[1]; - x0r = a[0] + a[8]; - x0i = a[1] + a[9]; - x1r = a[0] - a[8]; - x1i = a[1] - a[9]; - x2r = a[4] + a[12]; - x2i = a[5] + a[13]; - x3r = a[4] - a[12]; - x3i = a[5] - a[13]; - y0r = x0r + x2r; - y0i = x0i + x2i; - y2r = x0r - x2r; - y2i = x0i - x2i; - y1r = x1r - x3i; - y1i = x1i + x3r; - y3r = x1r + x3i; - y3i = x1i - x3r; - x0r = a[2] + a[10]; - x0i = a[3] + a[11]; - x1r = a[2] - a[10]; - x1i = a[3] - a[11]; - x2r = a[6] + a[14]; - x2i = a[7] + a[15]; - x3r = a[6] - a[14]; - x3i = a[7] - a[15]; - y4r = x0r + x2r; - y4i = x0i + x2i; - y6r = x0r - x2r; - y6i = x0i - x2i; - x0r = x1r - x3i; - x0i = x1i + x3r; - x2r = x1r + x3i; - x2i = x1i - x3r; - y5r = wn4r * (x0r - x0i); - y5i = wn4r * (x0r + x0i); - y7r = wn4r * (x2r - x2i); - y7i = wn4r * (x2r + x2i); - a[8] = y1r + y5r; - a[9] = y1i + y5i; - a[10] = y1r - y5r; - a[11] = y1i - y5i; - a[12] = y3r - y7i; - a[13] = y3i + y7r; - a[14] = y3r + y7i; - a[15] = y3i - y7r; - a[0] = y0r + y4r; - a[1] = y0i + y4i; - a[2] = y0r - y4r; - a[3] = y0i - y4i; - a[4] = y2r - y6i; - a[5] = y2i + y6r; - a[6] = y2r + y6i; - a[7] = y2i - y6r; -} - - -void cftf082(double *a, double *w) -{ - double wn4r, wk1r, wk1i, x0r, x0i, x1r, x1i, - y0r, y0i, y1r, y1i, y2r, y2i, y3r, y3i, - y4r, y4i, y5r, y5i, y6r, y6i, y7r, y7i; - - wn4r = w[1]; - wk1r = w[2]; - wk1i = w[3]; - y0r = a[0] - a[9]; - y0i = a[1] + a[8]; - y1r = a[0] + a[9]; - y1i = a[1] - a[8]; - x0r = a[4] - a[13]; - x0i = a[5] + a[12]; - y2r = wn4r * (x0r - x0i); - y2i = wn4r * (x0i + x0r); - x0r = a[4] + a[13]; - x0i = a[5] - a[12]; - y3r = wn4r * (x0r - x0i); - y3i = wn4r * (x0i + x0r); - x0r = a[2] - a[11]; - x0i = a[3] + a[10]; - y4r = wk1r * x0r - wk1i * x0i; - y4i = wk1r * x0i + wk1i * x0r; - x0r = a[2] + a[11]; - x0i = a[3] - a[10]; - y5r = wk1i * x0r - wk1r * x0i; - y5i = wk1i * x0i + wk1r * x0r; - x0r = a[6] - a[15]; - x0i = a[7] + a[14]; - y6r = wk1i * x0r - wk1r * x0i; - y6i = wk1i * x0i + wk1r * x0r; - x0r = a[6] + a[15]; - x0i = a[7] - a[14]; - y7r = wk1r * x0r - wk1i * x0i; - y7i = wk1r * x0i + wk1i * x0r; - x0r = y0r + y2r; - x0i = y0i + y2i; - x1r = y4r + y6r; - x1i = y4i + y6i; - a[0] = x0r + x1r; - a[1] = x0i + x1i; - a[2] = x0r - x1r; - a[3] = x0i - x1i; - x0r = y0r - y2r; - x0i = y0i - y2i; - x1r = y4r - y6r; - x1i = y4i - y6i; - a[4] = x0r - x1i; - a[5] = x0i + x1r; - a[6] = x0r + x1i; - a[7] = x0i - x1r; - x0r = y1r - y3i; - x0i = y1i + y3r; - x1r = y5r - y7r; - x1i = y5i - y7i; - a[8] = x0r + x1r; - a[9] = x0i + x1i; - a[10] = x0r - x1r; - a[11] = x0i - x1i; - x0r = y1r + y3i; - x0i = y1i - y3r; - x1r = y5r + y7r; - x1i = y5i + y7i; - a[12] = x0r - x1i; - a[13] = x0i + x1r; - a[14] = x0r + x1i; - a[15] = x0i - x1r; -} - - -void cftf040(double *a) -{ - double x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i; - - x0r = a[0] + a[4]; - x0i = a[1] + a[5]; - x1r = a[0] - a[4]; - x1i = a[1] - a[5]; - x2r = a[2] + a[6]; - x2i = a[3] + a[7]; - x3r = a[2] - a[6]; - x3i = a[3] - a[7]; - a[0] = x0r + x2r; - a[1] = x0i + x2i; - a[2] = x1r - x3i; - a[3] = x1i + x3r; - a[4] = x0r - x2r; - a[5] = x0i - x2i; - a[6] = x1r + x3i; - a[7] = x1i - x3r; -} - - -void cftb040(double *a) -{ - double x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i; - - x0r = a[0] + a[4]; - x0i = a[1] + a[5]; - x1r = a[0] - a[4]; - x1i = a[1] - a[5]; - x2r = a[2] + a[6]; - x2i = a[3] + a[7]; - x3r = a[2] - a[6]; - x3i = a[3] - a[7]; - a[0] = x0r + x2r; - a[1] = x0i + x2i; - a[2] = x1r + x3i; - a[3] = x1i - x3r; - a[4] = x0r - x2r; - a[5] = x0i - x2i; - a[6] = x1r - x3i; - a[7] = x1i + x3r; -} - - -void cftx020(double *a) -{ - double x0r, x0i; - - x0r = a[0] - a[2]; - x0i = a[1] - a[3]; - a[0] += a[2]; - a[1] += a[3]; - a[2] = x0r; - a[3] = x0i; -} - - -void rftfsub(int n, double *a, int nc, double *c) -{ - int j, k, kk, ks, m; - double wkr, wki, xr, xi, yr, yi; - - m = n >> 1; - ks = 2 * nc / m; - kk = 0; - for (j = 2; j < m; j += 2) - { - k = n - j; - kk += ks; - wkr = 0.5 - c[nc - kk]; - wki = c[kk]; - xr = a[j] - a[k]; - xi = a[j + 1] + a[k + 1]; - yr = wkr * xr - wki * xi; - yi = wkr * xi + wki * xr; - a[j] -= yr; - a[j + 1] -= yi; - a[k] += yr; - a[k + 1] -= yi; - } -} - - -void rftbsub(int n, double *a, int nc, double *c) -{ - int j, k, kk, ks, m; - double wkr, wki, xr, xi, yr, yi; - - m = n >> 1; - ks = 2 * nc / m; - kk = 0; - for (j = 2; j < m; j += 2) - { - k = n - j; - kk += ks; - wkr = 0.5 - c[nc - kk]; - wki = c[kk]; - xr = a[j] - a[k]; - xi = a[j + 1] + a[k + 1]; - yr = wkr * xr + wki * xi; - yi = wkr * xi - wki * xr; - a[j] -= yr; - a[j + 1] -= yi; - a[k] += yr; - a[k + 1] -= yi; - } -} - - -void dctsub(int n, double *a, int nc, double *c) -{ - int j, k, kk, ks, m; - double wkr, wki, xr; - - m = n >> 1; - ks = nc / n; - kk = 0; - for (j = 1; j < m; j++) - { - k = n - j; - kk += ks; - wkr = c[kk] - c[nc - kk]; - wki = c[kk] + c[nc - kk]; - xr = wki * a[j] - wkr * a[k]; - a[j] = wkr * a[j] + wki * a[k]; - a[k] = xr; - } - a[m] *= c[0]; -} - - -void dstsub(int n, double *a, int nc, double *c) -{ - int j, k, kk, ks, m; - double wkr, wki, xr; - - m = n >> 1; - ks = nc / n; - kk = 0; - for (j = 1; j < m; j++) - { - k = n - j; - kk += ks; - wkr = c[kk] - c[nc - kk]; - wki = c[kk] + c[nc - kk]; - xr = wki * a[k] - wkr * a[j]; - a[k] = wkr * a[k] + wki * a[j]; - a[j] = xr; - } - a[m] *= c[0]; -} - diff --git a/src/fftsg.h b/src/fftsg.h deleted file mode 100644 index eb7fc8a..0000000 --- a/src/fftsg.h +++ /dev/null @@ -1,21 +0,0 @@ -/* FFT functions */ -void cdft(int n, int isgn, double *a, int *ip, double *w); -void rdft(int n, int isgn, double *a, int *ip, double *w); -void ddct(int n, int isgn, double *a, int *ip, double *w); -void ddst(int n, int isgn, double *a, int *ip, double *w); -void dfct(int n, double *a, double *t, int *ip, double *w); -void dfst(int n, double *a, double *t, int *ip, double *w); - -/* Auxiliary functions */ -void makewt(int nw, int *ip, double *w); -void bitrv2(int n, int *ip, double *a); -void bitrv2conj(int n, int *ip, double *a); -void cftfsub(int n, double *a, double *w); -void cftbsub(int n, double *a, double *w); -void makect(int nc, int *ip, double *c); -void rftfsub(int n, double *a, int nc, double *c); -void rftbsub(int n, double *a, int nc, double *c); -void dctsub(int n, double *a, int nc, double *c); -void dstsub(int n, double *a, int nc, double *c); -void cft1st(int n, double *a, double *w); -void cftmdl(int n, int l, double *a, double *w); diff --git a/src/helper.c b/src/helper.c index de33ffa..cf63bbc 100644 --- a/src/helper.c +++ b/src/helper.c @@ -23,8 +23,6 @@ /* helper.c: helper functions. */ -#include <config.h> - #include <stdio.h> #include "xtract/libxtract.h" @@ -24,14 +24,14 @@ /* init.c: defines initialisation and free functions. Also contains library constructor routine. */ #ifdef HAVE_CONFIG_H -# include <config.h> +#include <config.h> #endif #include <math.h> #include <stdlib.h> #include <stdio.h> -#include "fftsg.h" +#include "fft.h" #include "xtract/libxtract.h" #include "xtract_window_private.h" @@ -39,6 +39,215 @@ #include "xtract_globals_private.h" + +#ifdef USE_OOURA +void xtract_init_ooura_data(xtract_ooura_data *ooura_data, unsigned int N) +{ + ooura_data->ooura_ip = (int *)calloc((2 + sqrt(N)), sizeof(int)); + ooura_data->ooura_w = (double *)calloc((N - 1), sizeof(double)); + ooura_data->initialised = true; +} + +void xtract_free_ooura_data(xtract_ooura_data *ooura_data) +{ + free(ooura_data->ooura_ip); + free(ooura_data->ooura_w); + ooura_data->ooura_ip = NULL; + ooura_data->ooura_w = NULL; + ooura_data->initialised = false; +} + +int xtract_init_ooura_(int N, int feature_name) +{ + + int M = N >> 1; + + if(feature_name == XTRACT_AUTOCORRELATION_FFT) + { + M = N; /* allow for zero padding */ + } + + switch(feature_name) + { + case XTRACT_SPECTRUM: + if(ooura_data_spectrum.initialised) + { + xtract_free_ooura_data(&ooura_data_spectrum); + } + xtract_init_ooura_data(&ooura_data_spectrum, M); + break; + case XTRACT_AUTOCORRELATION_FFT: + if(ooura_data_autocorrelation_fft.initialised) + { + xtract_free_ooura_data(&ooura_data_autocorrelation_fft); + } + xtract_init_ooura_data(&ooura_data_autocorrelation_fft, M); + break; + case XTRACT_DCT: + if(ooura_data_dct.initialised) + { + xtract_free_ooura_data(&ooura_data_dct); + } + xtract_init_ooura_data(&ooura_data_dct, M); + case XTRACT_MFCC: + if(ooura_data_mfcc.initialised) + { + xtract_free_ooura_data(&ooura_data_mfcc); + } + xtract_init_ooura_data(&ooura_data_mfcc, M); + break; + } + + return XTRACT_SUCCESS; +} + +void xtract_free_ooura_(void) +{ + if(ooura_data_spectrum.initialised) + { + xtract_free_ooura_data(&ooura_data_spectrum); + } + if(ooura_data_autocorrelation_fft.initialised) + { + xtract_free_ooura_data(&ooura_data_autocorrelation_fft); + } + if(ooura_data_dct.initialised) + { + xtract_free_ooura_data(&ooura_data_dct); + } + if(ooura_data_mfcc.initialised) + { + xtract_free_ooura_data(&ooura_data_mfcc); + } +} + +#else + +void xtract_init_vdsp_data(xtract_vdsp_data *vdsp_data, unsigned int N) +{ + vdsp_data->setup = vDSP_create_fftsetupD(log2f(N), FFT_RADIX2); + vdsp_data->fft.realp = (double *) malloc((N >> 1) * sizeof(double)); + vdsp_data->fft.imagp = (double *) malloc((N >> 1) * sizeof(double)); + vdsp_data->log2N = log2f(N); + vdsp_data->initialised = true; +} + +void xtract_free_vdsp_data(xtract_vdsp_data *vdsp_data) +{ + free(vdsp_data->fft.realp); + free(vdsp_data->fft.imagp); + vDSP_destroy_fftsetupD(vdsp_data->setup); + vdsp_data->fft.realp = NULL; + vdsp_data->fft.imagp = NULL; + vdsp_data->initialised = false; +} + +int xtract_init_vdsp_(int N, int feature_name) +{ + + int M = N >> 1; + + if(feature_name == XTRACT_AUTOCORRELATION_FFT) + { + M = N; /* allow for zero padding */ + } + + switch(feature_name) + { + case XTRACT_SPECTRUM: + if(vdsp_data_spectrum.initialised) + { + xtract_free_vdsp_data(&vdsp_data_spectrum); + } + xtract_init_vdsp_data(&vdsp_data_spectrum, M); + break; + case XTRACT_AUTOCORRELATION_FFT: + if(vdsp_data_autocorrelation_fft.initialised) + { + xtract_free_vdsp_data(&vdsp_data_autocorrelation_fft); + } + xtract_init_vdsp_data(&vdsp_data_autocorrelation_fft, M); + break; + case XTRACT_DCT: + if(vdsp_data_dct.initialised) + { + xtract_free_vdsp_data(&vdsp_data_dct); + } + xtract_init_vdsp_data(&vdsp_data_dct, M); + case XTRACT_MFCC: + if(vdsp_data_mfcc.initialised) + { + xtract_free_vdsp_data(&vdsp_data_mfcc); + } + xtract_init_vdsp_data(&vdsp_data_mfcc, M); + break; + } + + return XTRACT_SUCCESS; +} + +void xtract_free_vdsp_(void) +{ + if(vdsp_data_spectrum.initialised) + { + xtract_free_vdsp_data(&vdsp_data_spectrum); + } + if(vdsp_data_autocorrelation_fft.initialised) + { + xtract_free_vdsp_data(&vdsp_data_autocorrelation_fft); + } + if(vdsp_data_dct.initialised) + { + xtract_free_vdsp_data(&vdsp_data_dct); + } + if(vdsp_data_mfcc.initialised) + { + xtract_free_vdsp_data(&vdsp_data_mfcc); + } +} + + +#endif + +int xtract_init_fft(int N, int feature_name) +{ + if(!xtract_is_poweroftwo(N)) + { + fprintf(stderr, + "libxtract: error: only power-of-two FFT sizes are supported by Ooura FFT.\n"); + exit(EXIT_FAILURE); + } +#ifdef USE_OOURA + return xtract_init_ooura_(N, feature_name); +#else + return xtract_init_vdsp_(N, feature_name); +#endif +} + +void xtract_free_fft(void) +{ +#ifdef USE_OOURA + xtract_free_ooura_(); +#else + xtract_free_vdsp_(); +#endif +} + + +int xtract_init_bark(int N, double sr, int *band_limits) +{ + + double edges[] = {0, 100, 200, 300, 400, 510, 630, 770, 920, 1080, 1270, 1480, 1720, 2000, 2320, 2700, 3150, 3700, 4400, 5300, 6400, 7700, 9500, 12000, 15500, 20500, 27000}; /* Takes us up to sr = 54kHz (CCRMA: JOS)*/ + + int bands = XTRACT_BARK_BANDS; + + while(bands--) + band_limits[bands] = edges[bands] / sr * N; + /*FIX shohuld use rounding, but couldn't get it to work */ + + return XTRACT_SUCCESS; +} + int xtract_init_mfcc(int N, double nyquist, int style, double freq_min, double freq_max, int freq_bands, double **fft_tables) { @@ -152,107 +361,6 @@ int xtract_init_mfcc(int N, double nyquist, int style, double freq_min, double f } -void xtract_init_ooura(xtract_ooura_data *ooura_data, unsigned int N) -{ - ooura_data->ooura_ip = (int *)calloc((2 + sqrt(N)), sizeof(int)); - ooura_data->ooura_w = (double *)calloc((N - 1), sizeof(double)); - ooura_data->initialised = true; -} - -void xtract_free_ooura(xtract_ooura_data *ooura_data) -{ - free(ooura_data->ooura_ip); - free(ooura_data->ooura_w); - ooura_data->ooura_ip = NULL; - ooura_data->ooura_w = NULL; - ooura_data->initialised = false; -} - -int xtract_init_fft(int N, int feature_name) -{ - - int M = N >> 1; - - if(!xtract_is_poweroftwo(N)) - { - fprintf(stderr, - "libxtract: error: only power-of-two FFT sizes are supported.\n"); - exit(EXIT_FAILURE); - } - - if(feature_name == XTRACT_AUTOCORRELATION_FFT) - { - M = N; /* allow for zero padding */ - } - - switch(feature_name) - { - case XTRACT_SPECTRUM: - if(ooura_data_spectrum.initialised) - { - xtract_free_ooura(&ooura_data_spectrum); - } - xtract_init_ooura(&ooura_data_spectrum, M); - break; - case XTRACT_AUTOCORRELATION_FFT: - if(ooura_data_autocorrelation_fft.initialised) - { - xtract_free_ooura(&ooura_data_autocorrelation_fft); - } - xtract_init_ooura(&ooura_data_autocorrelation_fft, M); - break; - case XTRACT_DCT: - if(ooura_data_dct.initialised) - { - xtract_free_ooura(&ooura_data_dct); - } - xtract_init_ooura(&ooura_data_dct, M); - case XTRACT_MFCC: - if(ooura_data_mfcc.initialised) - { - xtract_free_ooura(&ooura_data_mfcc); - } - xtract_init_ooura(&ooura_data_mfcc, M); - break; - } - - return XTRACT_SUCCESS; -} - -void xtract_free_fft(void) -{ - if(ooura_data_spectrum.initialised) - { - xtract_free_ooura(&ooura_data_spectrum); - } - if(ooura_data_autocorrelation_fft.initialised) - { - xtract_free_ooura(&ooura_data_autocorrelation_fft); - } - if(ooura_data_dct.initialised) - { - xtract_free_ooura(&ooura_data_dct); - } - if(ooura_data_mfcc.initialised) - { - xtract_free_ooura(&ooura_data_mfcc); - } -} - -int xtract_init_bark(int N, double sr, int *band_limits) -{ - - double edges[] = {0, 100, 200, 300, 400, 510, 630, 770, 920, 1080, 1270, 1480, 1720, 2000, 2320, 2700, 3150, 3700, 4400, 5300, 6400, 7700, 9500, 12000, 15500, 20500, 27000}; /* Takes us up to sr = 54kHz (CCRMA: JOS)*/ - - int bands = XTRACT_BARK_BANDS; - - while(bands--) - band_limits[bands] = edges[bands] / sr * N; - /*FIX shohuld use rounding, but couldn't get it to work */ - - return XTRACT_SUCCESS; -} - double *xtract_init_window(const int N, const int type) { double *window; @@ -307,8 +415,15 @@ __attribute__((constructor)) void init() void _init()ยท #endif { +#ifdef USE_OOURA ooura_data_dct.initialised = false; ooura_data_spectrum.initialised = false; ooura_data_autocorrelation_fft.initialised = false; ooura_data_mfcc.initialised = false; +#else + vdsp_data_dct.initialised = false; + vdsp_data_spectrum.initialised = false; + vdsp_data_autocorrelation_fft.initialised = false; + vdsp_data_mfcc.initialised = false; +#endif } diff --git a/src/scalar.c b/src/scalar.c index 80f0058..4377f00 100644 --- a/src/scalar.c +++ b/src/scalar.c @@ -23,8 +23,6 @@ /* scalar.c: defines functions that extract a feature as a single value from an input vector */ -#include <config.h> - #include <stdlib.h> #include <string.h> #include <stdio.h> diff --git a/src/vector.c b/src/vector.c index 5525c3f..4773797 100644 --- a/src/vector.c +++ b/src/vector.c @@ -44,11 +44,14 @@ int xtract_spectrum(const double *data, const int N, const void *argv, double *r double max = 0.0; double NxN = XTRACT_SQ(N); double *marker = NULL; - size_t bytes = N * sizeof(double); + double real = 0.0; + double imag = 0.0; unsigned int n = 0; unsigned int m = 0; - unsigned int nx2 = 0; unsigned int M = N >> 1; +#ifndef USE_OOURA + DSPDoubleSplitComplex *fft = NULL; +#endif q = *(double *)argv; vector = (int)*((double *)argv+1); @@ -56,8 +59,11 @@ int xtract_spectrum(const double *data, const int N, const void *argv, double *r normalise = (int)*((double *)argv+3); XTRACT_CHECK_q; - +#ifdef USE_OOURA if(!ooura_data_spectrum.initialised) +#else + if(!vdsp_data_spectrum.initialised) +#endif { fprintf(stderr, "libxtract: error: xtract_spectrum() failed, " @@ -65,12 +71,19 @@ int xtract_spectrum(const double *data, const int N, const void *argv, double *r return XTRACT_NO_RESULT; } +#ifdef USE_OOURA /* ooura is in-place * the output format is * a[0] - DC, a[1] - nyquist, a[2...N-1] - remaining bins */ rdft(N, 1, data, ooura_data_spectrum.ooura_ip, ooura_data_spectrum.ooura_w); +#else + fft = &vdsp_data_spectrum.fft; + vDSP_ctozD((DSPDoubleComplex *)data, 2, fft, 1, N >> 1); + vDSP_fft_zripD(vdsp_data_spectrum.setup, fft, 1, + vdsp_data_spectrum.log2N, FFT_FORWARD); +#endif switch(vector) { @@ -78,12 +91,29 @@ int xtract_spectrum(const double *data, const int N, const void *argv, double *r case XTRACT_LOG_MAGNITUDE_SPECTRUM: for(n = 0, m = 0; m < M; ++n, ++m) { - if(!withDC && n == 0) + marker = &result[m]; + + if(n==0 && !withDC) /* discard DC and keep Nyquist */ { - continue; + ++n; +#ifdef USE_OOURA + marker = &result[M-1]; +#endif } - nx2 = n * 2; - temp = XTRACT_SQ(data[nx2]) + XTRACT_SQ(data[nx2+1]); +#ifdef USE_OOURA + if(n==1 && withDC) /* discard Nyquist */ + { + ++n; + } + + real = data[n*2]; + imag = data[n*2+1]; +#else + real = fft->realp[n]; + imag = fft->realp[n]; +#endif + + temp = XTRACT_SQ(real) + XTRACT_SQ(imag); if (temp > XTRACT_LOG_LIMIT) { temp = log(sqrt(temp) / (double)N); @@ -105,11 +135,30 @@ int xtract_spectrum(const double *data, const int N, const void *argv, double *r case XTRACT_POWER_SPECTRUM: for(n = 0, m = 0; m < M; ++n, ++m) { - if(!withDC && n == 0) + + marker = &result[m]; + + if(n==0 && !withDC) /* discard DC and keep Nyquist */ + { + ++n; +#ifdef USE_OOURA + marker = &result[M-1]; +#endif + } +#ifdef USE_OOURA + if(n==1 && withDC) /* discard Nyquist */ { ++n; } - result[m] = (XTRACT_SQ(data[n]) + XTRACT_SQ(data[N - n])) / NxN; + + real = data[n*2]; + imag = data[n*2+1]; +#else + real = fft->realp[n]; + imag = fft->realp[n]; +#endif + + result[m] = (XTRACT_SQ(real) + XTRACT_SQ(imag)) / NxN; XTRACT_SET_FREQUENCY; XTRACT_GET_MAX; } @@ -118,11 +167,29 @@ int xtract_spectrum(const double *data, const int N, const void *argv, double *r case XTRACT_LOG_POWER_SPECTRUM: for(n = 0, m = 0; m < M; ++n, ++m) { - if(!withDC && n == 0) + marker = &result[m]; + + if(n==0 && !withDC) /* discard DC and keep Nyquist */ + { + ++n; +#ifdef USE_OOURA + marker = &result[M-1]; +#endif + } +#ifdef USE_OOURA + if(n==1 && withDC) /* discard Nyquist */ { ++n; } - if ((temp = XTRACT_SQ(data[n]) + XTRACT_SQ(data[N - n])) > + + real = data[n*2]; + imag = data[n*2+1]; +#else + real = fft->realp[n]; + imag = fft->realp[n]; +#endif + + if ((temp = XTRACT_SQ(real) + XTRACT_SQ(imag)) > XTRACT_LOG_LIMIT) temp = log(temp / NxN); else @@ -144,19 +211,25 @@ int xtract_spectrum(const double *data, const int N, const void *argv, double *r if(n==0 && !withDC) /* discard DC and keep Nyquist */ { ++n; +#ifdef USE_OOURA marker = &result[M-1]; +#endif } +#ifdef USE_OOURA if(n==1 && withDC) /* discard Nyquist */ { ++n; } - - *marker = (double)(sqrt(XTRACT_SQ(data[n*2]) + - XTRACT_SQ(data[n*2+1])) / (double)N); + real = data[n*2]; + imag = data[n*2+1]; +#else + real = fft->realp[n]; + imag = fft->realp[n]; +#endif + *marker = sqrt(XTRACT_SQ(real) + XTRACT_SQ(imag)) / (double)N; XTRACT_SET_FREQUENCY; XTRACT_GET_MAX; - } break; } @@ -176,6 +249,10 @@ int xtract_autocorrelation_fft(const double *data, const int N, const void *argv double *rfft = NULL; int n = 0; int M = 0; + double M_double = 0.0; +#ifndef USE_OOURA + DSPDoubleSplitComplex *fft = NULL; +#endif M = N << 1; @@ -183,10 +260,11 @@ int xtract_autocorrelation_fft(const double *data, const int N, const void *argv rfft = (double *)calloc(M, sizeof(double)); memcpy(rfft, data, N * sizeof(double)); +#ifdef USE_OOURA rdft(M, 1, rfft, ooura_data_autocorrelation_fft.ooura_ip, ooura_data_autocorrelation_fft.ooura_w); - for(n = 2; n < M; n++) + for(n = 2; n < M; ++n) { rfft[n*2] = XTRACT_SQ(rfft[n*2]) + XTRACT_SQ(rfft[n*2+1]); rfft[n*2+1] = 0.0; @@ -198,13 +276,37 @@ int xtract_autocorrelation_fft(const double *data, const int N, const void *argv rdft(M, -1, rfft, ooura_data_autocorrelation_fft.ooura_ip, ooura_data_autocorrelation_fft.ooura_w); +#else + /* vDSP has its own autocorrelation function, but it doesn't fit the + * LibXtract model, e.g. we can't guarantee it's going to use + * an FFT for all values of N */ + fft = &vdsp_data_autocorrelation_fft.fft; + vDSP_ctozD((DSPDoubleComplex *)data, 2, fft, 1, N); + vDSP_fft_zripD(vdsp_data_autocorrelation_fft.setup, fft, 1, + vdsp_data_autocorrelation_fft.log2N, FFT_FORWARD); + + for(n = 0; n < N; ++n) + { + fft->realp[n] = XTRACT_SQ(fft->realp[n]) + XTRACT_SQ(fft->imagp[n]); + fft->imagp[n] = 0.0; + } + + vDSP_fft_zripD(vdsp_data_autocorrelation_fft.setup, fft, 1, + vdsp_data_autocorrelation_fft.log2N, FFT_INVERSE); +#endif + /* Normalisation factor */ M = M * N; +#ifdef USE_OOURA for(n = 0; n < N; n++) result[n] = rfft[n] / (double)M; - free(rfft); +#else + M_double = (double)M; + vDSP_ztocD(fft, 1, (DOUBLE_COMPLEX *)result, 2, N); + vDSP_vsdivD(result, 1, &M_double, result, 1, N); +#endif return XTRACT_SUCCESS; } diff --git a/src/window.c b/src/window.c index 18a4e15..0d00402 100644 --- a/src/window.c +++ b/src/window.c @@ -56,7 +56,7 @@ void hamming(double *window, const int N) const double M = N - 1; for (n = 0; n < N; n++) - window[n] = 0.53836 - (0.46164 * cos(2.0 * PI * (double)n / M)); + window[n] = 0.53836 - (0.46164 * cos(2.0 * M_PI * (double)n / M)); } @@ -67,7 +67,7 @@ void hann(double *window, const int N) const double M = N - 1; for (n = 0; n < N; n++) - window[n] = 0.5 * (1.0 - cos(2.0 * PI * (double)n / M)); + window[n] = 0.5 * (1.0 - cos(2.0 * M_PI * (double)n / M)); } @@ -107,7 +107,7 @@ void bartlett_hann(double *window, const int N) { term1 = a1 * fabs(n / M - 0.5); - term2 = a2 * cos(2.0 * PI * (double)n / M); + term2 = a2 * cos(2.0 * M_PI * (double)n / M); window[n] = a0 - term1 - term2; } @@ -127,8 +127,8 @@ void blackman(double *window, const int N) for (n = 0; n < N; n++) { - term1 = a1 * cos(2.0 * PI * (double)n / M); - term2 = a2 * cos(4.0 * PI * (double)n / M); + term1 = a1 * cos(2.0 * M_PI * (double)n / M); + term2 = a2 * cos(4.0 * M_PI * (double)n / M); window[n] = a0 - term1 + term2; } @@ -193,9 +193,9 @@ void blackman_harris(double *window, const int N) for (n = 0; n < N; n++) { - term1 = a1 * cos(2.0 * PI * n / M); - term2 = a2 * cos(4.0 * PI * n / M); - term3 = a3 * cos(6.0 * PI * n / M); + term1 = a1 * cos(2.0 * M_PI * n / M); + term2 = a2 * cos(4.0 * M_PI * n / M); + term3 = a3 * cos(6.0 * M_PI * n / M); window[n] = a0 - term1 + term2 - term3; } diff --git a/src/xtract_globals_private.h b/src/xtract_globals_private.h index b7c26e4..e643ab0 100644 --- a/src/xtract_globals_private.h +++ b/src/xtract_globals_private.h @@ -26,12 +26,7 @@ #ifndef XTRACT_GLOBALS_PRIVATE_H #define XTRACT_GLOBALS_PRIVATE_H -typedef struct xtract_ooura_data_ -{ - int *ooura_ip; - double *ooura_w; - bool initialised; -} xtract_ooura_data; +#include "fft.h" #ifdef DEFINE_GLOBALS #define GLOBAL @@ -39,10 +34,17 @@ typedef struct xtract_ooura_data_ #define GLOBAL extern #endif +#ifdef USE_OOURA GLOBAL struct xtract_ooura_data_ ooura_data_dct; GLOBAL struct xtract_ooura_data_ ooura_data_mfcc; GLOBAL struct xtract_ooura_data_ ooura_data_spectrum; GLOBAL struct xtract_ooura_data_ ooura_data_autocorrelation_fft; +#else +GLOBAL xtract_vdsp_data vdsp_data_dct; +GLOBAL xtract_vdsp_data vdsp_data_mfcc; +GLOBAL xtract_vdsp_data vdsp_data_spectrum; +GLOBAL xtract_vdsp_data vdsp_data_autocorrelation_fft; +#endif #endif /* Header guard */ |