*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, write to the Free Software
- * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "../dsputil.h"
+#include "gcc_fixes.h"
+
#include "dsputil_altivec.h"
/*
/**
- * Do a complex FFT with the parameters defined in fft_init(). The
+ * Do a complex FFT with the parameters defined in ff_fft_init(). The
* input data must be permuted before with s->revtab table. No
* 1.0/sqrt(n) normalization is done.
* AltiVec-enabled
* that successive MUL + ADD/SUB have been merged into
* fused multiply-add ('vec_madd' in altivec)
*/
-void fft_calc_altivec(FFTContext *s, FFTComplex *z)
+void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z)
{
-POWERPC_TBL_DECLARE(altivec_fft_num, s->nbits >= 6);
+POWERPC_PERF_DECLARE(altivec_fft_num, s->nbits >= 6);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
int ln = s->nbits;
- int j, np, np2;
- int nblocks, nloops;
+ int j, np, np2;
+ int nblocks, nloops;
register FFTComplex *p, *q;
FFTComplex *exptab = s->exptab;
int l;
FFTSample tmp_re, tmp_im;
-
-POWERPC_TBL_START_COUNT(altivec_fft_num, s->nbits >= 6);
-
+
+POWERPC_PERF_START_COUNT(altivec_fft_num, s->nbits >= 6);
+
np = 1 << ln;
/* pass 0 */
p=&z[0];
j=(np >> 1);
do {
- BF(p[0].re, p[0].im, p[1].re, p[1].im,
+ BF(p[0].re, p[0].im, p[1].re, p[1].im,
p[0].re, p[0].im, p[1].re, p[1].im);
p+=2;
} while (--j != 0);
/* pass 1 */
-
+
p=&z[0];
j=np >> 2;
if (s->inverse) {
do {
- BF(p[0].re, p[0].im, p[2].re, p[2].im,
+ BF(p[0].re, p[0].im, p[2].re, p[2].im,
p[0].re, p[0].im, p[2].re, p[2].im);
- BF(p[1].re, p[1].im, p[3].re, p[3].im,
+ BF(p[1].re, p[1].im, p[3].re, p[3].im,
p[1].re, p[1].im, -p[3].im, p[3].re);
p+=4;
} while (--j != 0);
} else {
do {
- BF(p[0].re, p[0].im, p[2].re, p[2].im,
+ BF(p[0].re, p[0].im, p[2].re, p[2].im,
p[0].re, p[0].im, p[2].re, p[2].im);
- BF(p[1].re, p[1].im, p[3].re, p[3].im,
+ BF(p[1].re, p[1].im, p[3].re, p[3].im,
p[1].re, p[1].im, p[3].im, -p[3].re);
p+=4;
} while (--j != 0);
for (j = 0; j < nblocks; ++j) {
BF(p->re, p->im, q->re, q->im,
p->re, p->im, q->re, q->im);
-
+
p++;
q++;
for(l = nblocks; l < np2; l += nblocks) {
nloops = nloops << 1;
} while (nblocks != 0);
-POWERPC_TBL_STOP_COUNT(altivec_fft_num, s->nbits >= 6);
+POWERPC_PERF_STOP_COUNT(altivec_fft_num, s->nbits >= 6);
#else /* ALTIVEC_USE_REFERENCE_C_CODE */
+#ifdef CONFIG_DARWIN
register const vector float vczero = (const vector float)(0.);
-
+#else
+ register const vector float vczero = (const vector float){0.,0.,0.,0.};
+#endif
+
int ln = s->nbits;
- int j, np, np2;
- int nblocks, nloops;
+ int j, np, np2;
+ int nblocks, nloops;
register FFTComplex *p, *q;
FFTComplex *cptr, *cptr1;
int k;
-POWERPC_TBL_START_COUNT(altivec_fft_num, s->nbits >= 6);
+POWERPC_PERF_START_COUNT(altivec_fft_num, s->nbits >= 6);
np = 1 << ln;
r = (vector float *)&z[0];
c1 = vcii(p,p,n,n);
-
+
if (s->inverse)
{
c2 = vcii(p,p,n,p);
{
c2 = vcii(p,p,p,n);
}
-
+
j = (np >> 2);
do {
a = vec_ld(0, r);
a1 = vec_ld(sizeof(vector float), r);
-
+
b = vec_perm(a,a,vcprmle(1,0,3,2));
a = vec_madd(a,c1,b);
/* do the pass 0 butterfly */
-
+
b = vec_perm(a1,a1,vcprmle(1,0,3,2));
b = vec_madd(a1,c1,b);
/* do the pass 0 butterfly */
-
+
/* multiply third by -i */
b = vec_perm(b,b,vcprmle(2,3,1,0));
-
+
/* do the pass 1 butterfly */
vec_st(vec_madd(b,c2,a), 0, r);
vec_st(vec_nmsub(b,c2,a), sizeof(vector float), r);
-
+
r += 2;
} while (--j != 0);
}
a = vec_ld(0, (float*)p);
b = vec_ld(0, (float*)q);
-
+
/* complex mul */
c = vec_ld(0, (float*)cptr);
/* cre*re cim*re */
c = vec_ld(sizeof(vector float), (float*)cptr);
/* -cim*im cre*im */
b = vec_madd(c, vec_perm(b,b,vcprmle(3,3,1,1)),t1);
-
+
/* butterfly */
vec_st(vec_add(a,b), 0, (float*)p);
vec_st(vec_sub(a,b), 0, (float*)q);
-
+
p += 2;
q += 2;
cptr += 4;
} while (--k);
-
+
p += nloops;
q += nloops;
} while (--j);
nloops = nloops << 1;
} while (nblocks != 0);
-POWERPC_TBL_STOP_COUNT(altivec_fft_num, s->nbits >= 6);
+POWERPC_PERF_STOP_COUNT(altivec_fft_num, s->nbits >= 6);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}