• Main Page
  • Related Pages
  • Modules
  • Data Structures
  • Files
  • Examples
  • File List
  • Globals

libavcodec/ppc/fft_altivec.c

Go to the documentation of this file.
00001 /*
00002  * FFT/IFFT transforms
00003  * AltiVec-enabled
00004  * Copyright (c) 2009 Loren Merritt
00005  *
00006  * This file is part of FFmpeg.
00007  *
00008  * FFmpeg is free software; you can redistribute it and/or
00009  * modify it under the terms of the GNU Lesser General Public
00010  * License as published by the Free Software Foundation; either
00011  * version 2.1 of the License, or (at your option) any later version.
00012  *
00013  * FFmpeg is distributed in the hope that it will be useful,
00014  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00015  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00016  * Lesser General Public License for more details.
00017  *
00018  * You should have received a copy of the GNU Lesser General Public
00019  * License along with FFmpeg; if not, write to the Free Software
00020  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00021  */
00022 #include "libavcodec/fft.h"
00023 #include "util_altivec.h"
00024 #include "types_altivec.h"
00025 
00035 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
00036 void ff_fft_calc_interleave_altivec(FFTContext *s, FFTComplex *z);
00037 
00038 #if HAVE_GNU_AS
00039 static void ff_imdct_half_altivec(FFTContext *s, FFTSample *output, const FFTSample *input)
00040 {
00041     int j, k;
00042     int n = 1 << s->mdct_bits;
00043     int n4 = n >> 2;
00044     int n8 = n >> 3;
00045     int n32 = n >> 5;
00046     const uint16_t *revtabj = s->revtab;
00047     const uint16_t *revtabk = s->revtab+n4;
00048     const vec_f *tcos = (const vec_f*)(s->tcos+n8);
00049     const vec_f *tsin = (const vec_f*)(s->tsin+n8);
00050     const vec_f *pin = (const vec_f*)(input+n4);
00051     vec_f *pout = (vec_f*)(output+n4);
00052 
00053     /* pre rotation */
00054     k = n32-1;
00055     do {
00056         vec_f cos,sin,cos0,sin0,cos1,sin1,re,im,r0,i0,r1,i1,a,b,c,d;
00057 #define CMULA(p,o0,o1,o2,o3)\
00058         a = pin[ k*2+p];                       /* { z[k].re,    z[k].im,    z[k+1].re,  z[k+1].im  } */\
00059         b = pin[-k*2-p-1];                     /* { z[-k-2].re, z[-k-2].im, z[-k-1].re, z[-k-1].im } */\
00060         re = vec_perm(a, b, vcprm(0,2,s0,s2)); /* { z[k].re,    z[k+1].re,  z[-k-2].re, z[-k-1].re } */\
00061         im = vec_perm(a, b, vcprm(s3,s1,3,1)); /* { z[-k-1].im, z[-k-2].im, z[k+1].im,  z[k].im    } */\
00062         cos = vec_perm(cos0, cos1, vcprm(o0,o1,s##o2,s##o3)); /* { cos[k], cos[k+1], cos[-k-2], cos[-k-1] } */\
00063         sin = vec_perm(sin0, sin1, vcprm(o0,o1,s##o2,s##o3));\
00064         r##p = im*cos - re*sin;\
00065         i##p = re*cos + im*sin;
00066 #define STORE2(v,dst)\
00067         j = dst;\
00068         vec_ste(v, 0, output+j*2);\
00069         vec_ste(v, 4, output+j*2);
00070 #define STORE8(p)\
00071         a = vec_perm(r##p, i##p, vcprm(0,s0,0,s0));\
00072         b = vec_perm(r##p, i##p, vcprm(1,s1,1,s1));\
00073         c = vec_perm(r##p, i##p, vcprm(2,s2,2,s2));\
00074         d = vec_perm(r##p, i##p, vcprm(3,s3,3,s3));\
00075         STORE2(a, revtabk[ p*2-4]);\
00076         STORE2(b, revtabk[ p*2-3]);\
00077         STORE2(c, revtabj[-p*2+2]);\
00078         STORE2(d, revtabj[-p*2+3]);
00079 
00080         cos0 = tcos[k];
00081         sin0 = tsin[k];
00082         cos1 = tcos[-k-1];
00083         sin1 = tsin[-k-1];
00084         CMULA(0, 0,1,2,3);
00085         CMULA(1, 2,3,0,1);
00086         STORE8(0);
00087         STORE8(1);
00088         revtabj += 4;
00089         revtabk -= 4;
00090         k--;
00091     } while(k >= 0);
00092 
00093     ff_fft_calc_altivec(s, (FFTComplex*)output);
00094 
00095     /* post rotation + reordering */
00096     j = -n32;
00097     k = n32-1;
00098     do {
00099         vec_f cos,sin,re,im,a,b,c,d;
00100 #define CMULB(d0,d1,o)\
00101         re = pout[o*2];\
00102         im = pout[o*2+1];\
00103         cos = tcos[o];\
00104         sin = tsin[o];\
00105         d0 = im*sin - re*cos;\
00106         d1 = re*sin + im*cos;
00107 
00108         CMULB(a,b,j);
00109         CMULB(c,d,k);
00110         pout[2*j]   = vec_perm(a, d, vcprm(0,s3,1,s2));
00111         pout[2*j+1] = vec_perm(a, d, vcprm(2,s1,3,s0));
00112         pout[2*k]   = vec_perm(c, b, vcprm(0,s3,1,s2));
00113         pout[2*k+1] = vec_perm(c, b, vcprm(2,s1,3,s0));
00114         j++;
00115         k--;
00116     } while(k >= 0);
00117 }
00118 
00119 static void ff_imdct_calc_altivec(FFTContext *s, FFTSample *output, const FFTSample *input)
00120 {
00121     int k;
00122     int n = 1 << s->mdct_bits;
00123     int n4 = n >> 2;
00124     int n16 = n >> 4;
00125     vec_u32 sign = {1U<<31,1U<<31,1U<<31,1U<<31};
00126     vec_u32 *p0 = (vec_u32*)(output+n4);
00127     vec_u32 *p1 = (vec_u32*)(output+n4*3);
00128 
00129     ff_imdct_half_altivec(s, output+n4, input);
00130 
00131     for (k = 0; k < n16; k++) {
00132         vec_u32 a = p0[k] ^ sign;
00133         vec_u32 b = p1[-k-1];
00134         p0[-k-1] = vec_perm(a, a, vcprm(3,2,1,0));
00135         p1[k]    = vec_perm(b, b, vcprm(3,2,1,0));
00136     }
00137 }
00138 #endif /* HAVE_GNU_AS */
00139 
00140 av_cold void ff_fft_init_altivec(FFTContext *s)
00141 {
00142 #if HAVE_GNU_AS
00143     s->fft_calc   = ff_fft_calc_interleave_altivec;
00144     s->imdct_calc = ff_imdct_calc_altivec;
00145     s->imdct_half = ff_imdct_half_altivec;
00146 #endif
00147 }

Generated on Fri Feb 22 2013 07:24:28 for FFmpeg by  doxygen 1.7.1