2 * FFT/MDCT transform with Extended 3DNow! optimizations
3 * Copyright (c) 2006-2008 Zuxy MENG Jie, Loren Merritt
5 * This file is part of Libav.
7 * Libav is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * Libav is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 #include "libavutil/x86_cpu.h"
23 #include "libavcodec/dsputil.h"
26 DECLARE_ALIGNED(8, static const int, m1m1)[2] = { 1<<31, 1<<31 };
28 #ifdef EMULATE_3DNOWEXT
32 "punpckldq "#s","#d"\n"
33 #define ff_fft_calc_3dn2 ff_fft_calc_3dn
34 #define ff_fft_dispatch_3dn2 ff_fft_dispatch_3dn
35 #define ff_fft_dispatch_interleave_3dn2 ff_fft_dispatch_interleave_3dn
36 #define ff_imdct_calc_3dn2 ff_imdct_calc_3dn
37 #define ff_imdct_half_3dn2 ff_imdct_half_3dn
39 #define PSWAPD(s,d) "pswapd "#s","#d"\n"
42 void ff_fft_dispatch_3dn2(FFTComplex *z, int nbits);
43 void ff_fft_dispatch_interleave_3dn2(FFTComplex *z, int nbits);
45 void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z)
49 ff_fft_dispatch_interleave_3dn2(z, s->nbits);
50 __asm__ volatile("femms");
53 FFSWAP(FFTSample, z[i].im, z[i+1].re);
56 void ff_imdct_half_3dn2(FFTContext *s, FFTSample *output, const FFTSample *input)
59 long n = s->mdct_size;
63 const uint16_t *revtab = s->revtab;
64 const FFTSample *tcos = s->tcos;
65 const FFTSample *tsin = s->tsin;
66 const FFTSample *in1, *in2;
67 FFTComplex *z = (FFTComplex *)output;
72 #ifdef EMULATE_3DNOWEXT
73 __asm__ volatile("movd %0, %%mm7" ::"r"(1<<31));
75 for(k = 0; k < n4; k++) {
76 // FIXME a single block is faster, but gcc 2.95 and 3.4.x on 32bit can't compile it
80 "punpckldq %1, %%mm0 \n"
81 "punpckldq %3, %%mm1 \n"
82 "movq %%mm0, %%mm2 \n"
83 PSWAPD( %%mm1, %%mm3 )
84 "pfmul %%mm1, %%mm0 \n"
85 "pfmul %%mm3, %%mm2 \n"
86 #ifdef EMULATE_3DNOWEXT
87 "movq %%mm0, %%mm1 \n"
88 "punpckhdq %%mm2, %%mm0 \n"
89 "punpckldq %%mm2, %%mm1 \n"
90 "pxor %%mm7, %%mm0 \n"
91 "pfadd %%mm1, %%mm0 \n"
93 "pfpnacc %%mm2, %%mm0 \n"
95 ::"m"(in2[-2*k]), "m"(in1[2*k]),
96 "m"(tcos[k]), "m"(tsin[k])
104 ff_fft_dispatch_3dn2(z, s->nbits);
106 #define CMUL(j,mm0,mm1)\
107 "movq (%2,"#j",2), %%mm6 \n"\
108 "movq 8(%2,"#j",2), "#mm0"\n"\
109 "movq %%mm6, "#mm1"\n"\
110 "movq "#mm0",%%mm7 \n"\
111 "pfmul (%3,"#j"), %%mm6 \n"\
112 "pfmul (%4,"#j"), "#mm0"\n"\
113 "pfmul (%4,"#j"), "#mm1"\n"\
114 "pfmul (%3,"#j"), %%mm7 \n"\
115 "pfsub %%mm6, "#mm0"\n"\
116 "pfadd %%mm7, "#mm1"\n"
123 CMUL(%0, %%mm0, %%mm1)
124 CMUL(%1, %%mm2, %%mm3)
125 "movd %%mm0, (%2,%0,2) \n"
126 "movd %%mm1,12(%2,%1,2) \n"
127 "movd %%mm2, (%2,%1,2) \n"
128 "movd %%mm3,12(%2,%0,2) \n"
129 "psrlq $32, %%mm0 \n"
130 "psrlq $32, %%mm1 \n"
131 "psrlq $32, %%mm2 \n"
132 "psrlq $32, %%mm3 \n"
133 "movd %%mm0, 8(%2,%0,2) \n"
134 "movd %%mm1, 4(%2,%1,2) \n"
135 "movd %%mm2, 8(%2,%1,2) \n"
136 "movd %%mm3, 4(%2,%0,2) \n"
141 :"r"(z+n8), "r"(tcos+n8), "r"(tsin+n8)
144 __asm__ volatile("femms");
147 void ff_imdct_calc_3dn2(FFTContext *s, FFTSample *output, const FFTSample *input)
150 long n = s->mdct_size;
153 ff_imdct_half_3dn2(s, output+n4, input);
160 PSWAPD((%2,%1), %%mm0)
161 PSWAPD((%3,%0), %%mm1)
162 "pxor %%mm7, %%mm0 \n"
163 "movq %%mm1, (%3,%1) \n"
164 "movq %%mm0, (%2,%0) \n"
169 :"r"(output+n4), "r"(output+n4*3),
172 __asm__ volatile("femms");