annotate x86/fft_3dn2.c @ 9484:fb240cb99bcf libavcodec

Remove useless init from ff_mpeg4_set_direct_mv() found by CSA.
author michael
date Fri, 17 Apr 2009 18:24:39 +0000
parents 7768bdfd4f7b
children 5cf49858179a
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
8430
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
1 /*
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
2 * FFT/MDCT transform with Extended 3DNow! optimizations
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
3 * Copyright (c) 2006-2008 Zuxy MENG Jie, Loren Merritt
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
4 *
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
5 * This file is part of FFmpeg.
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
6 *
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
7 * FFmpeg is free software; you can redistribute it and/or
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
8 * modify it under the terms of the GNU Lesser General Public
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
9 * License as published by the Free Software Foundation; either
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
10 * version 2.1 of the License, or (at your option) any later version.
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
11 *
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
12 * FFmpeg is distributed in the hope that it will be useful,
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
15 * Lesser General Public License for more details.
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
16 *
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
17 * You should have received a copy of the GNU Lesser General Public
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
18 * License along with FFmpeg; if not, write to the Free Software
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
20 */
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
21
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
22 #include "libavutil/x86_cpu.h"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
23 #include "libavcodec/dsputil.h"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
24
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
25 DECLARE_ALIGNED_8(static const int, m1m1[2]) = { 1<<31, 1<<31 };
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
26
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
27 #ifdef EMULATE_3DNOWEXT
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
28 #define PSWAPD(s,d)\
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
29 "movq "#s","#d"\n"\
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
30 "psrlq $32,"#d"\n"\
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
31 "punpckldq "#s","#d"\n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
32 #define ff_fft_calc_3dn2 ff_fft_calc_3dn
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
33 #define ff_fft_dispatch_3dn2 ff_fft_dispatch_3dn
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
34 #define ff_fft_dispatch_interleave_3dn2 ff_fft_dispatch_interleave_3dn
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
35 #define ff_imdct_calc_3dn2 ff_imdct_calc_3dn
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
36 #define ff_imdct_half_3dn2 ff_imdct_half_3dn
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
37 #else
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
38 #define PSWAPD(s,d) "pswapd "#s","#d"\n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
39 #endif
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
40
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
41 void ff_fft_dispatch_3dn2(FFTComplex *z, int nbits);
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
42 void ff_fft_dispatch_interleave_3dn2(FFTComplex *z, int nbits);
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
43
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
44 void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z)
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
45 {
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
46 int n = 1<<s->nbits;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
47 int i;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
48 ff_fft_dispatch_interleave_3dn2(z, s->nbits);
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
49 __asm__ volatile("femms");
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
50 if(n <= 8)
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
51 for(i=0; i<n; i+=2)
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
52 FFSWAP(FFTSample, z[i].im, z[i+1].re);
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
53 }
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
54
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
55 void ff_imdct_half_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input)
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
56 {
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
57 x86_reg j, k;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
58 long n = 1 << s->nbits;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
59 long n2 = n >> 1;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
60 long n4 = n >> 2;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
61 long n8 = n >> 3;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
62 const uint16_t *revtab = s->fft.revtab;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
63 const FFTSample *tcos = s->tcos;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
64 const FFTSample *tsin = s->tsin;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
65 const FFTSample *in1, *in2;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
66 FFTComplex *z = (FFTComplex *)output;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
67
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
68 /* pre rotation */
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
69 in1 = input;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
70 in2 = input + n2 - 1;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
71 #ifdef EMULATE_3DNOWEXT
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
72 __asm__ volatile("movd %0, %%mm7" ::"r"(1<<31));
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
73 #endif
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
74 for(k = 0; k < n4; k++) {
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
75 // FIXME a single block is faster, but gcc 2.95 and 3.4.x on 32bit can't compile it
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
76 __asm__ volatile(
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
77 "movd %0, %%mm0 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
78 "movd %2, %%mm1 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
79 "punpckldq %1, %%mm0 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
80 "punpckldq %3, %%mm1 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
81 "movq %%mm0, %%mm2 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
82 PSWAPD( %%mm1, %%mm3 )
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
83 "pfmul %%mm1, %%mm0 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
84 "pfmul %%mm3, %%mm2 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
85 #ifdef EMULATE_3DNOWEXT
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
86 "movq %%mm0, %%mm1 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
87 "punpckhdq %%mm2, %%mm0 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
88 "punpckldq %%mm2, %%mm1 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
89 "pxor %%mm7, %%mm0 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
90 "pfadd %%mm1, %%mm0 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
91 #else
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
92 "pfpnacc %%mm2, %%mm0 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
93 #endif
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
94 ::"m"(in2[-2*k]), "m"(in1[2*k]),
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
95 "m"(tcos[k]), "m"(tsin[k])
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
96 );
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
97 __asm__ volatile(
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
98 "movq %%mm0, %0 \n\t"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
99 :"=m"(z[revtab[k]])
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
100 );
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
101 }
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
102
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
103 ff_fft_dispatch_3dn2(z, s->fft.nbits);
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
104
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
105 #define CMUL(j,mm0,mm1)\
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
106 "movq (%2,"#j",2), %%mm6 \n"\
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
107 "movq 8(%2,"#j",2), "#mm0"\n"\
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
108 "movq %%mm6, "#mm1"\n"\
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
109 "movq "#mm0",%%mm7 \n"\
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
110 "pfmul (%3,"#j"), %%mm6 \n"\
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
111 "pfmul (%4,"#j"), "#mm0"\n"\
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
112 "pfmul (%4,"#j"), "#mm1"\n"\
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
113 "pfmul (%3,"#j"), %%mm7 \n"\
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
114 "pfsub %%mm6, "#mm0"\n"\
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
115 "pfadd %%mm7, "#mm1"\n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
116
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
117 /* post rotation */
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
118 j = -n2;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
119 k = n2-8;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
120 __asm__ volatile(
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
121 "1: \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
122 CMUL(%0, %%mm0, %%mm1)
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
123 CMUL(%1, %%mm2, %%mm3)
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
124 "movd %%mm0, (%2,%0,2) \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
125 "movd %%mm1,12(%2,%1,2) \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
126 "movd %%mm2, (%2,%1,2) \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
127 "movd %%mm3,12(%2,%0,2) \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
128 "psrlq $32, %%mm0 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
129 "psrlq $32, %%mm1 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
130 "psrlq $32, %%mm2 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
131 "psrlq $32, %%mm3 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
132 "movd %%mm0, 8(%2,%0,2) \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
133 "movd %%mm1, 4(%2,%1,2) \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
134 "movd %%mm2, 8(%2,%1,2) \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
135 "movd %%mm3, 4(%2,%0,2) \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
136 "sub $8, %1 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
137 "add $8, %0 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
138 "jl 1b \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
139 :"+r"(j), "+r"(k)
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
140 :"r"(z+n8), "r"(tcos+n8), "r"(tsin+n8)
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
141 :"memory"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
142 );
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
143 __asm__ volatile("femms");
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
144 }
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
145
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
146 void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input)
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
147 {
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
148 x86_reg j, k;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
149 long n = 1 << s->nbits;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
150 long n4 = n >> 2;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
151
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
152 ff_imdct_half_3dn2(s, output+n4, input);
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
153
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
154 j = -n;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
155 k = n-8;
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
156 __asm__ volatile(
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
157 "movq %4, %%mm7 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
158 "1: \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
159 PSWAPD((%2,%1), %%mm0)
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
160 PSWAPD((%3,%0), %%mm1)
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
161 "pxor %%mm7, %%mm0 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
162 "movq %%mm1, (%3,%1) \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
163 "movq %%mm0, (%2,%0) \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
164 "sub $8, %1 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
165 "add $8, %0 \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
166 "jl 1b \n"
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
167 :"+r"(j), "+r"(k)
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
168 :"r"(output+n4), "r"(output+n4*3),
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
169 "m"(*m1m1)
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
170 );
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
171 __asm__ volatile("femms");
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
172 }
7768bdfd4f7b Rename libavcodec/i386/ --> libavcodec/x86/.
diego
parents:
diff changeset
173