mpegaudiodec_mmx.c
Go to the documentation of this file.
1 /*
2  * MMX optimized MP3 decoding functions
3  * Copyright (c) 2010 Vitor Sessak
4  *
5  * This file is part of Libav.
6  *
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.
11  *
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.
16  *
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
20  */
21 
22 #include "libavutil/cpu.h"
23 #include "libavutil/x86_cpu.h"
24 #include "libavcodec/dsputil.h"
26 
27 void ff_imdct36_float_sse(float *out, float *buf, float *in, float *win);
28 void ff_imdct36_float_sse2(float *out, float *buf, float *in, float *win);
29 void ff_imdct36_float_sse3(float *out, float *buf, float *in, float *win);
30 void ff_imdct36_float_ssse3(float *out, float *buf, float *in, float *win);
31 void ff_imdct36_float_avx(float *out, float *buf, float *in, float *win);
32 void ff_four_imdct36_float_sse(float *out, float *buf, float *in, float *win,
33  float *tmpbuf);
34 void ff_four_imdct36_float_avx(float *out, float *buf, float *in, float *win,
35  float *tmpbuf);
36 
37 DECLARE_ALIGNED(16, static float, mdct_win_sse)[2][4][4*40];
38 
39 #define MACS(rt, ra, rb) rt+=(ra)*(rb)
40 #define MLSS(rt, ra, rb) rt-=(ra)*(rb)
41 
42 #define SUM8(op, sum, w, p) \
43 { \
44  op(sum, (w)[0 * 64], (p)[0 * 64]); \
45  op(sum, (w)[1 * 64], (p)[1 * 64]); \
46  op(sum, (w)[2 * 64], (p)[2 * 64]); \
47  op(sum, (w)[3 * 64], (p)[3 * 64]); \
48  op(sum, (w)[4 * 64], (p)[4 * 64]); \
49  op(sum, (w)[5 * 64], (p)[5 * 64]); \
50  op(sum, (w)[6 * 64], (p)[6 * 64]); \
51  op(sum, (w)[7 * 64], (p)[7 * 64]); \
52 }
53 
54 static void apply_window(const float *buf, const float *win1,
55  const float *win2, float *sum1, float *sum2, int len)
56 {
57  x86_reg count = - 4*len;
58  const float *win1a = win1+len;
59  const float *win2a = win2+len;
60  const float *bufa = buf+len;
61  float *sum1a = sum1+len;
62  float *sum2a = sum2+len;
63 
64 
65 #define MULT(a, b) \
66  "movaps " #a "(%1,%0), %%xmm1 \n\t" \
67  "movaps " #a "(%3,%0), %%xmm2 \n\t" \
68  "mulps %%xmm2, %%xmm1 \n\t" \
69  "subps %%xmm1, %%xmm0 \n\t" \
70  "mulps " #b "(%2,%0), %%xmm2 \n\t" \
71  "subps %%xmm2, %%xmm4 \n\t" \
72 
73  __asm__ volatile(
74  "1: \n\t"
75  "xorps %%xmm0, %%xmm0 \n\t"
76  "xorps %%xmm4, %%xmm4 \n\t"
77 
78  MULT( 0, 0)
79  MULT( 256, 64)
80  MULT( 512, 128)
81  MULT( 768, 192)
82  MULT(1024, 256)
83  MULT(1280, 320)
84  MULT(1536, 384)
85  MULT(1792, 448)
86 
87  "movaps %%xmm0, (%4,%0) \n\t"
88  "movaps %%xmm4, (%5,%0) \n\t"
89  "add $16, %0 \n\t"
90  "jl 1b \n\t"
91  :"+&r"(count)
92  :"r"(win1a), "r"(win2a), "r"(bufa), "r"(sum1a), "r"(sum2a)
93  );
94 
95 #undef MULT
96 }
97 
98 static void apply_window_mp3(float *in, float *win, int *unused, float *out,
99  int incr)
100 {
101  LOCAL_ALIGNED_16(float, suma, [17]);
102  LOCAL_ALIGNED_16(float, sumb, [17]);
103  LOCAL_ALIGNED_16(float, sumc, [17]);
104  LOCAL_ALIGNED_16(float, sumd, [17]);
105 
106  float sum;
107 
108  /* copy to avoid wrap */
109  memcpy(in + 512, in, 32 * sizeof(*in));
110 
111  apply_window(in + 16, win , win + 512, suma, sumc, 16);
112  apply_window(in + 32, win + 48, win + 640, sumb, sumd, 16);
113 
114  SUM8(MACS, suma[0], win + 32, in + 48);
115 
116  sumc[ 0] = 0;
117  sumb[16] = 0;
118  sumd[16] = 0;
119 
120 #define SUMS(suma, sumb, sumc, sumd, out1, out2) \
121  "movups " #sumd "(%4), %%xmm0 \n\t" \
122  "shufps $0x1b, %%xmm0, %%xmm0 \n\t" \
123  "subps " #suma "(%1), %%xmm0 \n\t" \
124  "movaps %%xmm0," #out1 "(%0) \n\t" \
125 \
126  "movups " #sumc "(%3), %%xmm0 \n\t" \
127  "shufps $0x1b, %%xmm0, %%xmm0 \n\t" \
128  "addps " #sumb "(%2), %%xmm0 \n\t" \
129  "movaps %%xmm0," #out2 "(%0) \n\t"
130 
131  if (incr == 1) {
132  __asm__ volatile(
133  SUMS( 0, 48, 4, 52, 0, 112)
134  SUMS(16, 32, 20, 36, 16, 96)
135  SUMS(32, 16, 36, 20, 32, 80)
136  SUMS(48, 0, 52, 4, 48, 64)
137 
138  :"+&r"(out)
139  :"r"(&suma[0]), "r"(&sumb[0]), "r"(&sumc[0]), "r"(&sumd[0])
140  :"memory"
141  );
142  out += 16*incr;
143  } else {
144  int j;
145  float *out2 = out + 32 * incr;
146  out[0 ] = -suma[ 0];
147  out += incr;
148  out2 -= incr;
149  for(j=1;j<16;j++) {
150  *out = -suma[ j] + sumd[16-j];
151  *out2 = sumb[16-j] + sumc[ j];
152  out += incr;
153  out2 -= incr;
154  }
155  }
156 
157  sum = 0;
158  SUM8(MLSS, sum, win + 16 + 32, in + 32);
159  *out = sum;
160 }
161 
162 
163 #if HAVE_YASM
164 #define DECL_IMDCT_BLOCKS(CPU1, CPU2) \
165 static void imdct36_blocks_ ## CPU1(float *out, float *buf, float *in, \
166  int count, int switch_point, int block_type) \
167 { \
168  int align_end = count - (count & 3); \
169  int j; \
170  for (j = 0; j < align_end; j+= 4) { \
171  LOCAL_ALIGNED_16(float, tmpbuf, [1024]); \
172  float *win = mdct_win_sse[switch_point && j < 4][block_type]; \
173  /* apply window & overlap with previous buffer */ \
174  \
175  /* select window */ \
176  ff_four_imdct36_float_ ## CPU2(out, buf, in, win, tmpbuf); \
177  in += 4*18; \
178  buf += 4*18; \
179  out += 4; \
180  } \
181  for (; j < count; j++) { \
182  /* apply window & overlap with previous buffer */ \
183  \
184  /* select window */ \
185  int win_idx = (switch_point && j < 2) ? 0 : block_type; \
186  float *win = ff_mdct_win_float[win_idx + (4 & -(j & 1))]; \
187  \
188  ff_imdct36_float_ ## CPU1(out, buf, in, win); \
189  \
190  in += 18; \
191  buf++; \
192  out++; \
193  } \
194 }
195 
196 DECL_IMDCT_BLOCKS(sse,sse)
197 DECL_IMDCT_BLOCKS(sse2,sse)
198 DECL_IMDCT_BLOCKS(sse3,sse)
199 DECL_IMDCT_BLOCKS(ssse3,sse)
200 DECL_IMDCT_BLOCKS(avx,avx)
201 #endif /* HAVE_YASM */
202 
204 {
205  int mm_flags = av_get_cpu_flags();
206 
207  int i, j;
208  for (j = 0; j < 4; j++) {
209  for (i = 0; i < 40; i ++) {
210  mdct_win_sse[0][j][4*i ] = ff_mdct_win_float[j ][i];
211  mdct_win_sse[0][j][4*i + 1] = ff_mdct_win_float[j + 4][i];
212  mdct_win_sse[0][j][4*i + 2] = ff_mdct_win_float[j ][i];
213  mdct_win_sse[0][j][4*i + 3] = ff_mdct_win_float[j + 4][i];
214  mdct_win_sse[1][j][4*i ] = ff_mdct_win_float[0 ][i];
215  mdct_win_sse[1][j][4*i + 1] = ff_mdct_win_float[4 ][i];
216  mdct_win_sse[1][j][4*i + 2] = ff_mdct_win_float[j ][i];
217  mdct_win_sse[1][j][4*i + 3] = ff_mdct_win_float[j + 4][i];
218  }
219  }
220 
221  if (mm_flags & AV_CPU_FLAG_SSE2) {
223  }
224 #if HAVE_YASM
225  if (mm_flags & AV_CPU_FLAG_AVX && HAVE_AVX) {
226  s->imdct36_blocks_float = imdct36_blocks_avx;
227 #if HAVE_SSE
228  } else if (mm_flags & AV_CPU_FLAG_SSSE3) {
229  s->imdct36_blocks_float = imdct36_blocks_ssse3;
230  } else if (mm_flags & AV_CPU_FLAG_SSE3) {
231  s->imdct36_blocks_float = imdct36_blocks_sse3;
232  } else if (mm_flags & AV_CPU_FLAG_SSE2) {
233  s->imdct36_blocks_float = imdct36_blocks_sse2;
234  } else if (mm_flags & AV_CPU_FLAG_SSE) {
235  s->imdct36_blocks_float = imdct36_blocks_sse;
236 #endif /* HAVE_SSE */
237  }
238 #endif /* HAVE_YASM */
239 }