2 * MMX optimized MP3 decoding functions
3 * Copyright (c) 2010 Vitor Sessak
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/cpu.h"
23 #include "libavutil/x86/asm.h"
24 #include "libavutil/x86/cpu.h"
25 #include "libavcodec/dsputil.h"
26 #include "libavcodec/mpegaudiodsp.h"
28 void ff_imdct36_float_sse(float *out, float *buf, float *in, float *win);
29 void ff_imdct36_float_sse2(float *out, float *buf, float *in, float *win);
30 void ff_imdct36_float_sse3(float *out, float *buf, float *in, float *win);
31 void ff_imdct36_float_ssse3(float *out, float *buf, float *in, float *win);
32 void ff_imdct36_float_avx(float *out, float *buf, float *in, float *win);
33 void ff_four_imdct36_float_sse(float *out, float *buf, float *in, float *win,
35 void ff_four_imdct36_float_avx(float *out, float *buf, float *in, float *win,
38 DECLARE_ALIGNED(16, static float, mdct_win_sse)[2][4][4*40];
42 #define MACS(rt, ra, rb) rt+=(ra)*(rb)
43 #define MLSS(rt, ra, rb) rt-=(ra)*(rb)
45 #define SUM8(op, sum, w, p) \
47 op(sum, (w)[0 * 64], (p)[0 * 64]); \
48 op(sum, (w)[1 * 64], (p)[1 * 64]); \
49 op(sum, (w)[2 * 64], (p)[2 * 64]); \
50 op(sum, (w)[3 * 64], (p)[3 * 64]); \
51 op(sum, (w)[4 * 64], (p)[4 * 64]); \
52 op(sum, (w)[5 * 64], (p)[5 * 64]); \
53 op(sum, (w)[6 * 64], (p)[6 * 64]); \
54 op(sum, (w)[7 * 64], (p)[7 * 64]); \
57 static void apply_window(const float *buf, const float *win1,
58 const float *win2, float *sum1, float *sum2, int len)
60 x86_reg count = - 4*len;
61 const float *win1a = win1+len;
62 const float *win2a = win2+len;
63 const float *bufa = buf+len;
64 float *sum1a = sum1+len;
65 float *sum2a = sum2+len;
69 "movaps " #a "(%1,%0), %%xmm1 \n\t" \
70 "movaps " #a "(%3,%0), %%xmm2 \n\t" \
71 "mulps %%xmm2, %%xmm1 \n\t" \
72 "subps %%xmm1, %%xmm0 \n\t" \
73 "mulps " #b "(%2,%0), %%xmm2 \n\t" \
74 "subps %%xmm2, %%xmm4 \n\t" \
78 "xorps %%xmm0, %%xmm0 \n\t"
79 "xorps %%xmm4, %%xmm4 \n\t"
90 "movaps %%xmm0, (%4,%0) \n\t"
91 "movaps %%xmm4, (%5,%0) \n\t"
95 :"r"(win1a), "r"(win2a), "r"(bufa), "r"(sum1a), "r"(sum2a)
101 static void apply_window_mp3(float *in, float *win, int *unused, float *out,
104 LOCAL_ALIGNED_16(float, suma, [17]);
105 LOCAL_ALIGNED_16(float, sumb, [17]);
106 LOCAL_ALIGNED_16(float, sumc, [17]);
107 LOCAL_ALIGNED_16(float, sumd, [17]);
111 /* copy to avoid wrap */
113 "movaps 0(%0), %%xmm0 \n\t" \
114 "movaps 16(%0), %%xmm1 \n\t" \
115 "movaps 32(%0), %%xmm2 \n\t" \
116 "movaps 48(%0), %%xmm3 \n\t" \
117 "movaps %%xmm0, 0(%1) \n\t" \
118 "movaps %%xmm1, 16(%1) \n\t" \
119 "movaps %%xmm2, 32(%1) \n\t" \
120 "movaps %%xmm3, 48(%1) \n\t" \
121 "movaps 64(%0), %%xmm0 \n\t" \
122 "movaps 80(%0), %%xmm1 \n\t" \
123 "movaps 96(%0), %%xmm2 \n\t" \
124 "movaps 112(%0), %%xmm3 \n\t" \
125 "movaps %%xmm0, 64(%1) \n\t" \
126 "movaps %%xmm1, 80(%1) \n\t" \
127 "movaps %%xmm2, 96(%1) \n\t" \
128 "movaps %%xmm3, 112(%1) \n\t"
129 ::"r"(in), "r"(in+512)
133 apply_window(in + 16, win , win + 512, suma, sumc, 16);
134 apply_window(in + 32, win + 48, win + 640, sumb, sumd, 16);
136 SUM8(MACS, suma[0], win + 32, in + 48);
142 #define SUMS(suma, sumb, sumc, sumd, out1, out2) \
143 "movups " #sumd "(%4), %%xmm0 \n\t" \
144 "shufps $0x1b, %%xmm0, %%xmm0 \n\t" \
145 "subps " #suma "(%1), %%xmm0 \n\t" \
146 "movaps %%xmm0," #out1 "(%0) \n\t" \
148 "movups " #sumc "(%3), %%xmm0 \n\t" \
149 "shufps $0x1b, %%xmm0, %%xmm0 \n\t" \
150 "addps " #sumb "(%2), %%xmm0 \n\t" \
151 "movaps %%xmm0," #out2 "(%0) \n\t"
155 SUMS( 0, 48, 4, 52, 0, 112)
156 SUMS(16, 32, 20, 36, 16, 96)
157 SUMS(32, 16, 36, 20, 32, 80)
158 SUMS(48, 0, 52, 4, 48, 64)
161 :"r"(&suma[0]), "r"(&sumb[0]), "r"(&sumc[0]), "r"(&sumd[0])
167 float *out2 = out + 32 * incr;
172 *out = -suma[ j] + sumd[16-j];
173 *out2 = sumb[16-j] + sumc[ j];
180 SUM8(MLSS, sum, win + 16 + 32, in + 32);
184 #endif /* HAVE_SSE2_INLINE */
187 #define DECL_IMDCT_BLOCKS(CPU1, CPU2) \
188 static void imdct36_blocks_ ## CPU1(float *out, float *buf, float *in, \
189 int count, int switch_point, int block_type) \
191 int align_end = count - (count & 3); \
193 for (j = 0; j < align_end; j+= 4) { \
194 LOCAL_ALIGNED_16(float, tmpbuf, [1024]); \
195 float *win = mdct_win_sse[switch_point && j < 4][block_type]; \
196 /* apply window & overlap with previous buffer */ \
198 /* select window */ \
199 ff_four_imdct36_float_ ## CPU2(out, buf, in, win, tmpbuf); \
204 for (; j < count; j++) { \
205 /* apply window & overlap with previous buffer */ \
207 /* select window */ \
208 int win_idx = (switch_point && j < 2) ? 0 : block_type; \
209 float *win = ff_mdct_win_float[win_idx + (4 & -(j & 1))]; \
211 ff_imdct36_float_ ## CPU1(out, buf, in, win); \
219 DECL_IMDCT_BLOCKS(sse,sse)
220 DECL_IMDCT_BLOCKS(sse2,sse)
221 DECL_IMDCT_BLOCKS(sse3,sse)
222 DECL_IMDCT_BLOCKS(ssse3,sse)
223 DECL_IMDCT_BLOCKS(avx,avx)
224 #endif /* HAVE_YASM */
226 void ff_mpadsp_init_mmx(MPADSPContext *s)
228 int mm_flags = av_get_cpu_flags();
231 for (j = 0; j < 4; j++) {
232 for (i = 0; i < 40; i ++) {
233 mdct_win_sse[0][j][4*i ] = ff_mdct_win_float[j ][i];
234 mdct_win_sse[0][j][4*i + 1] = ff_mdct_win_float[j + 4][i];
235 mdct_win_sse[0][j][4*i + 2] = ff_mdct_win_float[j ][i];
236 mdct_win_sse[0][j][4*i + 3] = ff_mdct_win_float[j + 4][i];
237 mdct_win_sse[1][j][4*i ] = ff_mdct_win_float[0 ][i];
238 mdct_win_sse[1][j][4*i + 1] = ff_mdct_win_float[4 ][i];
239 mdct_win_sse[1][j][4*i + 2] = ff_mdct_win_float[j ][i];
240 mdct_win_sse[1][j][4*i + 3] = ff_mdct_win_float[j + 4][i];
245 if (mm_flags & AV_CPU_FLAG_SSE2) {
246 s->apply_window_float = apply_window_mp3;
248 #endif /* HAVE_SSE2_INLINE */
251 if (EXTERNAL_AVX(mm_flags)) {
252 s->imdct36_blocks_float = imdct36_blocks_avx;
253 } else if (EXTERNAL_SSSE3(mm_flags)) {
254 s->imdct36_blocks_float = imdct36_blocks_ssse3;
255 } else if (EXTERNAL_SSE3(mm_flags)) {
256 s->imdct36_blocks_float = imdct36_blocks_sse3;
257 } else if (EXTERNAL_SSE2(mm_flags)) {
258 s->imdct36_blocks_float = imdct36_blocks_sse2;
259 } else if (EXTERNAL_SSE(mm_flags)) {
260 s->imdct36_blocks_float = imdct36_blocks_sse;
262 #endif /* HAVE_YASM */