2 * SIMD-optimized MP3 decoding functions
3 * Copyright (c) 2010 Vitor Sessak
5 * This file is part of FFmpeg.
7 * FFmpeg 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 * FFmpeg 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 FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 #include "libavutil/attributes.h"
23 #include "libavutil/cpu.h"
24 #include "libavutil/internal.h"
25 #include "libavutil/x86/asm.h"
26 #include "libavutil/x86/cpu.h"
27 #include "libavcodec/mpegaudiodsp.h"
30 static void imdct36_blocks_ ## CPU(float *out, float *buf, float *in, int count, int switch_point, int block_type);\
31 void ff_imdct36_float_ ## CPU(float *out, float *buf, float *in, float *win);
41 #endif /* HAVE_X86ASM */
43 void ff_four_imdct36_float_sse(float *out, float *buf, float *in, float *win,
45 void ff_four_imdct36_float_avx(float *out, float *buf, float *in, float *win,
48 DECLARE_ALIGNED(16, static float, mdct_win_sse)[2][4][4*40];
50 #if HAVE_6REGS && HAVE_SSE_INLINE
52 #define MACS(rt, ra, rb) rt+=(ra)*(rb)
53 #define MLSS(rt, ra, rb) rt-=(ra)*(rb)
55 #define SUM8(op, sum, w, p) \
57 op(sum, (w)[0 * 64], (p)[0 * 64]); \
58 op(sum, (w)[1 * 64], (p)[1 * 64]); \
59 op(sum, (w)[2 * 64], (p)[2 * 64]); \
60 op(sum, (w)[3 * 64], (p)[3 * 64]); \
61 op(sum, (w)[4 * 64], (p)[4 * 64]); \
62 op(sum, (w)[5 * 64], (p)[5 * 64]); \
63 op(sum, (w)[6 * 64], (p)[6 * 64]); \
64 op(sum, (w)[7 * 64], (p)[7 * 64]); \
67 static void apply_window(const float *buf, const float *win1,
68 const float *win2, float *sum1, float *sum2, int len)
70 x86_reg count = - 4*len;
71 const float *win1a = win1+len;
72 const float *win2a = win2+len;
73 const float *bufa = buf+len;
74 float *sum1a = sum1+len;
75 float *sum2a = sum2+len;
79 "movaps " #a "(%1,%0), %%xmm1 \n\t" \
80 "movaps " #a "(%3,%0), %%xmm2 \n\t" \
81 "mulps %%xmm2, %%xmm1 \n\t" \
82 "subps %%xmm1, %%xmm0 \n\t" \
83 "mulps " #b "(%2,%0), %%xmm2 \n\t" \
84 "subps %%xmm2, %%xmm4 \n\t" \
88 "xorps %%xmm0, %%xmm0 \n\t"
89 "xorps %%xmm4, %%xmm4 \n\t"
100 "movaps %%xmm0, (%4,%0) \n\t"
101 "movaps %%xmm4, (%5,%0) \n\t"
105 :"r"(win1a), "r"(win2a), "r"(bufa), "r"(sum1a), "r"(sum2a)
111 static void apply_window_mp3(float *in, float *win, int *unused, float *out,
114 LOCAL_ALIGNED_16(float, suma, [17]);
115 LOCAL_ALIGNED_16(float, sumb, [17]);
116 LOCAL_ALIGNED_16(float, sumc, [17]);
117 LOCAL_ALIGNED_16(float, sumd, [17]);
121 /* copy to avoid wrap */
123 "movaps 0(%0), %%xmm0 \n\t" \
124 "movaps 16(%0), %%xmm1 \n\t" \
125 "movaps 32(%0), %%xmm2 \n\t" \
126 "movaps 48(%0), %%xmm3 \n\t" \
127 "movaps %%xmm0, 0(%1) \n\t" \
128 "movaps %%xmm1, 16(%1) \n\t" \
129 "movaps %%xmm2, 32(%1) \n\t" \
130 "movaps %%xmm3, 48(%1) \n\t" \
131 "movaps 64(%0), %%xmm0 \n\t" \
132 "movaps 80(%0), %%xmm1 \n\t" \
133 "movaps 96(%0), %%xmm2 \n\t" \
134 "movaps 112(%0), %%xmm3 \n\t" \
135 "movaps %%xmm0, 64(%1) \n\t" \
136 "movaps %%xmm1, 80(%1) \n\t" \
137 "movaps %%xmm2, 96(%1) \n\t" \
138 "movaps %%xmm3, 112(%1) \n\t"
139 ::"r"(in), "r"(in+512)
143 apply_window(in + 16, win , win + 512, suma, sumc, 16);
144 apply_window(in + 32, win + 48, win + 640, sumb, sumd, 16);
146 SUM8(MACS, suma[0], win + 32, in + 48);
152 #define SUMS(suma, sumb, sumc, sumd, out1, out2) \
153 "movups " #sumd "(%4), %%xmm0 \n\t" \
154 "shufps $0x1b, %%xmm0, %%xmm0 \n\t" \
155 "subps " #suma "(%1), %%xmm0 \n\t" \
156 "movaps %%xmm0," #out1 "(%0) \n\t" \
158 "movups " #sumc "(%3), %%xmm0 \n\t" \
159 "shufps $0x1b, %%xmm0, %%xmm0 \n\t" \
160 "addps " #sumb "(%2), %%xmm0 \n\t" \
161 "movaps %%xmm0," #out2 "(%0) \n\t"
165 SUMS( 0, 48, 4, 52, 0, 112)
166 SUMS(16, 32, 20, 36, 16, 96)
167 SUMS(32, 16, 36, 20, 32, 80)
168 SUMS(48, 0, 52, 4, 48, 64)
171 :"r"(&suma[0]), "r"(&sumb[0]), "r"(&sumc[0]), "r"(&sumd[0])
177 float *out2 = out + 32 * incr;
182 *out = -suma[ j] + sumd[16-j];
183 *out2 = sumb[16-j] + sumc[ j];
190 SUM8(MLSS, sum, win + 16 + 32, in + 32);
194 #endif /* HAVE_6REGS && HAVE_SSE_INLINE */
197 #define DECL_IMDCT_BLOCKS(CPU1, CPU2) \
198 static void imdct36_blocks_ ## CPU1(float *out, float *buf, float *in, \
199 int count, int switch_point, int block_type) \
201 int align_end = count - (count & 3); \
203 for (j = 0; j < align_end; j+= 4) { \
204 LOCAL_ALIGNED_16(float, tmpbuf, [1024]); \
205 float *win = mdct_win_sse[switch_point && j < 4][block_type]; \
206 /* apply window & overlap with previous buffer */ \
208 /* select window */ \
209 ff_four_imdct36_float_ ## CPU2(out, buf, in, win, tmpbuf); \
214 for (; j < count; j++) { \
215 /* apply window & overlap with previous buffer */ \
217 /* select window */ \
218 int win_idx = (switch_point && j < 2) ? 0 : block_type; \
219 float *win = ff_mdct_win_float[win_idx + (4 & -(j & 1))]; \
221 ff_imdct36_float_ ## CPU1(out, buf, in, win); \
231 DECL_IMDCT_BLOCKS(sse,sse)
233 DECL_IMDCT_BLOCKS(sse2,sse)
234 DECL_IMDCT_BLOCKS(sse3,sse)
235 DECL_IMDCT_BLOCKS(ssse3,sse)
237 #if HAVE_AVX_EXTERNAL
238 DECL_IMDCT_BLOCKS(avx,avx)
240 #endif /* HAVE_X86ASM */
242 av_cold void ff_mpadsp_init_x86(MPADSPContext *s)
244 av_unused int cpu_flags = av_get_cpu_flags();
247 for (j = 0; j < 4; j++) {
248 for (i = 0; i < 40; i ++) {
249 mdct_win_sse[0][j][4*i ] = ff_mdct_win_float[j ][i];
250 mdct_win_sse[0][j][4*i + 1] = ff_mdct_win_float[j + 4][i];
251 mdct_win_sse[0][j][4*i + 2] = ff_mdct_win_float[j ][i];
252 mdct_win_sse[0][j][4*i + 3] = ff_mdct_win_float[j + 4][i];
253 mdct_win_sse[1][j][4*i ] = ff_mdct_win_float[0 ][i];
254 mdct_win_sse[1][j][4*i + 1] = ff_mdct_win_float[4 ][i];
255 mdct_win_sse[1][j][4*i + 2] = ff_mdct_win_float[j ][i];
256 mdct_win_sse[1][j][4*i + 3] = ff_mdct_win_float[j + 4][i];
260 #if HAVE_6REGS && HAVE_SSE_INLINE
261 if (INLINE_SSE(cpu_flags)) {
262 s->apply_window_float = apply_window_mp3;
264 #endif /* HAVE_SSE_INLINE */
269 if (EXTERNAL_SSE(cpu_flags)) {
270 s->imdct36_blocks_float = imdct36_blocks_sse;
273 if (EXTERNAL_SSE2(cpu_flags)) {
274 s->imdct36_blocks_float = imdct36_blocks_sse2;
276 if (EXTERNAL_SSE3(cpu_flags)) {
277 s->imdct36_blocks_float = imdct36_blocks_sse3;
279 if (EXTERNAL_SSSE3(cpu_flags)) {
280 s->imdct36_blocks_float = imdct36_blocks_ssse3;
283 #if HAVE_AVX_EXTERNAL
284 if (EXTERNAL_AVX(cpu_flags)) {
285 s->imdct36_blocks_float = imdct36_blocks_avx;
288 #endif /* HAVE_X86ASM */