cd638b92288d980fea75e0a5802704d535496b0b
[libav.git] / libavcodec / x86 / ac3dsp_init.c
1 /*
2 * x86-optimized AC-3 DSP functions
3 * Copyright (c) 2011 Justin Ruggles
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/attributes.h"
23 #include "libavutil/mem.h"
24 #include "libavutil/x86/asm.h"
25 #include "libavutil/x86/cpu.h"
26 #include "libavcodec/ac3.h"
27 #include "libavcodec/ac3dsp.h"
28
29 void ff_ac3_exponent_min_mmx (uint8_t *exp, int num_reuse_blocks, int nb_coefs);
30 void ff_ac3_exponent_min_mmxext(uint8_t *exp, int num_reuse_blocks, int nb_coefs);
31 void ff_ac3_exponent_min_sse2 (uint8_t *exp, int num_reuse_blocks, int nb_coefs);
32
33 int ff_ac3_max_msb_abs_int16_mmx (const int16_t *src, int len);
34 int ff_ac3_max_msb_abs_int16_mmxext(const int16_t *src, int len);
35 int ff_ac3_max_msb_abs_int16_sse2 (const int16_t *src, int len);
36 int ff_ac3_max_msb_abs_int16_ssse3(const int16_t *src, int len);
37
38 void ff_ac3_lshift_int16_mmx (int16_t *src, unsigned int len, unsigned int shift);
39 void ff_ac3_lshift_int16_sse2(int16_t *src, unsigned int len, unsigned int shift);
40
41 void ff_ac3_rshift_int32_mmx (int32_t *src, unsigned int len, unsigned int shift);
42 void ff_ac3_rshift_int32_sse2(int32_t *src, unsigned int len, unsigned int shift);
43
44 void ff_float_to_fixed24_3dnow(int32_t *dst, const float *src, unsigned int len);
45 void ff_float_to_fixed24_sse (int32_t *dst, const float *src, unsigned int len);
46 void ff_float_to_fixed24_sse2 (int32_t *dst, const float *src, unsigned int len);
47
48 int ff_ac3_compute_mantissa_size_sse2(uint16_t mant_cnt[6][16]);
49
50 void ff_ac3_extract_exponents_3dnow(uint8_t *exp, int32_t *coef, int nb_coefs);
51 void ff_ac3_extract_exponents_sse2 (uint8_t *exp, int32_t *coef, int nb_coefs);
52 void ff_ac3_extract_exponents_ssse3(uint8_t *exp, int32_t *coef, int nb_coefs);
53
54 void ff_apply_window_int16_round_mmxext(int16_t *output, const int16_t *input,
55 const int16_t *window, unsigned int len);
56 void ff_apply_window_int16_round_sse2(int16_t *output, const int16_t *input,
57 const int16_t *window, unsigned int len);
58 void ff_apply_window_int16_mmxext(int16_t *output, const int16_t *input,
59 const int16_t *window, unsigned int len);
60 void ff_apply_window_int16_sse2(int16_t *output, const int16_t *input,
61 const int16_t *window, unsigned int len);
62 void ff_apply_window_int16_ssse3(int16_t *output, const int16_t *input,
63 const int16_t *window, unsigned int len);
64 void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input,
65 const int16_t *window, unsigned int len);
66
67 #if HAVE_SSE_INLINE && HAVE_7REGS
68
69 #define IF1(x) x
70 #define IF0(x)
71
72 #define MIX5(mono, stereo) \
73 __asm__ volatile ( \
74 "movss 0(%1), %%xmm5 \n" \
75 "movss 8(%1), %%xmm6 \n" \
76 "movss 24(%1), %%xmm7 \n" \
77 "shufps $0, %%xmm5, %%xmm5 \n" \
78 "shufps $0, %%xmm6, %%xmm6 \n" \
79 "shufps $0, %%xmm7, %%xmm7 \n" \
80 "1: \n" \
81 "movaps (%0, %2), %%xmm0 \n" \
82 "movaps (%0, %3), %%xmm1 \n" \
83 "movaps (%0, %4), %%xmm2 \n" \
84 "movaps (%0, %5), %%xmm3 \n" \
85 "movaps (%0, %6), %%xmm4 \n" \
86 "mulps %%xmm5, %%xmm0 \n" \
87 "mulps %%xmm6, %%xmm1 \n" \
88 "mulps %%xmm5, %%xmm2 \n" \
89 "mulps %%xmm7, %%xmm3 \n" \
90 "mulps %%xmm7, %%xmm4 \n" \
91 stereo("addps %%xmm1, %%xmm0 \n") \
92 "addps %%xmm1, %%xmm2 \n" \
93 "addps %%xmm3, %%xmm0 \n" \
94 "addps %%xmm4, %%xmm2 \n" \
95 mono("addps %%xmm2, %%xmm0 \n") \
96 "movaps %%xmm0, (%0, %2) \n" \
97 stereo("movaps %%xmm2, (%0, %3) \n") \
98 "add $16, %0 \n" \
99 "jl 1b \n" \
100 : "+&r"(i) \
101 : "r"(matrix), \
102 "r"(samples[0] + len), \
103 "r"(samples[1] + len), \
104 "r"(samples[2] + len), \
105 "r"(samples[3] + len), \
106 "r"(samples[4] + len) \
107 : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
108 "%xmm4", "%xmm5", "%xmm6", "%xmm7",) \
109 "memory" \
110 );
111
112 #define MIX_MISC(stereo) \
113 __asm__ volatile ( \
114 "mov %5, %2 \n" \
115 "1: \n" \
116 "mov -%c7(%6, %2, %c8), %3 \n" \
117 "movaps (%3, %0), %%xmm0 \n" \
118 stereo("movaps %%xmm0, %%xmm1 \n") \
119 "mulps %%xmm4, %%xmm0 \n" \
120 stereo("mulps %%xmm5, %%xmm1 \n") \
121 "2: \n" \
122 "mov (%6, %2, %c8), %1 \n" \
123 "movaps (%1, %0), %%xmm2 \n" \
124 stereo("movaps %%xmm2, %%xmm3 \n") \
125 "mulps (%4, %2, 8), %%xmm2 \n" \
126 stereo("mulps 16(%4, %2, 8), %%xmm3 \n") \
127 "addps %%xmm2, %%xmm0 \n" \
128 stereo("addps %%xmm3, %%xmm1 \n") \
129 "add $4, %2 \n" \
130 "jl 2b \n" \
131 "mov %5, %2 \n" \
132 stereo("mov (%6, %2, %c8), %1 \n") \
133 "movaps %%xmm0, (%3, %0) \n" \
134 stereo("movaps %%xmm1, (%1, %0) \n") \
135 "add $16, %0 \n" \
136 "jl 1b \n" \
137 : "+&r"(i), "=&r"(j), "=&r"(k), "=&r"(m) \
138 : "r"(matrix_simd + in_ch), \
139 "g"((intptr_t) - 4 * (in_ch - 1)), \
140 "r"(samp + in_ch), \
141 "i"(sizeof(float *)), "i"(sizeof(float *)/4) \
142 : "memory" \
143 );
144
145 static void ac3_downmix_sse(float **samples, float (*matrix)[2],
146 int out_ch, int in_ch, int len)
147 {
148 int (*matrix_cmp)[2] = (int(*)[2])matrix;
149 intptr_t i, j, k, m;
150
151 i = -len * sizeof(float);
152 if (in_ch == 5 && out_ch == 2 &&
153 !(matrix_cmp[0][1] | matrix_cmp[2][0] |
154 matrix_cmp[3][1] | matrix_cmp[4][0] |
155 (matrix_cmp[1][0] ^ matrix_cmp[1][1]) |
156 (matrix_cmp[0][0] ^ matrix_cmp[2][1]))) {
157 MIX5(IF0, IF1);
158 } else if (in_ch == 5 && out_ch == 1 &&
159 matrix_cmp[0][0] == matrix_cmp[2][0] &&
160 matrix_cmp[3][0] == matrix_cmp[4][0]) {
161 MIX5(IF1, IF0);
162 } else {
163 DECLARE_ALIGNED(16, float, matrix_simd)[AC3_MAX_CHANNELS][2][4];
164 float *samp[AC3_MAX_CHANNELS];
165
166 for (j = 0; j < in_ch; j++)
167 samp[j] = samples[j] + len;
168
169 j = 2 * in_ch * sizeof(float);
170 __asm__ volatile (
171 "1: \n"
172 "sub $8, %0 \n"
173 "movss (%2, %0), %%xmm4 \n"
174 "movss 4(%2, %0), %%xmm5 \n"
175 "shufps $0, %%xmm4, %%xmm4 \n"
176 "shufps $0, %%xmm5, %%xmm5 \n"
177 "movaps %%xmm4, (%1, %0, 4) \n"
178 "movaps %%xmm5, 16(%1, %0, 4) \n"
179 "jg 1b \n"
180 : "+&r"(j)
181 : "r"(matrix_simd), "r"(matrix)
182 : "memory"
183 );
184 if (out_ch == 2) {
185 MIX_MISC(IF1);
186 } else {
187 MIX_MISC(IF0);
188 }
189 }
190 }
191
192 #endif /* HAVE_SSE_INLINE && HAVE_7REGS */
193
194 av_cold void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact)
195 {
196 int cpu_flags = av_get_cpu_flags();
197
198 if (EXTERNAL_MMX(cpu_flags)) {
199 c->ac3_exponent_min = ff_ac3_exponent_min_mmx;
200 c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_mmx;
201 c->ac3_lshift_int16 = ff_ac3_lshift_int16_mmx;
202 c->ac3_rshift_int32 = ff_ac3_rshift_int32_mmx;
203 }
204 if (EXTERNAL_AMD3DNOW(cpu_flags)) {
205 if (!bit_exact) {
206 c->float_to_fixed24 = ff_float_to_fixed24_3dnow;
207 }
208 }
209 if (EXTERNAL_MMXEXT(cpu_flags)) {
210 c->ac3_exponent_min = ff_ac3_exponent_min_mmxext;
211 c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_mmxext;
212 if (bit_exact) {
213 c->apply_window_int16 = ff_apply_window_int16_mmxext;
214 } else {
215 c->apply_window_int16 = ff_apply_window_int16_round_mmxext;
216 }
217 }
218 if (EXTERNAL_SSE(cpu_flags)) {
219 c->float_to_fixed24 = ff_float_to_fixed24_sse;
220 }
221 if (EXTERNAL_SSE2(cpu_flags)) {
222 c->ac3_exponent_min = ff_ac3_exponent_min_sse2;
223 c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_sse2;
224 c->float_to_fixed24 = ff_float_to_fixed24_sse2;
225 c->compute_mantissa_size = ff_ac3_compute_mantissa_size_sse2;
226 c->extract_exponents = ff_ac3_extract_exponents_sse2;
227 if (!(cpu_flags & AV_CPU_FLAG_SSE2SLOW)) {
228 c->ac3_lshift_int16 = ff_ac3_lshift_int16_sse2;
229 c->ac3_rshift_int32 = ff_ac3_rshift_int32_sse2;
230 }
231 if (bit_exact) {
232 c->apply_window_int16 = ff_apply_window_int16_sse2;
233 } else if (!(cpu_flags & AV_CPU_FLAG_SSE2SLOW)) {
234 c->apply_window_int16 = ff_apply_window_int16_round_sse2;
235 }
236 }
237 if (EXTERNAL_SSSE3(cpu_flags)) {
238 c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_ssse3;
239 if (cpu_flags & AV_CPU_FLAG_ATOM) {
240 c->apply_window_int16 = ff_apply_window_int16_ssse3_atom;
241 } else {
242 c->extract_exponents = ff_ac3_extract_exponents_ssse3;
243 c->apply_window_int16 = ff_apply_window_int16_ssse3;
244 }
245 }
246
247 #if HAVE_SSE_INLINE && HAVE_7REGS
248 if (INLINE_SSE(cpu_flags)) {
249 c->downmix = ac3_downmix_sse;
250 }
251 #endif
252 }