Libav
ac3dsp_init.c
Go to the documentation of this file.
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_sse2 (uint8_t *exp, int32_t *coef, int nb_coefs);
51 void ff_ac3_extract_exponents_ssse3(uint8_t *exp, int32_t *coef, int nb_coefs);
52 
53 void ff_apply_window_int16_round_mmxext(int16_t *output, const int16_t *input,
54  const int16_t *window, unsigned int len);
55 void ff_apply_window_int16_round_sse2(int16_t *output, const int16_t *input,
56  const int16_t *window, unsigned int len);
57 void ff_apply_window_int16_mmxext(int16_t *output, const int16_t *input,
58  const int16_t *window, unsigned int len);
59 void ff_apply_window_int16_sse2(int16_t *output, const int16_t *input,
60  const int16_t *window, unsigned int len);
61 void ff_apply_window_int16_ssse3(int16_t *output, const int16_t *input,
62  const int16_t *window, unsigned int len);
63 void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input,
64  const int16_t *window, unsigned int len);
65 
66 #if HAVE_SSE_INLINE && HAVE_7REGS
67 
68 #define IF1(x) x
69 #define IF0(x)
70 
71 #define MIX5(mono, stereo) \
72  __asm__ volatile ( \
73  "movss 0(%1), %%xmm5 \n" \
74  "movss 8(%1), %%xmm6 \n" \
75  "movss 24(%1), %%xmm7 \n" \
76  "shufps $0, %%xmm5, %%xmm5 \n" \
77  "shufps $0, %%xmm6, %%xmm6 \n" \
78  "shufps $0, %%xmm7, %%xmm7 \n" \
79  "1: \n" \
80  "movaps (%0, %2), %%xmm0 \n" \
81  "movaps (%0, %3), %%xmm1 \n" \
82  "movaps (%0, %4), %%xmm2 \n" \
83  "movaps (%0, %5), %%xmm3 \n" \
84  "movaps (%0, %6), %%xmm4 \n" \
85  "mulps %%xmm5, %%xmm0 \n" \
86  "mulps %%xmm6, %%xmm1 \n" \
87  "mulps %%xmm5, %%xmm2 \n" \
88  "mulps %%xmm7, %%xmm3 \n" \
89  "mulps %%xmm7, %%xmm4 \n" \
90  stereo("addps %%xmm1, %%xmm0 \n") \
91  "addps %%xmm1, %%xmm2 \n" \
92  "addps %%xmm3, %%xmm0 \n" \
93  "addps %%xmm4, %%xmm2 \n" \
94  mono("addps %%xmm2, %%xmm0 \n") \
95  "movaps %%xmm0, (%0, %2) \n" \
96  stereo("movaps %%xmm2, (%0, %3) \n") \
97  "add $16, %0 \n" \
98  "jl 1b \n" \
99  : "+&r"(i) \
100  : "r"(matrix), \
101  "r"(samples[0] + len), \
102  "r"(samples[1] + len), \
103  "r"(samples[2] + len), \
104  "r"(samples[3] + len), \
105  "r"(samples[4] + len) \
106  : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
107  "%xmm4", "%xmm5", "%xmm6", "%xmm7",) \
108  "memory" \
109  );
110 
111 #define MIX_MISC(stereo) \
112  __asm__ volatile ( \
113  "mov %5, %2 \n" \
114  "1: \n" \
115  "mov -%c7(%6, %2, %c8), %3 \n" \
116  "movaps (%3, %0), %%xmm0 \n" \
117  stereo("movaps %%xmm0, %%xmm1 \n") \
118  "mulps %%xmm4, %%xmm0 \n" \
119  stereo("mulps %%xmm5, %%xmm1 \n") \
120  "2: \n" \
121  "mov (%6, %2, %c8), %1 \n" \
122  "movaps (%1, %0), %%xmm2 \n" \
123  stereo("movaps %%xmm2, %%xmm3 \n") \
124  "mulps (%4, %2, 8), %%xmm2 \n" \
125  stereo("mulps 16(%4, %2, 8), %%xmm3 \n") \
126  "addps %%xmm2, %%xmm0 \n" \
127  stereo("addps %%xmm3, %%xmm1 \n") \
128  "add $4, %2 \n" \
129  "jl 2b \n" \
130  "mov %5, %2 \n" \
131  stereo("mov (%6, %2, %c8), %1 \n") \
132  "movaps %%xmm0, (%3, %0) \n" \
133  stereo("movaps %%xmm1, (%1, %0) \n") \
134  "add $16, %0 \n" \
135  "jl 1b \n" \
136  : "+&r"(i), "=&r"(j), "=&r"(k), "=&r"(m) \
137  : "r"(matrix_simd + in_ch), \
138  "g"((intptr_t) - 4 * (in_ch - 1)), \
139  "r"(samp + in_ch), \
140  "i"(sizeof(float *)), "i"(sizeof(float *)/4) \
141  : "memory" \
142  );
143 
144 static void ac3_downmix_sse(float **samples, float (*matrix)[2],
145  int out_ch, int in_ch, int len)
146 {
147  int (*matrix_cmp)[2] = (int(*)[2])matrix;
148  intptr_t i, j, k, m;
149 
150  i = -len * sizeof(float);
151  if (in_ch == 5 && out_ch == 2 &&
152  !(matrix_cmp[0][1] | matrix_cmp[2][0] |
153  matrix_cmp[3][1] | matrix_cmp[4][0] |
154  (matrix_cmp[1][0] ^ matrix_cmp[1][1]) |
155  (matrix_cmp[0][0] ^ matrix_cmp[2][1]))) {
156  MIX5(IF0, IF1);
157  } else if (in_ch == 5 && out_ch == 1 &&
158  matrix_cmp[0][0] == matrix_cmp[2][0] &&
159  matrix_cmp[3][0] == matrix_cmp[4][0]) {
160  MIX5(IF1, IF0);
161  } else {
162  DECLARE_ALIGNED(16, float, matrix_simd)[AC3_MAX_CHANNELS][2][4];
163  float *samp[AC3_MAX_CHANNELS];
164 
165  for (j = 0; j < in_ch; j++)
166  samp[j] = samples[j] + len;
167 
168  j = 2 * in_ch * sizeof(float);
169  __asm__ volatile (
170  "1: \n"
171  "sub $8, %0 \n"
172  "movss (%2, %0), %%xmm4 \n"
173  "movss 4(%2, %0), %%xmm5 \n"
174  "shufps $0, %%xmm4, %%xmm4 \n"
175  "shufps $0, %%xmm5, %%xmm5 \n"
176  "movaps %%xmm4, (%1, %0, 4) \n"
177  "movaps %%xmm5, 16(%1, %0, 4) \n"
178  "jg 1b \n"
179  : "+&r"(j)
180  : "r"(matrix_simd), "r"(matrix)
181  : "memory"
182  );
183  if (out_ch == 2) {
184  MIX_MISC(IF1);
185  } else {
186  MIX_MISC(IF0);
187  }
188  }
189 }
190 
191 #endif /* HAVE_SSE_INLINE && HAVE_7REGS */
192 
193 av_cold void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact)
194 {
195  int cpu_flags = av_get_cpu_flags();
196 
197  if (EXTERNAL_MMX(cpu_flags)) {
202  }
203  if (EXTERNAL_AMD3DNOW(cpu_flags)) {
204  if (!bit_exact) {
206  }
207  }
208  if (EXTERNAL_MMXEXT(cpu_flags)) {
211  if (bit_exact) {
213  } else {
215  }
216  }
217  if (EXTERNAL_SSE(cpu_flags)) {
219  }
220  if (EXTERNAL_SSE2(cpu_flags)) {
226  if (bit_exact) {
228  }
229  }
230 
231  if (EXTERNAL_SSE2_FAST(cpu_flags)) {
234  if (!bit_exact) {
236  }
237  }
238 
239  if (EXTERNAL_SSSE3(cpu_flags)) {
241  if (cpu_flags & AV_CPU_FLAG_ATOM) {
243  } else {
246  }
247  }
248 
249 #if HAVE_SSE_INLINE && HAVE_7REGS
250  if (INLINE_SSE(cpu_flags)) {
251  c->downmix = ac3_downmix_sse;
252  }
253 #endif
254 }
#define EXTERNAL_MMX(flags)
Definition: cpu.h:55
int ff_ac3_max_msb_abs_int16_sse2(const int16_t *src, int len)
memory handling functions
int ff_ac3_max_msb_abs_int16_mmx(const int16_t *src, int len)
#define IF1(x)
#define DECLARE_ALIGNED(n, t, v)
Definition: mem.h:58
#define EXTERNAL_SSE(flags)
Definition: cpu.h:57
void ff_float_to_fixed24_3dnow(int32_t *dst, const float *src, unsigned int len)
Macro definitions for various function/variable attributes.
void(* extract_exponents)(uint8_t *exp, int32_t *coef, int nb_coefs)
Definition: ac3dsp.h:127
void ff_float_to_fixed24_sse2(int32_t *dst, const float *src, unsigned int len)
void(* apply_window_int16)(int16_t *output, const int16_t *input, const int16_t *window, unsigned int len)
Apply symmetric window in 16-bit fixed-point.
Definition: ac3dsp.h:143
uint8_t
#define av_cold
Definition: attributes.h:66
void ff_ac3_extract_exponents_ssse3(uint8_t *exp, int32_t *coef, int nb_coefs)
void ff_ac3_rshift_int32_mmx(int32_t *src, unsigned int len, unsigned int shift)
void ff_ac3_lshift_int16_mmx(int16_t *src, unsigned int len, unsigned int shift)
void ff_ac3_lshift_int16_sse2(int16_t *src, unsigned int len, unsigned int shift)
#define AV_CPU_FLAG_ATOM
Atom processor, some SSSE3 instructions are slower.
Definition: cpu.h:42
#define EXTERNAL_SSE2_FAST(flags)
Definition: cpu.h:59
#define EXTERNAL_SSE2(flags)
Definition: cpu.h:58
void ff_ac3_exponent_min_sse2(uint8_t *exp, int num_reuse_blocks, int nb_coefs)
void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input, const int16_t *window, unsigned int len)
#define src
Definition: vp8dsp.c:254
av_cold void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact)
Definition: ac3dsp_init.c:193
void ff_apply_window_int16_round_mmxext(int16_t *output, const int16_t *input, const int16_t *window, unsigned int len)
#define AC3_MAX_CHANNELS
maximum number of channels, including coupling channel
Definition: ac3.h:31
void ff_ac3_rshift_int32_sse2(int32_t *src, unsigned int len, unsigned int shift)
int(* ac3_max_msb_abs_int16)(const int16_t *src, int len)
Calculate the maximum MSB of the absolute value of each element in an array of int16_t.
Definition: ac3dsp.h:54
void(* ac3_lshift_int16)(int16_t *src, unsigned int len, unsigned int shift)
Left-shift each value in an array of int16_t by a specified amount.
Definition: ac3dsp.h:65
void(* ac3_rshift_int32)(int32_t *src, unsigned int len, unsigned int shift)
Right-shift each value in an array of int32_t by a specified amount.
Definition: ac3dsp.h:76
void ff_apply_window_int16_round_sse2(int16_t *output, const int16_t *input, const int16_t *window, unsigned int len)
int ff_ac3_max_msb_abs_int16_ssse3(const int16_t *src, int len)
#define IF0(x)
int32_t
void ff_apply_window_int16_sse2(int16_t *output, const int16_t *input, const int16_t *window, unsigned int len)
#define INLINE_SSE(flags)
Definition: cpu.h:81
void ff_ac3_exponent_min_mmx(uint8_t *exp, int num_reuse_blocks, int nb_coefs)
void ff_ac3_exponent_min_mmxext(uint8_t *exp, int num_reuse_blocks, int nb_coefs)
void ff_apply_window_int16_mmxext(int16_t *output, const int16_t *input, const int16_t *window, unsigned int len)
void ff_apply_window_int16_ssse3(int16_t *output, const int16_t *input, const int16_t *window, unsigned int len)
void(* downmix)(float **samples, float(*matrix)[2], int out_ch, int in_ch, int len)
Definition: ac3dsp.h:129
int ff_ac3_max_msb_abs_int16_mmxext(const int16_t *src, int len)
void ff_float_to_fixed24_sse(int32_t *dst, const float *src, unsigned int len)
#define EXTERNAL_SSSE3(flags)
Definition: cpu.h:64
int(* compute_mantissa_size)(uint16_t mant_cnt[6][16])
Calculate the number of bits needed to encode a set of mantissas.
Definition: ac3dsp.h:125
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:47
void(* ac3_exponent_min)(uint8_t *exp, int num_reuse_blocks, int nb_coefs)
Set each encoded exponent in a block to the minimum of itself and the exponents in the same frequency...
Definition: ac3dsp.h:43
#define EXTERNAL_MMXEXT(flags)
Definition: cpu.h:56
int ff_ac3_compute_mantissa_size_sse2(uint16_t mant_cnt[6][16])
int len
void ff_ac3_extract_exponents_sse2(uint8_t *exp, int32_t *coef, int nb_coefs)
#define EXTERNAL_AMD3DNOW(flags)
Definition: cpu.h:53
Common code between the AC-3 encoder and decoder.
void(* float_to_fixed24)(int32_t *dst, const float *src, unsigned int len)
Convert an array of float in range [-1.0,1.0] to int32_t with range [-(1<<24),(1<<24)].
Definition: ac3dsp.h:89