2 * x86-optimized AC-3 DSP utils
3 * Copyright (c) 2011 Justin Ruggles
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/mem.h"
23 #include "libavutil/x86/asm.h"
24 #include "libavutil/x86/cpu.h"
25 #include "dsputil_mmx.h"
26 #include "libavcodec/ac3.h"
27 #include "libavcodec/ac3dsp.h"
29 extern void ff_ac3_exponent_min_mmx (uint8_t *exp, int num_reuse_blocks, int nb_coefs);
30 extern void ff_ac3_exponent_min_mmxext(uint8_t *exp, int num_reuse_blocks, int nb_coefs);
31 extern void ff_ac3_exponent_min_sse2 (uint8_t *exp, int num_reuse_blocks, int nb_coefs);
33 extern int ff_ac3_max_msb_abs_int16_mmx (const int16_t *src, int len);
34 extern int ff_ac3_max_msb_abs_int16_mmxext(const int16_t *src, int len);
35 extern int ff_ac3_max_msb_abs_int16_sse2 (const int16_t *src, int len);
36 extern int ff_ac3_max_msb_abs_int16_ssse3(const int16_t *src, int len);
38 extern void ff_ac3_lshift_int16_mmx (int16_t *src, unsigned int len, unsigned int shift);
39 extern void ff_ac3_lshift_int16_sse2(int16_t *src, unsigned int len, unsigned int shift);
41 extern void ff_ac3_rshift_int32_mmx (int32_t *src, unsigned int len, unsigned int shift);
42 extern void ff_ac3_rshift_int32_sse2(int32_t *src, unsigned int len, unsigned int shift);
44 extern void ff_float_to_fixed24_3dnow(int32_t *dst, const float *src, unsigned int len);
45 extern void ff_float_to_fixed24_sse (int32_t *dst, const float *src, unsigned int len);
46 extern void ff_float_to_fixed24_sse2 (int32_t *dst, const float *src, unsigned int len);
48 extern int ff_ac3_compute_mantissa_size_sse2(uint16_t mant_cnt[6][16]);
50 extern void ff_ac3_extract_exponents_3dnow(uint8_t *exp, int32_t *coef, int nb_coefs);
51 extern void ff_ac3_extract_exponents_sse2 (uint8_t *exp, int32_t *coef, int nb_coefs);
52 extern void ff_ac3_extract_exponents_ssse3(uint8_t *exp, int32_t *coef, int nb_coefs);
54 #if HAVE_SSE_INLINE && HAVE_7REGS
59 #define MIX5(mono, stereo) \
61 "movss 0(%1), %%xmm5 \n" \
62 "movss 8(%1), %%xmm6 \n" \
63 "movss 24(%1), %%xmm7 \n" \
64 "shufps $0, %%xmm5, %%xmm5 \n" \
65 "shufps $0, %%xmm6, %%xmm6 \n" \
66 "shufps $0, %%xmm7, %%xmm7 \n" \
68 "movaps (%0, %2), %%xmm0 \n" \
69 "movaps (%0, %3), %%xmm1 \n" \
70 "movaps (%0, %4), %%xmm2 \n" \
71 "movaps (%0, %5), %%xmm3 \n" \
72 "movaps (%0, %6), %%xmm4 \n" \
73 "mulps %%xmm5, %%xmm0 \n" \
74 "mulps %%xmm6, %%xmm1 \n" \
75 "mulps %%xmm5, %%xmm2 \n" \
76 "mulps %%xmm7, %%xmm3 \n" \
77 "mulps %%xmm7, %%xmm4 \n" \
78 stereo("addps %%xmm1, %%xmm0 \n") \
79 "addps %%xmm1, %%xmm2 \n" \
80 "addps %%xmm3, %%xmm0 \n" \
81 "addps %%xmm4, %%xmm2 \n" \
82 mono("addps %%xmm2, %%xmm0 \n") \
83 "movaps %%xmm0, (%0, %2) \n" \
84 stereo("movaps %%xmm2, (%0, %3) \n") \
89 "r"(samples[0] + len), \
90 "r"(samples[1] + len), \
91 "r"(samples[2] + len), \
92 "r"(samples[3] + len), \
93 "r"(samples[4] + len) \
94 : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
95 "%xmm4", "%xmm5", "%xmm6", "%xmm7",) \
99 #define MIX_MISC(stereo) \
103 "mov -%c7(%6, %2, %c8), %3 \n" \
104 "movaps (%3, %0), %%xmm0 \n" \
105 stereo("movaps %%xmm0, %%xmm1 \n") \
106 "mulps %%xmm4, %%xmm0 \n" \
107 stereo("mulps %%xmm5, %%xmm1 \n") \
109 "mov (%6, %2, %c8), %1 \n" \
110 "movaps (%1, %0), %%xmm2 \n" \
111 stereo("movaps %%xmm2, %%xmm3 \n") \
112 "mulps (%4, %2, 8), %%xmm2 \n" \
113 stereo("mulps 16(%4, %2, 8), %%xmm3 \n") \
114 "addps %%xmm2, %%xmm0 \n" \
115 stereo("addps %%xmm3, %%xmm1 \n") \
119 stereo("mov (%6, %2, %c8), %1 \n") \
120 "movaps %%xmm0, (%3, %0) \n" \
121 stereo("movaps %%xmm1, (%1, %0) \n") \
124 : "+&r"(i), "=&r"(j), "=&r"(k), "=&r"(m) \
125 : "r"(matrix_simd + in_ch), \
126 "g"((intptr_t) - 4 * (in_ch - 1)), \
128 "i"(sizeof(float *)), "i"(sizeof(float *)/4) \
132 static void ac3_downmix_sse(float **samples, float (*matrix)[2],
133 int out_ch, int in_ch, int len)
135 int (*matrix_cmp)[2] = (int(*)[2])matrix;
138 i = -len * sizeof(float);
139 if (in_ch == 5 && out_ch == 2 &&
140 !(matrix_cmp[0][1] | matrix_cmp[2][0] |
141 matrix_cmp[3][1] | matrix_cmp[4][0] |
142 (matrix_cmp[1][0] ^ matrix_cmp[1][1]) |
143 (matrix_cmp[0][0] ^ matrix_cmp[2][1]))) {
145 } else if (in_ch == 5 && out_ch == 1 &&
146 matrix_cmp[0][0] == matrix_cmp[2][0] &&
147 matrix_cmp[3][0] == matrix_cmp[4][0]) {
150 DECLARE_ALIGNED(16, float, matrix_simd)[AC3_MAX_CHANNELS][2][4];
151 float *samp[AC3_MAX_CHANNELS];
153 for (j = 0; j < in_ch; j++)
154 samp[j] = samples[j] + len;
156 j = 2 * in_ch * sizeof(float);
160 "movss (%2, %0), %%xmm4 \n"
161 "movss 4(%2, %0), %%xmm5 \n"
162 "shufps $0, %%xmm4, %%xmm4 \n"
163 "shufps $0, %%xmm5, %%xmm5 \n"
164 "movaps %%xmm4, (%1, %0, 4) \n"
165 "movaps %%xmm5, 16(%1, %0, 4) \n"
168 : "r"(matrix_simd), "r"(matrix)
179 #endif /* HAVE_SSE_INLINE && HAVE_7REGS */
181 av_cold void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact)
183 int mm_flags = av_get_cpu_flags();
185 if (EXTERNAL_MMX(mm_flags)) {
186 c->ac3_exponent_min = ff_ac3_exponent_min_mmx;
187 c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_mmx;
188 c->ac3_lshift_int16 = ff_ac3_lshift_int16_mmx;
189 c->ac3_rshift_int32 = ff_ac3_rshift_int32_mmx;
191 if (EXTERNAL_AMD3DNOW(mm_flags)) {
192 c->extract_exponents = ff_ac3_extract_exponents_3dnow;
194 c->float_to_fixed24 = ff_float_to_fixed24_3dnow;
197 if (EXTERNAL_MMXEXT(mm_flags)) {
198 c->ac3_exponent_min = ff_ac3_exponent_min_mmxext;
199 c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_mmxext;
201 if (EXTERNAL_SSE(mm_flags)) {
202 c->float_to_fixed24 = ff_float_to_fixed24_sse;
204 if (EXTERNAL_SSE2(mm_flags)) {
205 c->ac3_exponent_min = ff_ac3_exponent_min_sse2;
206 c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_sse2;
207 c->float_to_fixed24 = ff_float_to_fixed24_sse2;
208 c->compute_mantissa_size = ff_ac3_compute_mantissa_size_sse2;
209 c->extract_exponents = ff_ac3_extract_exponents_sse2;
210 if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) {
211 c->ac3_lshift_int16 = ff_ac3_lshift_int16_sse2;
212 c->ac3_rshift_int32 = ff_ac3_rshift_int32_sse2;
215 if (EXTERNAL_SSSE3(mm_flags)) {
216 c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_ssse3;
217 if (!(mm_flags & AV_CPU_FLAG_ATOM)) {
218 c->extract_exponents = ff_ac3_extract_exponents_ssse3;
222 #if HAVE_SSE_INLINE && HAVE_7REGS
223 if (INLINE_SSE(mm_flags)) {
224 c->downmix = ac3_downmix_sse;