ac3dsp_init.c 9.59 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
/*
 * x86-optimized AC-3 DSP utils
 * Copyright (c) 2011 Justin Ruggles
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * 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 28 29 30 31 32
#include "libavcodec/ac3dsp.h"

extern void ff_ac3_exponent_min_mmx   (uint8_t *exp, int num_reuse_blocks, int nb_coefs);
extern void ff_ac3_exponent_min_mmxext(uint8_t *exp, int num_reuse_blocks, int nb_coefs);
extern void ff_ac3_exponent_min_sse2  (uint8_t *exp, int num_reuse_blocks, int nb_coefs);

33 34 35 36
extern int ff_ac3_max_msb_abs_int16_mmx  (const int16_t *src, int len);
extern int ff_ac3_max_msb_abs_int16_mmx2 (const int16_t *src, int len);
extern int ff_ac3_max_msb_abs_int16_sse2 (const int16_t *src, int len);
extern int ff_ac3_max_msb_abs_int16_ssse3(const int16_t *src, int len);
37

38 39 40 41 42 43
extern void ff_ac3_lshift_int16_mmx (int16_t *src, unsigned int len, unsigned int shift);
extern void ff_ac3_lshift_int16_sse2(int16_t *src, unsigned int len, unsigned int shift);

extern void ff_ac3_rshift_int32_mmx (int32_t *src, unsigned int len, unsigned int shift);
extern void ff_ac3_rshift_int32_sse2(int32_t *src, unsigned int len, unsigned int shift);

44 45 46 47
extern void ff_float_to_fixed24_3dnow(int32_t *dst, const float *src, unsigned int len);
extern void ff_float_to_fixed24_sse  (int32_t *dst, const float *src, unsigned int len);
extern void ff_float_to_fixed24_sse2 (int32_t *dst, const float *src, unsigned int len);

48 49
extern int ff_ac3_compute_mantissa_size_sse2(uint16_t mant_cnt[6][16]);

50 51 52 53
extern void ff_ac3_extract_exponents_3dnow(uint8_t *exp, int32_t *coef, int nb_coefs);
extern void ff_ac3_extract_exponents_sse2 (uint8_t *exp, int32_t *coef, int nb_coefs);
extern void ff_ac3_extract_exponents_ssse3(uint8_t *exp, int32_t *coef, int nb_coefs);

54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166
#if HAVE_SSE_INLINE

#define IF1(x) x
#define IF0(x)

#define MIX5(mono, stereo)                                      \
    __asm__ volatile (                                          \
        "movss           0(%2), %%xmm5          \n"             \
        "movss           8(%2), %%xmm6          \n"             \
        "movss          24(%2), %%xmm7          \n"             \
        "shufps     $0, %%xmm5, %%xmm5          \n"             \
        "shufps     $0, %%xmm6, %%xmm6          \n"             \
        "shufps     $0, %%xmm7, %%xmm7          \n"             \
        "1:                                     \n"             \
        "movaps       (%0, %1), %%xmm0          \n"             \
        "movaps  0x400(%0, %1), %%xmm1          \n"             \
        "movaps  0x800(%0, %1), %%xmm2          \n"             \
        "movaps  0xc00(%0, %1), %%xmm3          \n"             \
        "movaps 0x1000(%0, %1), %%xmm4          \n"             \
        "mulps          %%xmm5, %%xmm0          \n"             \
        "mulps          %%xmm6, %%xmm1          \n"             \
        "mulps          %%xmm5, %%xmm2          \n"             \
        "mulps          %%xmm7, %%xmm3          \n"             \
        "mulps          %%xmm7, %%xmm4          \n"             \
 stereo("addps          %%xmm1, %%xmm0          \n")            \
        "addps          %%xmm1, %%xmm2          \n"             \
        "addps          %%xmm3, %%xmm0          \n"             \
        "addps          %%xmm4, %%xmm2          \n"             \
   mono("addps          %%xmm2, %%xmm0          \n")            \
        "movaps         %%xmm0, (%0, %1)        \n"             \
 stereo("movaps         %%xmm2, 0x400(%0, %1)   \n")            \
        "add               $16, %0              \n"             \
        "jl                 1b                  \n"             \
        : "+&r"(i)                                              \
        : "r"(samples[0] + len), "r"(matrix)                    \
        : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3",      \
                      "%xmm4", "%xmm5", "%xmm6", "%xmm7",)      \
         "memory"                                               \
    );

#define MIX_MISC(stereo)                                        \
    __asm__ volatile (                                          \
        "1:                                 \n"                 \
        "movaps     (%3, %0), %%xmm0        \n"                 \
 stereo("movaps       %%xmm0, %%xmm1        \n")                \
        "mulps        %%xmm4, %%xmm0        \n"                 \
 stereo("mulps        %%xmm5, %%xmm1        \n")                \
        "lea    1024(%3, %0), %1            \n"                 \
        "mov              %5, %2            \n"                 \
        "2:                                 \n"                 \
        "movaps         (%1), %%xmm2        \n"                 \
 stereo("movaps       %%xmm2, %%xmm3        \n")                \
        "mulps      (%4, %2), %%xmm2        \n"                 \
 stereo("mulps    16(%4, %2), %%xmm3        \n")                \
        "addps        %%xmm2, %%xmm0        \n"                 \
 stereo("addps        %%xmm3, %%xmm1        \n")                \
        "add           $1024, %1            \n"                 \
        "add             $32, %2            \n"                 \
        "jl               2b                \n"                 \
        "movaps       %%xmm0,     (%3, %0)  \n"                 \
 stereo("movaps       %%xmm1, 1024(%3, %0)  \n")                \
        "add             $16, %0            \n"                 \
        "jl               1b                \n"                 \
        : "+&r"(i), "=&r"(j), "=&r"(k)                          \
        : "r"(samples[0] + len), "r"(matrix_simd + in_ch),      \
          "g"((intptr_t) - 32 * (in_ch - 1))                    \
        : "memory"                                              \
    );

static void ac3_downmix_sse(float (*samples)[256], float (*matrix)[2],
                            int out_ch, int in_ch, int len)
{
    int (*matrix_cmp)[2] = (int(*)[2])matrix;
    intptr_t i, j, k;

    i = -len * sizeof(float);
    if (in_ch == 5 && out_ch == 2 &&
        !(matrix_cmp[0][1] | matrix_cmp[2][0]   |
          matrix_cmp[3][1] | matrix_cmp[4][0]   |
          (matrix_cmp[1][0] ^ matrix_cmp[1][1]) |
          (matrix_cmp[0][0] ^ matrix_cmp[2][1]))) {
        MIX5(IF0, IF1);
    } else if (in_ch == 5 && out_ch == 1 &&
               matrix_cmp[0][0] == matrix_cmp[2][0] &&
               matrix_cmp[3][0] == matrix_cmp[4][0]) {
        MIX5(IF1, IF0);
    } else {
        DECLARE_ALIGNED(16, float, matrix_simd)[AC3_MAX_CHANNELS][2][4];
        j = 2 * in_ch * sizeof(float);
        __asm__ volatile (
            "1:                                 \n"
            "sub             $8, %0             \n"
            "movss     (%2, %0), %%xmm4         \n"
            "movss    4(%2, %0), %%xmm5         \n"
            "shufps          $0, %%xmm4, %%xmm4 \n"
            "shufps          $0, %%xmm5, %%xmm5 \n"
            "movaps      %%xmm4,   (%1, %0, 4)  \n"
            "movaps      %%xmm5, 16(%1, %0, 4)  \n"
            "jg              1b                 \n"
            : "+&r"(j)
            : "r"(matrix_simd), "r"(matrix)
            : "memory"
        );
        if (out_ch == 2) {
            MIX_MISC(IF1);
        } else {
            MIX_MISC(IF0);
        }
    }
}

#endif /* HAVE_SSE_INLINE */

167
av_cold void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact)
168 169 170
{
    int mm_flags = av_get_cpu_flags();

171
    if (EXTERNAL_MMX(mm_flags)) {
172
        c->ac3_exponent_min = ff_ac3_exponent_min_mmx;
173
        c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_mmx;
174 175
        c->ac3_lshift_int16 = ff_ac3_lshift_int16_mmx;
        c->ac3_rshift_int32 = ff_ac3_rshift_int32_mmx;
176
    }
177
    if (EXTERNAL_AMD3DNOW(mm_flags)) {
178
        c->extract_exponents = ff_ac3_extract_exponents_3dnow;
179 180 181 182
        if (!bit_exact) {
            c->float_to_fixed24 = ff_float_to_fixed24_3dnow;
        }
    }
183
    if (EXTERNAL_MMXEXT(mm_flags)) {
184
        c->ac3_exponent_min = ff_ac3_exponent_min_mmxext;
185
        c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_mmx2;
186
    }
187
    if (EXTERNAL_SSE(mm_flags)) {
188 189
        c->float_to_fixed24 = ff_float_to_fixed24_sse;
    }
190
    if (EXTERNAL_SSE2(mm_flags)) {
191
        c->ac3_exponent_min = ff_ac3_exponent_min_sse2;
192
        c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_sse2;
193
        c->float_to_fixed24 = ff_float_to_fixed24_sse2;
194
        c->compute_mantissa_size = ff_ac3_compute_mantissa_size_sse2;
195
        c->extract_exponents = ff_ac3_extract_exponents_sse2;
196 197 198 199
        if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) {
            c->ac3_lshift_int16 = ff_ac3_lshift_int16_sse2;
            c->ac3_rshift_int32 = ff_ac3_rshift_int32_sse2;
        }
200
    }
201
    if (EXTERNAL_SSSE3(mm_flags)) {
202
        c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_ssse3;
203 204 205
        if (!(mm_flags & AV_CPU_FLAG_ATOM)) {
            c->extract_exponents = ff_ac3_extract_exponents_ssse3;
        }
206
    }
207

208
#if HAVE_SSE_INLINE
209 210 211
    if (INLINE_SSE(mm_flags)) {
        c->downmix = ac3_downmix_sse;
    }
212
#endif
213
}