X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fmpegaudiodec.c;h=ff1f1113e77c3c288ad83473800adb49c29cece8;hb=79b0b66ba07890c4f71b42d2554d73db537bec97;hp=97b41ac658ed7b211f112ba18d4ecf5525a3eb9f;hpb=59d3e367a85f5faa1f8890bc69799e4d3bf2811e;p=ffmpeg diff --git a/libavcodec/mpegaudiodec.c b/libavcodec/mpegaudiodec.c index 97b41ac658e..ff1f1113e77 100644 --- a/libavcodec/mpegaudiodec.c +++ b/libavcodec/mpegaudiodec.c @@ -1,24 +1,31 @@ /* * MPEG Audio decoder - * Copyright (c) 2001, 2002 Gerard Lantau. + * Copyright (c) 2001, 2002 Fabrice Bellard. * - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or - * (at your option) any later version. + * This library 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 of the License, or (at your option) any later version. * - * This program is distributed in the hope that it will be useful, + * This library 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 General Public License for more details. + * 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 General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ + +/** + * @file mpegaudiodec.c + * MPEG Audio decoder. + */ + //#define DEBUG #include "avcodec.h" -#include "mpegaudio.h" +#include "bitstream.h" +#include "dsputil.h" /* * TODO: @@ -28,45 +35,43 @@ /* define USE_HIGHPRECISION to have a bit exact (but slower) mpeg audio decoder */ -//#define USE_HIGHPRECISION - -#ifdef USE_HIGHPRECISION -#define FRAC_BITS 23 /* fractional bits for sb_samples and dct */ -#define WFRAC_BITS 16 /* fractional bits for window */ -#else -#define FRAC_BITS 15 /* fractional bits for sb_samples and dct */ -#define WFRAC_BITS 14 /* fractional bits for window */ +#ifdef CONFIG_MPEGAUDIO_HP +#define USE_HIGHPRECISION #endif +#include "mpegaudio.h" + #define FRAC_ONE (1 << FRAC_BITS) -#define MULL(a,b) (((INT64)(a) * (INT64)(b)) >> FRAC_BITS) -#define MUL64(a,b) ((INT64)(a) * (INT64)(b)) +#define MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS) +#define MUL64(a,b) ((int64_t)(a) * (int64_t)(b)) #define FIX(a) ((int)((a) * FRAC_ONE)) /* WARNING: only correct for posititive numbers */ #define FIXR(a) ((int)((a) * FRAC_ONE + 0.5)) #define FRAC_RND(a) (((a) + (FRAC_ONE/2)) >> FRAC_BITS) -#if FRAC_BITS <= 15 -typedef INT16 MPA_INT; -#else -typedef INT32 MPA_INT; -#endif +#define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5)) +//#define MULH(a,b) (((int64_t)(a) * (int64_t)(b))>>32) //gcc 3.4 creates an incredibly bloated mess out of this +static always_inline int MULH(int a, int b){ + return ((int64_t)(a) * (int64_t)(b))>>32; +} /****************/ #define HEADER_SIZE 4 #define BACKSTEP_SIZE 512 +struct GranuleDef; + typedef struct MPADecodeContext { - UINT8 inbuf1[2][MPA_MAX_CODED_FRAME_SIZE + BACKSTEP_SIZE]; /* input buffer */ + uint8_t inbuf1[2][MPA_MAX_CODED_FRAME_SIZE + BACKSTEP_SIZE]; /* input buffer */ int inbuf_index; - UINT8 *inbuf_ptr, *inbuf; + uint8_t *inbuf_ptr, *inbuf; int frame_size; int free_format_frame_size; /* frame size in case of free format (zero if currently unknown) */ /* next header (used in free format parsing) */ - UINT32 free_format_next_header; + uint32_t free_format_next_header; int error_protection; int layer; int sample_rate; @@ -78,33 +83,45 @@ typedef struct MPADecodeContext { int mode; int mode_ext; int lsf; - MPA_INT synth_buf[MPA_MAX_CHANNELS][512 * 2]; + MPA_INT synth_buf[MPA_MAX_CHANNELS][512 * 2] __attribute__((aligned(16))); int synth_buf_offset[MPA_MAX_CHANNELS]; - INT32 sb_samples[MPA_MAX_CHANNELS][36][SBLIMIT]; - INT32 mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */ + int32_t sb_samples[MPA_MAX_CHANNELS][36][SBLIMIT] __attribute__((aligned(16))); + int32_t mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */ #ifdef DEBUG int frame_count; #endif + void (*compute_antialias)(struct MPADecodeContext *s, struct GranuleDef *g); + int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3 + unsigned int dither_state; } MPADecodeContext; +/** + * Context for MP3On4 decoder + */ +typedef struct MP3On4DecodeContext { + int frames; ///< number of mp3 frames per block (number of mp3 decoder instances) + int chan_cfg; ///< channel config number + MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance +} MP3On4DecodeContext; + /* layer 3 "granule" */ typedef struct GranuleDef { - UINT8 scfsi; + uint8_t scfsi; int part2_3_length; int big_values; int global_gain; int scalefac_compress; - UINT8 block_type; - UINT8 switch_point; + uint8_t block_type; + uint8_t switch_point; int table_select[3]; int subblock_gain[3]; - UINT8 scalefac_scale; - UINT8 count1table_select; + uint8_t scalefac_scale; + uint8_t count1table_select; int region_size[3]; /* number of huffman codes in each region */ int preflag; int short_start, long_end; /* long/short band indexes */ - UINT8 scale_factors[40]; - INT32 sb_hybrid[SBLIMIT * 18]; /* 576 samples */ + uint8_t scale_factors[40]; + int32_t sb_hybrid[SBLIMIT * 18]; /* 576 samples */ } GranuleDef; #define MODE_EXT_MS_STEREO 2 @@ -113,70 +130,64 @@ typedef struct GranuleDef { /* layer 3 huffman tables */ typedef struct HuffTable { int xsize; - const UINT8 *bits; - const UINT16 *codes; + const uint8_t *bits; + const uint16_t *codes; } HuffTable; #include "mpegaudiodectab.h" +static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g); +static void compute_antialias_float(MPADecodeContext *s, GranuleDef *g); + /* vlc structure for decoding layer 3 huffman tables */ -static VLC huff_vlc[16]; -static UINT8 *huff_code_table[16]; +static VLC huff_vlc[16]; +static uint8_t *huff_code_table[16]; static VLC huff_quad_vlc[2]; /* computed from band_size_long */ -static UINT16 band_index_long[9][23]; +static uint16_t band_index_long[9][23]; /* XXX: free when all decoders are closed */ -#define TABLE_4_3_SIZE (8191 + 16) -static INT8 *table_4_3_exp; -#if FRAC_BITS <= 15 -static UINT16 *table_4_3_value; -#else -static UINT32 *table_4_3_value; -#endif +#define TABLE_4_3_SIZE (8191 + 16)*4 +static int8_t *table_4_3_exp; +static uint32_t *table_4_3_value; /* intensity stereo coef table */ -static INT32 is_table[2][16]; -static INT32 is_table_lsf[2][2][16]; -static INT32 csa_table[8][2]; -static INT32 mdct_win[8][36]; +static int32_t is_table[2][16]; +static int32_t is_table_lsf[2][2][16]; +static int32_t csa_table[8][4]; +static float csa_table_float[8][4]; +static int32_t mdct_win[8][36]; /* lower 2 bits: modulo 3, higher bits: shift */ -static UINT16 scale_factor_modshift[64]; +static uint16_t scale_factor_modshift[64]; /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */ -static INT32 scale_factor_mult[15][3]; +static int32_t scale_factor_mult[15][3]; /* mult table for layer 2 group quantization */ #define SCALE_GEN(v) \ { FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) } -static INT32 scale_factor_mult2[3][3] = { - SCALE_GEN(1.0 / 3.0), /* 3 steps */ - SCALE_GEN(1.0 / 5.0), /* 5 steps */ - SCALE_GEN(1.0 / 9.0), /* 9 steps */ +static const int32_t scale_factor_mult2[3][3] = { + SCALE_GEN(4.0 / 3.0), /* 3 steps */ + SCALE_GEN(4.0 / 5.0), /* 5 steps */ + SCALE_GEN(4.0 / 9.0), /* 9 steps */ }; -/* 2^(n/4) */ -static UINT32 scale_factor_mult3[4] = { - FIXR(1.0), - FIXR(1.18920711500272106671), - FIXR(1.41421356237309504880), - FIXR(1.68179283050742908605), -}; +void ff_mpa_synth_init(MPA_INT *window); +static MPA_INT window[512] __attribute__((aligned(16))); -static MPA_INT window[512]; - /* layer 1 unscaling */ /* n = number of bits of the mantissa minus 1 */ static inline int l1_unscale(int n, int mant, int scale_factor) { int shift, mod; - INT64 val; + int64_t val; shift = scale_factor_modshift[scale_factor]; mod = shift & 3; shift >>= 2; val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]); shift += n; - return (int)((val + (1 << (shift - 1))) >> shift); + /* NOTE: at this point, 1 <= shift >= 21 + 15 */ + return (int)((val + (1LL << (shift - 1))) >> shift); } static inline int l2_unscale_group(int steps, int mant, int scale_factor) @@ -186,38 +197,29 @@ static inline int l2_unscale_group(int steps, int mant, int scale_factor) shift = scale_factor_modshift[scale_factor]; mod = shift & 3; shift >>= 2; - /* XXX: store the result directly */ - val = (2 * (mant - (steps >> 1))) * scale_factor_mult2[steps >> 2][mod]; - return (val + (1 << (shift - 1))) >> shift; + + val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod]; + /* NOTE: at this point, 0 <= shift <= 21 */ + if (shift > 0) + val = (val + (1 << (shift - 1))) >> shift; + return val; } /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */ static inline int l3_unscale(int value, int exponent) { -#if FRAC_BITS <= 15 unsigned int m; -#else - UINT64 m; -#endif int e; - e = table_4_3_exp[value]; - e += (exponent >> 2); - e = FRAC_BITS - e; -#if FRAC_BITS <= 15 + e = table_4_3_exp [4*value + (exponent&3)]; + m = table_4_3_value[4*value + (exponent&3)]; + e -= (exponent >> 2); + assert(e>=1); if (e > 31) - e = 31; -#endif - m = table_4_3_value[value]; -#if FRAC_BITS <= 15 - m = (m * scale_factor_mult3[exponent & 3]); + return 0; m = (m + (1 << (e-1))) >> e; + return m; -#else - m = MUL64(m, scale_factor_mult3[exponent & 3]); - m = (m + (UINT64_C(1) << (e-1))) >> e; - return m; -#endif } /* all integer n^(4/3) computation code */ @@ -226,15 +228,17 @@ static inline int l3_unscale(int value, int exponent) #define POW_FRAC_BITS 24 #define POW_FRAC_ONE (1 << POW_FRAC_BITS) #define POW_FIX(a) ((int)((a) * POW_FRAC_ONE)) -#define POW_MULL(a,b) (((INT64)(a) * (INT64)(b)) >> POW_FRAC_BITS) +#define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS) static int dev_4_3_coefs[DEV_ORDER]; +#if 0 /* unused */ static int pow_mult3[3] = { POW_FIX(1.0), POW_FIX(1.25992104989487316476), POW_FIX(1.58740105196819947474), }; +#endif static void int_pow_init(void) { @@ -247,12 +251,13 @@ static void int_pow_init(void) } } +#if 0 /* unused, remove? */ /* return the mantissa and the binary exponent */ static int int_pow(int i, int *exp_ptr) { int e, er, eq, j; int a, a1; - + /* renormalize */ a = i; e = POW_FRAC_BITS; @@ -280,7 +285,7 @@ static int int_pow(int i, int *exp_ptr) eq--; } /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */ -#if (POW_FRAC_BITS - 1) > FRAC_BITS +#if POW_FRAC_BITS > FRAC_BITS a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS); /* correct overflow */ if (a >= 2 * (1 << FRAC_BITS)) { @@ -291,24 +296,32 @@ static int int_pow(int i, int *exp_ptr) *exp_ptr = eq; return a; } +#endif static int decode_init(AVCodecContext * avctx) { MPADecodeContext *s = avctx->priv_data; - static int init; + static int init=0; int i, j, k; - if(!init) { +#if defined(USE_HIGHPRECISION) && defined(CONFIG_AUDIO_NONSHORT) + avctx->sample_fmt= SAMPLE_FMT_S32; +#else + avctx->sample_fmt= SAMPLE_FMT_S16; +#endif + + if(avctx->antialias_algo != FF_AA_FLOAT) + s->compute_antialias= compute_antialias_integer; + else + s->compute_antialias= compute_antialias_float; + + if (!init && !avctx->parse_only) { /* scale factors table for layer 1/2 */ for(i=0;i<64;i++) { int shift, mod; /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */ - shift = (i / 3) - 1; + shift = (i / 3); mod = i % 3; -#if FRAC_BITS <= 15 - if (shift > 31) - shift = 31; -#endif scale_factor_modshift[i] = mod | (shift << 2); } @@ -316,45 +329,33 @@ static int decode_init(AVCodecContext * avctx) for(i=0;i<15;i++) { int n, norm; n = i + 2; - norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1); - scale_factor_mult[i][0] = MULL(FIXR(1.0), norm); - scale_factor_mult[i][1] = MULL(FIXR(0.7937005259), norm); - scale_factor_mult[i][2] = MULL(FIXR(0.6299605249), norm); + norm = ((int64_t_C(1) << n) * FRAC_ONE) / ((1 << n) - 1); + scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm); + scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm); + scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm); dprintf("%d: norm=%x s=%x %x %x\n", - i, norm, + i, norm, scale_factor_mult[i][0], scale_factor_mult[i][1], scale_factor_mult[i][2]); } - - /* window */ - /* max = 18760, max sum over all 16 coefs : 44736 */ - for(i=0;i<257;i++) { - int v; - v = mpa_enwindow[i]; -#if WFRAC_BITS < 16 - v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS); -#endif - window[i] = v; - if ((i & 63) != 0) - v = -v; - if (i != 0) - window[512 - i] = v; - } - + + ff_mpa_synth_init(window); + /* huffman decode tables */ huff_code_table[0] = NULL; for(i=1;i<16;i++) { const HuffTable *h = &mpa_huff_tables[i]; - int xsize, n, x, y; - UINT8 *code_table; + int xsize, x, y; + unsigned int n; + uint8_t *code_table; xsize = h->xsize; n = xsize * xsize; /* XXX: fail test */ - init_vlc(&huff_vlc[i], 8, n, - h->bits, 1, 1, h->codes, 2, 2); - + init_vlc(&huff_vlc[i], 8, n, + h->bits, 1, 1, h->codes, 2, 2, 1); + code_table = av_mallocz(n); j = 0; for(x=0;x> 1; - e1++; - } -#endif - e1--; - if (m != m1 || e != e1) { - printf("%4d: m=%x m1=%x e=%d e1=%d\n", - i, m, m1, e, e1); - } - } -#endif + f = pow((double)(i/4), 4.0 / 3.0) * pow(2, (i&3)*0.25); + fm = frexp(f, &e); + m = (uint32_t)(fm*(1LL<<31) + 0.5); + e+= FRAC_BITS - 31 + 5; + /* normalized to FRAC_BITS */ table_4_3_value[i] = m; - table_4_3_exp[i] = e; +// av_log(NULL, AV_LOG_DEBUG, "%d %d %f\n", i, m, pow((double)i, 4.0 / 3.0)); + table_4_3_exp[i] = -e; } - + for(i=0;i<7;i++) { float f; int v; @@ -445,7 +427,7 @@ static int decode_init(AVCodecContext * avctx) k = i & 1; is_table_lsf[j][k ^ 1][i] = FIXR(f); is_table_lsf[j][k][i] = FIXR(1.0); - dprintf("is_table_lsf %d %d: %x %x\n", + dprintf("is_table_lsf %d %d: %x %x\n", i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]); } } @@ -455,31 +437,47 @@ static int decode_init(AVCodecContext * avctx) ci = ci_table[i]; cs = 1.0 / sqrt(1.0 + ci * ci); ca = cs * ci; - csa_table[i][0] = FIX(cs); - csa_table[i][1] = FIX(ca); + csa_table[i][0] = FIXHR(cs/4); + csa_table[i][1] = FIXHR(ca/4); + csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4); + csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4); + csa_table_float[i][0] = cs; + csa_table_float[i][1] = ca; + csa_table_float[i][2] = ca + cs; + csa_table_float[i][3] = ca - cs; +// printf("%d %d %d %d\n", FIX(cs), FIX(cs-1), FIX(ca), FIX(cs)-FIX(ca)); +// av_log(NULL, AV_LOG_DEBUG,"%f %f %f %f\n", cs, ca, ca+cs, ca-cs); } /* compute mdct windows */ for(i=0;i<36;i++) { - int v; - v = FIXR(sin(M_PI * (i + 0.5) / 36.0)); - mdct_win[0][i] = v; - mdct_win[1][i] = v; - mdct_win[3][i] = v; - } - for(i=0;i<6;i++) { - mdct_win[1][18 + i] = FIXR(1.0); - mdct_win[1][24 + i] = FIXR(sin(M_PI * ((i + 6) + 0.5) / 12.0)); - mdct_win[1][30 + i] = FIXR(0.0); - - mdct_win[3][i] = FIXR(0.0); - mdct_win[3][6 + i] = FIXR(sin(M_PI * (i + 0.5) / 12.0)); - mdct_win[3][12 + i] = FIXR(1.0); + for(j=0; j<4; j++){ + double d; + + if(j==2 && i%3 != 1) + continue; + + d= sin(M_PI * (i + 0.5) / 36.0); + if(j==1){ + if (i>=30) d= 0; + else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0); + else if(i>=18) d= 1; + }else if(j==3){ + if (i< 6) d= 0; + else if(i< 12) d= sin(M_PI * (i - 6 + 0.5) / 12.0); + else if(i< 18) d= 1; + } + //merge last stage of imdct into the window coefficients + d*= 0.5 / cos(M_PI*(2*i + 19)/72); + + if(j==2) + mdct_win[j][i/3] = FIXHR((d / (1<<5))); + else + mdct_win[j][i ] = FIXHR((d / (1<<5))); +// av_log(NULL, AV_LOG_DEBUG, "%2d %d %f\n", i,j,d / (1<<5)); + } } - for(i=0;i<12;i++) - mdct_win[2][i] = FIXR(sin(M_PI * (i + 0.5) / 12.0)); - /* NOTE: we do frequency inversion adter the MDCT by changing the sign of the right window coefs */ for(j=0;j<4;j++) { @@ -506,10 +504,12 @@ static int decode_init(AVCodecContext * avctx) #ifdef DEBUG s->frame_count = 0; #endif + if (avctx->codec_id == CODEC_ID_MP3ADU) + s->adu_mode = 1; return 0; } -/* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */; +/* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */ /* cos(i*pi/64) */ @@ -578,7 +578,7 @@ static int decode_init(AVCodecContext * avctx) #define ADD(a, b) tab[a] += tab[b] /* DCT32 without 1/sqrt(2) coef zero scaling. */ -static void dct32(INT32 *out, INT32 *tab) +static void dct32(int32_t *out, int32_t *tab) { int tmp0, tmp1; @@ -609,7 +609,7 @@ static void dct32(INT32 *out, INT32 *tab) BF(5, 10, COS1_5); BF(6, 9, COS1_6); BF(7, 8, COS1_7); - + BF(16, 31, -COS1_0); BF(17, 30, -COS1_1); BF(18, 29, -COS1_2); @@ -618,23 +618,23 @@ static void dct32(INT32 *out, INT32 *tab) BF(21, 26, -COS1_5); BF(22, 25, -COS1_6); BF(23, 24, -COS1_7); - + /* pass 3 */ BF(0, 7, COS2_0); BF(1, 6, COS2_1); BF(2, 5, COS2_2); BF(3, 4, COS2_3); - + BF(8, 15, -COS2_0); BF(9, 14, -COS2_1); BF(10, 13, -COS2_2); BF(11, 12, -COS2_3); - + BF(16, 23, COS2_0); BF(17, 22, COS2_1); BF(18, 21, COS2_2); BF(19, 20, COS2_3); - + BF(24, 31, -COS2_0); BF(25, 30, -COS2_1); BF(26, 29, -COS2_2); @@ -643,28 +643,28 @@ static void dct32(INT32 *out, INT32 *tab) /* pass 4 */ BF(0, 3, COS3_0); BF(1, 2, COS3_1); - + BF(4, 7, -COS3_0); BF(5, 6, -COS3_1); - + BF(8, 11, COS3_0); BF(9, 10, COS3_1); - + BF(12, 15, -COS3_0); BF(13, 14, -COS3_1); - + BF(16, 19, COS3_0); BF(17, 18, COS3_1); - + BF(20, 23, -COS3_0); BF(21, 22, -COS3_1); - + BF(24, 27, COS3_0); BF(25, 26, COS3_1); - + BF(28, 31, -COS3_0); BF(29, 30, -COS3_1); - + /* pass 5 */ BF1(0, 1, 2, 3); BF2(4, 5, 6, 7); @@ -674,9 +674,9 @@ static void dct32(INT32 *out, INT32 *tab) BF2(20, 21, 22, 23); BF1(24, 25, 26, 27); BF2(28, 29, 30, 31); - + /* pass 6 */ - + ADD( 8, 12); ADD(12, 10); ADD(10, 14); @@ -701,7 +701,7 @@ static void dct32(INT32 *out, INT32 *tab) out[22] = tab[13]; out[14] = tab[14]; out[30] = tab[15]; - + ADD(24, 28); ADD(28, 26); ADD(26, 30); @@ -728,87 +728,147 @@ static void dct32(INT32 *out, INT32 *tab) out[31] = tab[31]; } -#define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 15) - #if FRAC_BITS <= 15 -#define OUT_SAMPLE(sum)\ -{\ - int sum1;\ - sum1 = (sum + (1 << (OUT_SHIFT - 1))) >> OUT_SHIFT;\ - if (sum1 < -32768)\ - sum1 = -32768;\ - else if (sum1 > 32767)\ - sum1 = 32767;\ - *samples = sum1;\ - samples += incr;\ +static inline int round_sample(int *sum) +{ + int sum1; + sum1 = (*sum) >> OUT_SHIFT; + *sum &= (1< OUT_MAX) + sum1 = OUT_MAX; + return sum1; } -#define SUM8(off, op) \ -{ \ - sum op w[0 * 64 + off] * p[0 * 64];\ - sum op w[1 * 64 + off] * p[1 * 64];\ - sum op w[2 * 64 + off] * p[2 * 64];\ - sum op w[3 * 64 + off] * p[3 * 64];\ - sum op w[4 * 64 + off] * p[4 * 64];\ - sum op w[5 * 64 + off] * p[5 * 64];\ - sum op w[6 * 64 + off] * p[6 * 64];\ - sum op w[7 * 64 + off] * p[7 * 64];\ -} +#if defined(ARCH_POWERPC_405) + +/* signed 16x16 -> 32 multiply add accumulate */ +#define MACS(rt, ra, rb) \ + asm ("maclhw %0, %2, %3" : "=r" (rt) : "0" (rt), "r" (ra), "r" (rb)); + +/* signed 16x16 -> 32 multiply */ +#define MULS(ra, rb) \ + ({ int __rt; asm ("mullhw %0, %1, %2" : "=r" (__rt) : "r" (ra), "r" (rb)); __rt; }) #else -#define OUT_SAMPLE(sum)\ -{\ - int sum1;\ - sum1 = (int)((sum + (INT64_C(1) << (OUT_SHIFT - 1))) >> OUT_SHIFT);\ - if (sum1 < -32768)\ - sum1 = -32768;\ - else if (sum1 > 32767)\ - sum1 = 32767;\ - *samples = sum1;\ - samples += incr;\ +/* signed 16x16 -> 32 multiply add accumulate */ +#define MACS(rt, ra, rb) rt += (ra) * (rb) + +/* signed 16x16 -> 32 multiply */ +#define MULS(ra, rb) ((ra) * (rb)) + +#endif + +#else + +static inline int round_sample(int64_t *sum) +{ + int sum1; + sum1 = (int)((*sum) >> OUT_SHIFT); + *sum &= (1< OUT_MAX) + sum1 = OUT_MAX; + return sum1; +} + +#define MULS(ra, rb) MUL64(ra, rb) + +#endif + +#define SUM8(sum, op, w, p) \ +{ \ + sum op MULS((w)[0 * 64], p[0 * 64]);\ + sum op MULS((w)[1 * 64], p[1 * 64]);\ + sum op MULS((w)[2 * 64], p[2 * 64]);\ + sum op MULS((w)[3 * 64], p[3 * 64]);\ + sum op MULS((w)[4 * 64], p[4 * 64]);\ + sum op MULS((w)[5 * 64], p[5 * 64]);\ + sum op MULS((w)[6 * 64], p[6 * 64]);\ + sum op MULS((w)[7 * 64], p[7 * 64]);\ } -#define SUM8(off, op) \ +#define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \ { \ - sum op MUL64(w[0 * 64 + off], p[0 * 64]);\ - sum op MUL64(w[1 * 64 + off], p[1 * 64]);\ - sum op MUL64(w[2 * 64 + off], p[2 * 64]);\ - sum op MUL64(w[3 * 64 + off], p[3 * 64]);\ - sum op MUL64(w[4 * 64 + off], p[4 * 64]);\ - sum op MUL64(w[5 * 64 + off], p[5 * 64]);\ - sum op MUL64(w[6 * 64 + off], p[6 * 64]);\ - sum op MUL64(w[7 * 64 + off], p[7 * 64]);\ + int tmp;\ + tmp = p[0 * 64];\ + sum1 op1 MULS((w1)[0 * 64], tmp);\ + sum2 op2 MULS((w2)[0 * 64], tmp);\ + tmp = p[1 * 64];\ + sum1 op1 MULS((w1)[1 * 64], tmp);\ + sum2 op2 MULS((w2)[1 * 64], tmp);\ + tmp = p[2 * 64];\ + sum1 op1 MULS((w1)[2 * 64], tmp);\ + sum2 op2 MULS((w2)[2 * 64], tmp);\ + tmp = p[3 * 64];\ + sum1 op1 MULS((w1)[3 * 64], tmp);\ + sum2 op2 MULS((w2)[3 * 64], tmp);\ + tmp = p[4 * 64];\ + sum1 op1 MULS((w1)[4 * 64], tmp);\ + sum2 op2 MULS((w2)[4 * 64], tmp);\ + tmp = p[5 * 64];\ + sum1 op1 MULS((w1)[5 * 64], tmp);\ + sum2 op2 MULS((w2)[5 * 64], tmp);\ + tmp = p[6 * 64];\ + sum1 op1 MULS((w1)[6 * 64], tmp);\ + sum2 op2 MULS((w2)[6 * 64], tmp);\ + tmp = p[7 * 64];\ + sum1 op1 MULS((w1)[7 * 64], tmp);\ + sum2 op2 MULS((w2)[7 * 64], tmp);\ } +void ff_mpa_synth_init(MPA_INT *window) +{ + int i; + + /* max = 18760, max sum over all 16 coefs : 44736 */ + for(i=0;i<257;i++) { + int v; + v = mpa_enwindow[i]; +#if WFRAC_BITS < 16 + v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS); #endif + window[i] = v; + if ((i & 63) != 0) + v = -v; + if (i != 0) + window[512 - i] = v; + } +} /* 32 sub band synthesis filter. Input: 32 sub band samples, Output: 32 samples. */ /* XXX: optimize by avoiding ring buffer usage */ -static void synth_filter(MPADecodeContext *s1, - int ch, INT16 *samples, int incr, - INT32 sb_samples[SBLIMIT]) +void ff_mpa_synth_filter(MPA_INT *synth_buf_ptr, int *synth_buf_offset, + MPA_INT *window, int *dither_state, + OUT_INT *samples, int incr, + int32_t sb_samples[SBLIMIT]) { - INT32 tmp[32]; - register MPA_INT *synth_buf, *p; - register MPA_INT *w; + int32_t tmp[32]; + register MPA_INT *synth_buf; + register const MPA_INT *w, *w2, *p; int j, offset, v; + OUT_INT *samples2; #if FRAC_BITS <= 15 - int sum; + int sum, sum2; #else - INT64 sum; + int64_t sum, sum2; #endif dct32(tmp, sb_samples); - - offset = s1->synth_buf_offset[ch]; - synth_buf = s1->synth_buf[ch] + offset; + + offset = *synth_buf_offset; + synth_buf = synth_buf_ptr + offset; for(j=0;j<32;j++) { v = tmp[j]; #if FRAC_BITS <= 15 + /* NOTE: can cause a loss in precision if very high amplitude + sound */ if (v > 32767) v = 32767; else if (v < -32768) @@ -819,142 +879,120 @@ static void synth_filter(MPADecodeContext *s1, /* copy to avoid wrap */ memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT)); + samples2 = samples + 31 * incr; w = window; - for(j=0;j<16;j++) { - sum = 0; - p = synth_buf + 16 + j; /* 0-15 */ - SUM8(0, +=); - p = synth_buf + 48 - j; /* 32-47 */ - SUM8(32, -=); - OUT_SAMPLE(sum); - w++; - } - - p = synth_buf + 32; /* 48 */ - sum = 0; - SUM8(32, -=); - OUT_SAMPLE(sum); + w2 = window + 31; + + sum = *dither_state; + p = synth_buf + 16; + SUM8(sum, +=, w, p); + p = synth_buf + 48; + SUM8(sum, -=, w + 32, p); + *samples = round_sample(&sum); + samples += incr; w++; - for(j=17;j<32;j++) { - sum = 0; - p = synth_buf + 48 - j; /* 17-31 */ - SUM8(0, -=); - p = synth_buf + 16 + j; /* 49-63 */ - SUM8(32, -=); - OUT_SAMPLE(sum); + /* we calculate two samples at the same time to avoid one memory + access per two sample */ + for(j=1;j<16;j++) { + sum2 = 0; + p = synth_buf + 16 + j; + SUM8P2(sum, +=, sum2, -=, w, w2, p); + p = synth_buf + 48 - j; + SUM8P2(sum, -=, sum2, -=, w + 32, w2 + 32, p); + + *samples = round_sample(&sum); + samples += incr; + sum += sum2; + *samples2 = round_sample(&sum); + samples2 -= incr; w++; + w2--; } - offset = (offset - 32) & 511; - s1->synth_buf_offset[ch] = offset; -} -/* cos(pi*i/24) */ -#define C1 FIXR(0.99144486137381041114) -#define C3 FIXR(0.92387953251128675612) -#define C5 FIXR(0.79335334029123516458) -#define C7 FIXR(0.60876142900872063941) -#define C9 FIXR(0.38268343236508977173) -#define C11 FIXR(0.13052619222005159154) + p = synth_buf + 32; + SUM8(sum, -=, w + 32, p); + *samples = round_sample(&sum); + *dither_state= sum; -/* 12 points IMDCT. We compute it "by hand" by factorizing obvious - cases. */ -static void imdct12(int *out, int *in) -{ - int tmp; - INT64 in1_3, in1_9, in4_3, in4_9; - - in1_3 = MUL64(in[1], C3); - in1_9 = MUL64(in[1], C9); - in4_3 = MUL64(in[4], C3); - in4_9 = MUL64(in[4], C9); - - tmp = FRAC_RND(MUL64(in[0], C7) - in1_3 - MUL64(in[2], C11) + - MUL64(in[3], C1) - in4_9 - MUL64(in[5], C5)); - out[0] = tmp; - out[5] = -tmp; - tmp = FRAC_RND(MUL64(in[0] - in[3], C9) - in1_3 + - MUL64(in[2] + in[5], C3) - in4_9); - out[1] = tmp; - out[4] = -tmp; - tmp = FRAC_RND(MUL64(in[0], C11) - in1_9 + MUL64(in[2], C7) - - MUL64(in[3], C5) + in4_3 - MUL64(in[5], C1)); - out[2] = tmp; - out[3] = -tmp; - tmp = FRAC_RND(MUL64(-in[0], C5) + in1_9 + MUL64(in[2], C1) + - MUL64(in[3], C11) - in4_3 - MUL64(in[5], C7)); - out[6] = tmp; - out[11] = tmp; - tmp = FRAC_RND(MUL64(-in[0] + in[3], C3) - in1_9 + - MUL64(in[2] + in[5], C9) + in4_3); - out[7] = tmp; - out[10] = tmp; - tmp = FRAC_RND(-MUL64(in[0], C1) - in1_3 - MUL64(in[2], C5) - - MUL64(in[3], C7) - in4_9 - MUL64(in[5], C11)); - out[8] = tmp; - out[9] = tmp; + offset = (offset - 32) & 511; + *synth_buf_offset = offset; } -#undef C1 -#undef C3 -#undef C5 -#undef C7 -#undef C9 -#undef C11 - -/* cos(pi*i/18) */ -#define C1 FIXR(0.98480775301220805936) -#define C2 FIXR(0.93969262078590838405) -#define C3 FIXR(0.86602540378443864676) -#define C4 FIXR(0.76604444311897803520) -#define C5 FIXR(0.64278760968653932632) -#define C6 FIXR(0.5) -#define C7 FIXR(0.34202014332566873304) -#define C8 FIXR(0.17364817766693034885) +#define C3 FIXHR(0.86602540378443864676/2) /* 0.5 / cos(pi*(2*i+1)/36) */ static const int icos36[9] = { FIXR(0.50190991877167369479), - FIXR(0.51763809020504152469), + FIXR(0.51763809020504152469), //0 FIXR(0.55168895948124587824), FIXR(0.61038729438072803416), - FIXR(0.70710678118654752439), + FIXR(0.70710678118654752439), //1 FIXR(0.87172339781054900991), FIXR(1.18310079157624925896), - FIXR(1.93185165257813657349), + FIXR(1.93185165257813657349), //2 FIXR(5.73685662283492756461), }; -static const int icos72[18] = { - /* 0.5 / cos(pi*(2*i+19)/72) */ - FIXR(0.74009361646113053152), - FIXR(0.82133981585229078570), - FIXR(0.93057949835178895673), - FIXR(1.08284028510010010928), - FIXR(1.30656296487637652785), - FIXR(1.66275476171152078719), - FIXR(2.31011315767264929558), - FIXR(3.83064878777019433457), - FIXR(11.46279281302667383546), - - /* 0.5 / cos(pi*(2*(i + 18) +19)/72) */ - FIXR(-0.67817085245462840086), - FIXR(-0.63023620700513223342), - FIXR(-0.59284452371708034528), - FIXR(-0.56369097343317117734), - FIXR(-0.54119610014619698439), - FIXR(-0.52426456257040533932), - FIXR(-0.51213975715725461845), - FIXR(-0.50431448029007636036), - FIXR(-0.50047634258165998492), -}; +/* 12 points IMDCT. We compute it "by hand" by factorizing obvious + cases. */ +static void imdct12(int *out, int *in) +{ + int in0, in1, in2, in3, in4, in5, t1, t2; + + in0= in[0*3]; + in1= in[1*3] + in[0*3]; + in2= in[2*3] + in[1*3]; + in3= in[3*3] + in[2*3]; + in4= in[4*3] + in[3*3]; + in5= in[5*3] + in[4*3]; + in5 += in3; + in3 += in1; + + in2= MULH(2*in2, C3); + in3= MULH(2*in3, C3); + + t1 = in0 - in4; + t2 = MULL(in1 - in5, icos36[4]); + + out[ 7]= + out[10]= t1 + t2; + out[ 1]= + out[ 4]= t1 - t2; + + in0 += in4>>1; + in4 = in0 + in2; + in1 += in5>>1; + in5 = MULL(in1 + in3, icos36[1]); + out[ 8]= + out[ 9]= in4 + in5; + out[ 2]= + out[ 3]= in4 - in5; + + in0 -= in2; + in1 = MULL(in1 - in3, icos36[7]); + out[ 0]= + out[ 5]= in0 - in1; + out[ 6]= + out[11]= in0 + in1; +} + +/* cos(pi*i/18) */ +#define C1 FIXHR(0.98480775301220805936/2) +#define C2 FIXHR(0.93969262078590838405/2) +#define C3 FIXHR(0.86602540378443864676/2) +#define C4 FIXHR(0.76604444311897803520/2) +#define C5 FIXHR(0.64278760968653932632/2) +#define C6 FIXHR(0.5/2) +#define C7 FIXHR(0.34202014332566873304/2) +#define C8 FIXHR(0.17364817766693034885/2) + /* using Lee like decomposition followed by hand coded 9 points DCT */ -static void imdct36(int *out, int *in) +static void imdct36(int *out, int *buf, int *in, int *win) { int i, j, t0, t1, t2, t3, s0, s1, s2, s3; int tmp[18], *tmp1, *in1; - INT64 in3_3, in6_6; for(i=17;i>=1;i--) in[i] += in[i-1]; @@ -964,30 +1002,61 @@ static void imdct36(int *out, int *in) for(j=0;j<2;j++) { tmp1 = tmp + j; in1 = in + j; +#if 0 +//more accurate but slower + int64_t t0, t1, t2, t3; + t2 = in1[2*4] + in1[2*8] - in1[2*2]; - in3_3 = MUL64(in1[2*3], C3); - in6_6 = MUL64(in1[2*6], C6); - - tmp1[0] = FRAC_RND(MUL64(in1[2*1], C1) + in3_3 + - MUL64(in1[2*5], C5) + MUL64(in1[2*7], C7)); - tmp1[2] = in1[2*0] + FRAC_RND(MUL64(in1[2*2], C2) + - MUL64(in1[2*4], C4) + in6_6 + - MUL64(in1[2*8], C8)); - tmp1[4] = FRAC_RND(MUL64(in1[2*1] - in1[2*5] - in1[2*7], C3)); - tmp1[6] = FRAC_RND(MUL64(in1[2*2] - in1[2*4] - in1[2*8], C6)) - - in1[2*6] + in1[2*0]; - tmp1[8] = FRAC_RND(MUL64(in1[2*1], C5) - in3_3 - - MUL64(in1[2*5], C7) + MUL64(in1[2*7], C1)); - tmp1[10] = in1[2*0] + FRAC_RND(MUL64(-in1[2*2], C8) - - MUL64(in1[2*4], C2) + in6_6 + - MUL64(in1[2*8], C4)); - tmp1[12] = FRAC_RND(MUL64(in1[2*1], C7) - in3_3 + - MUL64(in1[2*5], C1) - - MUL64(in1[2*7], C5)); - tmp1[14] = in1[2*0] + FRAC_RND(MUL64(-in1[2*2], C4) + - MUL64(in1[2*4], C8) + in6_6 - - MUL64(in1[2*8], C2)); - tmp1[16] = in1[2*0] - in1[2*2] + in1[2*4] - in1[2*6] + in1[2*8]; + t3 = (in1[2*0] + (int64_t)(in1[2*6]>>1))<<32; + t1 = in1[2*0] - in1[2*6]; + tmp1[ 6] = t1 - (t2>>1); + tmp1[16] = t1 + t2; + + t0 = MUL64(2*(in1[2*2] + in1[2*4]), C2); + t1 = MUL64( in1[2*4] - in1[2*8] , -2*C8); + t2 = MUL64(2*(in1[2*2] + in1[2*8]), -C4); + + tmp1[10] = (t3 - t0 - t2) >> 32; + tmp1[ 2] = (t3 + t0 + t1) >> 32; + tmp1[14] = (t3 + t2 - t1) >> 32; + + tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3); + t2 = MUL64(2*(in1[2*1] + in1[2*5]), C1); + t3 = MUL64( in1[2*5] - in1[2*7] , -2*C7); + t0 = MUL64(2*in1[2*3], C3); + + t1 = MUL64(2*(in1[2*1] + in1[2*7]), -C5); + + tmp1[ 0] = (t2 + t3 + t0) >> 32; + tmp1[12] = (t2 + t1 - t0) >> 32; + tmp1[ 8] = (t3 - t1 - t0) >> 32; +#else + t2 = in1[2*4] + in1[2*8] - in1[2*2]; + + t3 = in1[2*0] + (in1[2*6]>>1); + t1 = in1[2*0] - in1[2*6]; + tmp1[ 6] = t1 - (t2>>1); + tmp1[16] = t1 + t2; + + t0 = MULH(2*(in1[2*2] + in1[2*4]), C2); + t1 = MULH( in1[2*4] - in1[2*8] , -2*C8); + t2 = MULH(2*(in1[2*2] + in1[2*8]), -C4); + + tmp1[10] = t3 - t0 - t2; + tmp1[ 2] = t3 + t0 + t1; + tmp1[14] = t3 + t2 - t1; + + tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3); + t2 = MULH(2*(in1[2*1] + in1[2*5]), C1); + t3 = MULH( in1[2*5] - in1[2*7] , -2*C7); + t0 = MULH(2*in1[2*3], C3); + + t1 = MULH(2*(in1[2*1] + in1[2*7]), -C5); + + tmp1[ 0] = t2 + t3 + t0; + tmp1[12] = t2 + t1 - t0; + tmp1[ 8] = t3 - t1 - t0; +#endif } i = 0; @@ -1001,59 +1070,37 @@ static void imdct36(int *out, int *in) t3 = tmp[i + 3]; s1 = MULL(t3 + t2, icos36[j]); s3 = MULL(t3 - t2, icos36[8 - j]); - - t0 = MULL(s0 + s1, icos72[9 + 8 - j]); - t1 = MULL(s0 - s1, icos72[8 - j]); - out[18 + 9 + j] = t0; - out[18 + 8 - j] = t0; - out[9 + j] = -t1; - out[8 - j] = t1; - - t0 = MULL(s2 + s3, icos72[9+j]); - t1 = MULL(s2 - s3, icos72[j]); - out[18 + 9 + (8 - j)] = t0; - out[18 + j] = t0; - out[9 + (8 - j)] = -t1; - out[j] = t1; + + t0 = s0 + s1; + t1 = s0 - s1; + out[(9 + j)*SBLIMIT] = MULH(t1, win[9 + j]) + buf[9 + j]; + out[(8 - j)*SBLIMIT] = MULH(t1, win[8 - j]) + buf[8 - j]; + buf[9 + j] = MULH(t0, win[18 + 9 + j]); + buf[8 - j] = MULH(t0, win[18 + 8 - j]); + + t0 = s2 + s3; + t1 = s2 - s3; + out[(9 + 8 - j)*SBLIMIT] = MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j]; + out[( j)*SBLIMIT] = MULH(t1, win[ j]) + buf[ j]; + buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]); + buf[ + j] = MULH(t0, win[18 + j]); i += 4; } s0 = tmp[16]; s1 = MULL(tmp[17], icos36[4]); - t0 = MULL(s0 + s1, icos72[9 + 4]); - t1 = MULL(s0 - s1, icos72[4]); - out[18 + 9 + 4] = t0; - out[18 + 8 - 4] = t0; - out[9 + 4] = -t1; - out[8 - 4] = t1; -} - -/* fast header check for resync */ -static int check_header(UINT32 header) -{ - /* header */ - if ((header & 0xffe00000) != 0xffe00000) - return -1; - /* layer check */ - if (((header >> 17) & 3) == 0) - return -1; - /* bit rate */ - if (((header >> 12) & 0xf) == 0xf) - return -1; - /* frequency */ - if (((header >> 10) & 3) == 3) - return -1; - return 0; + t0 = s0 + s1; + t1 = s0 - s1; + out[(9 + 4)*SBLIMIT] = MULH(t1, win[9 + 4]) + buf[9 + 4]; + out[(8 - 4)*SBLIMIT] = MULH(t1, win[8 - 4]) + buf[8 - 4]; + buf[9 + 4] = MULH(t0, win[18 + 9 + 4]); + buf[8 - 4] = MULH(t0, win[18 + 8 - 4]); } -/* header + layer + bitrate + freq + lsf/mpeg25 */ -#define SAME_HEADER_MASK \ - (0xffe00000 | (3 << 17) | (0xf << 12) | (3 << 10) | (3 << 19)) - /* header decoding. MUST check the header before because no consistency check is done there. Return 1 if free format found and that the frame size must be computed externally */ -static int decode_header(MPADecodeContext *s, UINT32 header) +static int decode_header(MPADecodeContext *s, uint32_t header) { int sample_rate, frame_size, mpeg25, padding; int sample_rate_index, bitrate_index; @@ -1064,16 +1111,15 @@ static int decode_header(MPADecodeContext *s, UINT32 header) s->lsf = 1; mpeg25 = 1; } - + s->layer = 4 - ((header >> 17) & 3); /* extract frequency */ sample_rate_index = (header >> 10) & 3; sample_rate = mpa_freq_tab[sample_rate_index] >> (s->lsf + mpeg25); - if (sample_rate == 0) - return 1; sample_rate_index += 3 * (s->lsf + mpeg25); s->sample_rate_index = sample_rate_index; s->error_protection = ((header >> 16) & 1) ^ 1; + s->sample_rate = sample_rate; bitrate_index = (header >> 12) & 0xf; padding = (header >> 9) & 1; @@ -1088,7 +1134,7 @@ static int decode_header(MPADecodeContext *s, UINT32 header) s->nb_channels = 1; else s->nb_channels = 2; - + if (bitrate_index != 0) { frame_size = mpa_bitrate_tab[s->lsf][s->layer - 1][bitrate_index]; s->bit_rate = frame_size * 1000; @@ -1131,9 +1177,8 @@ static int decode_header(MPADecodeContext *s, UINT32 header) break; } } - s->sample_rate = sample_rate; - -#ifdef DEBUG + +#if defined(DEBUG) printf("layer%d, %d Hz, %d kbits/s, ", s->layer, s->sample_rate, s->bit_rate); if (s->nb_channels == 2) { @@ -1152,14 +1197,51 @@ static int decode_header(MPADecodeContext *s, UINT32 header) return 0; } +/* useful helper to get mpeg audio stream infos. Return -1 if error in + header, otherwise the coded frame size in bytes */ +int mpa_decode_header(AVCodecContext *avctx, uint32_t head) +{ + MPADecodeContext s1, *s = &s1; + memset( s, 0, sizeof(MPADecodeContext) ); + + if (ff_mpa_check_header(head) != 0) + return -1; + + if (decode_header(s, head) != 0) { + return -1; + } + + switch(s->layer) { + case 1: + avctx->frame_size = 384; + break; + case 2: + avctx->frame_size = 1152; + break; + default: + case 3: + if (s->lsf) + avctx->frame_size = 576; + else + avctx->frame_size = 1152; + break; + } + + avctx->sample_rate = s->sample_rate; + avctx->channels = s->nb_channels; + avctx->bit_rate = s->bit_rate; + avctx->sub_id = s->layer; + return s->frame_size; +} + /* return the number of decoded frames */ static int mp_decode_layer1(MPADecodeContext *s) { int bound, i, v, n, ch, j, mant; - UINT8 allocation[MPA_MAX_CHANNELS][SBLIMIT]; - UINT8 scale_factors[MPA_MAX_CHANNELS][SBLIMIT]; + uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT]; + uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT]; - if (s->mode == MPA_JSTEREO) + if (s->mode == MPA_JSTEREO) bound = (s->mode_ext + 1) * 4; else bound = SBLIMIT; @@ -1187,7 +1269,7 @@ static int mp_decode_layer1(MPADecodeContext *s) scale_factors[1][i] = get_bits(&s->gb, 6); } } - + /* compute samples */ for(j=0;j<12;j++) { for(i=0;i= 56) || - (ch_bitrate >= 56 && ch_bitrate <= 80)) + (ch_bitrate >= 56 && ch_bitrate <= 80)) table = 0; - else if (freq != 48000 && ch_bitrate >= 96) + else if (freq != 48000 && ch_bitrate >= 96) table = 1; - else if (freq != 32000 && ch_bitrate <= 48) + else if (freq != 32000 && ch_bitrate <= 48) table = 2; - else + else table = 3; } else { table = 4; @@ -1252,17 +1334,21 @@ static int mp_decode_layer2(MPADecodeContext *s) int scale, qindex, bits, steps, k, l, m, b; /* select decoding table */ - table = l2_select_table(s->bit_rate / 1000, s->nb_channels, + table = l2_select_table(s->bit_rate / 1000, s->nb_channels, s->sample_rate, s->lsf); sblimit = sblimit_table[table]; alloc_table = alloc_tables[table]; - if (s->mode == MPA_JSTEREO) + if (s->mode == MPA_JSTEREO) bound = (s->mode_ext + 1) * 4; else bound = sblimit; dprintf("bound=%d sblimit=%d\n", bound, sblimit); + + /* sanity check */ + if( bound > sblimit ) bound = sblimit; + /* parse bit allocation */ j = 0; for(i=0;inb_channels;ch++) { - if (bit_alloc[ch][i]) + if (bit_alloc[ch][i]) scale_code[ch][i] = get_bits(&s->gb, 2); } } - + /* scale factors */ for(i=0;inb_channels;ch++) { @@ -1360,13 +1446,13 @@ static int mp_decode_layer2(MPADecodeContext *s) /* 3 values at the same time */ v = get_bits(&s->gb, -bits); steps = quant_steps[qindex]; - s->sb_samples[ch][k * 12 + l + 0][i] = + s->sb_samples[ch][k * 12 + l + 0][i] = l2_unscale_group(steps, v % steps, scale); v = v / steps; - s->sb_samples[ch][k * 12 + l + 1][i] = + s->sb_samples[ch][k * 12 + l + 1][i] = l2_unscale_group(steps, v % steps, scale); v = v / steps; - s->sb_samples[ch][k * 12 + l + 2][i] = + s->sb_samples[ch][k * 12 + l + 2][i] = l2_unscale_group(steps, v, scale); } else { for(m=0;m<3;m++) { @@ -1382,7 +1468,7 @@ static int mp_decode_layer2(MPADecodeContext *s) } } /* next subband in alloc table */ - j += 1 << bit_alloc_bits; + j += 1 << bit_alloc_bits; } /* XXX: find a way to avoid this duplication of code */ for(i=bound;isb_samples[0][k * 12 + l + 0][i] = + s->sb_samples[0][k * 12 + l + 0][i] = l2_unscale_group(steps, mant, scale0); - s->sb_samples[1][k * 12 + l + 0][i] = + s->sb_samples[1][k * 12 + l + 0][i] = l2_unscale_group(steps, mant, scale1); mant = v % steps; v = v / steps; - s->sb_samples[0][k * 12 + l + 1][i] = + s->sb_samples[0][k * 12 + l + 1][i] = l2_unscale_group(steps, mant, scale0); - s->sb_samples[1][k * 12 + l + 1][i] = + s->sb_samples[1][k * 12 + l + 1][i] = l2_unscale_group(steps, mant, scale1); - s->sb_samples[0][k * 12 + l + 2][i] = + s->sb_samples[0][k * 12 + l + 2][i] = l2_unscale_group(steps, v, scale0); - s->sb_samples[1][k * 12 + l + 2][i] = + s->sb_samples[1][k * 12 + l + 2][i] = l2_unscale_group(steps, v, scale1); } else { for(m=0;m<3;m++) { mant = get_bits(&s->gb, bits); - s->sb_samples[0][k * 12 + l + m][i] = + s->sb_samples[0][k * 12 + l + m][i] = l1_unscale(bits - 1, mant, scale0); - s->sb_samples[1][k * 12 + l + m][i] = + s->sb_samples[1][k * 12 + l + m][i] = l1_unscale(bits - 1, mant, scale1); } } @@ -1432,7 +1518,7 @@ static int mp_decode_layer2(MPADecodeContext *s) s->sb_samples[1][k * 12 + l + 2][i] = 0; } /* next subband in alloc table */ - j += 1 << bit_alloc_bits; + j += 1 << bit_alloc_bits; } /* fill remaining samples to zero */ for(i=sblimit;igb.buffer + (s->gb.index>>3); -#else - ptr = s->gb.buf_ptr - (s->gb.bit_cnt >> 3); -#endif + ptr = (uint8_t *)(s->gb.buffer + (get_bits_count(&s->gb)>>3)); + /* copy old data before current one */ ptr -= backstep; - memcpy(ptr, s->inbuf1[s->inbuf_index ^ 1] + + memcpy(ptr, s->inbuf1[s->inbuf_index ^ 1] + BACKSTEP_SIZE + s->old_frame_size - backstep, backstep); /* init get bits again */ - init_get_bits(&s->gb, ptr, s->frame_size + backstep); + init_get_bits(&s->gb, ptr, (s->frame_size + backstep)*8); /* prepare next buffer */ s->inbuf_index ^= 1; @@ -1493,13 +1576,13 @@ static inline void lsf_sf_expand(int *slen, slen[0] = sf; } -static void exponents_from_scale_factors(MPADecodeContext *s, +static void exponents_from_scale_factors(MPADecodeContext *s, GranuleDef *g, - INT16 *exponents) + int16_t *exponents) { - const UINT8 *bstab, *pretab; + const uint8_t *bstab, *pretab; int len, i, j, k, l, v0, shift, gain, gains[3]; - INT16 *exp_ptr; + int16_t *exp_ptr; exp_ptr = exponents; gain = g->global_gain - 210; @@ -1541,15 +1624,13 @@ static inline int get_bitsz(GetBitContext *s, int n) } static int huffman_decode(MPADecodeContext *s, GranuleDef *g, - INT16 *exponents, int end_pos) + int16_t *exponents, int end_pos) { int s_index; int linbits, code, x, y, l, v, i, j, k, pos; - UINT8 *last_buf_ptr; - UINT32 last_bit_buf; - int last_bit_cnt; + GetBitContext last_gb; VLC *vlc; - UINT8 *code_table; + uint8_t *code_table; /* low frequencies (called big values) */ s_index = 0; @@ -1569,7 +1650,7 @@ static int huffman_decode(MPADecodeContext *s, GranuleDef *g, if (get_bits_count(&s->gb) >= end_pos) break; if (code_table) { - code = get_vlc(&s->gb, vlc); + code = get_vlc2(&s->gb, vlc->table, 8, 2); if (code < 0) return -1; y = code_table[code]; @@ -1579,7 +1660,7 @@ static int huffman_decode(MPADecodeContext *s, GranuleDef *g, x = 0; y = 0; } - dprintf("region=%d n=%d x=%d y=%d exp=%d\n", + dprintf("region=%d n=%d x=%d y=%d exp=%d\n", i, g->region_size[i] - j, x, y, exponents[s_index]); if (x) { if (x == 15) @@ -1603,40 +1684,24 @@ static int huffman_decode(MPADecodeContext *s, GranuleDef *g, g->sb_hybrid[s_index++] = v; } } - + /* high frequencies */ vlc = &huff_quad_vlc[g->count1table_select]; - last_buf_ptr = NULL; - last_bit_buf = 0; - last_bit_cnt = 0; + last_gb.buffer = NULL; while (s_index <= 572) { pos = get_bits_count(&s->gb); if (pos >= end_pos) { - if (pos > end_pos && last_buf_ptr != NULL) { + if (pos > end_pos && last_gb.buffer != NULL) { /* some encoders generate an incorrect size for this part. We must go back into the data */ s_index -= 4; -#ifdef ALT_BITSTREAM_READER - s->gb.buffer = last_buf_ptr; - s->gb.index = last_bit_cnt; -#else - s->gb.buf_ptr = last_buf_ptr; - s->gb.bit_buf = last_bit_buf; - s->gb.bit_cnt = last_bit_cnt; -#endif + s->gb = last_gb; } break; } -#ifdef ALT_BITSTREAM_READER - last_buf_ptr = s->gb.buffer; - last_bit_cnt = s->gb.index; -#else - last_buf_ptr = s->gb.buf_ptr; - last_bit_buf = s->gb.bit_buf; - last_bit_cnt = s->gb.bit_cnt; -#endif - - code = get_vlc(&s->gb, vlc); + last_gb= s->gb; + + code = get_vlc2(&s->gb, vlc->table, vlc->bits, 2); dprintf("t=%d code=%d\n", g->count1table_select, code); if (code < 0) return -1; @@ -1664,8 +1729,8 @@ static int huffman_decode(MPADecodeContext *s, GranuleDef *g, static void reorder_block(MPADecodeContext *s, GranuleDef *g) { int i, j, k, len; - INT32 *ptr, *dst, *ptr1; - INT32 tmp[576]; + int32_t *ptr, *dst, *ptr1; + int32_t tmp[576]; if (g->block_type != 2) return; @@ -1679,7 +1744,7 @@ static void reorder_block(MPADecodeContext *s, GranuleDef *g) } else { ptr = g->sb_hybrid; } - + for(i=g->short_start;i<13;i++) { len = band_size_short[s->sample_rate_index][i]; ptr1 = ptr; @@ -1690,7 +1755,7 @@ static void reorder_block(MPADecodeContext *s, GranuleDef *g) dst += 3; } } - memcpy(ptr1, tmp, len * 3 * sizeof(INT32)); + memcpy(ptr1, tmp, len * 3 * sizeof(int32_t)); } } @@ -1700,10 +1765,10 @@ static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1) { int i, j, k, l; - INT32 v1, v2; + int32_t v1, v2; int sf_max, tmp0, tmp1, sf, len, non_zero_found; - INT32 (*is_tab)[16]; - INT32 *tab0, *tab1; + int32_t (*is_tab)[16]; + int32_t *tab0, *tab1; int non_zero_found_short[3]; /* intensity stereo */ @@ -1715,7 +1780,7 @@ static void compute_stereo(MPADecodeContext *s, is_tab = is_table_lsf[g1->scalefac_compress & 1]; sf_max = 16; } - + tab0 = g0->sb_hybrid + 576; tab1 = g1->sb_hybrid + 576; @@ -1766,8 +1831,8 @@ static void compute_stereo(MPADecodeContext *s, } } - non_zero_found = non_zero_found_short[0] | - non_zero_found_short[1] | + non_zero_found = non_zero_found_short[0] | + non_zero_found_short[1] | non_zero_found_short[2]; for(i = g1->long_end - 1;i >= 0;i--) { @@ -1823,11 +1888,11 @@ static void compute_stereo(MPADecodeContext *s, } } -static void compute_antialias(MPADecodeContext *s, +static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g) { - INT32 *ptr, *p0, *p1, *csa; - int n, tmp0, tmp1, i, j; + int32_t *ptr, *csa; + int n, i; /* we antialias only "long" bands */ if (g->block_type == 2) { @@ -1838,35 +1903,78 @@ static void compute_antialias(MPADecodeContext *s, } else { n = SBLIMIT - 1; } - + ptr = g->sb_hybrid + 18; for(i = n;i > 0;i--) { - p0 = ptr - 1; - p1 = ptr; + int tmp0, tmp1, tmp2; csa = &csa_table[0][0]; - for(j=0;j<8;j++) { - tmp0 = *p0; - tmp1 = *p1; - *p0 = FRAC_RND(MUL64(tmp0, csa[0]) - MUL64(tmp1, csa[1])); - *p1 = FRAC_RND(MUL64(tmp0, csa[1]) + MUL64(tmp1, csa[0])); - p0--; - p1++; - csa += 2; - } +#define INT_AA(j) \ + tmp0 = ptr[-1-j];\ + tmp1 = ptr[ j];\ + tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\ + ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\ + ptr[ j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j])); + + INT_AA(0) + INT_AA(1) + INT_AA(2) + INT_AA(3) + INT_AA(4) + INT_AA(5) + INT_AA(6) + INT_AA(7) + + ptr += 18; + } +} + +static void compute_antialias_float(MPADecodeContext *s, + GranuleDef *g) +{ + int32_t *ptr; + int n, i; + + /* we antialias only "long" bands */ + if (g->block_type == 2) { + if (!g->switch_point) + return; + /* XXX: check this for 8000Hz case */ + n = 1; + } else { + n = SBLIMIT - 1; + } + + ptr = g->sb_hybrid + 18; + for(i = n;i > 0;i--) { + float tmp0, tmp1; + float *csa = &csa_table_float[0][0]; +#define FLOAT_AA(j)\ + tmp0= ptr[-1-j];\ + tmp1= ptr[ j];\ + ptr[-1-j] = lrintf(tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j]);\ + ptr[ j] = lrintf(tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j]); + + FLOAT_AA(0) + FLOAT_AA(1) + FLOAT_AA(2) + FLOAT_AA(3) + FLOAT_AA(4) + FLOAT_AA(5) + FLOAT_AA(6) + FLOAT_AA(7) + ptr += 18; } } static void compute_imdct(MPADecodeContext *s, - GranuleDef *g, - INT32 *sb_samples, - INT32 *mdct_buf) + GranuleDef *g, + int32_t *sb_samples, + int32_t *mdct_buf) { - INT32 *ptr, *win, *win1, *buf, *buf2, *out_ptr, *ptr1; - INT32 in[6]; - INT32 out[36]; - INT32 out2[12]; - int i, j, k, mdct_long_end, v, sblimit; + int32_t *ptr, *win, *win1, *buf, *out_ptr, *ptr1; + int32_t out2[12]; + int i, j, mdct_long_end, v, sblimit; /* find last non zero block */ ptr = g->sb_hybrid + 576; @@ -1892,7 +2000,6 @@ static void compute_imdct(MPADecodeContext *s, buf = mdct_buf; ptr = g->sb_hybrid; for(j=0;jblock_type]; /* select frequency inversion */ win = win1 + ((4 * 36) & -(j & 1)); - for(i=0;i<18;i++) { - *out_ptr = MULL(out[i], win[i]) + buf[i]; - buf[i] = MULL(out[i + 18], win[i + 18]); - out_ptr += SBLIMIT; - } + imdct36(out_ptr, buf, ptr, win); + out_ptr += 18*SBLIMIT; ptr += 18; buf += 18; } for(j=mdct_long_end;jlsf) { @@ -2017,7 +2128,7 @@ static int mp_decode_layer3(MPADecodeContext *s) granules[ch][1].scfsi = get_bits(&s->gb, 4); } } - + for(gr=0;grnb_channels;ch++) { dprintf("gr=%d ch=%d: side_info\n", gr, ch); @@ -2027,7 +2138,7 @@ static int mp_decode_layer3(MPADecodeContext *s) g->global_gain = get_bits(&s->gb, 8); /* if MS stereo only is selected, we precompute the 1/sqrt(2) renormalization factor */ - if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) == + if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) == MODE_EXT_MS_STEREO) g->global_gain -= 2; if (s->lsf) @@ -2042,15 +2153,15 @@ static int mp_decode_layer3(MPADecodeContext *s) g->switch_point = get_bits(&s->gb, 1); for(i=0;i<2;i++) g->table_select[i] = get_bits(&s->gb, 5); - for(i=0;i<3;i++) + for(i=0;i<3;i++) g->subblock_gain[i] = get_bits(&s->gb, 3); /* compute huffman coded region sizes */ if (g->block_type == 2) g->region_size[0] = (36 / 2); else { - if (s->sample_rate_index <= 2) + if (s->sample_rate_index <= 2) g->region_size[0] = (36 / 2); - else if (s->sample_rate_index != 8) + else if (s->sample_rate_index != 8) g->region_size[0] = (54 / 2); else g->region_size[0] = (108 / 2); @@ -2065,15 +2176,15 @@ static int mp_decode_layer3(MPADecodeContext *s) /* compute huffman coded region sizes */ region_address1 = get_bits(&s->gb, 4); region_address2 = get_bits(&s->gb, 3); - dprintf("region1=%d region2=%d\n", + dprintf("region1=%d region2=%d\n", region_address1, region_address2); - g->region_size[0] = + g->region_size[0] = band_index_long[s->sample_rate_index][region_address1 + 1] >> 1; l = region_address1 + region_address2 + 2; /* should not overflow */ if (l > 22) l = 22; - g->region_size[1] = + g->region_size[1] = band_index_long[s->sample_rate_index][l] >> 1; } /* convert region offsets to region sizes and truncate @@ -2100,11 +2211,11 @@ static int mp_decode_layer3(MPADecodeContext *s) g->long_end = 6; else g->long_end = 4; /* 8000 Hz */ - + if (s->sample_rate_index != 8) g->short_start = 3; else - g->short_start = 2; + g->short_start = 2; } else { g->long_end = 0; g->short_start = 0; @@ -2113,7 +2224,7 @@ static int mp_decode_layer3(MPADecodeContext *s) g->short_start = 13; g->long_end = 22; } - + g->preflag = 0; if (!s->lsf) g->preflag = get_bits(&s->gb, 1); @@ -2124,18 +2235,20 @@ static int mp_decode_layer3(MPADecodeContext *s) } } + if (!s->adu_mode) { /* now we get bits from the main_data_begin offset */ dprintf("seekback: %d\n", main_data_begin); seek_to_maindata(s, main_data_begin); + } for(gr=0;grnb_channels;ch++) { g = &granules[ch][gr]; - + bits_pos = get_bits_count(&s->gb); - + if (!s->lsf) { - UINT8 *sc; + uint8_t *sc; int slen, slen1, slen2; /* MPEG1 scale factors */ @@ -2172,7 +2285,7 @@ static int mp_decode_layer3(MPADecodeContext *s) } #if defined(DEBUG) { - printf("scfsi=%x gr=%d ch=%d scale_factors:\n", + printf("scfsi=%x gr=%d ch=%d scale_factors:\n", g->scfsi, gr, ch); for(i=0;iscale_factors[i]); @@ -2229,7 +2342,7 @@ static int mp_decode_layer3(MPADecodeContext *s) g->scale_factors[j] = 0; #if defined(DEBUG) { - printf("gr=%d ch=%d scale_factors:\n", + printf("gr=%d ch=%d scale_factors:\n", gr, ch); for(i=0;i<40;i++) printf(" %d", g->scale_factors[i]); @@ -2272,12 +2385,12 @@ static int mp_decode_layer3(MPADecodeContext *s) #if defined(DEBUG) sample_dump(0, g->sb_hybrid, 576); #endif - compute_antialias(s, g); -#ifdef DEBUG + s->compute_antialias(s, g); +#if defined(DEBUG) sample_dump(1, g->sb_hybrid, 576); #endif - compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]); -#ifdef DEBUG + compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]); +#if defined(DEBUG) sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576); #endif } @@ -2285,15 +2398,15 @@ static int mp_decode_layer3(MPADecodeContext *s) return nb_granules * 18; } -static int mp_decode_frame(MPADecodeContext *s, - short *samples) +static int mp_decode_frame(MPADecodeContext *s, + OUT_INT *samples) { int i, nb_frames, ch; - short *samples_ptr; + OUT_INT *samples_ptr; + + init_get_bits(&s->gb, s->inbuf + HEADER_SIZE, + (s->inbuf_ptr - s->inbuf - HEADER_SIZE)*8); - init_get_bits(&s->gb, s->inbuf + HEADER_SIZE, - s->inbuf_ptr - s->inbuf - HEADER_SIZE); - /* skip error protection field */ if (s->error_protection) get_bits(&s->gb, 16); @@ -2326,32 +2439,33 @@ static int mp_decode_frame(MPADecodeContext *s, for(ch=0;chnb_channels;ch++) { samples_ptr = samples + ch; for(i=0;inb_channels, + ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]), + window, &s->dither_state, + samples_ptr, s->nb_channels, s->sb_samples[ch][i]); samples_ptr += 32 * s->nb_channels; } } #ifdef DEBUG - s->frame_count++; + s->frame_count++; #endif - return nb_frames * 32 * sizeof(short) * s->nb_channels; + return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels; } static int decode_frame(AVCodecContext * avctx, - void *data, int *data_size, - UINT8 * buf, int buf_size) + void *data, int *data_size, + uint8_t * buf, int buf_size) { MPADecodeContext *s = avctx->priv_data; - UINT32 header; - UINT8 *buf_ptr; + uint32_t header; + uint8_t *buf_ptr; int len, out_size; - short *out_samples = data; + OUT_INT *out_samples = data; - *data_size = 0; buf_ptr = buf; while (buf_size > 0) { - len = s->inbuf_ptr - s->inbuf; - if (s->frame_size == 0) { + len = s->inbuf_ptr - s->inbuf; + if (s->frame_size == 0) { /* special case for next header for first frame in free format case (XXX: find a simpler method) */ if (s->free_format_next_header != 0) { @@ -2363,56 +2477,69 @@ static int decode_frame(AVCodecContext * avctx, s->free_format_next_header = 0; goto got_header; } - /* no header seen : find one. We need at least HEADER_SIZE + /* no header seen : find one. We need at least HEADER_SIZE bytes to parse it */ - len = HEADER_SIZE - len; - if (len > buf_size) - len = buf_size; - if (len > 0) { - memcpy(s->inbuf_ptr, buf_ptr, len); - buf_ptr += len; - buf_size -= len; - s->inbuf_ptr += len; - } - if ((s->inbuf_ptr - s->inbuf) >= HEADER_SIZE) { + len = HEADER_SIZE - len; + if (len > buf_size) + len = buf_size; + if (len > 0) { + memcpy(s->inbuf_ptr, buf_ptr, len); + buf_ptr += len; + buf_size -= len; + s->inbuf_ptr += len; + } + if ((s->inbuf_ptr - s->inbuf) >= HEADER_SIZE) { got_header: - header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) | - (s->inbuf[2] << 8) | s->inbuf[3]; + header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) | + (s->inbuf[2] << 8) | s->inbuf[3]; - if (check_header(header) < 0) { - /* no sync found : move by one byte (inefficient, but simple!) */ - memcpy(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1); - s->inbuf_ptr--; + if (ff_mpa_check_header(header) < 0) { + /* no sync found : move by one byte (inefficient, but simple!) */ + memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1); + s->inbuf_ptr--; dprintf("skip %x\n", header); /* reset free format frame size to give a chance to get a new bitrate */ s->free_format_frame_size = 0; - } else { - if (decode_header(s, header) == 1) { - /* free format: compute frame size */ - s->frame_size = -1; - memcpy(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1); - s->inbuf_ptr--; - } else { - /* update codec info */ - avctx->sample_rate = s->sample_rate; - avctx->channels = s->nb_channels; - avctx->bit_rate = s->bit_rate; - avctx->frame_size = s->frame_size; + } else { + if (decode_header(s, header) == 1) { + /* free format: prepare to compute frame size */ + s->frame_size = -1; } - } - } + /* update codec info */ + avctx->sample_rate = s->sample_rate; + avctx->channels = s->nb_channels; + avctx->bit_rate = s->bit_rate; + avctx->sub_id = s->layer; + switch(s->layer) { + case 1: + avctx->frame_size = 384; + break; + case 2: + avctx->frame_size = 1152; + break; + case 3: + if (s->lsf) + avctx->frame_size = 576; + else + avctx->frame_size = 1152; + break; + } + } + } } else if (s->frame_size == -1) { /* free format : find next sync to compute frame size */ - len = MPA_MAX_CODED_FRAME_SIZE - len; - if (len > buf_size) - len = buf_size; + len = MPA_MAX_CODED_FRAME_SIZE - len; + if (len > buf_size) + len = buf_size; if (len == 0) { /* frame too long: resync */ s->frame_size = 0; + memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1); + s->inbuf_ptr--; } else { - UINT8 *p, *pend; - UINT32 header1; + uint8_t *p, *pend; + uint32_t header1; int padding; memcpy(s->inbuf_ptr, buf_ptr, len); @@ -2441,7 +2568,7 @@ static int decode_frame(AVCodecContext * avctx, s->free_format_frame_size -= padding * 4; else s->free_format_frame_size -= padding; - dprintf("free frame size=%d padding=%d\n", + dprintf("free frame size=%d padding=%d\n", s->free_format_frame_size, padding); decode_header(s, header1); goto next_data; @@ -2453,30 +2580,259 @@ static int decode_frame(AVCodecContext * avctx, s->inbuf_ptr += len; buf_size -= len; } - } else if (len < s->frame_size) { + } else if (len < s->frame_size) { if (s->frame_size > MPA_MAX_CODED_FRAME_SIZE) s->frame_size = MPA_MAX_CODED_FRAME_SIZE; - len = s->frame_size - len; - if (len > buf_size) - len = buf_size; - else if (len < 4) - len = buf_size > 4 ? 4 : buf_size; - memcpy(s->inbuf_ptr, buf_ptr, len); - buf_ptr += len; - s->inbuf_ptr += len; - buf_size -= len; - } else { - out_size = mp_decode_frame(s, out_samples); - s->inbuf_ptr = s->inbuf; - s->frame_size = 0; - *data_size = out_size; - break; - } + len = s->frame_size - len; + if (len > buf_size) + len = buf_size; + memcpy(s->inbuf_ptr, buf_ptr, len); + buf_ptr += len; + s->inbuf_ptr += len; + buf_size -= len; + } next_data: + if (s->frame_size > 0 && + (s->inbuf_ptr - s->inbuf) >= s->frame_size) { + if (avctx->parse_only) { + /* simply return the frame data */ + *(uint8_t **)data = s->inbuf; + out_size = s->inbuf_ptr - s->inbuf; + } else { + out_size = mp_decode_frame(s, out_samples); + } + s->inbuf_ptr = s->inbuf; + s->frame_size = 0; + if(out_size>=0) + *data_size = out_size; + else + av_log(avctx, AV_LOG_DEBUG, "Error while decoding mpeg audio frame\n"); //FIXME return -1 / but also return the number of bytes consumed + break; + } } return buf_ptr - buf; } + +static int decode_frame_adu(AVCodecContext * avctx, + void *data, int *data_size, + uint8_t * buf, int buf_size) +{ + MPADecodeContext *s = avctx->priv_data; + uint32_t header; + int len, out_size; + OUT_INT *out_samples = data; + + len = buf_size; + + // Discard too short frames + if (buf_size < HEADER_SIZE) { + *data_size = 0; + return buf_size; + } + + + if (len > MPA_MAX_CODED_FRAME_SIZE) + len = MPA_MAX_CODED_FRAME_SIZE; + + memcpy(s->inbuf, buf, len); + s->inbuf_ptr = s->inbuf + len; + + // Get header and restore sync word + header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) | + (s->inbuf[2] << 8) | s->inbuf[3] | 0xffe00000; + + if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame + *data_size = 0; + return buf_size; + } + + decode_header(s, header); + /* update codec info */ + avctx->sample_rate = s->sample_rate; + avctx->channels = s->nb_channels; + avctx->bit_rate = s->bit_rate; + avctx->sub_id = s->layer; + + avctx->frame_size=s->frame_size = len; + + if (avctx->parse_only) { + /* simply return the frame data */ + *(uint8_t **)data = s->inbuf; + out_size = s->inbuf_ptr - s->inbuf; + } else { + out_size = mp_decode_frame(s, out_samples); + } + + *data_size = out_size; + return buf_size; +} + + +/* Next 3 arrays are indexed by channel config number (passed via codecdata) */ +static int mp3Frames[16] = {0,1,1,2,3,3,4,5,2}; /* number of mp3 decoder instances */ +static int mp3Channels[16] = {0,1,2,3,4,5,6,8,4}; /* total output channels */ +/* offsets into output buffer, assume output order is FL FR BL BR C LFE */ +static int chan_offset[9][5] = { + {0}, + {0}, // C + {0}, // FLR + {2,0}, // C FLR + {2,0,3}, // C FLR BS + {4,0,2}, // C FLR BLRS + {4,0,2,5}, // C FLR BLRS LFE + {4,0,2,6,5}, // C FLR BLRS BLR LFE + {0,2} // FLR BLRS +}; + + +static int decode_init_mp3on4(AVCodecContext * avctx) +{ + MP3On4DecodeContext *s = avctx->priv_data; + int i; + + if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) { + av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n"); + return -1; + } + + s->chan_cfg = (((unsigned char *)avctx->extradata)[1] >> 3) & 0x0f; + s->frames = mp3Frames[s->chan_cfg]; + if(!s->frames) { + av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n"); + return -1; + } + avctx->channels = mp3Channels[s->chan_cfg]; + + /* Init the first mp3 decoder in standard way, so that all tables get builded + * We replace avctx->priv_data with the context of the first decoder so that + * decode_init() does not have to be changed. + * Other decoders will be inited here copying data from the first context + */ + // Allocate zeroed memory for the first decoder context + s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext)); + // Put decoder context in place to make init_decode() happy + avctx->priv_data = s->mp3decctx[0]; + decode_init(avctx); + // Restore mp3on4 context pointer + avctx->priv_data = s; + s->mp3decctx[0]->adu_mode = 1; // Set adu mode + + /* Create a separate codec/context for each frame (first is already ok). + * Each frame is 1 or 2 channels - up to 5 frames allowed + */ + for (i = 1; i < s->frames; i++) { + s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext)); + s->mp3decctx[i]->compute_antialias = s->mp3decctx[0]->compute_antialias; + s->mp3decctx[i]->inbuf = &s->mp3decctx[i]->inbuf1[0][BACKSTEP_SIZE]; + s->mp3decctx[i]->inbuf_ptr = s->mp3decctx[i]->inbuf; + s->mp3decctx[i]->adu_mode = 1; + } + + return 0; +} + + +static int decode_close_mp3on4(AVCodecContext * avctx) +{ + MP3On4DecodeContext *s = avctx->priv_data; + int i; + + for (i = 0; i < s->frames; i++) + if (s->mp3decctx[i]) + av_free(s->mp3decctx[i]); + + return 0; +} + + +static int decode_frame_mp3on4(AVCodecContext * avctx, + void *data, int *data_size, + uint8_t * buf, int buf_size) +{ + MP3On4DecodeContext *s = avctx->priv_data; + MPADecodeContext *m; + int len, out_size = 0; + uint32_t header; + OUT_INT *out_samples = data; + OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS]; + OUT_INT *outptr, *bp; + int fsize; + unsigned char *start2 = buf, *start; + int fr, i, j, n; + int off = avctx->channels; + int *coff = chan_offset[s->chan_cfg]; + + len = buf_size; + + // Discard too short frames + if (buf_size < HEADER_SIZE) { + *data_size = 0; + return buf_size; + } + + // If only one decoder interleave is not needed + outptr = s->frames == 1 ? out_samples : decoded_buf; + + for (fr = 0; fr < s->frames; fr++) { + start = start2; + fsize = (start[0] << 4) | (start[1] >> 4); + start2 += fsize; + if (fsize > len) + fsize = len; + len -= fsize; + if (fsize > MPA_MAX_CODED_FRAME_SIZE) + fsize = MPA_MAX_CODED_FRAME_SIZE; + m = s->mp3decctx[fr]; + assert (m != NULL); + /* copy original to new */ + m->inbuf_ptr = m->inbuf + fsize; + memcpy(m->inbuf, start, fsize); + + // Get header + header = (m->inbuf[0] << 24) | (m->inbuf[1] << 16) | + (m->inbuf[2] << 8) | m->inbuf[3] | 0xfff00000; + + if (ff_mpa_check_header(header) < 0) { // Bad header, discard block + *data_size = 0; + return buf_size; + } + + decode_header(m, header); + mp_decode_frame(m, decoded_buf); + + n = MPA_FRAME_SIZE * m->nb_channels; + out_size += n * sizeof(OUT_INT); + if(s->frames > 1) { + /* interleave output data */ + bp = out_samples + coff[fr]; + if(m->nb_channels == 1) { + for(j = 0; j < n; j++) { + *bp = decoded_buf[j]; + bp += off; + } + } else { + for(j = 0; j < n; j++) { + bp[0] = decoded_buf[j++]; + bp[1] = decoded_buf[j]; + bp += off; + } + } + } + } + + /* update codec info */ + avctx->sample_rate = s->mp3decctx[0]->sample_rate; + avctx->frame_size= buf_size; + avctx->bit_rate = 0; + for (i = 0; i < s->frames; i++) + avctx->bit_rate += s->mp3decctx[i]->bit_rate; + + *data_size = out_size; + return buf_size; +} + + AVCodec mp2_decoder = { "mp2", @@ -2487,16 +2843,44 @@ AVCodec mp2_decoder = NULL, NULL, decode_frame, + CODEC_CAP_PARSE_ONLY, }; AVCodec mp3_decoder = { "mp3", CODEC_TYPE_AUDIO, - CODEC_ID_MP3LAME, + CODEC_ID_MP3, sizeof(MPADecodeContext), decode_init, NULL, NULL, decode_frame, + CODEC_CAP_PARSE_ONLY, +}; + +AVCodec mp3adu_decoder = +{ + "mp3adu", + CODEC_TYPE_AUDIO, + CODEC_ID_MP3ADU, + sizeof(MPADecodeContext), + decode_init, + NULL, + NULL, + decode_frame_adu, + CODEC_CAP_PARSE_ONLY, +}; + +AVCodec mp3on4_decoder = +{ + "mp3on4", + CODEC_TYPE_AUDIO, + CODEC_ID_MP3ON4, + sizeof(MP3On4DecodeContext), + decode_init_mp3on4, + NULL, + decode_close_mp3on4, + decode_frame_mp3on4, + 0 };