X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Falac.c;h=5c48a4b28f4a105a8a34dcc602e8bddf788ef90a;hb=b07781b6e4f6267f20aec4e7f1d390ddb223af46;hp=3dd0ab9edb131e9b40f30c6c46c2e1ff2f68a8cc;hpb=a1301f2934adfebf46d8226094b7d5b90eeb04f0;p=ffmpeg diff --git a/libavcodec/alac.c b/libavcodec/alac.c index 3dd0ab9edb1..5c48a4b28f4 100644 --- a/libavcodec/alac.c +++ b/libavcodec/alac.c @@ -20,7 +20,7 @@ */ /** - * @file alac.c + * @file libavcodec/alac.c * ALAC (Apple Lossless Audio Codec) decoder * @author 2005 David Hammerton * @@ -53,8 +53,10 @@ #include "avcodec.h" -#include "bitstream.h" +#include "get_bits.h" #include "bytestream.h" +#include "unary.h" +#include "mathops.h" #define ALAC_EXTRADATA_SIZE 36 #define MAX_CHANNELS 2 @@ -67,7 +69,6 @@ typedef struct { * set this to 1 */ int context_initialized; - int samplesize; int numchannels; int bytespersample; @@ -78,16 +79,10 @@ typedef struct { /* stuff from setinfo */ uint32_t setinfo_max_samples_per_frame; /* 0x1000 = 4096 */ /* max samples per frame? */ - uint8_t setinfo_7a; /* 0x00 */ uint8_t setinfo_sample_size; /* 0x10 */ uint8_t setinfo_rice_historymult; /* 0x28 */ uint8_t setinfo_rice_initialhistory; /* 0x0a */ uint8_t setinfo_rice_kmodifier; /* 0x0e */ - uint8_t setinfo_7f; /* 0x02 */ - uint16_t setinfo_80; /* 0x00ff */ - uint32_t setinfo_82; /* 0x000020e7 */ /* max sample size?? */ - uint32_t setinfo_86; /* 0x00069fe4 */ /* bit rate (average)?? */ - uint32_t setinfo_8a_rate; /* 0x0000ac44 */ /* end setinfo stuff */ } ALACContext; @@ -106,7 +101,7 @@ static void allocate_buffers(ALACContext *alac) static int alac_set_info(ALACContext *alac) { - unsigned char *ptr = alac->avctx->extradata; + const unsigned char *ptr = alac->avctx->extradata; ptr += 4; /* size */ ptr += 4; /* alac */ @@ -119,37 +114,51 @@ static int alac_set_info(ALACContext *alac) /* buffer size / 2 ? */ alac->setinfo_max_samples_per_frame = bytestream_get_be32(&ptr); - alac->setinfo_7a = *ptr++; + ptr++; /* ??? */ alac->setinfo_sample_size = *ptr++; + if (alac->setinfo_sample_size > 32) { + av_log(alac->avctx, AV_LOG_ERROR, "setinfo_sample_size too large\n"); + return -1; + } alac->setinfo_rice_historymult = *ptr++; alac->setinfo_rice_initialhistory = *ptr++; alac->setinfo_rice_kmodifier = *ptr++; - /* channels? */ - alac->setinfo_7f = *ptr++; - alac->setinfo_80 = bytestream_get_be16(&ptr); - /* max coded frame size */ - alac->setinfo_82 = bytestream_get_be32(&ptr); - /* bitrate ? */ - alac->setinfo_86 = bytestream_get_be32(&ptr); - /* samplerate */ - alac->setinfo_8a_rate = bytestream_get_be32(&ptr); + ptr++; /* channels? */ + bytestream_get_be16(&ptr); /* ??? */ + bytestream_get_be32(&ptr); /* max coded frame size */ + bytestream_get_be32(&ptr); /* bitrate ? */ + bytestream_get_be32(&ptr); /* samplerate */ allocate_buffers(alac); return 0; } -/* hideously inefficient. could use a bitmask search, - * alternatively bsr on x86, - */ -static int count_leading_zeros(int32_t input) -{ - int i = 0; - while (!(0x80000000 & input) && i < 32) { - i++; - input = input << 1; +static inline int decode_scalar(GetBitContext *gb, int k, int limit, int readsamplesize){ + /* read x - number of 1s before 0 represent the rice */ + int x = get_unary_0_9(gb); + + if (x > 8) { /* RICE THRESHOLD */ + /* use alternative encoding */ + x = get_bits(gb, readsamplesize); + } else { + if (k >= limit) + k = limit; + + if (k != 1) { + int extrabits = show_bits(gb, k); + + /* multiply x by 2^k - 1, as part of their strange algorithm */ + x = (x << k) - x; + + if (extrabits > 1) { + x += extrabits - 1; + skip_bits(gb, k); + } else + skip_bits(gb, k - 1); + } } - return i; + return x; } static void bastardized_rice_decompress(ALACContext *alac, @@ -167,54 +176,16 @@ static void bastardized_rice_decompress(ALACContext *alac, int sign_modifier = 0; for (output_count = 0; output_count < output_size; output_count++) { - int32_t x = 0; + int32_t x; int32_t x_modified; int32_t final_val; - /* read x - number of 1s before 0 represent the rice */ - while (x <= 8 && get_bits1(&alac->gb)) { - x++; - } - - - if (x > 8) { /* RICE THRESHOLD */ - /* use alternative encoding */ - int32_t value; - - value = get_bits(&alac->gb, readsamplesize); - - /* mask value to readsamplesize size */ - if (readsamplesize != 32) - value &= (0xffffffff >> (32 - readsamplesize)); - - x = value; - } else { - /* standard rice encoding */ - int extrabits; - int k; /* size of extra bits */ - - /* read k, that is bits as is */ - k = 31 - rice_kmodifier - count_leading_zeros((history >> 9) + 3); + /* standard rice encoding */ + int k; /* size of extra bits */ - if (k < 0) - k += rice_kmodifier; - else - k = rice_kmodifier; - - if (k != 1) { - extrabits = show_bits(&alac->gb, k); - - /* multiply x by 2^k - 1, as part of their strange algorithm */ - x = (x << k) - x; - - if (extrabits > 1) { - x += extrabits - 1; - get_bits(&alac->gb, k); - } else { - get_bits(&alac->gb, k - 1); - } - } - } + /* read k, that is bits as is */ + k = av_log2((history >> 9) + 3); + x= decode_scalar(&alac->gb, k, rice_kmodifier, readsamplesize); x_modified = sign_modifier + x; final_val = (x_modified + 1) / 2; @@ -225,50 +196,30 @@ static void bastardized_rice_decompress(ALACContext *alac, sign_modifier = 0; /* now update the history */ - history += (x_modified * rice_historymult) - - ((history * rice_historymult) >> 9); + history += x_modified * rice_historymult + - ((history * rice_historymult) >> 9); if (x_modified > 0xffff) history = 0xffff; /* special case: there may be compressed blocks of 0 */ if ((history < 128) && (output_count+1 < output_size)) { - int block_size; + int k; + unsigned int block_size; sign_modifier = 1; - x = 0; - while (x <= 8 && get_bits1(&alac->gb)) { - x++; - } - - if (x > 8) { - block_size = get_bits(&alac->gb, 16); - block_size &= 0xffff; - } else { - int k; - int extrabits; + k = 7 - av_log2(history) + ((history + 16) >> 6 /* / 64 */); - k = count_leading_zeros(history) + ((history + 16) >> 6 /* / 64 */) - 24; - - extrabits = show_bits(&alac->gb, k); - - block_size = (((1 << k) - 1) & rice_kmodifier_mask) * x - + extrabits - 1; - - if (extrabits < 2) { - x = 1 - extrabits; - block_size += x; - get_bits(&alac->gb, k - 1); - } else { - get_bits(&alac->gb, k); - } - } + block_size= decode_scalar(&alac->gb, k, rice_kmodifier, 16); if (block_size > 0) { + if(block_size >= output_size - output_count){ + av_log(alac->avctx, AV_LOG_ERROR, "invalid zero block size of %d %d %d\n", block_size, output_size, output_count); + block_size= output_size - output_count - 1; + } memset(&output_buffer[output_count+1], 0, block_size * 4); output_count += block_size; - } if (block_size > 0xffff) @@ -279,12 +230,10 @@ static void bastardized_rice_decompress(ALACContext *alac, } } -#define SIGN_EXTENDED32(val, bits) ((val << (32 - bits)) >> (32 - bits)) - -#define SIGN_ONLY(v) \ - ((v < 0) ? (-1) : \ - ((v > 0) ? (1) : \ - (0))) +static inline int sign_only(int v) +{ + return v ? FFSIGN(v) : 0; +} static void predictor_decompress_fir_adapt(int32_t *error_buffer, int32_t *buffer_out, @@ -300,7 +249,9 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer, *buffer_out = *error_buffer; if (!predictor_coef_num) { - if (output_size <= 1) return; + if (output_size <= 1) + return; + memcpy(buffer_out+1, error_buffer+1, (output_size-1) * 4); return; } @@ -309,53 +260,48 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer, /* second-best case scenario for fir decompression, * error describes a small difference from the previous sample only */ - if (output_size <= 1) return; + if (output_size <= 1) + return; for (i = 0; i < output_size - 1; i++) { int32_t prev_value; int32_t error_value; prev_value = buffer_out[i]; error_value = error_buffer[i+1]; - buffer_out[i+1] = SIGN_EXTENDED32((prev_value + error_value), readsamplesize); + buffer_out[i+1] = + sign_extend((prev_value + error_value), readsamplesize); } return; } /* read warm-up samples */ - if (predictor_coef_num > 0) { - int i; + if (predictor_coef_num > 0) for (i = 0; i < predictor_coef_num; i++) { int32_t val; val = buffer_out[i] + error_buffer[i+1]; - - val = SIGN_EXTENDED32(val, readsamplesize); - + val = sign_extend(val, readsamplesize); buffer_out[i+1] = val; } - } #if 0 /* 4 and 8 are very common cases (the only ones i've seen). these - * should be unrolled and optimised + * should be unrolled and optimized */ if (predictor_coef_num == 4) { - /* FIXME: optimised general case */ + /* FIXME: optimized general case */ return; } if (predictor_coef_table == 8) { - /* FIXME: optimised general case */ + /* FIXME: optimized general case */ return; } #endif - /* general case */ if (predictor_coef_num > 0) { - for (i = predictor_coef_num + 1; - i < output_size; - i++) { + for (i = predictor_coef_num + 1; i < output_size; i++) { int j; int sum = 0; int outval; @@ -369,7 +315,7 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer, outval = (1 << (predictor_quantitization-1)) + sum; outval = outval >> predictor_quantitization; outval = outval + buffer_out[0] + error_val; - outval = SIGN_EXTENDED32(outval, readsamplesize); + outval = sign_extend(outval, readsamplesize); buffer_out[predictor_coef_num+1] = outval; @@ -378,7 +324,7 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer, while (predictor_num >= 0 && error_val > 0) { int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num]; - int sign = SIGN_ONLY(val); + int sign = sign_only(val); predictor_coef_table[predictor_num] -= sign; @@ -394,7 +340,7 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer, while (predictor_num >= 0 && error_val < 0) { int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num]; - int sign = - SIGN_ONLY(val); + int sign = - sign_only(val); predictor_coef_table[predictor_num] -= sign; @@ -412,32 +358,29 @@ static void predictor_decompress_fir_adapt(int32_t *error_buffer, } } -static void deinterlace_16(int32_t *buffer_a, int32_t *buffer_b, - int16_t *buffer_out, - int numchannels, int numsamples, - uint8_t interlacing_shift, - uint8_t interlacing_leftweight) +static void reconstruct_stereo_16(int32_t *buffer[MAX_CHANNELS], + int16_t *buffer_out, + int numchannels, int numsamples, + uint8_t interlacing_shift, + uint8_t interlacing_leftweight) { int i; - if (numsamples <= 0) return; + if (numsamples <= 0) + return; /* weighted interlacing */ if (interlacing_leftweight) { for (i = 0; i < numsamples; i++) { - int32_t difference, midright; - int16_t left; - int16_t right; + int32_t a, b; - midright = buffer_a[i]; - difference = buffer_b[i]; + a = buffer[0][i]; + b = buffer[1][i]; + a -= (b * interlacing_leftweight) >> interlacing_shift; + b += a; - right = midright - ((difference * interlacing_leftweight) >> interlacing_shift); - left = (midright - ((difference * interlacing_leftweight) >> interlacing_shift)) - + difference; - - buffer_out[i*numchannels] = left; - buffer_out[i*numchannels + 1] = right; + buffer_out[i*numchannels] = b; + buffer_out[i*numchannels + 1] = a; } return; @@ -447,8 +390,8 @@ static void deinterlace_16(int32_t *buffer_a, int32_t *buffer_b, for (i = 0; i < numsamples; i++) { int16_t left, right; - left = buffer_a[i]; - right = buffer_b[i]; + left = buffer[0][i]; + right = buffer[1][i]; buffer_out[i*numchannels] = left; buffer_out[i*numchannels + 1] = right; @@ -457,14 +400,16 @@ static void deinterlace_16(int32_t *buffer_a, int32_t *buffer_b, static int alac_decode_frame(AVCodecContext *avctx, void *outbuffer, int *outputsize, - uint8_t *inbuffer, int input_buffer_size) + AVPacket *avpkt) { + const uint8_t *inbuffer = avpkt->data; + int input_buffer_size = avpkt->size; ALACContext *alac = avctx->priv_data; int channels; - int32_t outputsamples; + unsigned int outputsamples; int hassize; - int readsamplesize; + unsigned int readsamplesize; int wasted_bytes; int isnotcompressed; uint8_t interlacing_shift; @@ -491,44 +436,62 @@ static int alac_decode_frame(AVCodecContext *avctx, init_get_bits(&alac->gb, inbuffer, input_buffer_size * 8); channels = get_bits(&alac->gb, 3) + 1; + if (channels > MAX_CHANNELS) { + av_log(avctx, AV_LOG_ERROR, "channels > %d not supported\n", + MAX_CHANNELS); + return input_buffer_size; + } - /* 2^result = something to do with output waiting. - * perhaps matters if we read > 1 frame in a pass? - */ - get_bits(&alac->gb, 4); + /* 2^result = something to do with output waiting. + * perhaps matters if we read > 1 frame in a pass? + */ + skip_bits(&alac->gb, 4); - get_bits(&alac->gb, 12); /* unknown, skip 12 bits */ + skip_bits(&alac->gb, 12); /* unknown, skip 12 bits */ - hassize = get_bits(&alac->gb, 1); /* the output sample size is stored soon */ + /* the output sample size is stored soon */ + hassize = get_bits1(&alac->gb); - wasted_bytes = get_bits(&alac->gb, 2); /* unknown ? */ + wasted_bytes = get_bits(&alac->gb, 2); /* unknown ? */ - isnotcompressed = get_bits(&alac->gb, 1); /* whether the frame is compressed */ + /* whether the frame is compressed */ + isnotcompressed = get_bits1(&alac->gb); - if (hassize) { - /* now read the number of samples, - * as a 32bit integer */ - outputsamples = get_bits(&alac->gb, 32); - } else - outputsamples = alac->setinfo_max_samples_per_frame; + if (hassize) { + /* now read the number of samples as a 32bit integer */ + outputsamples = get_bits_long(&alac->gb, 32); + if(outputsamples > alac->setinfo_max_samples_per_frame){ + av_log(avctx, AV_LOG_ERROR, "outputsamples %d > %d\n", outputsamples, alac->setinfo_max_samples_per_frame); + return -1; + } + } else + outputsamples = alac->setinfo_max_samples_per_frame; - *outputsize = outputsamples * alac->bytespersample; - readsamplesize = alac->setinfo_sample_size - (wasted_bytes * 8) + channels - 1; + if(outputsamples > *outputsize / alac->bytespersample){ + av_log(avctx, AV_LOG_ERROR, "sample buffer too small\n"); + return -1; + } - if (!isnotcompressed) { - /* so it is compressed */ - int16_t predictor_coef_table[channels][32]; - int predictor_coef_num[channels]; - int prediction_type[channels]; - int prediction_quantitization[channels]; - int ricemodifier[channels]; + *outputsize = outputsamples * alac->bytespersample; + readsamplesize = alac->setinfo_sample_size - (wasted_bytes * 8) + channels - 1; + if (readsamplesize > MIN_CACHE_BITS) { + av_log(avctx, AV_LOG_ERROR, "readsamplesize too big (%d)\n", readsamplesize); + return -1; + } - int i, chan; + if (!isnotcompressed) { + /* so it is compressed */ + int16_t predictor_coef_table[channels][32]; + int predictor_coef_num[channels]; + int prediction_type[channels]; + int prediction_quantitization[channels]; + int ricemodifier[channels]; + int i, chan; - interlacing_shift = get_bits(&alac->gb, 8); - interlacing_leftweight = get_bits(&alac->gb, 8); + interlacing_shift = get_bits(&alac->gb, 8); + interlacing_leftweight = get_bits(&alac->gb, 8); - for (chan = 0; chan < channels; chan++) { + for (chan = 0; chan < channels; chan++) { prediction_type[chan] = get_bits(&alac->gb, 4); prediction_quantitization[chan] = get_bits(&alac->gb, 4); @@ -536,16 +499,14 @@ static int alac_decode_frame(AVCodecContext *avctx, predictor_coef_num[chan] = get_bits(&alac->gb, 5); /* read the predictor table */ - for (i = 0; i < predictor_coef_num[chan]; i++) { + for (i = 0; i < predictor_coef_num[chan]; i++) predictor_coef_table[chan][i] = (int16_t)get_bits(&alac->gb, 16); - } - } + } - if (wasted_bytes) { - av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented, unhandling of wasted_bytes\n"); - } + if (wasted_bytes) + av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented, unhandling of wasted_bytes\n"); - for (chan = 0; chan < channels; chan++) { + for (chan = 0; chan < channels; chan++) { bastardized_rice_decompress(alac, alac->predicterror_buffer[chan], outputsamples, @@ -556,7 +517,7 @@ static int alac_decode_frame(AVCodecContext *avctx, (1 << alac->setinfo_rice_kmodifier) - 1); if (prediction_type[chan] == 0) { - /* adaptive fir */ + /* adaptive fir */ predictor_decompress_fir_adapt(alac->predicterror_buffer[chan], alac->outputsamples_buffer[chan], outputsamples, @@ -566,96 +527,80 @@ static int alac_decode_frame(AVCodecContext *avctx, prediction_quantitization[chan]); } else { av_log(avctx, AV_LOG_ERROR, "FIXME: unhandled prediction type: %i\n", prediction_type[chan]); - /* i think the only other prediction type (or perhaps this is just a - * boolean?) runs adaptive fir twice.. like: + /* I think the only other prediction type (or perhaps this is + * just a boolean?) runs adaptive fir twice.. like: * predictor_decompress_fir_adapt(predictor_error, tempout, ...) * predictor_decompress_fir_adapt(predictor_error, outputsamples ...) * little strange.. */ } - } - } else { - /* not compressed, easy case */ - if (alac->setinfo_sample_size <= 16) { - int i, chan; - for (chan = 0; chan < channels; chan++) { - for (i = 0; i < outputsamples; i++) { - int32_t audiobits; + } + } else { + /* not compressed, easy case */ + int i, chan; + for (i = 0; i < outputsamples; i++) + for (chan = 0; chan < channels; chan++) { + int32_t audiobits; - audiobits = get_bits(&alac->gb, alac->setinfo_sample_size); - audiobits = SIGN_EXTENDED32(audiobits, readsamplesize); + audiobits = get_sbits_long(&alac->gb, alac->setinfo_sample_size); - alac->outputsamples_buffer[chan][i] = audiobits; - } - } - } else { - int i, chan; - for (chan = 0; chan < channels; chan++) { - for (i = 0; i < outputsamples; i++) { - int32_t audiobits; - - audiobits = get_bits(&alac->gb, 16); - /* special case of sign extension.. - * as we'll be ORing the low 16bits into this */ - audiobits = audiobits << 16; - audiobits = audiobits >> (32 - alac->setinfo_sample_size); - audiobits |= get_bits(&alac->gb, alac->setinfo_sample_size - 16); - - alac->outputsamples_buffer[chan][i] = audiobits; - } - } + alac->outputsamples_buffer[chan][i] = audiobits; + } + /* wasted_bytes = 0; */ + interlacing_shift = 0; + interlacing_leftweight = 0; + } + if (get_bits(&alac->gb, 3) != 7) + av_log(avctx, AV_LOG_ERROR, "Error : Wrong End Of Frame\n"); + + switch(alac->setinfo_sample_size) { + case 16: + if (channels == 2) { + reconstruct_stereo_16(alac->outputsamples_buffer, + (int16_t*)outbuffer, + alac->numchannels, + outputsamples, + interlacing_shift, + interlacing_leftweight); + } else { + int i; + for (i = 0; i < outputsamples; i++) { + int16_t sample = alac->outputsamples_buffer[0][i]; + ((int16_t*)outbuffer)[i * alac->numchannels] = sample; } - /* wasted_bytes = 0; */ - interlacing_shift = 0; - interlacing_leftweight = 0; - } - - switch(alac->setinfo_sample_size) { - case 16: { - if (channels == 2) { - deinterlace_16(alac->outputsamples_buffer[0], - alac->outputsamples_buffer[1], - (int16_t*)outbuffer, - alac->numchannels, - outputsamples, - interlacing_shift, - interlacing_leftweight); - } else { - int i; - for (i = 0; i < outputsamples; i++) { - int16_t sample = alac->outputsamples_buffer[0][i]; - ((int16_t*)outbuffer)[i * alac->numchannels] = sample; - } - } - break; - } - case 20: - case 24: - case 32: - av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size); - break; - default: - break; } + break; + case 20: + case 24: + // It is not clear if there exist any encoder that creates 24 bit ALAC + // files. iTunes convert 24 bit raw files to 16 bit before encoding. + case 32: + av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size); + break; + default: + break; + } + if (input_buffer_size * 8 - get_bits_count(&alac->gb) > 8) + av_log(avctx, AV_LOG_ERROR, "Error : %d bits left\n", input_buffer_size * 8 - get_bits_count(&alac->gb)); return input_buffer_size; } -static int alac_decode_init(AVCodecContext * avctx) +static av_cold int alac_decode_init(AVCodecContext * avctx) { ALACContext *alac = avctx->priv_data; alac->avctx = avctx; alac->context_initialized = 0; - alac->samplesize = alac->avctx->bits_per_sample; alac->numchannels = alac->avctx->channels; - alac->bytespersample = (alac->samplesize / 8) * alac->numchannels; + alac->bytespersample = 2 * alac->numchannels; + avctx->sample_fmt = SAMPLE_FMT_S16; return 0; } -static int alac_decode_close(AVCodecContext *avctx) +static av_cold int alac_decode_close(AVCodecContext *avctx) { ALACContext *alac = avctx->priv_data; @@ -677,4 +622,5 @@ AVCodec alac_decoder = { NULL, alac_decode_close, alac_decode_frame, + .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"), };