[FFmpeg-devel] [PATCH] ALS decoder
Michael Niedermayer
michaelni
Sun Aug 23 20:29:54 CEST 2009
On Sun, Aug 23, 2009 at 02:00:22PM +0200, Thilo Borgmann wrote:
> Revision 6 attached.
>
> Depends on ceillog2.rev1.patch.
[...]
> +/** Reads an ALSSpecificConfig from a buffer into the output struct.
> + */
> +static av_cold int read_specific_config(ALSDecContext *ctx)
> +{
> + GetBitContext gb;
> + uint64_t ht_size;
> + int i, config_offset, crc_enabled;
> + MPEG4AudioConfig m4ac;
> + ALSSpecificConfig *sconf = &ctx->sconf;
> + AVCodecContext *avctx = ctx->avctx;
> + const uint8_t *buffer = avctx->extradata;
> + int buffer_size = avctx->extradata_size;
> + uint32_t samples;
> +
> + init_get_bits(&gb, buffer, buffer_size * 8);
> +
> + config_offset = ff_mpeg4audio_get_config(&m4ac, buffer, buffer_size);
> +
> + if (config_offset < 0)
> + return -1;
> +
> + skip_bits_long(&gb, config_offset);
> + buffer_size -= config_offset >> 3;
> +
> + if (buffer_size < 22)
> + return -1;
> +
> + // read the fixed items
> + sconf->als_id = get_bits_long(&gb, 32);
> + avctx->sample_rate = m4ac.sample_rate;
> + skip_bits_long(&gb, 32); // sample rate already known
> + samples = get_bits_long(&gb, 32);
> + avctx->channels = m4ac.channels;
> + skip_bits(&gb, 16); // number of channels already knwon
> + skip_bits(&gb, 3); // skip file_type
> + sconf->resolution = get_bits(&gb, 3);
> + sconf->floating = get_bits1(&gb);
> + sconf->msb_first = get_bits1(&gb);
> + sconf->frame_length = get_bits(&gb, 16) + 1;
> + sconf->ra_distance = get_bits(&gb, 8);
> + sconf->ra_flag = get_bits(&gb, 2);
> + sconf->adapt_order = get_bits1(&gb);
> + sconf->coef_table = get_bits(&gb, 2);
> + sconf->long_term_prediction = get_bits1(&gb);
> + sconf->max_order = get_bits(&gb, 10);
> + sconf->block_switching = get_bits(&gb, 2);
> + sconf->bgmc = get_bits1(&gb);
> + sconf->sb_part = get_bits1(&gb);
> + sconf->joint_stereo = get_bits1(&gb);
> + sconf->mc_coding = get_bits1(&gb);
> + sconf->chan_config = get_bits1(&gb);
> + sconf->chan_sort = get_bits1(&gb);
> + crc_enabled = get_bits1(&gb);
> + sconf->rlslms = get_bits1(&gb);
> + skip_bits(&gb, 5); // skip 5 reserved bits
> + sconf->aux_data_enabled = get_bits1(&gb);
> + buffer_size -= 22;
> +
> +
> + // check for ALSSpecificConfig struct
> + if (sconf->als_id != MKBETAG('A','L','S','\0'))
> + return -1;
> +
> + ctx->cur_frame_length = sconf->frame_length;
> +
> + // allocate quantized parcor coefficient buffer
> + if (!(ctx->quant_cof = av_malloc(sizeof(int64_t) * sconf->max_order)) ||
> + !(ctx->lpc_cof = av_malloc(sizeof(int64_t) * sconf->max_order))) {
sizeof(*ctx->lpc_cof)
[...]
> +
> +/** Parses the bs_info field to extract the block partitioning used in block switching mode,
> + * refer to ISO/IEC 14496-3, section 11.6.2.
> + */
> +static void parse_bs_info(uint32_t bs_info, unsigned int n, unsigned int div,
> + unsigned int **div_blocks, unsigned int *num_blocks)
> +{
> + if (n < 31 && ((bs_info >> (30 - n)) & 1)) {
> + // if the level is valid and the investigated bit n is set
> + // then recursively check both children at bits (2n+1) and (2n+2)
> + n *= 2;
> + div += 1;
this sounds like it could use get_bits()
> + parse_bs_info(bs_info, n + 1, div, div_blocks, num_blocks);
> + parse_bs_info(bs_info, n + 2, div, div_blocks, num_blocks);
> + } else {
> + // else the bit is not set or the last level has been reached
> + // (bit implicitly not set)
> + **div_blocks = div;
> + (*div_blocks)++;
> + (*num_blocks)++;
> + }
> +}
> +
> +
> +/** Reads and decodes a Rice codeword.
> + */
> +static int64_t decode_rice(GetBitContext *gb, unsigned int k)
> +{
> + int max = gb->size_in_bits - get_bits_count(gb) - k;
> +
> + if (k > 1) {
[...]
> + q = get_unary(gb, 0, max);
[...]
> + } else {
> + int q;
> + q = get_unary(gb, 0, max);
[...]
please factorize this out, also please factorize anything else out
if there is something that can be factored out
> +/** Converts PARCOR coefficient k to direct filter coefficient.
> + */
> +static void parcor_to_lpc(unsigned int k, int64_t *par, int64_t *cof)
> +{
> + int i;
> +
> + for (i = 0; i < (k+1) >> 1; i++) {
> + int32_t tmp1, tmp2;
> +
> + tmp1 = cof[ i ] + ((par[k] * cof[k - i - 1] + (1 << 19)) >> 20);
> + tmp2 = cof[k - i - 1] + ((par[k] * cof[ i ] + (1 << 19)) >> 20);
> + cof[k - i - 1] = tmp2;
> + cof[ i ] = tmp1;
as the intermediate tmp* are 32bit so likely can be cof, making
multiplications faster
> + }
> +
> + cof[k] = par[k];
> +}
> +
> +
> +/** Reformat block sizes from log2 format to direct form. Also assure that the
> + * block sizes of the last frame correspond to the actual number of samples.
> + */
> +static void reconstruct_block_sizes(ALSDecContext *ctx, uint32_t *div_blocks)
> +{
> + unsigned int b;
> +
> + // The last frame may have an overdetermined block structure given in
> + // the bitstream. In that case the defined block structure would need
> + // more samples than available to be consistent.
> + // The block structure is actually used but the block sizes are adapted
> + // to fit the actual number of available samples.
> + // Example: 5 samples, 2nd level block sizes: 2 2 2 2.
> + // This results in the actual block sizes: 2 2 1 0.
> + // This is not specified in 14496-3 but actually done by the reference
> + // codec RM22 revision 2.
> + // This appears to happen in case of an odd number of samples in the last
> + // frame which is actually not allowed by the block length switching part
> + // of 14496-3.
> + // The ALS conformance files feature an odd number of samples in the last
> + // frame.
> + if (ctx->cur_frame_length == ctx->last_frame_length) {
> + unsigned int remaining = ctx->cur_frame_length;
> +
> + for (b = 0; b < ctx->num_blocks; b++) {
> + div_blocks[b] = ctx->sconf.frame_length >> div_blocks[b];
[...]
> + }
> + } else {
> + for (b = 0; b < ctx->num_blocks; b++)
> + div_blocks[b] = ctx->sconf.frame_length >> div_blocks[b];
duplicate
[...]
> +/** Reads the block data for a non-constant block
> + */
> +static int read_var_block(ALSDecContext *ctx, unsigned int ra_block,
> + int64_t *raw_samples, unsigned int block_length,
> + unsigned int *js_blocks, int64_t *raw_other,
> + unsigned int *shift_lsbs)
> +{
> + ALSSpecificConfig *sconf = &ctx->sconf;
> + AVCodecContext *avctx = ctx->avctx;
> + GetBitContext *gb = &ctx->gb;
> + unsigned int k;
> + unsigned int s[8];
> + unsigned int sub_blocks, sb_length;
> + unsigned int opt_order = 1;
> + int64_t *quant_cof = ctx->quant_cof;
> + int64_t *lpc_cof = ctx->lpc_cof;
> + unsigned int start = 0;
> + int sb, smp;
> + int64_t y;
> +
> + *js_blocks = get_bits1(gb);
> +
> + // determine the number of sub blocks for entropy decoding
> + if (!sconf->bgmc && !sconf->sb_part)
> + sub_blocks = 1;
> + else if (sconf->bgmc && sconf->sb_part)
> + sub_blocks = 1 << get_bits(gb, 2);
> + else
> + sub_blocks = 1 << (2 * get_bits1(gb));
> +
> + // do not continue in case of a damaged stream since
> + // block_length must be evenly divisible by sub_blocks
> + if (block_length % sub_blocks) {
> + av_log(avctx, AV_LOG_WARNING,
> + "Block length is not evenly divisible by the number of sub blocks.\n");
> + return -1;
> + }
> +
> + sb_length = block_length / sub_blocks;
> +
> +
> + if (sconf->bgmc) {
> + // TODO: BGMC mode
> + } else {
> + s[0] = get_bits(gb, (sconf->resolution > 1) ? 5 : 4);
> + for (k = 1; k < sub_blocks; k++)
> + s[k] = s[k - 1] + decode_rice(gb, 0);
> + }
> +
> + if (get_bits1(gb)) {
> + *shift_lsbs = get_bits(gb, 4) + 1;
> + }
> +
> +
> + if (!sconf->rlslms) {
> + int64_t quant_index;
> +
> + if (sconf->adapt_order) {
> + int opt_order_length =
> + FFMIN(
> + av_ceil_log2(sconf->max_order+1),
> + FFMAX(av_ceil_log2((block_length >> 3) - 1), 1)
> + );
> + opt_order = get_bits(gb, opt_order_length);
> + } else {
> + opt_order = sconf->max_order;
> + }
> +
> + if (opt_order) {
> + if (sconf->coef_table == 3) {
> +
> + // read coefficient 0
> + quant_index = get_bits(gb, 7);
> + quant_cof[0] = parcor_scaled_values[quant_index];;
> +
> + // read coefficient 1
> + quant_index = get_bits(gb, 7);
> + quant_cof[1] = -parcor_scaled_values[quant_index];
> +
> + // read coefficients 2 to opt_order
> + for (k = 2; k < opt_order; k++) {
> + quant_index = get_bits(gb, 7);
> + quant_cof[k] = (quant_index << 14) - (0x7F << 13);
> + }
> + } else {
> + int offset, rice_param, k_max;
> +
> + // read coefficient 0
> + offset = parcor_rice_table[sconf->coef_table][0][0];
> + rice_param = parcor_rice_table[sconf->coef_table][0][1];
> + quant_index = decode_rice(gb, rice_param) + offset;
> + quant_cof[0] = parcor_scaled_values[quant_index + 64];
> +
> + // read coefficient 1
> + offset = parcor_rice_table[sconf->coef_table][1][0];
> + rice_param = parcor_rice_table[sconf->coef_table][1][1];
> + quant_index = decode_rice(gb, rice_param) + offset;
> + quant_cof[1] = -parcor_scaled_values[quant_index + 64];
> +
> + // read coefficients 2 to 19
> + k_max = FFMIN(20, opt_order);
> + for (k = 2; k < k_max; k++) {
> + offset = parcor_rice_table[sconf->coef_table][k][0];
> + rice_param = parcor_rice_table[sconf->coef_table][k][1];
> + quant_index = decode_rice(gb, rice_param) + offset;
> + quant_cof[k] = (quant_index << 14) + (1 << 13);
> + }
the 3 first lines of these 3 blocks of code are duplicated and can be factorized
[...]
> +/** Reads the block data.
> + */
> +static int read_block_data(ALSDecContext *ctx, unsigned int ra_block,
> + int64_t *raw_samples, unsigned int block_length,
> + unsigned int *js_blocks, int64_t *raw_other)
> +{
> + ALSSpecificConfig *sconf = &ctx->sconf;
> + GetBitContext *gb = &ctx->gb;
> + unsigned int shift_lsbs = 0;
> + unsigned int block_type;
> + unsigned int k;
> +
> + block_type = get_bits1(gb);
> +
> + if (block_type == 0) {
the temporary variable block_type is unneeded
> + read_const_block(ctx, raw_samples, block_length, js_blocks);
> + } else {
> + if (read_var_block(ctx, ra_block, raw_samples, block_length, js_blocks,
> + raw_other, &shift_lsbs))
> + return -1;
> + }
> +
> + if (sconf->rlslms) {
> + // TODO: read RLSLMS extension data
> + }
> +
> + if (!sconf->mc_coding || ctx->js_switch) {
> + align_get_bits(gb);
> + }
> +
> + if (shift_lsbs) {
> + for (k = 0; k < block_length; k++)
> + raw_samples[k] <<= shift_lsbs;
> + }
> +
> + return 0;
> +}
> +
> +
> +/** Decodes blocks independently.
> + */
> +static int decode_blocks_ind(ALSDecContext *ctx, unsigned int ra_frame,
> + unsigned int c, unsigned int *div_blocks,
> + unsigned int *js_blocks)
> +{
> + ALSSpecificConfig *sconf = &ctx->sconf;
> + int64_t *raw_sample;
> + unsigned int b, ra_block;
> + raw_sample = ctx->raw_samples[c];
> +
> + for (b = 0; b < ctx->num_blocks; b++) {
> + ra_block = !b && ra_frame;
id use ra_frame rename it to ra_block and set it to 0 at the end of the loop
this simplification can likely be done for more code
> + if (read_block_data(ctx, ra_block, raw_sample,
> + div_blocks[b], &js_blocks[0], NULL)) {
> + // damaged block, write zero for the rest of the frame
> + while (b < ctx->num_blocks) {
> + memset(raw_sample, 0, div_blocks[b]);
> + raw_sample += div_blocks[b];
> + b++;
> + }
> + return -1;
> + }
> + raw_sample += div_blocks[b];
> + }
> +
> + // store carryover raw samples
> + memmove((ctx->raw_samples[c]) - sconf->max_order,
> + (ctx->raw_samples[c]) - sconf->max_order + sconf->frame_length,
> + sizeof(int64_t) * sconf->max_order);
sizeof(an entry of the array) is much better because if the type changes
you donz have to update all that code
> +
> + return 0;
> +}
> +
> +
> +/** Decodes blocks dependently.
> + */
> +static int decode_blocks(ALSDecContext *ctx, unsigned int ra_frame,
> + unsigned int c, unsigned int *div_blocks,
> + unsigned int *js_blocks)
> +{
> + ALSSpecificConfig *sconf = &ctx->sconf;
> + unsigned int offset = 0;
> + int64_t *raw_samples_R;
> + int64_t *raw_samples_L;
> + unsigned int b, ra_block;
> +
> + // decode all blocks
> + for (b = 0; b < ctx->num_blocks; b++) {
> + unsigned int s;
> + raw_samples_L = ctx->raw_samples[c ] + offset;
> + raw_samples_R = ctx->raw_samples[c + 1] + offset;
> + ra_block = !b && ra_frame;
> + if (read_block_data(ctx, ra_block, raw_samples_L, div_blocks[b],
> + &js_blocks[0], raw_samples_R) ||
> + read_block_data(ctx, ra_block, raw_samples_R, div_blocks[b],
> + &js_blocks[1], raw_samples_L)) {
> + // damaged block, write zero for the rest of the frame
> + while (b < ctx->num_blocks) {
> + memset(raw_samples_L, 0, div_blocks[b]);
> + memset(raw_samples_R, 0, div_blocks[b]);
> + raw_samples_L += div_blocks[b];
> + raw_samples_R += div_blocks[b];
> + b++;
> + }
> + return -1;
> + }
> +
> + // reconstruct joint-stereo blocks
> + if (js_blocks[0]) {
> + if (js_blocks[1])
> + av_log(ctx->avctx, AV_LOG_WARNING, "Invalid channel pair!\n");
> +
> + for (s = 0; s < div_blocks[b]; s++)
> + raw_samples_L[s] = raw_samples_R[s] - raw_samples_L[s];
> + } else if (js_blocks[1]) {
> + for (s = 0; s < div_blocks[b]; s++)
> + raw_samples_R[s] = raw_samples_R[s] + raw_samples_L[s];
> + }
> +
> + offset += div_blocks[b];
> + }
> +
> + // store carryover raw samples
> + memmove((ctx->raw_samples[c]) - sconf->max_order,
> + (ctx->raw_samples[c]) - sconf->max_order + sconf->frame_length,
> + sizeof(int64_t) * sconf->max_order);
> +
> + memmove((ctx->raw_samples[c + 1]) - sconf->max_order,
> + (ctx->raw_samples[c + 1]) - sconf->max_order + sconf->frame_length,
> + sizeof(int64_t) * sconf->max_order);
> +
> + return 0;
> +}
this looks similar to decode_blocks_ind() i guess some parts could be
factored
[...]
> +/** Initializes the ALS decoder.
> + */
> +static av_cold int decode_init(AVCodecContext *avctx)
> +{
> + unsigned int c;
> + unsigned int channel_size;
> + ALSDecContext *ctx = avctx->priv_data;
> + ALSSpecificConfig *sconf = &ctx->sconf;
> + ctx->avctx = avctx;
> +
> + if (!avctx->extradata) {
> + av_log(avctx, AV_LOG_ERROR, "Missing required ALS extradata.\n");
> + return -1;
> + }
> +
> + if (read_specific_config(ctx)) {
> + av_log(avctx, AV_LOG_ERROR, "Reading ALSSpecificConfig failed.\n");
> + decode_end(avctx);
> + return -1;
> + }
> +
> + if (check_specific_config(ctx)) {
> + decode_end(avctx);
> + return -1;
> + }
> +
> + if (sconf->floating) {
> + avctx->sample_fmt = SAMPLE_FMT_FLT;
> + avctx->bits_per_raw_sample = 32;
why is this not set to 24 for simplifying that if(), i dont think
bits_per_raw_sample has a meaning currently for floats but maybe i
forgot something
[...]
--
Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
Asymptotically faster algorithms should always be preferred if you have
asymptotical amounts of data
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20090823/651897c3/attachment.pgp>
More information about the ffmpeg-devel
mailing list