X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Ftwinvq.c;h=940def45f579298fd56d609898ed28f8dcd3acf8;hb=851960f6f8cf1f946fe42fa36cf6598fac68072c;hp=7b56140653e8a3b59d4ba33d11d0c619d4513de0;hpb=6c145ecf785dc3d26ba3fed3ea9892cc80244625;p=ffmpeg diff --git a/libavcodec/twinvq.c b/libavcodec/twinvq.c index 7b56140653e..940def45f57 100644 --- a/libavcodec/twinvq.c +++ b/libavcodec/twinvq.c @@ -25,212 +25,17 @@ #include "libavutil/channel_layout.h" #include "libavutil/float_dsp.h" #include "avcodec.h" -#include "get_bits.h" #include "fft.h" #include "internal.h" #include "lsp.h" #include "sinewin.h" -#include "twinvq_data.h" - -enum FrameType { - FT_SHORT = 0, ///< Short frame (divided in n sub-blocks) - FT_MEDIUM, ///< Medium frame (divided in mmtab; - int size_s = mtab->size / mtab->fmode[FT_SHORT].sub; + const TwinVQModeTab *mtab = tctx->mtab; + int size_s = mtab->size / mtab->fmode[TWINVQ_FT_SHORT].sub; for (i = 0; i < size_s / 2; i++) { float cos_i = tctx->cos_tabs[0][i]; @@ -302,19 +107,19 @@ static inline float get_cos(int idx, int part, const float *cos_tab, int size) * unexplained condition. * * @param step the size of a block "siiiibiiii" - * @param in the cosinus of the LSP data - * @param part is 0 for 0...PI (positive cossinus values) and 1 for PI...2PI - * (negative cossinus values) + * @param in the cosine of the LSP data + * @param part is 0 for 0...PI (positive cosine values) and 1 for PI...2PI + * (negative cosine values) * @param size the size of the whole output */ -static inline void eval_lpcenv_or_interp(TwinContext *tctx, - enum FrameType ftype, +static inline void eval_lpcenv_or_interp(TwinVQContext *tctx, + enum TwinVQFrameType ftype, float *out, const float *in, int size, int step, int part) { int i; - const ModeTab *mtab = tctx->mtab; - const float *cos_tab = tctx->cos_tabs[ftype]; + const TwinVQModeTab *mtab = tctx->mtab; + const float *cos_tab = tctx->cos_tabs[ftype]; // Fill the 's' for (i = 0; i < size; i += step) @@ -344,7 +149,7 @@ static inline void eval_lpcenv_or_interp(TwinContext *tctx, out[size - 2 * step], step - 1); } -static void eval_lpcenv_2parts(TwinContext *tctx, enum FrameType ftype, +static void eval_lpcenv_2parts(TwinVQContext *tctx, enum TwinVQFrameType ftype, const float *buf, float *lpc, int size, int step) { @@ -355,7 +160,8 @@ static void eval_lpcenv_2parts(TwinContext *tctx, enum FrameType ftype, interpolate(lpc + size / 2 - step + 1, lpc[size / 2], lpc[size / 2 - step], step); - memset_float(lpc + size - 2 * step + 1, lpc[size - 2 * step], 2 * step - 1); + twinvq_memset_float(lpc + size - 2 * step + 1, lpc[size - 2 * step], + 2 * step - 1); } /** @@ -363,8 +169,8 @@ static void eval_lpcenv_2parts(TwinContext *tctx, enum FrameType ftype, * bitstream, sum the corresponding vectors and write the result to *out * after permutation. */ -static void dequant(TwinContext *tctx, GetBitContext *gb, float *out, - enum FrameType ftype, +static void dequant(TwinVQContext *tctx, const uint8_t *cb_bits, float *out, + enum TwinVQFrameType ftype, const int16_t *cb0, const int16_t *cb1, int cb_len) { int pos = 0; @@ -379,22 +185,20 @@ static void dequant(TwinContext *tctx, GetBitContext *gb, float *out, int bitstream_second_part = (i >= tctx->bits_main_spec_change[ftype]); int bits = tctx->bits_main_spec[0][ftype][bitstream_second_part]; + tmp0 = *cb_bits++; if (bits == 7) { - if (get_bits1(gb)) + if (tmp0 & 0x40) sign0 = -1; - bits = 6; + tmp0 &= 0x3F; } - tmp0 = get_bits(gb, bits); bits = tctx->bits_main_spec[1][ftype][bitstream_second_part]; - + tmp1 = *cb_bits++; if (bits == 7) { - if (get_bits1(gb)) + if (tmp1 & 0x40) sign1 = -1; - - bits = 6; + tmp1 &= 0x3F; } - tmp1 = get_bits(gb, bits); tab0 = cb0 + tmp0 * cb_len; tab1 = cb1 + tmp1 * cb_len; @@ -407,129 +211,32 @@ static void dequant(TwinContext *tctx, GetBitContext *gb, float *out, } } -static inline float mulawinv(float y, float clip, float mu) -{ - y = av_clipf(y / clip, -1, 1); - return clip * FFSIGN(y) * (exp(log(1 + mu) * fabs(y)) - 1) / mu; -} - -/** - * Evaluate a * b / 400 rounded to the nearest integer. When, for example, - * a * b == 200 and the nearest integer is ill-defined, use a table to emulate - * the following broken float-based implementation used by the binary decoder: - * - * @code - * static int very_broken_op(int a, int b) - * { - * static float test; // Ugh, force gcc to do the division first... - * - * test = a / 400.; - * return b * test + 0.5; - * } - * @endcode - * - * @note if this function is replaced by just ROUNDED_DIV(a * b, 400.), the - * stddev between the original file (before encoding with Yamaha encoder) and - * the decoded output increases, which leads one to believe that the encoder - * expects exactly this broken calculation. - */ -static int very_broken_op(int a, int b) -{ - int x = a * b + 200; - int size; - const uint8_t *rtab; - - if (x % 400 || b % 5) - return x / 400; - - x /= 400; - - size = tabs[b / 5].size; - rtab = tabs[b / 5].tab; - return x - rtab[size * av_log2(2 * (x - 1) / size) + (x - 1) % size]; -} - -/** - * Sum to data a periodic peak of a given period, width and shape. - * - * @param period the period of the peak divised by 400.0 - */ -static void add_peak(int period, int width, const float *shape, - float ppc_gain, float *speech, int len) +static void dec_gain(TwinVQContext *tctx, + enum TwinVQFrameType ftype, float *out) { - int i, j; - - const float *shape_end = shape + len; - int center; - - // First peak centered around zero - for (i = 0; i < width / 2; i++) - speech[i] += ppc_gain * *shape++; - - for (i = 1; i < ROUNDED_DIV(len, width); i++) { - center = very_broken_op(period, i); - for (j = -width / 2; j < (width + 1) / 2; j++) - speech[j + center] += ppc_gain * *shape++; - } - - // For the last block, be careful not to go beyond the end of the buffer - center = very_broken_op(period, i); - for (j = -width / 2; j < (width + 1) / 2 && shape < shape_end; j++) - speech[j + center] += ppc_gain * *shape++; -} - -static void decode_ppc(TwinContext *tctx, int period_coef, const float *shape, - float ppc_gain, float *speech) -{ - const ModeTab *mtab = tctx->mtab; - int isampf = tctx->avctx->sample_rate / 1000; - int ibps = tctx->avctx->bit_rate / (1000 * tctx->avctx->channels); - int min_period = ROUNDED_DIV(40 * 2 * mtab->size, isampf); - int max_period = ROUNDED_DIV(40 * 2 * mtab->size * 6, isampf); - int period_range = max_period - min_period; - - // This is actually the period multiplied by 400. It is just linearly coded - // between its maximum and minimum value. - int period = min_period + - ROUNDED_DIV(period_coef * period_range, - (1 << mtab->ppc_period_bit) - 1); - int width; - - if (isampf == 22 && ibps == 32) { - // For some unknown reason, NTT decided to code this case differently... - width = ROUNDED_DIV((period + 800) * mtab->peak_per2wid, - 400 * mtab->size); - } else - width = period * mtab->peak_per2wid / (400 * mtab->size); - - add_peak(period, width, shape, ppc_gain, speech, mtab->ppc_shape_len); -} - -static void dec_gain(TwinContext *tctx, GetBitContext *gb, enum FrameType ftype, - float *out) -{ - const ModeTab *mtab = tctx->mtab; + const TwinVQModeTab *mtab = tctx->mtab; + const TwinVQFrameData *bits = &tctx->bits[tctx->cur_frame]; int i, j; int sub = mtab->fmode[ftype].sub; - float step = AMP_MAX / ((1 << GAIN_BITS) - 1); - float sub_step = SUB_AMP_MAX / ((1 << SUB_GAIN_BITS) - 1); + float step = TWINVQ_AMP_MAX / ((1 << TWINVQ_GAIN_BITS) - 1); + float sub_step = TWINVQ_SUB_AMP_MAX / ((1 << TWINVQ_SUB_GAIN_BITS) - 1); - if (ftype == FT_LONG) { + if (ftype == TWINVQ_FT_LONG) { for (i = 0; i < tctx->avctx->channels; i++) - out[i] = (1. / (1 << 13)) * - mulawinv(step * 0.5 + step * get_bits(gb, GAIN_BITS), - AMP_MAX, MULAW_MU); + out[i] = (1.0 / (1 << 13)) * + twinvq_mulawinv(step * 0.5 + step * bits->gain_bits[i], + TWINVQ_AMP_MAX, TWINVQ_MULAW_MU); } else { for (i = 0; i < tctx->avctx->channels; i++) { - float val = (1. / (1 << 23)) * - mulawinv(step * 0.5 + step * get_bits(gb, GAIN_BITS), - AMP_MAX, MULAW_MU); + float val = (1.0 / (1 << 23)) * + twinvq_mulawinv(step * 0.5 + step * bits->gain_bits[i], + TWINVQ_AMP_MAX, TWINVQ_MULAW_MU); for (j = 0; j < sub; j++) out[i * sub + j] = - val * mulawinv(sub_step * 0.5 + - sub_step * get_bits(gb, SUB_GAIN_BITS), - SUB_AMP_MAX, MULAW_MU); + val * twinvq_mulawinv(sub_step * 0.5 + + sub_step * bits->sub_gain_bits[i * sub + j], + TWINVQ_SUB_AMP_MAX, TWINVQ_MULAW_MU); } } } @@ -553,10 +260,10 @@ static void rearrange_lsp(int order, float *lsp, float min_dist) } } -static void decode_lsp(TwinContext *tctx, int lpc_idx1, uint8_t *lpc_idx2, +static void decode_lsp(TwinVQContext *tctx, int lpc_idx1, uint8_t *lpc_idx2, int lpc_hist_idx, float *lsp, float *hist) { - const ModeTab *mtab = tctx->mtab; + const TwinVQModeTab *mtab = tctx->mtab; int i, j; const float *cb = mtab->lspcodebook; @@ -582,7 +289,7 @@ static void decode_lsp(TwinContext *tctx, int lpc_idx1, uint8_t *lpc_idx2, rearrange_lsp(mtab->n_lsp, lsp, 0.0001); for (i = 0; i < mtab->n_lsp; i++) { - float tmp1 = 1. - cb3[lpc_hist_idx * mtab->n_lsp + i]; + float tmp1 = 1.0 - cb3[lpc_hist_idx * mtab->n_lsp + i]; float tmp2 = hist[i] * cb3[lpc_hist_idx * mtab->n_lsp + i]; hist[i] = lsp[i]; lsp[i] = lsp[i] * tmp1 + tmp2; @@ -593,8 +300,8 @@ static void decode_lsp(TwinContext *tctx, int lpc_idx1, uint8_t *lpc_idx2, ff_sort_nearly_sorted_floats(lsp, mtab->n_lsp); } -static void dec_lpc_spectrum_inv(TwinContext *tctx, float *lsp, - enum FrameType ftype, float *lpc) +static void dec_lpc_spectrum_inv(TwinVQContext *tctx, float *lsp, + enum TwinVQFrameType ftype, float *lpc) { int i; int size = tctx->mtab->size / tctx->mtab->fmode[ftype].sub; @@ -603,13 +310,13 @@ static void dec_lpc_spectrum_inv(TwinContext *tctx, float *lsp, lsp[i] = 2 * cos(lsp[i]); switch (ftype) { - case FT_LONG: + case TWINVQ_FT_LONG: eval_lpcenv_2parts(tctx, ftype, lsp, lpc, size, 8); break; - case FT_MEDIUM: + case TWINVQ_FT_MEDIUM: eval_lpcenv_2parts(tctx, ftype, lsp, lpc, size, 2); break; - case FT_SHORT: + case TWINVQ_FT_SHORT: eval_lpcenv(tctx, lsp, lpc); break; } @@ -617,22 +324,22 @@ static void dec_lpc_spectrum_inv(TwinContext *tctx, float *lsp, static const uint8_t wtype_to_wsize[] = { 0, 0, 2, 2, 2, 1, 0, 1, 1 }; -static void imdct_and_window(TwinContext *tctx, enum FrameType ftype, int wtype, - float *in, float *prev, int ch) +static void imdct_and_window(TwinVQContext *tctx, enum TwinVQFrameType ftype, + int wtype, float *in, float *prev, int ch) { - FFTContext *mdct = &tctx->mdct_ctx[ftype]; - const ModeTab *mtab = tctx->mtab; - int bsize = mtab->size / mtab->fmode[ftype].sub; - int size = mtab->size; - float *buf1 = tctx->tmp_buf; + FFTContext *mdct = &tctx->mdct_ctx[ftype]; + const TwinVQModeTab *mtab = tctx->mtab; + int bsize = mtab->size / mtab->fmode[ftype].sub; + int size = mtab->size; + float *buf1 = tctx->tmp_buf; int j, first_wsize, wsize; // Window size float *out = tctx->curr_frame + 2 * ch * mtab->size; float *out2 = out; float *prev_buf; int types_sizes[] = { - mtab->size / mtab->fmode[FT_LONG].sub, - mtab->size / mtab->fmode[FT_MEDIUM].sub, - mtab->size / (mtab->fmode[FT_SHORT].sub * 2), + mtab->size / mtab->fmode[TWINVQ_FT_LONG].sub, + mtab->size / mtab->fmode[TWINVQ_FT_MEDIUM].sub, + mtab->size / (mtab->fmode[TWINVQ_FT_SHORT].sub * 2), }; wsize = types_sizes[wtype_to_wsize[wtype]]; @@ -640,7 +347,7 @@ static void imdct_and_window(TwinContext *tctx, enum FrameType ftype, int wtype, prev_buf = prev + (size - bsize) / 2; for (j = 0; j < mtab->fmode[ftype].sub; j++) { - int sub_wtype = ftype == FT_MEDIUM ? 8 : wtype; + int sub_wtype = ftype == TWINVQ_FT_MEDIUM ? 8 : wtype; if (!j && wtype == 4) sub_wtype = 4; @@ -660,7 +367,7 @@ static void imdct_and_window(TwinContext *tctx, enum FrameType ftype, int wtype, memcpy(out2, buf1 + bsize * j + wsize / 2, (bsize - wsize / 2) * sizeof(float)); - out2 += ftype == FT_MEDIUM ? (bsize - wsize) / 2 : bsize - wsize; + out2 += ftype == TWINVQ_FT_MEDIUM ? (bsize - wsize) / 2 : bsize - wsize; prev_buf = buf1 + bsize * j + bsize / 2; } @@ -668,12 +375,13 @@ static void imdct_and_window(TwinContext *tctx, enum FrameType ftype, int wtype, tctx->last_block_pos[ch] = (size + first_wsize) / 2; } -static void imdct_output(TwinContext *tctx, enum FrameType ftype, int wtype, - float **out) +static void imdct_output(TwinVQContext *tctx, enum TwinVQFrameType ftype, + int wtype, float **out, int offset) { - const ModeTab *mtab = tctx->mtab; - float *prev_buf = tctx->prev_frame + tctx->last_block_pos[0]; + const TwinVQModeTab *mtab = tctx->mtab; + float *prev_buf = tctx->prev_frame + tctx->last_block_pos[0]; int size1, size2, i; + float *out1, *out2; for (i = 0; i < tctx->avctx->channels; i++) imdct_and_window(tctx, ftype, wtype, @@ -687,121 +395,68 @@ static void imdct_output(TwinContext *tctx, enum FrameType ftype, int wtype, size2 = tctx->last_block_pos[0]; size1 = mtab->size - size2; - memcpy(&out[0][0], prev_buf, size1 * sizeof(out[0][0])); - memcpy(&out[0][size1], tctx->curr_frame, size2 * sizeof(out[0][0])); + out1 = &out[0][0] + offset; + memcpy(out1, prev_buf, size1 * sizeof(*out1)); + memcpy(out1 + size1, tctx->curr_frame, size2 * sizeof(*out1)); if (tctx->avctx->channels == 2) { - memcpy(&out[1][0], &prev_buf[2 * mtab->size], - size1 * sizeof(out[1][0])); - memcpy(&out[1][size1], &tctx->curr_frame[2 * mtab->size], - size2 * sizeof(out[1][0])); - tctx->fdsp.butterflies_float(out[0], out[1], mtab->size); + out2 = &out[1][0] + offset; + memcpy(out2, &prev_buf[2 * mtab->size], + size1 * sizeof(*out2)); + memcpy(out2 + size1, &tctx->curr_frame[2 * mtab->size], + size2 * sizeof(*out2)); + tctx->fdsp.butterflies_float(out1, out2, mtab->size); } } -static void dec_bark_env(TwinContext *tctx, const uint8_t *in, int use_hist, - int ch, float *out, float gain, enum FrameType ftype) +static void read_and_decode_spectrum(TwinVQContext *tctx, float *out, + enum TwinVQFrameType ftype) { - const ModeTab *mtab = tctx->mtab; + const TwinVQModeTab *mtab = tctx->mtab; + TwinVQFrameData *bits = &tctx->bits[tctx->cur_frame]; + int channels = tctx->avctx->channels; + int sub = mtab->fmode[ftype].sub; + int block_size = mtab->size / sub; + float gain[TWINVQ_CHANNELS_MAX * TWINVQ_SUBBLOCKS_MAX]; + float ppc_shape[TWINVQ_PPC_SHAPE_LEN_MAX * TWINVQ_CHANNELS_MAX * 4]; + int i, j; - float *hist = tctx->bark_hist[ftype][ch]; - float val = ((const float []) { 0.4, 0.35, 0.28 })[ftype]; - int bark_n_coef = mtab->fmode[ftype].bark_n_coef; - int fw_cb_len = mtab->fmode[ftype].bark_env_size / bark_n_coef; - int idx = 0; - - for (i = 0; i < fw_cb_len; i++) - for (j = 0; j < bark_n_coef; j++, idx++) { - float tmp2 = mtab->fmode[ftype].bark_cb[fw_cb_len * in[j] + i] * - (1. / 4096); - float st = use_hist ? (1. - val) * tmp2 + val * hist[idx] + 1. - : tmp2 + 1.; - - hist[idx] = tmp2; - if (st < -1.) - st = 1.; - - memset_float(out, st * gain, mtab->fmode[ftype].bark_tab[idx]); - out += mtab->fmode[ftype].bark_tab[idx]; - } -} -static void read_and_decode_spectrum(TwinContext *tctx, GetBitContext *gb, - float *out, enum FrameType ftype) -{ - const ModeTab *mtab = tctx->mtab; - int channels = tctx->avctx->channels; - int sub = mtab->fmode[ftype].sub; - int block_size = mtab->size / sub; - float gain[CHANNELS_MAX * SUBBLOCKS_MAX]; - float ppc_shape[PPC_SHAPE_LEN_MAX * CHANNELS_MAX * 4]; - uint8_t bark1[CHANNELS_MAX][SUBBLOCKS_MAX][BARK_N_COEF_MAX]; - uint8_t bark_use_hist[CHANNELS_MAX][SUBBLOCKS_MAX]; - - uint8_t lpc_idx1[CHANNELS_MAX]; - uint8_t lpc_idx2[CHANNELS_MAX][LSP_SPLIT_MAX]; - uint8_t lpc_hist_idx[CHANNELS_MAX]; - - int i, j, k; - - dequant(tctx, gb, out, ftype, + dequant(tctx, bits->main_coeffs, out, ftype, mtab->fmode[ftype].cb0, mtab->fmode[ftype].cb1, mtab->fmode[ftype].cb_len_read); - for (i = 0; i < channels; i++) - for (j = 0; j < sub; j++) - for (k = 0; k < mtab->fmode[ftype].bark_n_coef; k++) - bark1[i][j][k] = - get_bits(gb, mtab->fmode[ftype].bark_n_bit); - - for (i = 0; i < channels; i++) - for (j = 0; j < sub; j++) - bark_use_hist[i][j] = get_bits1(gb); - - dec_gain(tctx, gb, ftype, gain); + dec_gain(tctx, ftype, gain); - for (i = 0; i < channels; i++) { - lpc_hist_idx[i] = get_bits(gb, tctx->mtab->lsp_bit0); - lpc_idx1[i] = get_bits(gb, tctx->mtab->lsp_bit1); - - for (j = 0; j < tctx->mtab->lsp_split; j++) - lpc_idx2[i][j] = get_bits(gb, tctx->mtab->lsp_bit2); - } - - if (ftype == FT_LONG) { + if (ftype == TWINVQ_FT_LONG) { int cb_len_p = (tctx->n_div[3] + mtab->ppc_shape_len * channels - 1) / tctx->n_div[3]; - dequant(tctx, gb, ppc_shape, FT_PPC, mtab->ppc_shape_cb, - mtab->ppc_shape_cb + cb_len_p * PPC_SHAPE_CB_SIZE, cb_len_p); + dequant(tctx, bits->ppc_coeffs, ppc_shape, + TWINVQ_FT_PPC, mtab->ppc_shape_cb, + mtab->ppc_shape_cb + cb_len_p * TWINVQ_PPC_SHAPE_CB_SIZE, + cb_len_p); } for (i = 0; i < channels; i++) { float *chunk = out + mtab->size * i; - float lsp[LSP_COEFS_MAX]; + float lsp[TWINVQ_LSP_COEFS_MAX]; for (j = 0; j < sub; j++) { - dec_bark_env(tctx, bark1[i][j], bark_use_hist[i][j], i, - tctx->tmp_buf, gain[sub * i + j], ftype); + tctx->dec_bark_env(tctx, bits->bark1[i][j], + bits->bark_use_hist[i][j], i, + tctx->tmp_buf, gain[sub * i + j], ftype); tctx->fdsp.vector_fmul(chunk + block_size * j, chunk + block_size * j, tctx->tmp_buf, block_size); } - if (ftype == FT_LONG) { - float pgain_step = 25000. / ((1 << mtab->pgain_bit) - 1); - int p_coef = get_bits(gb, tctx->mtab->ppc_period_bit); - int g_coef = get_bits(gb, tctx->mtab->pgain_bit); - float v = 1. / 8192 * - mulawinv(pgain_step * g_coef + pgain_step / 2, - 25000., PGAIN_MU); - - decode_ppc(tctx, p_coef, ppc_shape + i * mtab->ppc_shape_len, v, - chunk); - } + if (ftype == TWINVQ_FT_LONG) + tctx->decode_ppc(tctx, bits->p_coef[i], bits->g_coef[i], + ppc_shape + i * mtab->ppc_shape_len, chunk); - decode_lsp(tctx, lpc_idx1[i], lpc_idx2[i], lpc_hist_idx[i], lsp, - tctx->lsp_hist[i]); + decode_lsp(tctx, bits->lpc_idx1[i], bits->lpc_idx2[i], + bits->lpc_hist_idx[i], lsp, tctx->lsp_hist[i]); dec_lpc_spectrum_inv(tctx, lsp, ftype, tctx->tmp_buf); @@ -812,32 +467,26 @@ static void read_and_decode_spectrum(TwinContext *tctx, GetBitContext *gb, } } -static int twin_decode_frame(AVCodecContext *avctx, void *data, - int *got_frame_ptr, AVPacket *avpkt) +const enum TwinVQFrameType ff_twinvq_wtype_to_ftype_table[] = { + TWINVQ_FT_LONG, TWINVQ_FT_LONG, TWINVQ_FT_SHORT, TWINVQ_FT_LONG, + TWINVQ_FT_MEDIUM, TWINVQ_FT_LONG, TWINVQ_FT_LONG, TWINVQ_FT_MEDIUM, + TWINVQ_FT_MEDIUM +}; + +int ff_twinvq_decode_frame(AVCodecContext *avctx, void *data, + int *got_frame_ptr, AVPacket *avpkt) { AVFrame *frame = data; const uint8_t *buf = avpkt->data; int buf_size = avpkt->size; - TwinContext *tctx = avctx->priv_data; - GetBitContext gb; - const ModeTab *mtab = tctx->mtab; - float **out = NULL; - enum FrameType ftype; - int window_type, ret; - static const enum FrameType wtype_to_ftype_table[] = { - FT_LONG, FT_LONG, FT_SHORT, FT_LONG, - FT_MEDIUM, FT_LONG, FT_LONG, FT_MEDIUM, FT_MEDIUM - }; - - if (buf_size * 8 < avctx->bit_rate * mtab->size / avctx->sample_rate + 8) { - av_log(avctx, AV_LOG_ERROR, - "Frame too small (%d bytes). Truncated file?\n", buf_size); - return AVERROR(EINVAL); - } + TwinVQContext *tctx = avctx->priv_data; + const TwinVQModeTab *mtab = tctx->mtab; + float **out = NULL; + int ret; /* get output buffer */ if (tctx->discarded_packets >= 2) { - frame->nb_samples = mtab->size; + frame->nb_samples = mtab->size * tctx->frames_per_packet; if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) { av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n"); return ret; @@ -845,22 +494,26 @@ static int twin_decode_frame(AVCodecContext *avctx, void *data, out = (float **)frame->extended_data; } - init_get_bits(&gb, buf, buf_size * 8); - skip_bits(&gb, get_bits(&gb, 8)); - window_type = get_bits(&gb, WINDOW_TYPE_BITS); - - if (window_type > 8) { - av_log(avctx, AV_LOG_ERROR, "Invalid window type, broken sample?\n"); - return -1; + if (buf_size < avctx->block_align) { + av_log(avctx, AV_LOG_ERROR, + "Frame too small (%d bytes). Truncated file?\n", buf_size); + return AVERROR(EINVAL); } - ftype = wtype_to_ftype_table[window_type]; + if ((ret = tctx->read_bitstream(avctx, tctx, buf, buf_size)) < 0) + return ret; - read_and_decode_spectrum(tctx, &gb, tctx->spectrum, ftype); + for (tctx->cur_frame = 0; tctx->cur_frame < tctx->frames_per_packet; + tctx->cur_frame++) { + read_and_decode_spectrum(tctx, tctx->spectrum, + tctx->bits[tctx->cur_frame].ftype); - imdct_output(tctx, ftype, window_type, out); + imdct_output(tctx, tctx->bits[tctx->cur_frame].ftype, + tctx->bits[tctx->cur_frame].window_type, out, + tctx->cur_frame * mtab->size); - FFSWAP(float *, tctx->curr_frame, tctx->prev_frame); + FFSWAP(float *, tctx->curr_frame, tctx->prev_frame); + } if (tctx->discarded_packets < 2) { tctx->discarded_packets++; @@ -870,20 +523,23 @@ static int twin_decode_frame(AVCodecContext *avctx, void *data, *got_frame_ptr = 1; - return buf_size; + // VQF can deliver packets 1 byte greater than block align + if (buf_size == avctx->block_align + 1) + return buf_size; + return avctx->block_align; } /** * Init IMDCT and windowing tables */ -static av_cold int init_mdct_win(TwinContext *tctx) +static av_cold int init_mdct_win(TwinVQContext *tctx) { int i, j, ret; - const ModeTab *mtab = tctx->mtab; - int size_s = mtab->size / mtab->fmode[FT_SHORT].sub; - int size_m = mtab->size / mtab->fmode[FT_MEDIUM].sub; - int channels = tctx->avctx->channels; - float norm = channels == 1 ? 2. : 1.; + const TwinVQModeTab *mtab = tctx->mtab; + int size_s = mtab->size / mtab->fmode[TWINVQ_FT_SHORT].sub; + int size_m = mtab->size / mtab->fmode[TWINVQ_FT_MEDIUM].sub; + int channels = tctx->avctx->channels; + float norm = channels == 1 ? 2.0 : 1.0; for (i = 0; i < 3; i++) { int bsize = tctx->mtab->size / tctx->mtab->fmode[i].sub; @@ -936,19 +592,19 @@ alloc_fail: static void permutate_in_line(int16_t *tab, int num_vect, int num_blocks, int block_size, const uint8_t line_len[2], int length_div, - enum FrameType ftype) + enum TwinVQFrameType ftype) { int i, j; for (i = 0; i < line_len[0]; i++) { int shift; - if (num_blocks == 1 || - (ftype == FT_LONG && num_vect % num_blocks) || - (ftype != FT_LONG && num_vect & 1) || + if (num_blocks == 1 || + (ftype == TWINVQ_FT_LONG && num_vect % num_blocks) || + (ftype != TWINVQ_FT_LONG && num_vect & 1) || i == line_len[1]) { shift = 0; - } else if (ftype == FT_LONG) { + } else if (ftype == TWINVQ_FT_LONG) { shift = i; } else shift = i * i; @@ -993,14 +649,14 @@ static void linear_perm(int16_t *out, int16_t *in, int n_blocks, int size) out[i] = block_size * (in[i] % n_blocks) + in[i] / n_blocks; } -static av_cold void construct_perm_table(TwinContext *tctx, - enum FrameType ftype) +static av_cold void construct_perm_table(TwinVQContext *tctx, + enum TwinVQFrameType ftype) { int block_size, size; - const ModeTab *mtab = tctx->mtab; + const TwinVQModeTab *mtab = tctx->mtab; int16_t *tmp_perm = (int16_t *)tctx->tmp_buf; - if (ftype == FT_PPC) { + if (ftype == TWINVQ_FT_PPC) { size = tctx->avctx->channels; block_size = mtab->ppc_shape_len; } else { @@ -1019,12 +675,12 @@ static av_cold void construct_perm_table(TwinContext *tctx, size * block_size); } -static av_cold void init_bitstream_params(TwinContext *tctx) +static av_cold void init_bitstream_params(TwinVQContext *tctx) { - const ModeTab *mtab = tctx->mtab; - int n_ch = tctx->avctx->channels; - int total_fr_bits = tctx->avctx->bit_rate * mtab->size / - tctx->avctx->sample_rate; + const TwinVQModeTab *mtab = tctx->mtab; + int n_ch = tctx->avctx->channels; + int total_fr_bits = tctx->avctx->bit_rate * mtab->size / + tctx->avctx->sample_rate; int lsp_bits_per_block = n_ch * (mtab->lsp_bit0 + mtab->lsp_bit1 + mtab->lsp_split * mtab->lsp_bit2); @@ -1033,7 +689,7 @@ static av_cold void init_bitstream_params(TwinContext *tctx) mtab->ppc_period_bit); int bsize_no_main_cb[3], bse_bits[3], i; - enum FrameType frametype; + enum TwinVQFrameType frametype; for (i = 0; i < 3; i++) // +1 for history usage switch @@ -1042,12 +698,18 @@ static av_cold void init_bitstream_params(TwinContext *tctx) mtab->fmode[i].bark_n_bit + 1); bsize_no_main_cb[2] = bse_bits[2] + lsp_bits_per_block + ppc_bits + - WINDOW_TYPE_BITS + n_ch * GAIN_BITS; + TWINVQ_WINDOW_TYPE_BITS + n_ch * TWINVQ_GAIN_BITS; for (i = 0; i < 2; i++) bsize_no_main_cb[i] = - lsp_bits_per_block + n_ch * GAIN_BITS + WINDOW_TYPE_BITS + - mtab->fmode[i].sub * (bse_bits[i] + n_ch * SUB_GAIN_BITS); + lsp_bits_per_block + n_ch * TWINVQ_GAIN_BITS + + TWINVQ_WINDOW_TYPE_BITS + + mtab->fmode[i].sub * (bse_bits[i] + n_ch * TWINVQ_SUB_GAIN_BITS); + + if (tctx->codec == TWINVQ_CODEC_METASOUND && !tctx->is_6kbps) { + bsize_no_main_cb[1] += 2; + bsize_no_main_cb[2] += 2; + } // The remaining bits are all used for the main spectrum coefficients for (i = 0; i < 4; i++) { @@ -1084,13 +746,13 @@ static av_cold void init_bitstream_params(TwinContext *tctx) tctx->length_change[i] = num_rounded_up; } - for (frametype = FT_SHORT; frametype <= FT_PPC; frametype++) + for (frametype = TWINVQ_FT_SHORT; frametype <= TWINVQ_FT_PPC; frametype++) construct_perm_table(tctx, frametype); } -static av_cold int twin_decode_close(AVCodecContext *avctx) +av_cold int ff_twinvq_decode_close(AVCodecContext *avctx) { - TwinContext *tctx = avctx->priv_data; + TwinVQContext *tctx = avctx->priv_data; int i; for (i = 0; i < 3; i++) { @@ -1106,109 +768,38 @@ static av_cold int twin_decode_close(AVCodecContext *avctx) return 0; } -static av_cold int twin_decode_init(AVCodecContext *avctx) +av_cold int ff_twinvq_decode_init(AVCodecContext *avctx) { - int ret, isampf, ibps; - TwinContext *tctx = avctx->priv_data; + int ret; + TwinVQContext *tctx = avctx->priv_data; tctx->avctx = avctx; avctx->sample_fmt = AV_SAMPLE_FMT_FLTP; - if (!avctx->extradata || avctx->extradata_size < 12) { - av_log(avctx, AV_LOG_ERROR, "Missing or incomplete extradata\n"); + if (!avctx->block_align) { + avctx->block_align = tctx->frame_size + 7 >> 3; + } else if (avctx->block_align * 8 < tctx->frame_size) { + av_log(avctx, AV_LOG_ERROR, "Block align is %d bits, expected %d\n", + avctx->block_align * 8, tctx->frame_size); return AVERROR_INVALIDDATA; } - avctx->channels = AV_RB32(avctx->extradata) + 1; - avctx->bit_rate = AV_RB32(avctx->extradata + 4) * 1000; - isampf = AV_RB32(avctx->extradata + 8); - - if (isampf < 8 || isampf > 44) { - av_log(avctx, AV_LOG_ERROR, "Unsupported sample rate\n"); + tctx->frames_per_packet = avctx->block_align * 8 / tctx->frame_size; + if (tctx->frames_per_packet > TWINVQ_MAX_FRAMES_PER_PACKET) { + av_log(avctx, AV_LOG_ERROR, "Too many frames per packet (%d)\n", + tctx->frames_per_packet); return AVERROR_INVALIDDATA; } - switch (isampf) { - case 44: - avctx->sample_rate = 44100; - break; - case 22: - avctx->sample_rate = 22050; - break; - case 11: - avctx->sample_rate = 11025; - break; - default: - avctx->sample_rate = isampf * 1000; - break; - } - - if (avctx->channels <= 0 || avctx->channels > CHANNELS_MAX) { - av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %i\n", - avctx->channels); - return -1; - } - avctx->channel_layout = avctx->channels == 1 ? AV_CH_LAYOUT_MONO - : AV_CH_LAYOUT_STEREO; - - ibps = avctx->bit_rate / (1000 * avctx->channels); - switch ((isampf << 8) + ibps) { - case (8 << 8) + 8: - tctx->mtab = &mode_08_08; - break; - case (11 << 8) + 8: - tctx->mtab = &mode_11_08; - break; - case (11 << 8) + 10: - tctx->mtab = &mode_11_10; - break; - case (16 << 8) + 16: - tctx->mtab = &mode_16_16; - break; - case (22 << 8) + 20: - tctx->mtab = &mode_22_20; - break; - case (22 << 8) + 24: - tctx->mtab = &mode_22_24; - break; - case (22 << 8) + 32: - tctx->mtab = &mode_22_32; - break; - case (44 << 8) + 40: - tctx->mtab = &mode_44_40; - break; - case (44 << 8) + 48: - tctx->mtab = &mode_44_48; - break; - default: - av_log(avctx, AV_LOG_ERROR, - "This version does not support %d kHz - %d kbit/s/ch mode.\n", - isampf, isampf); - return -1; - } - - avpriv_float_dsp_init(&tctx->fdsp, avctx->flags & CODEC_FLAG_BITEXACT); + avpriv_float_dsp_init(&tctx->fdsp, avctx->flags & AV_CODEC_FLAG_BITEXACT); if ((ret = init_mdct_win(tctx))) { av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n"); - twin_decode_close(avctx); + ff_twinvq_decode_close(avctx); return ret; } init_bitstream_params(tctx); - memset_float(tctx->bark_hist[0][0], 0.1, FF_ARRAY_ELEMS(tctx->bark_hist)); + twinvq_memset_float(tctx->bark_hist[0][0], 0.1, + FF_ARRAY_ELEMS(tctx->bark_hist)); return 0; } - -AVCodec ff_twinvq_decoder = { - .name = "twinvq", - .type = AVMEDIA_TYPE_AUDIO, - .id = AV_CODEC_ID_TWINVQ, - .priv_data_size = sizeof(TwinContext), - .init = twin_decode_init, - .close = twin_decode_close, - .decode = twin_decode_frame, - .capabilities = CODEC_CAP_DR1, - .long_name = NULL_IF_CONFIG_SMALL("VQF TwinVQ"), - .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP, - AV_SAMPLE_FMT_NONE }, -};