yading@10: /* yading@10: * RV30/40 decoder common data yading@10: * Copyright (c) 2007 Mike Melanson, Konstantin Shishkov yading@10: * yading@10: * This file is part of FFmpeg. yading@10: * yading@10: * FFmpeg is free software; you can redistribute it and/or yading@10: * modify it under the terms of the GNU Lesser General Public yading@10: * License as published by the Free Software Foundation; either yading@10: * version 2.1 of the License, or (at your option) any later version. yading@10: * yading@10: * FFmpeg is distributed in the hope that it will be useful, yading@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of yading@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU yading@10: * Lesser General Public License for more details. yading@10: * yading@10: * You should have received a copy of the GNU Lesser General Public yading@10: * License along with FFmpeg; if not, write to the Free Software yading@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA yading@10: */ yading@10: yading@10: /** yading@10: * @file yading@10: * RV30/40 decoder common data yading@10: */ yading@10: yading@10: #include "libavutil/imgutils.h" yading@10: #include "libavutil/internal.h" yading@10: yading@10: #include "avcodec.h" yading@10: #include "error_resilience.h" yading@10: #include "mpegvideo.h" yading@10: #include "golomb.h" yading@10: #include "internal.h" yading@10: #include "mathops.h" yading@10: #include "rectangle.h" yading@10: #include "thread.h" yading@10: yading@10: #include "rv34vlc.h" yading@10: #include "rv34data.h" yading@10: #include "rv34.h" yading@10: yading@10: //#define DEBUG yading@10: yading@10: static inline void ZERO8x2(void* dst, int stride) yading@10: { yading@10: fill_rectangle(dst, 1, 2, stride, 0, 4); yading@10: fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4); yading@10: } yading@10: yading@10: /** translation of RV30/40 macroblock types to lavc ones */ yading@10: static const int rv34_mb_type_to_lavc[12] = { yading@10: MB_TYPE_INTRA, yading@10: MB_TYPE_INTRA16x16 | MB_TYPE_SEPARATE_DC, yading@10: MB_TYPE_16x16 | MB_TYPE_L0, yading@10: MB_TYPE_8x8 | MB_TYPE_L0, yading@10: MB_TYPE_16x16 | MB_TYPE_L0, yading@10: MB_TYPE_16x16 | MB_TYPE_L1, yading@10: MB_TYPE_SKIP, yading@10: MB_TYPE_DIRECT2 | MB_TYPE_16x16, yading@10: MB_TYPE_16x8 | MB_TYPE_L0, yading@10: MB_TYPE_8x16 | MB_TYPE_L0, yading@10: MB_TYPE_16x16 | MB_TYPE_L0L1, yading@10: MB_TYPE_16x16 | MB_TYPE_L0 | MB_TYPE_SEPARATE_DC yading@10: }; yading@10: yading@10: yading@10: static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES]; yading@10: yading@10: static int rv34_decode_mv(RV34DecContext *r, int block_type); yading@10: yading@10: /** yading@10: * @name RV30/40 VLC generating functions yading@10: * @{ yading@10: */ yading@10: yading@10: static const int table_offs[] = { yading@10: 0, 1818, 3622, 4144, 4698, 5234, 5804, 5868, 5900, 5932, yading@10: 5996, 6252, 6316, 6348, 6380, 7674, 8944, 10274, 11668, 12250, yading@10: 14060, 15846, 16372, 16962, 17512, 18148, 18180, 18212, 18244, 18308, yading@10: 18564, 18628, 18660, 18692, 20036, 21314, 22648, 23968, 24614, 26384, yading@10: 28190, 28736, 29366, 29938, 30608, 30640, 30672, 30704, 30768, 31024, yading@10: 31088, 31120, 31184, 32570, 33898, 35236, 36644, 37286, 39020, 40802, yading@10: 41368, 42052, 42692, 43348, 43380, 43412, 43444, 43476, 43604, 43668, yading@10: 43700, 43732, 45100, 46430, 47778, 49160, 49802, 51550, 53340, 53972, yading@10: 54648, 55348, 55994, 56122, 56154, 56186, 56218, 56346, 56410, 56442, yading@10: 56474, 57878, 59290, 60636, 62036, 62682, 64460, 64524, 64588, 64716, yading@10: 64844, 66076, 67466, 67978, 68542, 69064, 69648, 70296, 72010, 72074, yading@10: 72138, 72202, 72330, 73572, 74936, 75454, 76030, 76566, 77176, 77822, yading@10: 79582, 79646, 79678, 79742, 79870, 81180, 82536, 83064, 83672, 84242, yading@10: 84934, 85576, 87384, 87448, 87480, 87544, 87672, 88982, 90340, 90902, yading@10: 91598, 92182, 92846, 93488, 95246, 95278, 95310, 95374, 95502, 96878, yading@10: 98266, 98848, 99542, 100234, 100884, 101524, 103320, 103352, 103384, 103416, yading@10: 103480, 104874, 106222, 106910, 107584, 108258, 108902, 109544, 111366, 111398, yading@10: 111430, 111462, 111494, 112878, 114320, 114988, 115660, 116310, 116950, 117592 yading@10: }; yading@10: yading@10: static VLC_TYPE table_data[117592][2]; yading@10: yading@10: /** yading@10: * Generate VLC from codeword lengths. yading@10: * @param bits codeword lengths (zeroes are accepted) yading@10: * @param size length of input data yading@10: * @param vlc output VLC yading@10: * @param insyms symbols for input codes (NULL for default ones) yading@10: * @param num VLC table number (for static initialization) yading@10: */ yading@10: static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms, yading@10: const int num) yading@10: { yading@10: int i; yading@10: int counts[17] = {0}, codes[17]; yading@10: uint16_t cw[MAX_VLC_SIZE], syms[MAX_VLC_SIZE]; yading@10: uint8_t bits2[MAX_VLC_SIZE]; yading@10: int maxbits = 0, realsize = 0; yading@10: yading@10: for(i = 0; i < size; i++){ yading@10: if(bits[i]){ yading@10: bits2[realsize] = bits[i]; yading@10: syms[realsize] = insyms ? insyms[i] : i; yading@10: realsize++; yading@10: maxbits = FFMAX(maxbits, bits[i]); yading@10: counts[bits[i]]++; yading@10: } yading@10: } yading@10: yading@10: codes[0] = 0; yading@10: for(i = 0; i < 16; i++) yading@10: codes[i+1] = (codes[i] + counts[i]) << 1; yading@10: for(i = 0; i < realsize; i++) yading@10: cw[i] = codes[bits2[i]]++; yading@10: yading@10: vlc->table = &table_data[table_offs[num]]; yading@10: vlc->table_allocated = table_offs[num + 1] - table_offs[num]; yading@10: ff_init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize, yading@10: bits2, 1, 1, yading@10: cw, 2, 2, yading@10: syms, 2, 2, INIT_VLC_USE_NEW_STATIC); yading@10: } yading@10: yading@10: /** yading@10: * Initialize all tables. yading@10: */ yading@10: static av_cold void rv34_init_tables(void) yading@10: { yading@10: int i, j, k; yading@10: yading@10: for(i = 0; i < NUM_INTRA_TABLES; i++){ yading@10: for(j = 0; j < 2; j++){ yading@10: rv34_gen_vlc(rv34_table_intra_cbppat [i][j], CBPPAT_VLC_SIZE, &intra_vlcs[i].cbppattern[j], NULL, 19*i + 0 + j); yading@10: rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL, 19*i + 2 + j); yading@10: rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j], NULL, 19*i + 4 + j); yading@10: for(k = 0; k < 4; k++){ yading@10: rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2], CBP_VLC_SIZE, &intra_vlcs[i].cbp[j][k], rv34_cbp_code, 19*i + 6 + j*4 + k); yading@10: } yading@10: } yading@10: for(j = 0; j < 4; j++){ yading@10: rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL, 19*i + 14 + j); yading@10: } yading@10: rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL, 19*i + 18); yading@10: } yading@10: yading@10: for(i = 0; i < NUM_INTER_TABLES; i++){ yading@10: rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL, i*12 + 95); yading@10: for(j = 0; j < 4; j++){ yading@10: rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code, i*12 + 96 + j); yading@10: } yading@10: for(j = 0; j < 2; j++){ yading@10: rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j], NULL, i*12 + 100 + j); yading@10: rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL, i*12 + 102 + j); yading@10: rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j], NULL, i*12 + 104 + j); yading@10: } yading@10: rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL, i*12 + 106); yading@10: } yading@10: } yading@10: yading@10: /** @} */ // vlc group yading@10: yading@10: /** yading@10: * @name RV30/40 4x4 block decoding functions yading@10: * @{ yading@10: */ yading@10: yading@10: /** yading@10: * Decode coded block pattern. yading@10: */ yading@10: static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table) yading@10: { yading@10: int pattern, code, cbp=0; yading@10: int ones; yading@10: static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000}; yading@10: static const int shifts[4] = { 0, 2, 8, 10 }; yading@10: const int *curshift = shifts; yading@10: int i, t, mask; yading@10: yading@10: code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2); yading@10: pattern = code & 0xF; yading@10: code >>= 4; yading@10: yading@10: ones = rv34_count_ones[pattern]; yading@10: yading@10: for(mask = 8; mask; mask >>= 1, curshift++){ yading@10: if(pattern & mask) yading@10: cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0]; yading@10: } yading@10: yading@10: for(i = 0; i < 4; i++){ yading@10: t = (modulo_three_table[code] >> (6 - 2*i)) & 3; yading@10: if(t == 1) yading@10: cbp |= cbp_masks[get_bits1(gb)] << i; yading@10: if(t == 2) yading@10: cbp |= cbp_masks[2] << i; yading@10: } yading@10: return cbp; yading@10: } yading@10: yading@10: /** yading@10: * Get one coefficient value from the bitstream and store it. yading@10: */ yading@10: static inline void decode_coeff(int16_t *dst, int coef, int esc, GetBitContext *gb, VLC* vlc, int q) yading@10: { yading@10: if(coef){ yading@10: if(coef == esc){ yading@10: coef = get_vlc2(gb, vlc->table, 9, 2); yading@10: if(coef > 23){ yading@10: coef -= 23; yading@10: coef = 22 + ((1 << coef) | get_bits(gb, coef)); yading@10: } yading@10: coef += esc; yading@10: } yading@10: if(get_bits1(gb)) yading@10: coef = -coef; yading@10: *dst = (coef*q + 8) >> 4; yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * Decode 2x2 subblock of coefficients. yading@10: */ yading@10: static inline void decode_subblock(int16_t *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc, int q) yading@10: { yading@10: int flags = modulo_three_table[code]; yading@10: yading@10: decode_coeff( dst+0*4+0, (flags >> 6) , 3, gb, vlc, q); yading@10: if(is_block2){ yading@10: decode_coeff(dst+1*4+0, (flags >> 4) & 3, 2, gb, vlc, q); yading@10: decode_coeff(dst+0*4+1, (flags >> 2) & 3, 2, gb, vlc, q); yading@10: }else{ yading@10: decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q); yading@10: decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q); yading@10: } yading@10: decode_coeff( dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q); yading@10: } yading@10: yading@10: /** yading@10: * Decode a single coefficient. yading@10: */ yading@10: static inline void decode_subblock1(int16_t *dst, int code, GetBitContext *gb, VLC *vlc, int q) yading@10: { yading@10: int coeff = modulo_three_table[code] >> 6; yading@10: decode_coeff(dst, coeff, 3, gb, vlc, q); yading@10: } yading@10: yading@10: static inline void decode_subblock3(int16_t *dst, int code, GetBitContext *gb, VLC *vlc, yading@10: int q_dc, int q_ac1, int q_ac2) yading@10: { yading@10: int flags = modulo_three_table[code]; yading@10: yading@10: decode_coeff(dst+0*4+0, (flags >> 6) , 3, gb, vlc, q_dc); yading@10: decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q_ac1); yading@10: decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q_ac1); yading@10: decode_coeff(dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q_ac2); yading@10: } yading@10: yading@10: /** yading@10: * Decode coefficients for 4x4 block. yading@10: * yading@10: * This is done by filling 2x2 subblocks with decoded coefficients yading@10: * in this order (the same for subblocks and subblock coefficients): yading@10: * o--o yading@10: * / yading@10: * / yading@10: * o--o yading@10: */ yading@10: yading@10: static int rv34_decode_block(int16_t *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc, int q_dc, int q_ac1, int q_ac2) yading@10: { yading@10: int code, pattern, has_ac = 1; yading@10: yading@10: code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2); yading@10: yading@10: pattern = code & 0x7; yading@10: yading@10: code >>= 3; yading@10: yading@10: if (modulo_three_table[code] & 0x3F) { yading@10: decode_subblock3(dst, code, gb, &rvlc->coefficient, q_dc, q_ac1, q_ac2); yading@10: } else { yading@10: decode_subblock1(dst, code, gb, &rvlc->coefficient, q_dc); yading@10: if (!pattern) yading@10: return 0; yading@10: has_ac = 0; yading@10: } yading@10: yading@10: if(pattern & 4){ yading@10: code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2); yading@10: decode_subblock(dst + 4*0+2, code, 0, gb, &rvlc->coefficient, q_ac2); yading@10: } yading@10: if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block yading@10: code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2); yading@10: decode_subblock(dst + 4*2+0, code, 1, gb, &rvlc->coefficient, q_ac2); yading@10: } yading@10: if(pattern & 1){ yading@10: code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2); yading@10: decode_subblock(dst + 4*2+2, code, 0, gb, &rvlc->coefficient, q_ac2); yading@10: } yading@10: return has_ac | pattern; yading@10: } yading@10: yading@10: /** yading@10: * @name RV30/40 bitstream parsing yading@10: * @{ yading@10: */ yading@10: yading@10: /** yading@10: * Decode starting slice position. yading@10: * @todo Maybe replace with ff_h263_decode_mba() ? yading@10: */ yading@10: int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size) yading@10: { yading@10: int i; yading@10: for(i = 0; i < 5; i++) yading@10: if(rv34_mb_max_sizes[i] >= mb_size - 1) yading@10: break; yading@10: return rv34_mb_bits_sizes[i]; yading@10: } yading@10: yading@10: /** yading@10: * Select VLC set for decoding from current quantizer, modifier and frame type. yading@10: */ yading@10: static inline RV34VLC* choose_vlc_set(int quant, int mod, int type) yading@10: { yading@10: if(mod == 2 && quant < 19) quant += 10; yading@10: else if(mod && quant < 26) quant += 5; yading@10: return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]] yading@10: : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]]; yading@10: } yading@10: yading@10: /** yading@10: * Decode intra macroblock header and return CBP in case of success, -1 otherwise. yading@10: */ yading@10: static int rv34_decode_intra_mb_header(RV34DecContext *r, int8_t *intra_types) yading@10: { yading@10: MpegEncContext *s = &r->s; yading@10: GetBitContext *gb = &s->gb; yading@10: int mb_pos = s->mb_x + s->mb_y * s->mb_stride; yading@10: int t; yading@10: yading@10: r->is16 = get_bits1(gb); yading@10: if(r->is16){ yading@10: s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA16x16; yading@10: r->block_type = RV34_MB_TYPE_INTRA16x16; yading@10: t = get_bits(gb, 2); yading@10: fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0])); yading@10: r->luma_vlc = 2; yading@10: }else{ yading@10: if(!r->rv30){ yading@10: if(!get_bits1(gb)) yading@10: av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n"); yading@10: } yading@10: s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA; yading@10: r->block_type = RV34_MB_TYPE_INTRA; yading@10: if(r->decode_intra_types(r, gb, intra_types) < 0) yading@10: return -1; yading@10: r->luma_vlc = 1; yading@10: } yading@10: yading@10: r->chroma_vlc = 0; yading@10: r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0); yading@10: yading@10: return rv34_decode_cbp(gb, r->cur_vlcs, r->is16); yading@10: } yading@10: yading@10: /** yading@10: * Decode inter macroblock header and return CBP in case of success, -1 otherwise. yading@10: */ yading@10: static int rv34_decode_inter_mb_header(RV34DecContext *r, int8_t *intra_types) yading@10: { yading@10: MpegEncContext *s = &r->s; yading@10: GetBitContext *gb = &s->gb; yading@10: int mb_pos = s->mb_x + s->mb_y * s->mb_stride; yading@10: int i, t; yading@10: yading@10: r->block_type = r->decode_mb_info(r); yading@10: if(r->block_type == -1) yading@10: return -1; yading@10: s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type]; yading@10: r->mb_type[mb_pos] = r->block_type; yading@10: if(r->block_type == RV34_MB_SKIP){ yading@10: if(s->pict_type == AV_PICTURE_TYPE_P) yading@10: r->mb_type[mb_pos] = RV34_MB_P_16x16; yading@10: if(s->pict_type == AV_PICTURE_TYPE_B) yading@10: r->mb_type[mb_pos] = RV34_MB_B_DIRECT; yading@10: } yading@10: r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]); yading@10: rv34_decode_mv(r, r->block_type); yading@10: if(r->block_type == RV34_MB_SKIP){ yading@10: fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0])); yading@10: return 0; yading@10: } yading@10: r->chroma_vlc = 1; yading@10: r->luma_vlc = 0; yading@10: yading@10: if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){ yading@10: if(r->is16){ yading@10: t = get_bits(gb, 2); yading@10: fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0])); yading@10: r->luma_vlc = 2; yading@10: }else{ yading@10: if(r->decode_intra_types(r, gb, intra_types) < 0) yading@10: return -1; yading@10: r->luma_vlc = 1; yading@10: } yading@10: r->chroma_vlc = 0; yading@10: r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0); yading@10: }else{ yading@10: for(i = 0; i < 16; i++) yading@10: intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0; yading@10: r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1); yading@10: if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){ yading@10: r->is16 = 1; yading@10: r->chroma_vlc = 1; yading@10: r->luma_vlc = 2; yading@10: r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0); yading@10: } yading@10: } yading@10: yading@10: return rv34_decode_cbp(gb, r->cur_vlcs, r->is16); yading@10: } yading@10: yading@10: /** @} */ //bitstream functions yading@10: yading@10: /** yading@10: * @name motion vector related code (prediction, reconstruction, motion compensation) yading@10: * @{ yading@10: */ yading@10: yading@10: /** macroblock partition width in 8x8 blocks */ yading@10: static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 }; yading@10: yading@10: /** macroblock partition height in 8x8 blocks */ yading@10: static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 }; yading@10: yading@10: /** availability index for subblocks */ yading@10: static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 }; yading@10: yading@10: /** yading@10: * motion vector prediction yading@10: * yading@10: * Motion prediction performed for the block by using median prediction of yading@10: * motion vectors from the left, top and right top blocks but in corner cases yading@10: * some other vectors may be used instead. yading@10: */ yading@10: static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no) yading@10: { yading@10: MpegEncContext *s = &r->s; yading@10: int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride; yading@10: int A[2] = {0}, B[2], C[2]; yading@10: int i, j; yading@10: int mx, my; yading@10: int* avail = r->avail_cache + avail_indexes[subblock_no]; yading@10: int c_off = part_sizes_w[block_type]; yading@10: yading@10: mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride; yading@10: if(subblock_no == 3) yading@10: c_off = -1; yading@10: yading@10: if(avail[-1]){ yading@10: A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0]; yading@10: A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1]; yading@10: } yading@10: if(avail[-4]){ yading@10: B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0]; yading@10: B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1]; yading@10: }else{ yading@10: B[0] = A[0]; yading@10: B[1] = A[1]; yading@10: } yading@10: if(!avail[c_off-4]){ yading@10: if(avail[-4] && (avail[-1] || r->rv30)){ yading@10: C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0]; yading@10: C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1]; yading@10: }else{ yading@10: C[0] = A[0]; yading@10: C[1] = A[1]; yading@10: } yading@10: }else{ yading@10: C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0]; yading@10: C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1]; yading@10: } yading@10: mx = mid_pred(A[0], B[0], C[0]); yading@10: my = mid_pred(A[1], B[1], C[1]); yading@10: mx += r->dmv[dmv_no][0]; yading@10: my += r->dmv[dmv_no][1]; yading@10: for(j = 0; j < part_sizes_h[block_type]; j++){ yading@10: for(i = 0; i < part_sizes_w[block_type]; i++){ yading@10: s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx; yading@10: s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my; yading@10: } yading@10: } yading@10: } yading@10: yading@10: #define GET_PTS_DIFF(a, b) ((a - b + 8192) & 0x1FFF) yading@10: yading@10: /** yading@10: * Calculate motion vector component that should be added for direct blocks. yading@10: */ yading@10: static int calc_add_mv(RV34DecContext *r, int dir, int val) yading@10: { yading@10: int mul = dir ? -r->mv_weight2 : r->mv_weight1; yading@10: yading@10: return (val * mul + 0x2000) >> 14; yading@10: } yading@10: yading@10: /** yading@10: * Predict motion vector for B-frame macroblock. yading@10: */ yading@10: static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2], yading@10: int A_avail, int B_avail, int C_avail, yading@10: int *mx, int *my) yading@10: { yading@10: if(A_avail + B_avail + C_avail != 3){ yading@10: *mx = A[0] + B[0] + C[0]; yading@10: *my = A[1] + B[1] + C[1]; yading@10: if(A_avail + B_avail + C_avail == 2){ yading@10: *mx /= 2; yading@10: *my /= 2; yading@10: } yading@10: }else{ yading@10: *mx = mid_pred(A[0], B[0], C[0]); yading@10: *my = mid_pred(A[1], B[1], C[1]); yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * motion vector prediction for B-frames yading@10: */ yading@10: static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir) yading@10: { yading@10: MpegEncContext *s = &r->s; yading@10: int mb_pos = s->mb_x + s->mb_y * s->mb_stride; yading@10: int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride; yading@10: int A[2] = { 0 }, B[2] = { 0 }, C[2] = { 0 }; yading@10: int has_A = 0, has_B = 0, has_C = 0; yading@10: int mx, my; yading@10: int i, j; yading@10: Picture *cur_pic = s->current_picture_ptr; yading@10: const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0; yading@10: int type = cur_pic->mb_type[mb_pos]; yading@10: yading@10: if((r->avail_cache[6-1] & type) & mask){ yading@10: A[0] = cur_pic->motion_val[dir][mv_pos - 1][0]; yading@10: A[1] = cur_pic->motion_val[dir][mv_pos - 1][1]; yading@10: has_A = 1; yading@10: } yading@10: if((r->avail_cache[6-4] & type) & mask){ yading@10: B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0]; yading@10: B[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][1]; yading@10: has_B = 1; yading@10: } yading@10: if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){ yading@10: C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][0]; yading@10: C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][1]; yading@10: has_C = 1; yading@10: }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){ yading@10: C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][0]; yading@10: C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][1]; yading@10: has_C = 1; yading@10: } yading@10: yading@10: rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my); yading@10: yading@10: mx += r->dmv[dir][0]; yading@10: my += r->dmv[dir][1]; yading@10: yading@10: for(j = 0; j < 2; j++){ yading@10: for(i = 0; i < 2; i++){ yading@10: cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx; yading@10: cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my; yading@10: } yading@10: } yading@10: if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){ yading@10: ZERO8x2(cur_pic->motion_val[!dir][mv_pos], s->b8_stride); yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * motion vector prediction - RV3 version yading@10: */ yading@10: static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir) yading@10: { yading@10: MpegEncContext *s = &r->s; yading@10: int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride; yading@10: int A[2] = {0}, B[2], C[2]; yading@10: int i, j, k; yading@10: int mx, my; yading@10: int* avail = r->avail_cache + avail_indexes[0]; yading@10: yading@10: if(avail[-1]){ yading@10: A[0] = s->current_picture_ptr->motion_val[0][mv_pos - 1][0]; yading@10: A[1] = s->current_picture_ptr->motion_val[0][mv_pos - 1][1]; yading@10: } yading@10: if(avail[-4]){ yading@10: B[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][0]; yading@10: B[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][1]; yading@10: }else{ yading@10: B[0] = A[0]; yading@10: B[1] = A[1]; yading@10: } yading@10: if(!avail[-4 + 2]){ yading@10: if(avail[-4] && (avail[-1])){ yading@10: C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][0]; yading@10: C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][1]; yading@10: }else{ yading@10: C[0] = A[0]; yading@10: C[1] = A[1]; yading@10: } yading@10: }else{ yading@10: C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][0]; yading@10: C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][1]; yading@10: } yading@10: mx = mid_pred(A[0], B[0], C[0]); yading@10: my = mid_pred(A[1], B[1], C[1]); yading@10: mx += r->dmv[0][0]; yading@10: my += r->dmv[0][1]; yading@10: for(j = 0; j < 2; j++){ yading@10: for(i = 0; i < 2; i++){ yading@10: for(k = 0; k < 2; k++){ yading@10: s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx; yading@10: s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][1] = my; yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: static const int chroma_coeffs[3] = { 0, 3, 5 }; yading@10: yading@10: /** yading@10: * generic motion compensation function yading@10: * yading@10: * @param r decoder context yading@10: * @param block_type type of the current block yading@10: * @param xoff horizontal offset from the start of the current block yading@10: * @param yoff vertical offset from the start of the current block yading@10: * @param mv_off offset to the motion vector information yading@10: * @param width width of the current partition in 8x8 blocks yading@10: * @param height height of the current partition in 8x8 blocks yading@10: * @param dir motion compensation direction (i.e. from the last or the next reference frame) yading@10: * @param thirdpel motion vectors are specified in 1/3 of pixel yading@10: * @param qpel_mc a set of functions used to perform luma motion compensation yading@10: * @param chroma_mc a set of functions used to perform chroma motion compensation yading@10: */ yading@10: static inline void rv34_mc(RV34DecContext *r, const int block_type, yading@10: const int xoff, const int yoff, int mv_off, yading@10: const int width, const int height, int dir, yading@10: const int thirdpel, int weighted, yading@10: qpel_mc_func (*qpel_mc)[16], yading@10: h264_chroma_mc_func (*chroma_mc)) yading@10: { yading@10: MpegEncContext *s = &r->s; yading@10: uint8_t *Y, *U, *V, *srcY, *srcU, *srcV; yading@10: int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y; yading@10: int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off; yading@10: int is16x16 = 1; yading@10: yading@10: if(thirdpel){ yading@10: int chroma_mx, chroma_my; yading@10: mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24); yading@10: my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24); yading@10: lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3; yading@10: ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3; yading@10: chroma_mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2; yading@10: chroma_my = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2; yading@10: umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24); yading@10: umy = (chroma_my + (3 << 24)) / 3 - (1 << 24); yading@10: uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3]; yading@10: uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3]; yading@10: }else{ yading@10: int cx, cy; yading@10: mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2; yading@10: my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2; yading@10: lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3; yading@10: ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3; yading@10: cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2; yading@10: cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2; yading@10: umx = cx >> 2; yading@10: umy = cy >> 2; yading@10: uvmx = (cx & 3) << 1; yading@10: uvmy = (cy & 3) << 1; yading@10: //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3 yading@10: if(uvmx == 6 && uvmy == 6) yading@10: uvmx = uvmy = 4; yading@10: } yading@10: yading@10: if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME)) { yading@10: /* wait for the referenced mb row to be finished */ yading@10: int mb_row = s->mb_y + ((yoff + my + 5 + 8 * height) >> 4); yading@10: ThreadFrame *f = dir ? &s->next_picture_ptr->tf : &s->last_picture_ptr->tf; yading@10: ff_thread_await_progress(f, mb_row, 0); yading@10: } yading@10: yading@10: dxy = ly*4 + lx; yading@10: srcY = dir ? s->next_picture_ptr->f.data[0] : s->last_picture_ptr->f.data[0]; yading@10: srcU = dir ? s->next_picture_ptr->f.data[1] : s->last_picture_ptr->f.data[1]; yading@10: srcV = dir ? s->next_picture_ptr->f.data[2] : s->last_picture_ptr->f.data[2]; yading@10: src_x = s->mb_x * 16 + xoff + mx; yading@10: src_y = s->mb_y * 16 + yoff + my; yading@10: uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx; yading@10: uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy; yading@10: srcY += src_y * s->linesize + src_x; yading@10: srcU += uvsrc_y * s->uvlinesize + uvsrc_x; yading@10: srcV += uvsrc_y * s->uvlinesize + uvsrc_x; yading@10: if(s->h_edge_pos - (width << 3) < 6 || s->v_edge_pos - (height << 3) < 6 || yading@10: (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4 || yading@10: (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4) { yading@10: uint8_t *uvbuf = s->edge_emu_buffer + 22 * s->linesize; yading@10: yading@10: srcY -= 2 + 2*s->linesize; yading@10: s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6, yading@10: src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos); yading@10: srcY = s->edge_emu_buffer + 2 + 2*s->linesize; yading@10: s->vdsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1, yading@10: uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1); yading@10: s->vdsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1, yading@10: uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1); yading@10: srcU = uvbuf; yading@10: srcV = uvbuf + 16; yading@10: } yading@10: if(!weighted){ yading@10: Y = s->dest[0] + xoff + yoff *s->linesize; yading@10: U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize; yading@10: V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize; yading@10: }else{ yading@10: Y = r->tmp_b_block_y [dir] + xoff + yoff *s->linesize; yading@10: U = r->tmp_b_block_uv[dir*2] + (xoff>>1) + (yoff>>1)*s->uvlinesize; yading@10: V = r->tmp_b_block_uv[dir*2+1] + (xoff>>1) + (yoff>>1)*s->uvlinesize; yading@10: } yading@10: yading@10: if(block_type == RV34_MB_P_16x8){ yading@10: qpel_mc[1][dxy](Y, srcY, s->linesize); yading@10: Y += 8; yading@10: srcY += 8; yading@10: }else if(block_type == RV34_MB_P_8x16){ yading@10: qpel_mc[1][dxy](Y, srcY, s->linesize); yading@10: Y += 8 * s->linesize; yading@10: srcY += 8 * s->linesize; yading@10: } yading@10: is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16); yading@10: qpel_mc[!is16x16][dxy](Y, srcY, s->linesize); yading@10: chroma_mc[2-width] (U, srcU, s->uvlinesize, height*4, uvmx, uvmy); yading@10: chroma_mc[2-width] (V, srcV, s->uvlinesize, height*4, uvmx, uvmy); yading@10: } yading@10: yading@10: static void rv34_mc_1mv(RV34DecContext *r, const int block_type, yading@10: const int xoff, const int yoff, int mv_off, yading@10: const int width, const int height, int dir) yading@10: { yading@10: rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30, 0, yading@10: r->rdsp.put_pixels_tab, yading@10: r->rdsp.put_chroma_pixels_tab); yading@10: } yading@10: yading@10: static void rv4_weight(RV34DecContext *r) yading@10: { yading@10: r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][0](r->s.dest[0], yading@10: r->tmp_b_block_y[0], yading@10: r->tmp_b_block_y[1], yading@10: r->weight1, yading@10: r->weight2, yading@10: r->s.linesize); yading@10: r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[1], yading@10: r->tmp_b_block_uv[0], yading@10: r->tmp_b_block_uv[2], yading@10: r->weight1, yading@10: r->weight2, yading@10: r->s.uvlinesize); yading@10: r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[2], yading@10: r->tmp_b_block_uv[1], yading@10: r->tmp_b_block_uv[3], yading@10: r->weight1, yading@10: r->weight2, yading@10: r->s.uvlinesize); yading@10: } yading@10: yading@10: static void rv34_mc_2mv(RV34DecContext *r, const int block_type) yading@10: { yading@10: int weighted = !r->rv30 && block_type != RV34_MB_B_BIDIR && r->weight1 != 8192; yading@10: yading@10: rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30, weighted, yading@10: r->rdsp.put_pixels_tab, yading@10: r->rdsp.put_chroma_pixels_tab); yading@10: if(!weighted){ yading@10: rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 0, yading@10: r->rdsp.avg_pixels_tab, yading@10: r->rdsp.avg_chroma_pixels_tab); yading@10: }else{ yading@10: rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 1, yading@10: r->rdsp.put_pixels_tab, yading@10: r->rdsp.put_chroma_pixels_tab); yading@10: rv4_weight(r); yading@10: } yading@10: } yading@10: yading@10: static void rv34_mc_2mv_skip(RV34DecContext *r) yading@10: { yading@10: int i, j; yading@10: int weighted = !r->rv30 && r->weight1 != 8192; yading@10: yading@10: for(j = 0; j < 2; j++) yading@10: for(i = 0; i < 2; i++){ yading@10: rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30, yading@10: weighted, yading@10: r->rdsp.put_pixels_tab, yading@10: r->rdsp.put_chroma_pixels_tab); yading@10: rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30, yading@10: weighted, yading@10: weighted ? r->rdsp.put_pixels_tab : r->rdsp.avg_pixels_tab, yading@10: weighted ? r->rdsp.put_chroma_pixels_tab : r->rdsp.avg_chroma_pixels_tab); yading@10: } yading@10: if(weighted) yading@10: rv4_weight(r); yading@10: } yading@10: yading@10: /** number of motion vectors in each macroblock type */ yading@10: static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 }; yading@10: yading@10: /** yading@10: * Decode motion vector differences yading@10: * and perform motion vector reconstruction and motion compensation. yading@10: */ yading@10: static int rv34_decode_mv(RV34DecContext *r, int block_type) yading@10: { yading@10: MpegEncContext *s = &r->s; yading@10: GetBitContext *gb = &s->gb; yading@10: int i, j, k, l; yading@10: int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride; yading@10: int next_bt; yading@10: yading@10: memset(r->dmv, 0, sizeof(r->dmv)); yading@10: for(i = 0; i < num_mvs[block_type]; i++){ yading@10: r->dmv[i][0] = svq3_get_se_golomb(gb); yading@10: r->dmv[i][1] = svq3_get_se_golomb(gb); yading@10: } yading@10: switch(block_type){ yading@10: case RV34_MB_TYPE_INTRA: yading@10: case RV34_MB_TYPE_INTRA16x16: yading@10: ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride); yading@10: return 0; yading@10: case RV34_MB_SKIP: yading@10: if(s->pict_type == AV_PICTURE_TYPE_P){ yading@10: ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride); yading@10: rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0); yading@10: break; yading@10: } yading@10: case RV34_MB_B_DIRECT: yading@10: //surprisingly, it uses motion scheme from next reference frame yading@10: /* wait for the current mb row to be finished */ yading@10: if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME)) yading@10: ff_thread_await_progress(&s->next_picture_ptr->tf, FFMAX(0, s->mb_y-1), 0); yading@10: yading@10: next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride]; yading@10: if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){ yading@10: ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride); yading@10: ZERO8x2(s->current_picture_ptr->motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride); yading@10: }else yading@10: for(j = 0; j < 2; j++) yading@10: for(i = 0; i < 2; i++) yading@10: for(k = 0; k < 2; k++) yading@10: for(l = 0; l < 2; l++) yading@10: s->current_picture_ptr->motion_val[l][mv_pos + i + j*s->b8_stride][k] = calc_add_mv(r, l, s->next_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][k]); yading@10: if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC yading@10: rv34_mc_2mv(r, block_type); yading@10: else yading@10: rv34_mc_2mv_skip(r); yading@10: ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride); yading@10: break; yading@10: case RV34_MB_P_16x16: yading@10: case RV34_MB_P_MIX16x16: yading@10: rv34_pred_mv(r, block_type, 0, 0); yading@10: rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0); yading@10: break; yading@10: case RV34_MB_B_FORWARD: yading@10: case RV34_MB_B_BACKWARD: yading@10: r->dmv[1][0] = r->dmv[0][0]; yading@10: r->dmv[1][1] = r->dmv[0][1]; yading@10: if(r->rv30) yading@10: rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD); yading@10: else yading@10: rv34_pred_mv_b (r, block_type, block_type == RV34_MB_B_BACKWARD); yading@10: rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD); yading@10: break; yading@10: case RV34_MB_P_16x8: yading@10: case RV34_MB_P_8x16: yading@10: rv34_pred_mv(r, block_type, 0, 0); yading@10: rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1); yading@10: if(block_type == RV34_MB_P_16x8){ yading@10: rv34_mc_1mv(r, block_type, 0, 0, 0, 2, 1, 0); yading@10: rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0); yading@10: } yading@10: if(block_type == RV34_MB_P_8x16){ yading@10: rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0); yading@10: rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0); yading@10: } yading@10: break; yading@10: case RV34_MB_B_BIDIR: yading@10: rv34_pred_mv_b (r, block_type, 0); yading@10: rv34_pred_mv_b (r, block_type, 1); yading@10: rv34_mc_2mv (r, block_type); yading@10: break; yading@10: case RV34_MB_P_8x8: yading@10: for(i=0;i< 4;i++){ yading@10: rv34_pred_mv(r, block_type, i, i); yading@10: rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0); yading@10: } yading@10: break; yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: /** @} */ // mv group yading@10: yading@10: /** yading@10: * @name Macroblock reconstruction functions yading@10: * @{ yading@10: */ yading@10: /** mapping of RV30/40 intra prediction types to standard H.264 types */ yading@10: static const int ittrans[9] = { yading@10: DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED, yading@10: VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED, yading@10: }; yading@10: yading@10: /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */ yading@10: static const int ittrans16[4] = { yading@10: DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8, yading@10: }; yading@10: yading@10: /** yading@10: * Perform 4x4 intra prediction. yading@10: */ yading@10: static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right) yading@10: { yading@10: uint8_t *prev = dst - stride + 4; yading@10: uint32_t topleft; yading@10: yading@10: if(!up && !left) yading@10: itype = DC_128_PRED; yading@10: else if(!up){ yading@10: if(itype == VERT_PRED) itype = HOR_PRED; yading@10: if(itype == DC_PRED) itype = LEFT_DC_PRED; yading@10: }else if(!left){ yading@10: if(itype == HOR_PRED) itype = VERT_PRED; yading@10: if(itype == DC_PRED) itype = TOP_DC_PRED; yading@10: if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN; yading@10: } yading@10: if(!down){ yading@10: if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN; yading@10: if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN; yading@10: if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN; yading@10: } yading@10: if(!right && up){ yading@10: topleft = dst[-stride + 3] * 0x01010101u; yading@10: prev = (uint8_t*)&topleft; yading@10: } yading@10: r->h.pred4x4[itype](dst, prev, stride); yading@10: } yading@10: yading@10: static inline int adjust_pred16(int itype, int up, int left) yading@10: { yading@10: if(!up && !left) yading@10: itype = DC_128_PRED8x8; yading@10: else if(!up){ yading@10: if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8; yading@10: if(itype == VERT_PRED8x8) itype = HOR_PRED8x8; yading@10: if(itype == DC_PRED8x8) itype = LEFT_DC_PRED8x8; yading@10: }else if(!left){ yading@10: if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8; yading@10: if(itype == HOR_PRED8x8) itype = VERT_PRED8x8; yading@10: if(itype == DC_PRED8x8) itype = TOP_DC_PRED8x8; yading@10: } yading@10: return itype; yading@10: } yading@10: yading@10: static inline void rv34_process_block(RV34DecContext *r, yading@10: uint8_t *pdst, int stride, yading@10: int fc, int sc, int q_dc, int q_ac) yading@10: { yading@10: MpegEncContext *s = &r->s; yading@10: int16_t *ptr = s->block[0]; yading@10: int has_ac = rv34_decode_block(ptr, &s->gb, r->cur_vlcs, yading@10: fc, sc, q_dc, q_ac, q_ac); yading@10: if(has_ac){ yading@10: r->rdsp.rv34_idct_add(pdst, stride, ptr); yading@10: }else{ yading@10: r->rdsp.rv34_idct_dc_add(pdst, stride, ptr[0]); yading@10: ptr[0] = 0; yading@10: } yading@10: } yading@10: yading@10: static void rv34_output_i16x16(RV34DecContext *r, int8_t *intra_types, int cbp) yading@10: { yading@10: LOCAL_ALIGNED_16(int16_t, block16, [16]); yading@10: MpegEncContext *s = &r->s; yading@10: GetBitContext *gb = &s->gb; yading@10: int q_dc = rv34_qscale_tab[ r->luma_dc_quant_i[s->qscale] ], yading@10: q_ac = rv34_qscale_tab[s->qscale]; yading@10: uint8_t *dst = s->dest[0]; yading@10: int16_t *ptr = s->block[0]; yading@10: int i, j, itype, has_ac; yading@10: yading@10: memset(block16, 0, 16 * sizeof(*block16)); yading@10: yading@10: has_ac = rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac); yading@10: if(has_ac) yading@10: r->rdsp.rv34_inv_transform(block16); yading@10: else yading@10: r->rdsp.rv34_inv_transform_dc(block16); yading@10: yading@10: itype = ittrans16[intra_types[0]]; yading@10: itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]); yading@10: r->h.pred16x16[itype](dst, s->linesize); yading@10: yading@10: for(j = 0; j < 4; j++){ yading@10: for(i = 0; i < 4; i++, cbp >>= 1){ yading@10: int dc = block16[i + j*4]; yading@10: yading@10: if(cbp & 1){ yading@10: has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac); yading@10: }else yading@10: has_ac = 0; yading@10: yading@10: if(has_ac){ yading@10: ptr[0] = dc; yading@10: r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr); yading@10: }else yading@10: r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc); yading@10: } yading@10: yading@10: dst += 4*s->linesize; yading@10: } yading@10: yading@10: itype = ittrans16[intra_types[0]]; yading@10: if(itype == PLANE_PRED8x8) itype = DC_PRED8x8; yading@10: itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]); yading@10: yading@10: q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]]; yading@10: q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]]; yading@10: yading@10: for(j = 1; j < 3; j++){ yading@10: dst = s->dest[j]; yading@10: r->h.pred8x8[itype](dst, s->uvlinesize); yading@10: for(i = 0; i < 4; i++, cbp >>= 1){ yading@10: uint8_t *pdst; yading@10: if(!(cbp & 1)) continue; yading@10: pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize; yading@10: yading@10: rv34_process_block(r, pdst, s->uvlinesize, yading@10: r->chroma_vlc, 1, q_dc, q_ac); yading@10: } yading@10: } yading@10: } yading@10: yading@10: static void rv34_output_intra(RV34DecContext *r, int8_t *intra_types, int cbp) yading@10: { yading@10: MpegEncContext *s = &r->s; yading@10: uint8_t *dst = s->dest[0]; yading@10: int avail[6*8] = {0}; yading@10: int i, j, k; yading@10: int idx, q_ac, q_dc; yading@10: yading@10: // Set neighbour information. yading@10: if(r->avail_cache[1]) yading@10: avail[0] = 1; yading@10: if(r->avail_cache[2]) yading@10: avail[1] = avail[2] = 1; yading@10: if(r->avail_cache[3]) yading@10: avail[3] = avail[4] = 1; yading@10: if(r->avail_cache[4]) yading@10: avail[5] = 1; yading@10: if(r->avail_cache[5]) yading@10: avail[8] = avail[16] = 1; yading@10: if(r->avail_cache[9]) yading@10: avail[24] = avail[32] = 1; yading@10: yading@10: q_ac = rv34_qscale_tab[s->qscale]; yading@10: for(j = 0; j < 4; j++){ yading@10: idx = 9 + j*8; yading@10: for(i = 0; i < 4; i++, cbp >>= 1, dst += 4, idx++){ yading@10: rv34_pred_4x4_block(r, dst, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]); yading@10: avail[idx] = 1; yading@10: if(!(cbp & 1)) continue; yading@10: yading@10: rv34_process_block(r, dst, s->linesize, yading@10: r->luma_vlc, 0, q_ac, q_ac); yading@10: } yading@10: dst += s->linesize * 4 - 4*4; yading@10: intra_types += r->intra_types_stride; yading@10: } yading@10: yading@10: intra_types -= r->intra_types_stride * 4; yading@10: yading@10: q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]]; yading@10: q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]]; yading@10: yading@10: for(k = 0; k < 2; k++){ yading@10: dst = s->dest[1+k]; yading@10: fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4); yading@10: yading@10: for(j = 0; j < 2; j++){ yading@10: int* acache = r->avail_cache + 6 + j*4; yading@10: for(i = 0; i < 2; i++, cbp >>= 1, acache++){ yading@10: int itype = ittrans[intra_types[i*2+j*2*r->intra_types_stride]]; yading@10: rv34_pred_4x4_block(r, dst+4*i, s->uvlinesize, itype, acache[-4], acache[-1], !i && !j, acache[-3]); yading@10: acache[0] = 1; yading@10: yading@10: if(!(cbp&1)) continue; yading@10: yading@10: rv34_process_block(r, dst + 4*i, s->uvlinesize, yading@10: r->chroma_vlc, 1, q_dc, q_ac); yading@10: } yading@10: yading@10: dst += 4*s->uvlinesize; yading@10: } yading@10: } yading@10: } yading@10: yading@10: static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step) yading@10: { yading@10: int d; yading@10: d = motion_val[0][0] - motion_val[-step][0]; yading@10: if(d < -3 || d > 3) yading@10: return 1; yading@10: d = motion_val[0][1] - motion_val[-step][1]; yading@10: if(d < -3 || d > 3) yading@10: return 1; yading@10: return 0; yading@10: } yading@10: yading@10: static int rv34_set_deblock_coef(RV34DecContext *r) yading@10: { yading@10: MpegEncContext *s = &r->s; yading@10: int hmvmask = 0, vmvmask = 0, i, j; yading@10: int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride; yading@10: int16_t (*motion_val)[2] = &s->current_picture_ptr->motion_val[0][midx]; yading@10: for(j = 0; j < 16; j += 8){ yading@10: for(i = 0; i < 2; i++){ yading@10: if(is_mv_diff_gt_3(motion_val + i, 1)) yading@10: vmvmask |= 0x11 << (j + i*2); yading@10: if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride)) yading@10: hmvmask |= 0x03 << (j + i*2); yading@10: } yading@10: motion_val += s->b8_stride; yading@10: } yading@10: if(s->first_slice_line) yading@10: hmvmask &= ~0x000F; yading@10: if(!s->mb_x) yading@10: vmvmask &= ~0x1111; yading@10: if(r->rv30){ //RV30 marks both subblocks on the edge for filtering yading@10: vmvmask |= (vmvmask & 0x4444) >> 1; yading@10: hmvmask |= (hmvmask & 0x0F00) >> 4; yading@10: if(s->mb_x) yading@10: r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3; yading@10: if(!s->first_slice_line) yading@10: r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12; yading@10: } yading@10: return hmvmask | vmvmask; yading@10: } yading@10: yading@10: static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types) yading@10: { yading@10: MpegEncContext *s = &r->s; yading@10: GetBitContext *gb = &s->gb; yading@10: uint8_t *dst = s->dest[0]; yading@10: int16_t *ptr = s->block[0]; yading@10: int mb_pos = s->mb_x + s->mb_y * s->mb_stride; yading@10: int cbp, cbp2; yading@10: int q_dc, q_ac, has_ac; yading@10: int i, j; yading@10: int dist; yading@10: yading@10: // Calculate which neighbours are available. Maybe it's worth optimizing too. yading@10: memset(r->avail_cache, 0, sizeof(r->avail_cache)); yading@10: fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4); yading@10: dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width; yading@10: if(s->mb_x && dist) yading@10: r->avail_cache[5] = yading@10: r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1]; yading@10: if(dist >= s->mb_width) yading@10: r->avail_cache[2] = yading@10: r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride]; yading@10: if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1) yading@10: r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1]; yading@10: if(s->mb_x && dist > s->mb_width) yading@10: r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1]; yading@10: yading@10: s->qscale = r->si.quant; yading@10: cbp = cbp2 = rv34_decode_inter_mb_header(r, intra_types); yading@10: r->cbp_luma [mb_pos] = cbp; yading@10: r->cbp_chroma[mb_pos] = cbp >> 16; yading@10: r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos]; yading@10: s->current_picture_ptr->qscale_table[mb_pos] = s->qscale; yading@10: yading@10: if(cbp == -1) yading@10: return -1; yading@10: yading@10: if (IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){ yading@10: if(r->is16) rv34_output_i16x16(r, intra_types, cbp); yading@10: else rv34_output_intra(r, intra_types, cbp); yading@10: return 0; yading@10: } yading@10: yading@10: if(r->is16){ yading@10: // Only for RV34_MB_P_MIX16x16 yading@10: LOCAL_ALIGNED_16(int16_t, block16, [16]); yading@10: memset(block16, 0, 16 * sizeof(*block16)); yading@10: q_dc = rv34_qscale_tab[ r->luma_dc_quant_p[s->qscale] ]; yading@10: q_ac = rv34_qscale_tab[s->qscale]; yading@10: if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac)) yading@10: r->rdsp.rv34_inv_transform(block16); yading@10: else yading@10: r->rdsp.rv34_inv_transform_dc(block16); yading@10: yading@10: q_ac = rv34_qscale_tab[s->qscale]; yading@10: yading@10: for(j = 0; j < 4; j++){ yading@10: for(i = 0; i < 4; i++, cbp >>= 1){ yading@10: int dc = block16[i + j*4]; yading@10: yading@10: if(cbp & 1){ yading@10: has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac); yading@10: }else yading@10: has_ac = 0; yading@10: yading@10: if(has_ac){ yading@10: ptr[0] = dc; yading@10: r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr); yading@10: }else yading@10: r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc); yading@10: } yading@10: yading@10: dst += 4*s->linesize; yading@10: } yading@10: yading@10: r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1); yading@10: }else{ yading@10: q_ac = rv34_qscale_tab[s->qscale]; yading@10: yading@10: for(j = 0; j < 4; j++){ yading@10: for(i = 0; i < 4; i++, cbp >>= 1){ yading@10: if(!(cbp & 1)) continue; yading@10: yading@10: rv34_process_block(r, dst + 4*i, s->linesize, yading@10: r->luma_vlc, 0, q_ac, q_ac); yading@10: } yading@10: dst += 4*s->linesize; yading@10: } yading@10: } yading@10: yading@10: q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]]; yading@10: q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]]; yading@10: yading@10: for(j = 1; j < 3; j++){ yading@10: dst = s->dest[j]; yading@10: for(i = 0; i < 4; i++, cbp >>= 1){ yading@10: uint8_t *pdst; yading@10: if(!(cbp & 1)) continue; yading@10: pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize; yading@10: yading@10: rv34_process_block(r, pdst, s->uvlinesize, yading@10: r->chroma_vlc, 1, q_dc, q_ac); yading@10: } yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types) yading@10: { yading@10: MpegEncContext *s = &r->s; yading@10: int cbp, dist; yading@10: int mb_pos = s->mb_x + s->mb_y * s->mb_stride; yading@10: yading@10: // Calculate which neighbours are available. Maybe it's worth optimizing too. yading@10: memset(r->avail_cache, 0, sizeof(r->avail_cache)); yading@10: fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4); yading@10: dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width; yading@10: if(s->mb_x && dist) yading@10: r->avail_cache[5] = yading@10: r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1]; yading@10: if(dist >= s->mb_width) yading@10: r->avail_cache[2] = yading@10: r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride]; yading@10: if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1) yading@10: r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1]; yading@10: if(s->mb_x && dist > s->mb_width) yading@10: r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1]; yading@10: yading@10: s->qscale = r->si.quant; yading@10: cbp = rv34_decode_intra_mb_header(r, intra_types); yading@10: r->cbp_luma [mb_pos] = cbp; yading@10: r->cbp_chroma[mb_pos] = cbp >> 16; yading@10: r->deblock_coefs[mb_pos] = 0xFFFF; yading@10: s->current_picture_ptr->qscale_table[mb_pos] = s->qscale; yading@10: yading@10: if(cbp == -1) yading@10: return -1; yading@10: yading@10: if(r->is16){ yading@10: rv34_output_i16x16(r, intra_types, cbp); yading@10: return 0; yading@10: } yading@10: yading@10: rv34_output_intra(r, intra_types, cbp); yading@10: return 0; yading@10: } yading@10: yading@10: static int check_slice_end(RV34DecContext *r, MpegEncContext *s) yading@10: { yading@10: int bits; yading@10: if(s->mb_y >= s->mb_height) yading@10: return 1; yading@10: if(!s->mb_num_left) yading@10: return 1; yading@10: if(r->s.mb_skip_run > 1) yading@10: return 0; yading@10: bits = get_bits_left(&s->gb); yading@10: if(bits <= 0 || (bits < 8 && !show_bits(&s->gb, bits))) yading@10: return 1; yading@10: return 0; yading@10: } yading@10: yading@10: yading@10: static void rv34_decoder_free(RV34DecContext *r) yading@10: { yading@10: av_freep(&r->intra_types_hist); yading@10: r->intra_types = NULL; yading@10: av_freep(&r->tmp_b_block_base); yading@10: av_freep(&r->mb_type); yading@10: av_freep(&r->cbp_luma); yading@10: av_freep(&r->cbp_chroma); yading@10: av_freep(&r->deblock_coefs); yading@10: } yading@10: yading@10: yading@10: static int rv34_decoder_alloc(RV34DecContext *r) yading@10: { yading@10: r->intra_types_stride = r->s.mb_width * 4 + 4; yading@10: yading@10: r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height * yading@10: sizeof(*r->cbp_chroma)); yading@10: r->cbp_luma = av_malloc(r->s.mb_stride * r->s.mb_height * yading@10: sizeof(*r->cbp_luma)); yading@10: r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height * yading@10: sizeof(*r->deblock_coefs)); yading@10: r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 * yading@10: sizeof(*r->intra_types_hist)); yading@10: r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height * yading@10: sizeof(*r->mb_type)); yading@10: yading@10: if (!(r->cbp_chroma && r->cbp_luma && r->deblock_coefs && yading@10: r->intra_types_hist && r->mb_type)) { yading@10: rv34_decoder_free(r); yading@10: return AVERROR(ENOMEM); yading@10: } yading@10: yading@10: r->intra_types = r->intra_types_hist + r->intra_types_stride * 4; yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: yading@10: static int rv34_decoder_realloc(RV34DecContext *r) yading@10: { yading@10: rv34_decoder_free(r); yading@10: return rv34_decoder_alloc(r); yading@10: } yading@10: yading@10: yading@10: static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size) yading@10: { yading@10: MpegEncContext *s = &r->s; yading@10: GetBitContext *gb = &s->gb; yading@10: int mb_pos, slice_type; yading@10: int res; yading@10: yading@10: init_get_bits(&r->s.gb, buf, buf_size*8); yading@10: res = r->parse_slice_header(r, gb, &r->si); yading@10: if(res < 0){ yading@10: av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n"); yading@10: return -1; yading@10: } yading@10: yading@10: slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I; yading@10: if (slice_type != s->pict_type) { yading@10: av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n"); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: if (s->width != r->si.width || s->height != r->si.height) { yading@10: av_log(s->avctx, AV_LOG_ERROR, "Size mismatch\n"); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: yading@10: r->si.end = end; yading@10: s->qscale = r->si.quant; yading@10: s->mb_num_left = r->si.end - r->si.start; yading@10: r->s.mb_skip_run = 0; yading@10: yading@10: mb_pos = s->mb_x + s->mb_y * s->mb_width; yading@10: if(r->si.start != mb_pos){ yading@10: av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos); yading@10: s->mb_x = r->si.start % s->mb_width; yading@10: s->mb_y = r->si.start / s->mb_width; yading@10: } yading@10: memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist)); yading@10: s->first_slice_line = 1; yading@10: s->resync_mb_x = s->mb_x; yading@10: s->resync_mb_y = s->mb_y; yading@10: yading@10: ff_init_block_index(s); yading@10: while(!check_slice_end(r, s)) { yading@10: ff_update_block_index(s); yading@10: yading@10: if(r->si.type) yading@10: res = rv34_decode_inter_macroblock(r, r->intra_types + s->mb_x * 4 + 4); yading@10: else yading@10: res = rv34_decode_intra_macroblock(r, r->intra_types + s->mb_x * 4 + 4); yading@10: if(res < 0){ yading@10: ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_ERROR); yading@10: return -1; yading@10: } yading@10: if (++s->mb_x == s->mb_width) { yading@10: s->mb_x = 0; yading@10: s->mb_y++; yading@10: ff_init_block_index(s); yading@10: yading@10: memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist)); yading@10: memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist)); yading@10: yading@10: if(r->loop_filter && s->mb_y >= 2) yading@10: r->loop_filter(r, s->mb_y - 2); yading@10: yading@10: if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME)) yading@10: ff_thread_report_progress(&s->current_picture_ptr->tf, yading@10: s->mb_y - 2, 0); yading@10: yading@10: } yading@10: if(s->mb_x == s->resync_mb_x) yading@10: s->first_slice_line=0; yading@10: s->mb_num_left--; yading@10: } yading@10: ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END); yading@10: yading@10: return s->mb_y == s->mb_height; yading@10: } yading@10: yading@10: /** @} */ // recons group end yading@10: yading@10: /** yading@10: * Initialize decoder. yading@10: */ yading@10: av_cold int ff_rv34_decode_init(AVCodecContext *avctx) yading@10: { yading@10: RV34DecContext *r = avctx->priv_data; yading@10: MpegEncContext *s = &r->s; yading@10: int ret; yading@10: yading@10: ff_MPV_decode_defaults(s); yading@10: s->avctx = avctx; yading@10: s->out_format = FMT_H263; yading@10: s->codec_id = avctx->codec_id; yading@10: yading@10: s->width = avctx->width; yading@10: s->height = avctx->height; yading@10: yading@10: r->s.avctx = avctx; yading@10: avctx->flags |= CODEC_FLAG_EMU_EDGE; yading@10: r->s.flags |= CODEC_FLAG_EMU_EDGE; yading@10: avctx->pix_fmt = AV_PIX_FMT_YUV420P; yading@10: avctx->has_b_frames = 1; yading@10: s->low_delay = 0; yading@10: yading@10: if ((ret = ff_MPV_common_init(s)) < 0) yading@10: return ret; yading@10: yading@10: ff_h264_pred_init(&r->h, AV_CODEC_ID_RV40, 8, 1); yading@10: yading@10: #if CONFIG_RV30_DECODER yading@10: if (avctx->codec_id == AV_CODEC_ID_RV30) yading@10: ff_rv30dsp_init(&r->rdsp); yading@10: #endif yading@10: #if CONFIG_RV40_DECODER yading@10: if (avctx->codec_id == AV_CODEC_ID_RV40) yading@10: ff_rv40dsp_init(&r->rdsp); yading@10: #endif yading@10: yading@10: if ((ret = rv34_decoder_alloc(r)) < 0) yading@10: return ret; yading@10: yading@10: if(!intra_vlcs[0].cbppattern[0].bits) yading@10: rv34_init_tables(); yading@10: yading@10: avctx->internal->allocate_progress = 1; yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: int ff_rv34_decode_init_thread_copy(AVCodecContext *avctx) yading@10: { yading@10: int err; yading@10: RV34DecContext *r = avctx->priv_data; yading@10: yading@10: r->s.avctx = avctx; yading@10: yading@10: if (avctx->internal->is_copy) { yading@10: r->tmp_b_block_base = NULL; yading@10: if ((err = ff_MPV_common_init(&r->s)) < 0) yading@10: return err; yading@10: if ((err = rv34_decoder_alloc(r)) < 0) yading@10: return err; yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: int ff_rv34_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src) yading@10: { yading@10: RV34DecContext *r = dst->priv_data, *r1 = src->priv_data; yading@10: MpegEncContext * const s = &r->s, * const s1 = &r1->s; yading@10: int err; yading@10: yading@10: if (dst == src || !s1->context_initialized) yading@10: return 0; yading@10: yading@10: if (s->height != s1->height || s->width != s1->width) { yading@10: s->height = s1->height; yading@10: s->width = s1->width; yading@10: if ((err = ff_MPV_common_frame_size_change(s)) < 0) yading@10: return err; yading@10: if ((err = rv34_decoder_realloc(r)) < 0) yading@10: return err; yading@10: } yading@10: yading@10: if ((err = ff_mpeg_update_thread_context(dst, src))) yading@10: return err; yading@10: yading@10: r->cur_pts = r1->cur_pts; yading@10: r->last_pts = r1->last_pts; yading@10: r->next_pts = r1->next_pts; yading@10: yading@10: memset(&r->si, 0, sizeof(r->si)); yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n) yading@10: { yading@10: if(avctx->slice_count) return avctx->slice_offset[n]; yading@10: else return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) : AV_RB32(buf + n*8); yading@10: } yading@10: yading@10: static int finish_frame(AVCodecContext *avctx, AVFrame *pict) yading@10: { yading@10: RV34DecContext *r = avctx->priv_data; yading@10: MpegEncContext *s = &r->s; yading@10: int got_picture = 0, ret; yading@10: yading@10: ff_er_frame_end(&s->er); yading@10: ff_MPV_frame_end(s); yading@10: s->mb_num_left = 0; yading@10: yading@10: if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME)) yading@10: ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0); yading@10: yading@10: if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) { yading@10: if ((ret = av_frame_ref(pict, &s->current_picture_ptr->f)) < 0) yading@10: return ret; yading@10: ff_print_debug_info(s, s->current_picture_ptr, pict); yading@10: ff_mpv_export_qp_table(s, pict, s->current_picture_ptr, FF_QSCALE_TYPE_MPEG1); yading@10: got_picture = 1; yading@10: } else if (s->last_picture_ptr != NULL) { yading@10: if ((ret = av_frame_ref(pict, &s->last_picture_ptr->f)) < 0) yading@10: return ret; yading@10: ff_print_debug_info(s, s->last_picture_ptr, pict); yading@10: ff_mpv_export_qp_table(s, pict, s->last_picture_ptr, FF_QSCALE_TYPE_MPEG1); yading@10: got_picture = 1; yading@10: } yading@10: yading@10: return got_picture; yading@10: } yading@10: yading@10: static AVRational update_sar(int old_w, int old_h, AVRational sar, int new_w, int new_h) yading@10: { yading@10: // attempt to keep aspect during typical resolution switches yading@10: if (!sar.num) yading@10: sar = (AVRational){1, 1}; yading@10: yading@10: sar = av_mul_q(sar, (AVRational){new_h * old_w, new_w * old_h}); yading@10: return sar; yading@10: } yading@10: yading@10: int ff_rv34_decode_frame(AVCodecContext *avctx, yading@10: void *data, int *got_picture_ptr, yading@10: AVPacket *avpkt) yading@10: { yading@10: const uint8_t *buf = avpkt->data; yading@10: int buf_size = avpkt->size; yading@10: RV34DecContext *r = avctx->priv_data; yading@10: MpegEncContext *s = &r->s; yading@10: AVFrame *pict = data; yading@10: SliceInfo si; yading@10: int i, ret; yading@10: int slice_count; yading@10: const uint8_t *slices_hdr = NULL; yading@10: int last = 0; yading@10: yading@10: /* no supplementary picture */ yading@10: if (buf_size == 0) { yading@10: /* special case for last picture */ yading@10: if (s->low_delay==0 && s->next_picture_ptr) { yading@10: if ((ret = av_frame_ref(pict, &s->next_picture_ptr->f)) < 0) yading@10: return ret; yading@10: s->next_picture_ptr = NULL; yading@10: yading@10: *got_picture_ptr = 1; yading@10: } yading@10: return 0; yading@10: } yading@10: yading@10: if(!avctx->slice_count){ yading@10: slice_count = (*buf++) + 1; yading@10: slices_hdr = buf + 4; yading@10: buf += 8 * slice_count; yading@10: buf_size -= 1 + 8 * slice_count; yading@10: }else yading@10: slice_count = avctx->slice_count; yading@10: yading@10: //parse first slice header to check whether this frame can be decoded yading@10: if(get_slice_offset(avctx, slices_hdr, 0) < 0 || yading@10: get_slice_offset(avctx, slices_hdr, 0) > buf_size){ yading@10: av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n"); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), (buf_size-get_slice_offset(avctx, slices_hdr, 0))*8); yading@10: if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){ yading@10: av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n"); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: if ((!s->last_picture_ptr || !s->last_picture_ptr->f.data[0]) && yading@10: si.type == AV_PICTURE_TYPE_B) { yading@10: av_log(avctx, AV_LOG_ERROR, "Invalid decoder state: B-frame without " yading@10: "reference data.\n"); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: if( (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B) yading@10: || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I) yading@10: || avctx->skip_frame >= AVDISCARD_ALL) yading@10: return avpkt->size; yading@10: yading@10: /* first slice */ yading@10: if (si.start == 0) { yading@10: if (s->mb_num_left > 0) { yading@10: av_log(avctx, AV_LOG_ERROR, "New frame but still %d MB left.\n", yading@10: s->mb_num_left); yading@10: ff_er_frame_end(&s->er); yading@10: ff_MPV_frame_end(s); yading@10: } yading@10: yading@10: if (s->width != si.width || s->height != si.height) { yading@10: int err; yading@10: yading@10: av_log(s->avctx, AV_LOG_WARNING, "Changing dimensions to %dx%d\n", yading@10: si.width, si.height); yading@10: yading@10: if (av_image_check_size(si.width, si.height, 0, s->avctx)) yading@10: return AVERROR_INVALIDDATA; yading@10: yading@10: s->avctx->sample_aspect_ratio = update_sar( yading@10: s->width, s->height, s->avctx->sample_aspect_ratio, yading@10: si.width, si.height); yading@10: s->width = si.width; yading@10: s->height = si.height; yading@10: avcodec_set_dimensions(s->avctx, s->width, s->height); yading@10: if ((err = ff_MPV_common_frame_size_change(s)) < 0) yading@10: return err; yading@10: if ((err = rv34_decoder_realloc(r)) < 0) yading@10: return err; yading@10: } yading@10: s->pict_type = si.type ? si.type : AV_PICTURE_TYPE_I; yading@10: if (ff_MPV_frame_start(s, s->avctx) < 0) yading@10: return -1; yading@10: ff_mpeg_er_frame_start(s); yading@10: if (!r->tmp_b_block_base) { yading@10: int i; yading@10: yading@10: r->tmp_b_block_base = av_malloc(s->linesize * 48); yading@10: for (i = 0; i < 2; i++) yading@10: r->tmp_b_block_y[i] = r->tmp_b_block_base yading@10: + i * 16 * s->linesize; yading@10: for (i = 0; i < 4; i++) yading@10: r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize yading@10: + (i >> 1) * 8 * s->uvlinesize yading@10: + (i & 1) * 16; yading@10: } yading@10: r->cur_pts = si.pts; yading@10: if (s->pict_type != AV_PICTURE_TYPE_B) { yading@10: r->last_pts = r->next_pts; yading@10: r->next_pts = r->cur_pts; yading@10: } else { yading@10: int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts); yading@10: int dist0 = GET_PTS_DIFF(r->cur_pts, r->last_pts); yading@10: int dist1 = GET_PTS_DIFF(r->next_pts, r->cur_pts); yading@10: yading@10: if(!refdist){ yading@10: r->mv_weight1 = r->mv_weight2 = r->weight1 = r->weight2 = 8192; yading@10: r->scaled_weight = 0; yading@10: }else{ yading@10: r->mv_weight1 = (dist0 << 14) / refdist; yading@10: r->mv_weight2 = (dist1 << 14) / refdist; yading@10: if((r->mv_weight1|r->mv_weight2) & 511){ yading@10: r->weight1 = r->mv_weight1; yading@10: r->weight2 = r->mv_weight2; yading@10: r->scaled_weight = 0; yading@10: }else{ yading@10: r->weight1 = r->mv_weight1 >> 9; yading@10: r->weight2 = r->mv_weight2 >> 9; yading@10: r->scaled_weight = 1; yading@10: } yading@10: } yading@10: } yading@10: s->mb_x = s->mb_y = 0; yading@10: ff_thread_finish_setup(s->avctx); yading@10: } else if (HAVE_THREADS && yading@10: (s->avctx->active_thread_type & FF_THREAD_FRAME)) { yading@10: av_log(s->avctx, AV_LOG_ERROR, "Decoder needs full frames in frame " yading@10: "multithreading mode (start MB is %d).\n", si.start); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: yading@10: for(i = 0; i < slice_count; i++){ yading@10: int offset = get_slice_offset(avctx, slices_hdr, i); yading@10: int size; yading@10: if(i+1 == slice_count) yading@10: size = buf_size - offset; yading@10: else yading@10: size = get_slice_offset(avctx, slices_hdr, i+1) - offset; yading@10: yading@10: if(offset < 0 || offset > buf_size){ yading@10: av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n"); yading@10: break; yading@10: } yading@10: yading@10: r->si.end = s->mb_width * s->mb_height; yading@10: s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start; yading@10: yading@10: if(i+1 < slice_count){ yading@10: if (get_slice_offset(avctx, slices_hdr, i+1) < 0 || yading@10: get_slice_offset(avctx, slices_hdr, i+1) > buf_size) { yading@10: av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n"); yading@10: break; yading@10: } yading@10: init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, i+1), (buf_size-get_slice_offset(avctx, slices_hdr, i+1))*8); yading@10: if(r->parse_slice_header(r, &r->s.gb, &si) < 0){ yading@10: if(i+2 < slice_count) yading@10: size = get_slice_offset(avctx, slices_hdr, i+2) - offset; yading@10: else yading@10: size = buf_size - offset; yading@10: }else yading@10: r->si.end = si.start; yading@10: } yading@10: if (size < 0 || size > buf_size - offset) { yading@10: av_log(avctx, AV_LOG_ERROR, "Slice size is invalid\n"); yading@10: break; yading@10: } yading@10: last = rv34_decode_slice(r, r->si.end, buf + offset, size); yading@10: if(last) yading@10: break; yading@10: } yading@10: yading@10: if (s->current_picture_ptr) { yading@10: if (last) { yading@10: if(r->loop_filter) yading@10: r->loop_filter(r, s->mb_height - 1); yading@10: yading@10: ret = finish_frame(avctx, pict); yading@10: if (ret < 0) yading@10: return ret; yading@10: *got_picture_ptr = ret; yading@10: } else if (HAVE_THREADS && yading@10: (s->avctx->active_thread_type & FF_THREAD_FRAME)) { yading@10: av_log(avctx, AV_LOG_INFO, "marking unfished frame as finished\n"); yading@10: /* always mark the current frame as finished, frame-mt supports yading@10: * only complete frames */ yading@10: ff_er_frame_end(&s->er); yading@10: ff_MPV_frame_end(s); yading@10: s->mb_num_left = 0; yading@10: ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: } yading@10: yading@10: return avpkt->size; yading@10: } yading@10: yading@10: av_cold int ff_rv34_decode_end(AVCodecContext *avctx) yading@10: { yading@10: RV34DecContext *r = avctx->priv_data; yading@10: yading@10: ff_MPV_common_end(&r->s); yading@10: rv34_decoder_free(r); yading@10: yading@10: return 0; yading@10: }