yading@10: /* yading@10: * Copyright (c) 2003 The FFmpeg Project 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: * How to use this decoder: yading@10: * SVQ3 data is transported within Apple Quicktime files. Quicktime files yading@10: * have stsd atoms to describe media trak properties. A stsd atom for a yading@10: * video trak contains 1 or more ImageDescription atoms. These atoms begin yading@10: * with the 4-byte length of the atom followed by the codec fourcc. Some yading@10: * decoders need information in this atom to operate correctly. Such yading@10: * is the case with SVQ3. In order to get the best use out of this decoder, yading@10: * the calling app must make the SVQ3 ImageDescription atom available yading@10: * via the AVCodecContext's extradata[_size] field: yading@10: * yading@10: * AVCodecContext.extradata = pointer to ImageDescription, first characters yading@10: * are expected to be 'S', 'V', 'Q', and '3', NOT the 4-byte atom length yading@10: * AVCodecContext.extradata_size = size of ImageDescription atom memory yading@10: * buffer (which will be the same as the ImageDescription atom size field yading@10: * from the QT file, minus 4 bytes since the length is missing) yading@10: * yading@10: * You will know you have these parameters passed correctly when the decoder yading@10: * correctly decodes this file: yading@10: * http://samples.mplayerhq.hu/V-codecs/SVQ3/Vertical400kbit.sorenson3.mov yading@10: */ yading@10: #include "internal.h" yading@10: #include "avcodec.h" yading@10: #include "mpegvideo.h" yading@10: #include "h264.h" yading@10: yading@10: #include "h264data.h" // FIXME FIXME FIXME yading@10: yading@10: #include "h264_mvpred.h" yading@10: #include "golomb.h" yading@10: #include "hpeldsp.h" yading@10: #include "rectangle.h" yading@10: #include "vdpau_internal.h" yading@10: yading@10: #if CONFIG_ZLIB yading@10: #include yading@10: #endif yading@10: yading@10: #include "svq1.h" yading@10: #include "svq3.h" yading@10: yading@10: /** yading@10: * @file yading@10: * svq3 decoder. yading@10: */ yading@10: yading@10: typedef struct { yading@10: H264Context h; yading@10: HpelDSPContext hdsp; yading@10: Picture *cur_pic; yading@10: Picture *next_pic; yading@10: Picture *last_pic; yading@10: int halfpel_flag; yading@10: int thirdpel_flag; yading@10: int unknown_flag; yading@10: int next_slice_index; yading@10: uint32_t watermark_key; yading@10: uint8_t *buf; yading@10: int buf_size; yading@10: int adaptive_quant; yading@10: int next_p_frame_damaged; yading@10: int h_edge_pos; yading@10: int v_edge_pos; yading@10: int last_frame_output; yading@10: } SVQ3Context; yading@10: yading@10: #define FULLPEL_MODE 1 yading@10: #define HALFPEL_MODE 2 yading@10: #define THIRDPEL_MODE 3 yading@10: #define PREDICT_MODE 4 yading@10: yading@10: /* dual scan (from some older h264 draft) yading@10: * o-->o-->o o yading@10: * | /| yading@10: * o o o / o yading@10: * | / | |/ | yading@10: * o o o o yading@10: * / yading@10: * o-->o-->o-->o yading@10: */ yading@10: static const uint8_t svq3_scan[16] = { yading@10: 0 + 0 * 4, 1 + 0 * 4, 2 + 0 * 4, 2 + 1 * 4, yading@10: 2 + 2 * 4, 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, yading@10: 0 + 1 * 4, 0 + 2 * 4, 1 + 1 * 4, 1 + 2 * 4, yading@10: 0 + 3 * 4, 1 + 3 * 4, 2 + 3 * 4, 3 + 3 * 4, yading@10: }; yading@10: yading@10: static const uint8_t svq3_pred_0[25][2] = { yading@10: { 0, 0 }, yading@10: { 1, 0 }, { 0, 1 }, yading@10: { 0, 2 }, { 1, 1 }, { 2, 0 }, yading@10: { 3, 0 }, { 2, 1 }, { 1, 2 }, { 0, 3 }, yading@10: { 0, 4 }, { 1, 3 }, { 2, 2 }, { 3, 1 }, { 4, 0 }, yading@10: { 4, 1 }, { 3, 2 }, { 2, 3 }, { 1, 4 }, yading@10: { 2, 4 }, { 3, 3 }, { 4, 2 }, yading@10: { 4, 3 }, { 3, 4 }, yading@10: { 4, 4 } yading@10: }; yading@10: yading@10: static const int8_t svq3_pred_1[6][6][5] = { yading@10: { { 2, -1, -1, -1, -1 }, { 2, 1, -1, -1, -1 }, { 1, 2, -1, -1, -1 }, yading@10: { 2, 1, -1, -1, -1 }, { 1, 2, -1, -1, -1 }, { 1, 2, -1, -1, -1 } }, yading@10: { { 0, 2, -1, -1, -1 }, { 0, 2, 1, 4, 3 }, { 0, 1, 2, 4, 3 }, yading@10: { 0, 2, 1, 4, 3 }, { 2, 0, 1, 3, 4 }, { 0, 4, 2, 1, 3 } }, yading@10: { { 2, 0, -1, -1, -1 }, { 2, 1, 0, 4, 3 }, { 1, 2, 4, 0, 3 }, yading@10: { 2, 1, 0, 4, 3 }, { 2, 1, 4, 3, 0 }, { 1, 2, 4, 0, 3 } }, yading@10: { { 2, 0, -1, -1, -1 }, { 2, 0, 1, 4, 3 }, { 1, 2, 0, 4, 3 }, yading@10: { 2, 1, 0, 4, 3 }, { 2, 1, 3, 4, 0 }, { 2, 4, 1, 0, 3 } }, yading@10: { { 0, 2, -1, -1, -1 }, { 0, 2, 1, 3, 4 }, { 1, 2, 3, 0, 4 }, yading@10: { 2, 0, 1, 3, 4 }, { 2, 1, 3, 0, 4 }, { 2, 0, 4, 3, 1 } }, yading@10: { { 0, 2, -1, -1, -1 }, { 0, 2, 4, 1, 3 }, { 1, 4, 2, 0, 3 }, yading@10: { 4, 2, 0, 1, 3 }, { 2, 0, 1, 4, 3 }, { 4, 2, 1, 0, 3 } }, yading@10: }; yading@10: yading@10: static const struct { yading@10: uint8_t run; yading@10: uint8_t level; yading@10: } svq3_dct_tables[2][16] = { yading@10: { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 }, yading@10: { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } }, yading@10: { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 }, yading@10: { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } } yading@10: }; yading@10: yading@10: static const uint32_t svq3_dequant_coeff[32] = { yading@10: 3881, 4351, 4890, 5481, 6154, 6914, 7761, 8718, yading@10: 9781, 10987, 12339, 13828, 15523, 17435, 19561, 21873, yading@10: 24552, 27656, 30847, 34870, 38807, 43747, 49103, 54683, yading@10: 61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533 yading@10: }; yading@10: yading@10: void ff_svq3_luma_dc_dequant_idct_c(int16_t *output, int16_t *input, int qp) yading@10: { yading@10: const int qmul = svq3_dequant_coeff[qp]; yading@10: #define stride 16 yading@10: int i; yading@10: int temp[16]; yading@10: static const uint8_t x_offset[4] = { 0, 1 * stride, 4 * stride, 5 * stride }; yading@10: yading@10: for (i = 0; i < 4; i++) { yading@10: const int z0 = 13 * (input[4 * i + 0] + input[4 * i + 2]); yading@10: const int z1 = 13 * (input[4 * i + 0] - input[4 * i + 2]); yading@10: const int z2 = 7 * input[4 * i + 1] - 17 * input[4 * i + 3]; yading@10: const int z3 = 17 * input[4 * i + 1] + 7 * input[4 * i + 3]; yading@10: yading@10: temp[4 * i + 0] = z0 + z3; yading@10: temp[4 * i + 1] = z1 + z2; yading@10: temp[4 * i + 2] = z1 - z2; yading@10: temp[4 * i + 3] = z0 - z3; yading@10: } yading@10: yading@10: for (i = 0; i < 4; i++) { yading@10: const int offset = x_offset[i]; yading@10: const int z0 = 13 * (temp[4 * 0 + i] + temp[4 * 2 + i]); yading@10: const int z1 = 13 * (temp[4 * 0 + i] - temp[4 * 2 + i]); yading@10: const int z2 = 7 * temp[4 * 1 + i] - 17 * temp[4 * 3 + i]; yading@10: const int z3 = 17 * temp[4 * 1 + i] + 7 * temp[4 * 3 + i]; yading@10: yading@10: output[stride * 0 + offset] = (z0 + z3) * qmul + 0x80000 >> 20; yading@10: output[stride * 2 + offset] = (z1 + z2) * qmul + 0x80000 >> 20; yading@10: output[stride * 8 + offset] = (z1 - z2) * qmul + 0x80000 >> 20; yading@10: output[stride * 10 + offset] = (z0 - z3) * qmul + 0x80000 >> 20; yading@10: } yading@10: } yading@10: #undef stride yading@10: yading@10: void ff_svq3_add_idct_c(uint8_t *dst, int16_t *block, yading@10: int stride, int qp, int dc) yading@10: { yading@10: const int qmul = svq3_dequant_coeff[qp]; yading@10: int i; yading@10: yading@10: if (dc) { yading@10: dc = 13 * 13 * (dc == 1 ? 1538 * block[0] yading@10: : qmul * (block[0] >> 3) / 2); yading@10: block[0] = 0; yading@10: } yading@10: yading@10: for (i = 0; i < 4; i++) { yading@10: const int z0 = 13 * (block[0 + 4 * i] + block[2 + 4 * i]); yading@10: const int z1 = 13 * (block[0 + 4 * i] - block[2 + 4 * i]); yading@10: const int z2 = 7 * block[1 + 4 * i] - 17 * block[3 + 4 * i]; yading@10: const int z3 = 17 * block[1 + 4 * i] + 7 * block[3 + 4 * i]; yading@10: yading@10: block[0 + 4 * i] = z0 + z3; yading@10: block[1 + 4 * i] = z1 + z2; yading@10: block[2 + 4 * i] = z1 - z2; yading@10: block[3 + 4 * i] = z0 - z3; yading@10: } yading@10: yading@10: for (i = 0; i < 4; i++) { yading@10: const int z0 = 13 * (block[i + 4 * 0] + block[i + 4 * 2]); yading@10: const int z1 = 13 * (block[i + 4 * 0] - block[i + 4 * 2]); yading@10: const int z2 = 7 * block[i + 4 * 1] - 17 * block[i + 4 * 3]; yading@10: const int z3 = 17 * block[i + 4 * 1] + 7 * block[i + 4 * 3]; yading@10: const int rr = (dc + 0x80000); yading@10: yading@10: dst[i + stride * 0] = av_clip_uint8(dst[i + stride * 0] + ((z0 + z3) * qmul + rr >> 20)); yading@10: dst[i + stride * 1] = av_clip_uint8(dst[i + stride * 1] + ((z1 + z2) * qmul + rr >> 20)); yading@10: dst[i + stride * 2] = av_clip_uint8(dst[i + stride * 2] + ((z1 - z2) * qmul + rr >> 20)); yading@10: dst[i + stride * 3] = av_clip_uint8(dst[i + stride * 3] + ((z0 - z3) * qmul + rr >> 20)); yading@10: } yading@10: yading@10: memset(block, 0, 16 * sizeof(int16_t)); yading@10: } yading@10: yading@10: static inline int svq3_decode_block(GetBitContext *gb, int16_t *block, yading@10: int index, const int type) yading@10: { yading@10: static const uint8_t *const scan_patterns[4] = yading@10: { luma_dc_zigzag_scan, zigzag_scan, svq3_scan, chroma_dc_scan }; yading@10: yading@10: int run, level, sign, limit; yading@10: unsigned vlc; yading@10: const int intra = 3 * type >> 2; yading@10: const uint8_t *const scan = scan_patterns[type]; yading@10: yading@10: for (limit = (16 >> intra); index < 16; index = limit, limit += 8) { yading@10: for (; (vlc = svq3_get_ue_golomb(gb)) != 0; index++) { yading@10: if ((int32_t)vlc < 0) yading@10: return -1; yading@10: yading@10: sign = (vlc & 1) ? 0 : -1; yading@10: vlc = vlc + 1 >> 1; yading@10: yading@10: if (type == 3) { yading@10: if (vlc < 3) { yading@10: run = 0; yading@10: level = vlc; yading@10: } else if (vlc < 4) { yading@10: run = 1; yading@10: level = 1; yading@10: } else { yading@10: run = vlc & 0x3; yading@10: level = (vlc + 9 >> 2) - run; yading@10: } yading@10: } else { yading@10: if (vlc < 16U) { yading@10: run = svq3_dct_tables[intra][vlc].run; yading@10: level = svq3_dct_tables[intra][vlc].level; yading@10: } else if (intra) { yading@10: run = vlc & 0x7; yading@10: level = (vlc >> 3) + ((run == 0) ? 8 : ((run < 2) ? 2 : ((run < 5) ? 0 : -1))); yading@10: } else { yading@10: run = vlc & 0xF; yading@10: level = (vlc >> 4) + ((run == 0) ? 4 : ((run < 3) ? 2 : ((run < 10) ? 1 : 0))); yading@10: } yading@10: } yading@10: yading@10: yading@10: if ((index += run) >= limit) yading@10: return -1; yading@10: yading@10: block[scan[index]] = (level ^ sign) - sign; yading@10: } yading@10: yading@10: if (type != 2) { yading@10: break; yading@10: } yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static inline void svq3_mc_dir_part(SVQ3Context *s, yading@10: int x, int y, int width, int height, yading@10: int mx, int my, int dxy, yading@10: int thirdpel, int dir, int avg) yading@10: { yading@10: H264Context *h = &s->h; yading@10: const Picture *pic = (dir == 0) ? s->last_pic : s->next_pic; yading@10: uint8_t *src, *dest; yading@10: int i, emu = 0; yading@10: int blocksize = 2 - (width >> 3); // 16->0, 8->1, 4->2 yading@10: yading@10: mx += x; yading@10: my += y; yading@10: yading@10: if (mx < 0 || mx >= s->h_edge_pos - width - 1 || yading@10: my < 0 || my >= s->v_edge_pos - height - 1) { yading@10: emu = 1; yading@10: mx = av_clip(mx, -16, s->h_edge_pos - width + 15); yading@10: my = av_clip(my, -16, s->v_edge_pos - height + 15); yading@10: } yading@10: yading@10: /* form component predictions */ yading@10: dest = h->cur_pic.f.data[0] + x + y * h->linesize; yading@10: src = pic->f.data[0] + mx + my * h->linesize; yading@10: yading@10: if (emu) { yading@10: h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src, h->linesize, yading@10: width + 1, height + 1, yading@10: mx, my, s->h_edge_pos, s->v_edge_pos); yading@10: src = h->edge_emu_buffer; yading@10: } yading@10: if (thirdpel) yading@10: (avg ? h->dsp.avg_tpel_pixels_tab yading@10: : h->dsp.put_tpel_pixels_tab)[dxy](dest, src, h->linesize, yading@10: width, height); yading@10: else yading@10: (avg ? s->hdsp.avg_pixels_tab yading@10: : s->hdsp.put_pixels_tab)[blocksize][dxy](dest, src, h->linesize, yading@10: height); yading@10: yading@10: if (!(h->flags & CODEC_FLAG_GRAY)) { yading@10: mx = mx + (mx < (int) x) >> 1; yading@10: my = my + (my < (int) y) >> 1; yading@10: width = width >> 1; yading@10: height = height >> 1; yading@10: blocksize++; yading@10: yading@10: for (i = 1; i < 3; i++) { yading@10: dest = h->cur_pic.f.data[i] + (x >> 1) + (y >> 1) * h->uvlinesize; yading@10: src = pic->f.data[i] + mx + my * h->uvlinesize; yading@10: yading@10: if (emu) { yading@10: h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src, h->uvlinesize, yading@10: width + 1, height + 1, yading@10: mx, my, (s->h_edge_pos >> 1), yading@10: s->v_edge_pos >> 1); yading@10: src = h->edge_emu_buffer; yading@10: } yading@10: if (thirdpel) yading@10: (avg ? h->dsp.avg_tpel_pixels_tab yading@10: : h->dsp.put_tpel_pixels_tab)[dxy](dest, src, yading@10: h->uvlinesize, yading@10: width, height); yading@10: else yading@10: (avg ? s->hdsp.avg_pixels_tab yading@10: : s->hdsp.put_pixels_tab)[blocksize][dxy](dest, src, yading@10: h->uvlinesize, yading@10: height); yading@10: } yading@10: } yading@10: } yading@10: yading@10: static inline int svq3_mc_dir(SVQ3Context *s, int size, int mode, yading@10: int dir, int avg) yading@10: { yading@10: int i, j, k, mx, my, dx, dy, x, y; yading@10: H264Context *h = &s->h; yading@10: const int part_width = ((size & 5) == 4) ? 4 : 16 >> (size & 1); yading@10: const int part_height = 16 >> ((unsigned)(size + 1) / 3); yading@10: const int extra_width = (mode == PREDICT_MODE) ? -16 * 6 : 0; yading@10: const int h_edge_pos = 6 * (s->h_edge_pos - part_width) - extra_width; yading@10: const int v_edge_pos = 6 * (s->v_edge_pos - part_height) - extra_width; yading@10: yading@10: for (i = 0; i < 16; i += part_height) yading@10: for (j = 0; j < 16; j += part_width) { yading@10: const int b_xy = (4 * h->mb_x + (j >> 2)) + yading@10: (4 * h->mb_y + (i >> 2)) * h->b_stride; yading@10: int dxy; yading@10: x = 16 * h->mb_x + j; yading@10: y = 16 * h->mb_y + i; yading@10: k = (j >> 2 & 1) + (i >> 1 & 2) + yading@10: (j >> 1 & 4) + (i & 8); yading@10: yading@10: if (mode != PREDICT_MODE) { yading@10: pred_motion(h, k, part_width >> 2, dir, 1, &mx, &my); yading@10: } else { yading@10: mx = s->next_pic->motion_val[0][b_xy][0] << 1; yading@10: my = s->next_pic->motion_val[0][b_xy][1] << 1; yading@10: yading@10: if (dir == 0) { yading@10: mx = mx * h->frame_num_offset / yading@10: h->prev_frame_num_offset + 1 >> 1; yading@10: my = my * h->frame_num_offset / yading@10: h->prev_frame_num_offset + 1 >> 1; yading@10: } else { yading@10: mx = mx * (h->frame_num_offset - h->prev_frame_num_offset) / yading@10: h->prev_frame_num_offset + 1 >> 1; yading@10: my = my * (h->frame_num_offset - h->prev_frame_num_offset) / yading@10: h->prev_frame_num_offset + 1 >> 1; yading@10: } yading@10: } yading@10: yading@10: /* clip motion vector prediction to frame border */ yading@10: mx = av_clip(mx, extra_width - 6 * x, h_edge_pos - 6 * x); yading@10: my = av_clip(my, extra_width - 6 * y, v_edge_pos - 6 * y); yading@10: yading@10: /* get (optional) motion vector differential */ yading@10: if (mode == PREDICT_MODE) { yading@10: dx = dy = 0; yading@10: } else { yading@10: dy = svq3_get_se_golomb(&h->gb); yading@10: dx = svq3_get_se_golomb(&h->gb); yading@10: yading@10: if (dx == INVALID_VLC || dy == INVALID_VLC) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "invalid MV vlc\n"); yading@10: return -1; yading@10: } yading@10: } yading@10: yading@10: /* compute motion vector */ yading@10: if (mode == THIRDPEL_MODE) { yading@10: int fx, fy; yading@10: mx = (mx + 1 >> 1) + dx; yading@10: my = (my + 1 >> 1) + dy; yading@10: fx = (unsigned)(mx + 0x3000) / 3 - 0x1000; yading@10: fy = (unsigned)(my + 0x3000) / 3 - 0x1000; yading@10: dxy = (mx - 3 * fx) + 4 * (my - 3 * fy); yading@10: yading@10: svq3_mc_dir_part(s, x, y, part_width, part_height, yading@10: fx, fy, dxy, 1, dir, avg); yading@10: mx += mx; yading@10: my += my; yading@10: } else if (mode == HALFPEL_MODE || mode == PREDICT_MODE) { yading@10: mx = (unsigned)(mx + 1 + 0x3000) / 3 + dx - 0x1000; yading@10: my = (unsigned)(my + 1 + 0x3000) / 3 + dy - 0x1000; yading@10: dxy = (mx & 1) + 2 * (my & 1); yading@10: yading@10: svq3_mc_dir_part(s, x, y, part_width, part_height, yading@10: mx >> 1, my >> 1, dxy, 0, dir, avg); yading@10: mx *= 3; yading@10: my *= 3; yading@10: } else { yading@10: mx = (unsigned)(mx + 3 + 0x6000) / 6 + dx - 0x1000; yading@10: my = (unsigned)(my + 3 + 0x6000) / 6 + dy - 0x1000; yading@10: yading@10: svq3_mc_dir_part(s, x, y, part_width, part_height, yading@10: mx, my, 0, 0, dir, avg); yading@10: mx *= 6; yading@10: my *= 6; yading@10: } yading@10: yading@10: /* update mv_cache */ yading@10: if (mode != PREDICT_MODE) { yading@10: int32_t mv = pack16to32(mx, my); yading@10: yading@10: if (part_height == 8 && i < 8) { yading@10: AV_WN32A(h->mv_cache[dir][scan8[k] + 1 * 8], mv); yading@10: yading@10: if (part_width == 8 && j < 8) yading@10: AV_WN32A(h->mv_cache[dir][scan8[k] + 1 + 1 * 8], mv); yading@10: } yading@10: if (part_width == 8 && j < 8) yading@10: AV_WN32A(h->mv_cache[dir][scan8[k] + 1], mv); yading@10: if (part_width == 4 || part_height == 4) yading@10: AV_WN32A(h->mv_cache[dir][scan8[k]], mv); yading@10: } yading@10: yading@10: /* write back motion vectors */ yading@10: fill_rectangle(h->cur_pic.motion_val[dir][b_xy], yading@10: part_width >> 2, part_height >> 2, h->b_stride, yading@10: pack16to32(mx, my), 4); yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type) yading@10: { yading@10: H264Context *h = &s->h; yading@10: int i, j, k, m, dir, mode; yading@10: int cbp = 0; yading@10: uint32_t vlc; yading@10: int8_t *top, *left; yading@10: const int mb_xy = h->mb_xy; yading@10: const int b_xy = 4 * h->mb_x + 4 * h->mb_y * h->b_stride; yading@10: yading@10: h->top_samples_available = (h->mb_y == 0) ? 0x33FF : 0xFFFF; yading@10: h->left_samples_available = (h->mb_x == 0) ? 0x5F5F : 0xFFFF; yading@10: h->topright_samples_available = 0xFFFF; yading@10: yading@10: if (mb_type == 0) { /* SKIP */ yading@10: if (h->pict_type == AV_PICTURE_TYPE_P || yading@10: s->next_pic->mb_type[mb_xy] == -1) { yading@10: svq3_mc_dir_part(s, 16 * h->mb_x, 16 * h->mb_y, 16, 16, yading@10: 0, 0, 0, 0, 0, 0); yading@10: yading@10: if (h->pict_type == AV_PICTURE_TYPE_B) yading@10: svq3_mc_dir_part(s, 16 * h->mb_x, 16 * h->mb_y, 16, 16, yading@10: 0, 0, 0, 0, 1, 1); yading@10: yading@10: mb_type = MB_TYPE_SKIP; yading@10: } else { yading@10: mb_type = FFMIN(s->next_pic->mb_type[mb_xy], 6); yading@10: if (svq3_mc_dir(s, mb_type, PREDICT_MODE, 0, 0) < 0) yading@10: return -1; yading@10: if (svq3_mc_dir(s, mb_type, PREDICT_MODE, 1, 1) < 0) yading@10: return -1; yading@10: yading@10: mb_type = MB_TYPE_16x16; yading@10: } yading@10: } else if (mb_type < 8) { /* INTER */ yading@10: if (s->thirdpel_flag && s->halfpel_flag == !get_bits1(&h->gb)) yading@10: mode = THIRDPEL_MODE; yading@10: else if (s->halfpel_flag && yading@10: s->thirdpel_flag == !get_bits1(&h->gb)) yading@10: mode = HALFPEL_MODE; yading@10: else yading@10: mode = FULLPEL_MODE; yading@10: yading@10: /* fill caches */ yading@10: /* note ref_cache should contain here: yading@10: * ???????? yading@10: * ???11111 yading@10: * N??11111 yading@10: * N??11111 yading@10: * N??11111 yading@10: */ yading@10: yading@10: for (m = 0; m < 2; m++) { yading@10: if (h->mb_x > 0 && h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - 1] + 6] != -1) { yading@10: for (i = 0; i < 4; i++) yading@10: AV_COPY32(h->mv_cache[m][scan8[0] - 1 + i * 8], yading@10: h->cur_pic.motion_val[m][b_xy - 1 + i * h->b_stride]); yading@10: } else { yading@10: for (i = 0; i < 4; i++) yading@10: AV_ZERO32(h->mv_cache[m][scan8[0] - 1 + i * 8]); yading@10: } yading@10: if (h->mb_y > 0) { yading@10: memcpy(h->mv_cache[m][scan8[0] - 1 * 8], yading@10: h->cur_pic.motion_val[m][b_xy - h->b_stride], yading@10: 4 * 2 * sizeof(int16_t)); yading@10: memset(&h->ref_cache[m][scan8[0] - 1 * 8], yading@10: (h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride]] == -1) ? PART_NOT_AVAILABLE : 1, 4); yading@10: yading@10: if (h->mb_x < h->mb_width - 1) { yading@10: AV_COPY32(h->mv_cache[m][scan8[0] + 4 - 1 * 8], yading@10: h->cur_pic.motion_val[m][b_xy - h->b_stride + 4]); yading@10: h->ref_cache[m][scan8[0] + 4 - 1 * 8] = yading@10: (h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride + 1] + 6] == -1 || yading@10: h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride]] == -1) ? PART_NOT_AVAILABLE : 1; yading@10: } else yading@10: h->ref_cache[m][scan8[0] + 4 - 1 * 8] = PART_NOT_AVAILABLE; yading@10: if (h->mb_x > 0) { yading@10: AV_COPY32(h->mv_cache[m][scan8[0] - 1 - 1 * 8], yading@10: h->cur_pic.motion_val[m][b_xy - h->b_stride - 1]); yading@10: h->ref_cache[m][scan8[0] - 1 - 1 * 8] = yading@10: (h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride - 1] + 3] == -1) ? PART_NOT_AVAILABLE : 1; yading@10: } else yading@10: h->ref_cache[m][scan8[0] - 1 - 1 * 8] = PART_NOT_AVAILABLE; yading@10: } else yading@10: memset(&h->ref_cache[m][scan8[0] - 1 * 8 - 1], yading@10: PART_NOT_AVAILABLE, 8); yading@10: yading@10: if (h->pict_type != AV_PICTURE_TYPE_B) yading@10: break; yading@10: } yading@10: yading@10: /* decode motion vector(s) and form prediction(s) */ yading@10: if (h->pict_type == AV_PICTURE_TYPE_P) { yading@10: if (svq3_mc_dir(s, mb_type - 1, mode, 0, 0) < 0) yading@10: return -1; yading@10: } else { /* AV_PICTURE_TYPE_B */ yading@10: if (mb_type != 2) { yading@10: if (svq3_mc_dir(s, 0, mode, 0, 0) < 0) yading@10: return -1; yading@10: } else { yading@10: for (i = 0; i < 4; i++) yading@10: memset(h->cur_pic.motion_val[0][b_xy + i * h->b_stride], yading@10: 0, 4 * 2 * sizeof(int16_t)); yading@10: } yading@10: if (mb_type != 1) { yading@10: if (svq3_mc_dir(s, 0, mode, 1, mb_type == 3) < 0) yading@10: return -1; yading@10: } else { yading@10: for (i = 0; i < 4; i++) yading@10: memset(h->cur_pic.motion_val[1][b_xy + i * h->b_stride], yading@10: 0, 4 * 2 * sizeof(int16_t)); yading@10: } yading@10: } yading@10: yading@10: mb_type = MB_TYPE_16x16; yading@10: } else if (mb_type == 8 || mb_type == 33) { /* INTRA4x4 */ yading@10: memset(h->intra4x4_pred_mode_cache, -1, 8 * 5 * sizeof(int8_t)); yading@10: yading@10: if (mb_type == 8) { yading@10: if (h->mb_x > 0) { yading@10: for (i = 0; i < 4; i++) yading@10: h->intra4x4_pred_mode_cache[scan8[0] - 1 + i * 8] = h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - 1] + 6 - i]; yading@10: if (h->intra4x4_pred_mode_cache[scan8[0] - 1] == -1) yading@10: h->left_samples_available = 0x5F5F; yading@10: } yading@10: if (h->mb_y > 0) { yading@10: h->intra4x4_pred_mode_cache[4 + 8 * 0] = h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride] + 0]; yading@10: h->intra4x4_pred_mode_cache[5 + 8 * 0] = h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride] + 1]; yading@10: h->intra4x4_pred_mode_cache[6 + 8 * 0] = h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride] + 2]; yading@10: h->intra4x4_pred_mode_cache[7 + 8 * 0] = h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride] + 3]; yading@10: yading@10: if (h->intra4x4_pred_mode_cache[4 + 8 * 0] == -1) yading@10: h->top_samples_available = 0x33FF; yading@10: } yading@10: yading@10: /* decode prediction codes for luma blocks */ yading@10: for (i = 0; i < 16; i += 2) { yading@10: vlc = svq3_get_ue_golomb(&h->gb); yading@10: yading@10: if (vlc >= 25U) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "luma prediction:%d\n", vlc); yading@10: return -1; yading@10: } yading@10: yading@10: left = &h->intra4x4_pred_mode_cache[scan8[i] - 1]; yading@10: top = &h->intra4x4_pred_mode_cache[scan8[i] - 8]; yading@10: yading@10: left[1] = svq3_pred_1[top[0] + 1][left[0] + 1][svq3_pred_0[vlc][0]]; yading@10: left[2] = svq3_pred_1[top[1] + 1][left[1] + 1][svq3_pred_0[vlc][1]]; yading@10: yading@10: if (left[1] == -1 || left[2] == -1) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "weird prediction\n"); yading@10: return -1; yading@10: } yading@10: } yading@10: } else { /* mb_type == 33, DC_128_PRED block type */ yading@10: for (i = 0; i < 4; i++) yading@10: memset(&h->intra4x4_pred_mode_cache[scan8[0] + 8 * i], DC_PRED, 4); yading@10: } yading@10: yading@10: write_back_intra_pred_mode(h); yading@10: yading@10: if (mb_type == 8) { yading@10: ff_h264_check_intra4x4_pred_mode(h); yading@10: yading@10: h->top_samples_available = (h->mb_y == 0) ? 0x33FF : 0xFFFF; yading@10: h->left_samples_available = (h->mb_x == 0) ? 0x5F5F : 0xFFFF; yading@10: } else { yading@10: for (i = 0; i < 4; i++) yading@10: memset(&h->intra4x4_pred_mode_cache[scan8[0] + 8 * i], DC_128_PRED, 4); yading@10: yading@10: h->top_samples_available = 0x33FF; yading@10: h->left_samples_available = 0x5F5F; yading@10: } yading@10: yading@10: mb_type = MB_TYPE_INTRA4x4; yading@10: } else { /* INTRA16x16 */ yading@10: dir = i_mb_type_info[mb_type - 8].pred_mode; yading@10: dir = (dir >> 1) ^ 3 * (dir & 1) ^ 1; yading@10: yading@10: if ((h->intra16x16_pred_mode = ff_h264_check_intra_pred_mode(h, dir, 0)) == -1) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "check_intra_pred_mode = -1\n"); yading@10: return -1; yading@10: } yading@10: yading@10: cbp = i_mb_type_info[mb_type - 8].cbp; yading@10: mb_type = MB_TYPE_INTRA16x16; yading@10: } yading@10: yading@10: if (!IS_INTER(mb_type) && h->pict_type != AV_PICTURE_TYPE_I) { yading@10: for (i = 0; i < 4; i++) yading@10: memset(h->cur_pic.motion_val[0][b_xy + i * h->b_stride], yading@10: 0, 4 * 2 * sizeof(int16_t)); yading@10: if (h->pict_type == AV_PICTURE_TYPE_B) { yading@10: for (i = 0; i < 4; i++) yading@10: memset(h->cur_pic.motion_val[1][b_xy + i * h->b_stride], yading@10: 0, 4 * 2 * sizeof(int16_t)); yading@10: } yading@10: } yading@10: if (!IS_INTRA4x4(mb_type)) { yading@10: memset(h->intra4x4_pred_mode + h->mb2br_xy[mb_xy], DC_PRED, 8); yading@10: } yading@10: if (!IS_SKIP(mb_type) || h->pict_type == AV_PICTURE_TYPE_B) { yading@10: memset(h->non_zero_count_cache + 8, 0, 14 * 8 * sizeof(uint8_t)); yading@10: } yading@10: yading@10: if (!IS_INTRA16x16(mb_type) && yading@10: (!IS_SKIP(mb_type) || h->pict_type == AV_PICTURE_TYPE_B)) { yading@10: if ((vlc = svq3_get_ue_golomb(&h->gb)) >= 48U){ yading@10: av_log(h->avctx, AV_LOG_ERROR, "cbp_vlc=%d\n", vlc); yading@10: return -1; yading@10: } yading@10: yading@10: cbp = IS_INTRA(mb_type) ? golomb_to_intra4x4_cbp[vlc] yading@10: : golomb_to_inter_cbp[vlc]; yading@10: } yading@10: if (IS_INTRA16x16(mb_type) || yading@10: (h->pict_type != AV_PICTURE_TYPE_I && s->adaptive_quant && cbp)) { yading@10: h->qscale += svq3_get_se_golomb(&h->gb); yading@10: yading@10: if (h->qscale > 31u) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "qscale:%d\n", h->qscale); yading@10: return -1; yading@10: } yading@10: } yading@10: if (IS_INTRA16x16(mb_type)) { yading@10: AV_ZERO128(h->mb_luma_dc[0] + 0); yading@10: AV_ZERO128(h->mb_luma_dc[0] + 8); yading@10: if (svq3_decode_block(&h->gb, h->mb_luma_dc[0], 0, 1)) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "error while decoding intra luma dc\n"); yading@10: return -1; yading@10: } yading@10: } yading@10: yading@10: if (cbp) { yading@10: const int index = IS_INTRA16x16(mb_type) ? 1 : 0; yading@10: const int type = ((h->qscale < 24 && IS_INTRA4x4(mb_type)) ? 2 : 1); yading@10: yading@10: for (i = 0; i < 4; i++) yading@10: if ((cbp & (1 << i))) { yading@10: for (j = 0; j < 4; j++) { yading@10: k = index ? (1 * (j & 1) + 2 * (i & 1) + yading@10: 2 * (j & 2) + 4 * (i & 2)) yading@10: : (4 * i + j); yading@10: h->non_zero_count_cache[scan8[k]] = 1; yading@10: yading@10: if (svq3_decode_block(&h->gb, &h->mb[16 * k], index, type)) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "error while decoding block\n"); yading@10: return -1; yading@10: } yading@10: } yading@10: } yading@10: yading@10: if ((cbp & 0x30)) { yading@10: for (i = 1; i < 3; ++i) yading@10: if (svq3_decode_block(&h->gb, &h->mb[16 * 16 * i], 0, 3)) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "error while decoding chroma dc block\n"); yading@10: return -1; yading@10: } yading@10: yading@10: if ((cbp & 0x20)) { yading@10: for (i = 1; i < 3; i++) { yading@10: for (j = 0; j < 4; j++) { yading@10: k = 16 * i + j; yading@10: h->non_zero_count_cache[scan8[k]] = 1; yading@10: yading@10: if (svq3_decode_block(&h->gb, &h->mb[16 * k], 1, 1)) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "error while decoding chroma ac block\n"); yading@10: return -1; yading@10: } yading@10: } yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: h->cbp = cbp; yading@10: h->cur_pic.mb_type[mb_xy] = mb_type; yading@10: yading@10: if (IS_INTRA(mb_type)) yading@10: h->chroma_pred_mode = ff_h264_check_intra_pred_mode(h, DC_PRED8x8, 1); yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static int svq3_decode_slice_header(AVCodecContext *avctx) yading@10: { yading@10: SVQ3Context *s = avctx->priv_data; yading@10: H264Context *h = &s->h; yading@10: const int mb_xy = h->mb_xy; yading@10: int i, header; yading@10: unsigned slice_id; yading@10: yading@10: header = get_bits(&h->gb, 8); yading@10: yading@10: if (((header & 0x9F) != 1 && (header & 0x9F) != 2) || (header & 0x60) == 0) { yading@10: /* TODO: what? */ yading@10: av_log(avctx, AV_LOG_ERROR, "unsupported slice header (%02X)\n", header); yading@10: return -1; yading@10: } else { yading@10: int length = header >> 5 & 3; yading@10: yading@10: s->next_slice_index = get_bits_count(&h->gb) + yading@10: 8 * show_bits(&h->gb, 8 * length) + yading@10: 8 * length; yading@10: yading@10: if (s->next_slice_index > h->gb.size_in_bits) { yading@10: av_log(avctx, AV_LOG_ERROR, "slice after bitstream end\n"); yading@10: return -1; yading@10: } yading@10: yading@10: h->gb.size_in_bits = s->next_slice_index - 8 * (length - 1); yading@10: skip_bits(&h->gb, 8); yading@10: yading@10: if (s->watermark_key) { yading@10: uint32_t header = AV_RL32(&h->gb.buffer[(get_bits_count(&h->gb) >> 3) + 1]); yading@10: AV_WL32(&h->gb.buffer[(get_bits_count(&h->gb) >> 3) + 1], yading@10: header ^ s->watermark_key); yading@10: } yading@10: if (length > 0) { yading@10: memmove((uint8_t *) &h->gb.buffer[get_bits_count(&h->gb) >> 3], yading@10: &h->gb.buffer[h->gb.size_in_bits >> 3], length - 1); yading@10: } yading@10: skip_bits_long(&h->gb, 0); yading@10: } yading@10: yading@10: if ((slice_id = svq3_get_ue_golomb(&h->gb)) >= 3) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "illegal slice type %d \n", slice_id); yading@10: return -1; yading@10: } yading@10: yading@10: h->slice_type = golomb_to_pict_type[slice_id]; yading@10: yading@10: if ((header & 0x9F) == 2) { yading@10: i = (h->mb_num < 64) ? 6 : (1 + av_log2(h->mb_num - 1)); yading@10: h->mb_skip_run = get_bits(&h->gb, i) - yading@10: (h->mb_y * h->mb_width + h->mb_x); yading@10: } else { yading@10: skip_bits1(&h->gb); yading@10: h->mb_skip_run = 0; yading@10: } yading@10: yading@10: h->slice_num = get_bits(&h->gb, 8); yading@10: h->qscale = get_bits(&h->gb, 5); yading@10: s->adaptive_quant = get_bits1(&h->gb); yading@10: yading@10: /* unknown fields */ yading@10: skip_bits1(&h->gb); yading@10: yading@10: if (s->unknown_flag) yading@10: skip_bits1(&h->gb); yading@10: yading@10: skip_bits1(&h->gb); yading@10: skip_bits(&h->gb, 2); yading@10: yading@10: while (get_bits1(&h->gb)) yading@10: skip_bits(&h->gb, 8); yading@10: yading@10: /* reset intra predictors and invalidate motion vector references */ yading@10: if (h->mb_x > 0) { yading@10: memset(h->intra4x4_pred_mode + h->mb2br_xy[mb_xy - 1] + 3, yading@10: -1, 4 * sizeof(int8_t)); yading@10: memset(h->intra4x4_pred_mode + h->mb2br_xy[mb_xy - h->mb_x], yading@10: -1, 8 * sizeof(int8_t) * h->mb_x); yading@10: } yading@10: if (h->mb_y > 0) { yading@10: memset(h->intra4x4_pred_mode + h->mb2br_xy[mb_xy - h->mb_stride], yading@10: -1, 8 * sizeof(int8_t) * (h->mb_width - h->mb_x)); yading@10: yading@10: if (h->mb_x > 0) yading@10: h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride - 1] + 3] = -1; yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static av_cold int svq3_decode_init(AVCodecContext *avctx) yading@10: { yading@10: SVQ3Context *s = avctx->priv_data; yading@10: H264Context *h = &s->h; yading@10: int m; yading@10: unsigned char *extradata; yading@10: unsigned char *extradata_end; yading@10: unsigned int size; yading@10: int marker_found = 0; yading@10: yading@10: s->cur_pic = av_mallocz(sizeof(*s->cur_pic)); yading@10: s->last_pic = av_mallocz(sizeof(*s->last_pic)); yading@10: s->next_pic = av_mallocz(sizeof(*s->next_pic)); yading@10: if (!s->next_pic || !s->last_pic || !s->cur_pic) { yading@10: av_freep(&s->cur_pic); yading@10: av_freep(&s->last_pic); yading@10: av_freep(&s->next_pic); yading@10: return AVERROR(ENOMEM); yading@10: } yading@10: yading@10: if (ff_h264_decode_init(avctx) < 0) yading@10: return -1; yading@10: yading@10: ff_hpeldsp_init(&s->hdsp, avctx->flags); yading@10: h->flags = avctx->flags; yading@10: h->is_complex = 1; yading@10: h->sps.chroma_format_idc = 1; yading@10: h->picture_structure = PICT_FRAME; yading@10: avctx->pix_fmt = avctx->codec->pix_fmts[0]; yading@10: yading@10: h->chroma_qp[0] = h->chroma_qp[1] = 4; yading@10: h->chroma_x_shift = h->chroma_y_shift = 1; yading@10: yading@10: s->halfpel_flag = 1; yading@10: s->thirdpel_flag = 1; yading@10: s->unknown_flag = 0; yading@10: yading@10: /* prowl for the "SEQH" marker in the extradata */ yading@10: extradata = (unsigned char *)avctx->extradata; yading@10: extradata_end = avctx->extradata + avctx->extradata_size; yading@10: if (extradata) { yading@10: for (m = 0; m + 8 < avctx->extradata_size; m++) { yading@10: if (!memcmp(extradata, "SEQH", 4)) { yading@10: marker_found = 1; yading@10: break; yading@10: } yading@10: extradata++; yading@10: } yading@10: } yading@10: yading@10: /* if a match was found, parse the extra data */ yading@10: if (marker_found) { yading@10: GetBitContext gb; yading@10: int frame_size_code; yading@10: yading@10: size = AV_RB32(&extradata[4]); yading@10: if (size > extradata_end - extradata - 8) yading@10: return AVERROR_INVALIDDATA; yading@10: init_get_bits(&gb, extradata + 8, size * 8); yading@10: yading@10: /* 'frame size code' and optional 'width, height' */ yading@10: frame_size_code = get_bits(&gb, 3); yading@10: switch (frame_size_code) { yading@10: case 0: yading@10: avctx->width = 160; yading@10: avctx->height = 120; yading@10: break; yading@10: case 1: yading@10: avctx->width = 128; yading@10: avctx->height = 96; yading@10: break; yading@10: case 2: yading@10: avctx->width = 176; yading@10: avctx->height = 144; yading@10: break; yading@10: case 3: yading@10: avctx->width = 352; yading@10: avctx->height = 288; yading@10: break; yading@10: case 4: yading@10: avctx->width = 704; yading@10: avctx->height = 576; yading@10: break; yading@10: case 5: yading@10: avctx->width = 240; yading@10: avctx->height = 180; yading@10: break; yading@10: case 6: yading@10: avctx->width = 320; yading@10: avctx->height = 240; yading@10: break; yading@10: case 7: yading@10: avctx->width = get_bits(&gb, 12); yading@10: avctx->height = get_bits(&gb, 12); yading@10: break; yading@10: } yading@10: yading@10: s->halfpel_flag = get_bits1(&gb); yading@10: s->thirdpel_flag = get_bits1(&gb); yading@10: yading@10: /* unknown fields */ yading@10: skip_bits1(&gb); yading@10: skip_bits1(&gb); yading@10: skip_bits1(&gb); yading@10: skip_bits1(&gb); yading@10: yading@10: h->low_delay = get_bits1(&gb); yading@10: yading@10: /* unknown field */ yading@10: skip_bits1(&gb); yading@10: yading@10: while (get_bits1(&gb)) yading@10: skip_bits(&gb, 8); yading@10: yading@10: s->unknown_flag = get_bits1(&gb); yading@10: avctx->has_b_frames = !h->low_delay; yading@10: if (s->unknown_flag) { yading@10: #if CONFIG_ZLIB yading@10: unsigned watermark_width = svq3_get_ue_golomb(&gb); yading@10: unsigned watermark_height = svq3_get_ue_golomb(&gb); yading@10: int u1 = svq3_get_ue_golomb(&gb); yading@10: int u2 = get_bits(&gb, 8); yading@10: int u3 = get_bits(&gb, 2); yading@10: int u4 = svq3_get_ue_golomb(&gb); yading@10: unsigned long buf_len = watermark_width * yading@10: watermark_height * 4; yading@10: int offset = get_bits_count(&gb) + 7 >> 3; yading@10: uint8_t *buf; yading@10: yading@10: if (watermark_height <= 0 || (uint64_t)watermark_width*4 > UINT_MAX/watermark_height) yading@10: return -1; yading@10: yading@10: buf = av_malloc(buf_len); yading@10: av_log(avctx, AV_LOG_DEBUG, "watermark size: %dx%d\n", yading@10: watermark_width, watermark_height); yading@10: av_log(avctx, AV_LOG_DEBUG, yading@10: "u1: %x u2: %x u3: %x compressed data size: %d offset: %d\n", yading@10: u1, u2, u3, u4, offset); yading@10: if (uncompress(buf, &buf_len, extradata + 8 + offset, yading@10: size - offset) != Z_OK) { yading@10: av_log(avctx, AV_LOG_ERROR, yading@10: "could not uncompress watermark logo\n"); yading@10: av_free(buf); yading@10: return -1; yading@10: } yading@10: s->watermark_key = ff_svq1_packet_checksum(buf, buf_len, 0); yading@10: s->watermark_key = s->watermark_key << 16 | s->watermark_key; yading@10: av_log(avctx, AV_LOG_DEBUG, yading@10: "watermark key %#x\n", s->watermark_key); yading@10: av_free(buf); yading@10: #else yading@10: av_log(avctx, AV_LOG_ERROR, yading@10: "this svq3 file contains watermark which need zlib support compiled in\n"); yading@10: return -1; yading@10: #endif yading@10: } yading@10: } yading@10: yading@10: h->width = avctx->width; yading@10: h->height = avctx->height; yading@10: h->mb_width = (h->width + 15) / 16; yading@10: h->mb_height = (h->height + 15) / 16; yading@10: h->mb_stride = h->mb_width + 1; yading@10: h->mb_num = h->mb_width * h->mb_height; yading@10: h->b_stride = 4 * h->mb_width; yading@10: s->h_edge_pos = h->mb_width * 16; yading@10: s->v_edge_pos = h->mb_height * 16; yading@10: yading@10: if (ff_h264_alloc_tables(h) < 0) { yading@10: av_log(avctx, AV_LOG_ERROR, "svq3 memory allocation failed\n"); yading@10: return AVERROR(ENOMEM); yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static void free_picture(AVCodecContext *avctx, Picture *pic) yading@10: { yading@10: int i; yading@10: for (i = 0; i < 2; i++) { yading@10: av_buffer_unref(&pic->motion_val_buf[i]); yading@10: av_buffer_unref(&pic->ref_index_buf[i]); yading@10: } yading@10: av_buffer_unref(&pic->mb_type_buf); yading@10: yading@10: av_frame_unref(&pic->f); yading@10: } yading@10: yading@10: static int get_buffer(AVCodecContext *avctx, Picture *pic) yading@10: { yading@10: SVQ3Context *s = avctx->priv_data; yading@10: H264Context *h = &s->h; yading@10: const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1; yading@10: const int mb_array_size = h->mb_stride * h->mb_height; yading@10: const int b4_stride = h->mb_width * 4 + 1; yading@10: const int b4_array_size = b4_stride * h->mb_height * 4; yading@10: int ret; yading@10: yading@10: if (!pic->motion_val_buf[0]) { yading@10: int i; yading@10: yading@10: pic->mb_type_buf = av_buffer_allocz((big_mb_num + h->mb_stride) * sizeof(uint32_t)); yading@10: if (!pic->mb_type_buf) yading@10: return AVERROR(ENOMEM); yading@10: pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1; yading@10: yading@10: for (i = 0; i < 2; i++) { yading@10: pic->motion_val_buf[i] = av_buffer_allocz(2 * (b4_array_size + 4) * sizeof(int16_t)); yading@10: pic->ref_index_buf[i] = av_buffer_allocz(4 * mb_array_size); yading@10: if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i]) { yading@10: ret = AVERROR(ENOMEM); yading@10: goto fail; yading@10: } yading@10: yading@10: pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4; yading@10: pic->ref_index[i] = pic->ref_index_buf[i]->data; yading@10: } yading@10: } yading@10: pic->reference = !(h->pict_type == AV_PICTURE_TYPE_B); yading@10: yading@10: ret = ff_get_buffer(avctx, &pic->f, yading@10: pic->reference ? AV_GET_BUFFER_FLAG_REF : 0); yading@10: if (ret < 0) yading@10: goto fail; yading@10: yading@10: if (!h->edge_emu_buffer) { yading@10: h->edge_emu_buffer = av_mallocz(pic->f.linesize[0] * 17); yading@10: if (!h->edge_emu_buffer) yading@10: return AVERROR(ENOMEM); yading@10: } yading@10: yading@10: h->linesize = pic->f.linesize[0]; yading@10: h->uvlinesize = pic->f.linesize[1]; yading@10: yading@10: return 0; yading@10: fail: yading@10: free_picture(avctx, pic); yading@10: return ret; yading@10: } yading@10: yading@10: static int svq3_decode_frame(AVCodecContext *avctx, void *data, yading@10: int *got_frame, AVPacket *avpkt) yading@10: { yading@10: SVQ3Context *s = avctx->priv_data; yading@10: H264Context *h = &s->h; yading@10: int buf_size = avpkt->size; yading@10: int left; yading@10: uint8_t *buf; yading@10: int ret, m, i; yading@10: yading@10: /* special case for last picture */ yading@10: if (buf_size == 0) { yading@10: if (s->next_pic->f.data[0] && !h->low_delay && !s->last_frame_output) { yading@10: ret = av_frame_ref(data, &s->next_pic->f); yading@10: if (ret < 0) yading@10: return ret; yading@10: s->last_frame_output = 1; yading@10: *got_frame = 1; yading@10: } yading@10: return 0; yading@10: } yading@10: yading@10: h->mb_x = h->mb_y = h->mb_xy = 0; yading@10: yading@10: if (s->watermark_key) { yading@10: av_fast_malloc(&s->buf, &s->buf_size, yading@10: buf_size+FF_INPUT_BUFFER_PADDING_SIZE); yading@10: if (!s->buf) yading@10: return AVERROR(ENOMEM); yading@10: memcpy(s->buf, avpkt->data, buf_size); yading@10: buf = s->buf; yading@10: } else { yading@10: buf = avpkt->data; yading@10: } yading@10: yading@10: init_get_bits(&h->gb, buf, 8 * buf_size); yading@10: yading@10: if (svq3_decode_slice_header(avctx)) yading@10: return -1; yading@10: yading@10: h->pict_type = h->slice_type; yading@10: yading@10: if (h->pict_type != AV_PICTURE_TYPE_B) yading@10: FFSWAP(Picture*, s->next_pic, s->last_pic); yading@10: yading@10: av_frame_unref(&s->cur_pic->f); yading@10: yading@10: /* for skipping the frame */ yading@10: s->cur_pic->f.pict_type = h->pict_type; yading@10: s->cur_pic->f.key_frame = (h->pict_type == AV_PICTURE_TYPE_I); yading@10: yading@10: ret = get_buffer(avctx, s->cur_pic); yading@10: if (ret < 0) yading@10: return ret; yading@10: yading@10: h->cur_pic_ptr = s->cur_pic; yading@10: av_frame_unref(&h->cur_pic.f); yading@10: h->cur_pic = *s->cur_pic; yading@10: ret = av_frame_ref(&h->cur_pic.f, &s->cur_pic->f); yading@10: if (ret < 0) yading@10: return ret; yading@10: yading@10: for (i = 0; i < 16; i++) { yading@10: h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3); yading@10: h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3); yading@10: } yading@10: for (i = 0; i < 16; i++) { yading@10: h->block_offset[16 + i] = yading@10: h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3); yading@10: h->block_offset[48 + 16 + i] = yading@10: h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3); yading@10: } yading@10: yading@10: if (h->pict_type != AV_PICTURE_TYPE_I) { yading@10: if (!s->last_pic->f.data[0]) { yading@10: av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n"); yading@10: ret = get_buffer(avctx, s->last_pic); yading@10: if (ret < 0) yading@10: return ret; yading@10: memset(s->last_pic->f.data[0], 0, avctx->height * s->last_pic->f.linesize[0]); yading@10: memset(s->last_pic->f.data[1], 0x80, (avctx->height / 2) * yading@10: s->last_pic->f.linesize[1]); yading@10: memset(s->last_pic->f.data[2], 0x80, (avctx->height / 2) * yading@10: s->last_pic->f.linesize[2]); yading@10: } yading@10: yading@10: if (h->pict_type == AV_PICTURE_TYPE_B && !s->next_pic->f.data[0]) { yading@10: av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n"); yading@10: ret = get_buffer(avctx, s->next_pic); yading@10: if (ret < 0) yading@10: return ret; yading@10: memset(s->next_pic->f.data[0], 0, avctx->height * s->next_pic->f.linesize[0]); yading@10: memset(s->next_pic->f.data[1], 0x80, (avctx->height / 2) * yading@10: s->next_pic->f.linesize[1]); yading@10: memset(s->next_pic->f.data[2], 0x80, (avctx->height / 2) * yading@10: s->next_pic->f.linesize[2]); yading@10: } yading@10: } yading@10: yading@10: if (avctx->debug & FF_DEBUG_PICT_INFO) yading@10: av_log(h->avctx, AV_LOG_DEBUG, yading@10: "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n", yading@10: av_get_picture_type_char(h->pict_type), yading@10: s->halfpel_flag, s->thirdpel_flag, yading@10: s->adaptive_quant, h->qscale, h->slice_num); yading@10: yading@10: if (avctx->skip_frame >= AVDISCARD_NONREF && h->pict_type == AV_PICTURE_TYPE_B || yading@10: avctx->skip_frame >= AVDISCARD_NONKEY && h->pict_type != AV_PICTURE_TYPE_I || yading@10: avctx->skip_frame >= AVDISCARD_ALL) yading@10: return 0; yading@10: yading@10: if (s->next_p_frame_damaged) { yading@10: if (h->pict_type == AV_PICTURE_TYPE_B) yading@10: return 0; yading@10: else yading@10: s->next_p_frame_damaged = 0; yading@10: } yading@10: yading@10: if (h->pict_type == AV_PICTURE_TYPE_B) { yading@10: h->frame_num_offset = h->slice_num - h->prev_frame_num; yading@10: yading@10: if (h->frame_num_offset < 0) yading@10: h->frame_num_offset += 256; yading@10: if (h->frame_num_offset == 0 || yading@10: h->frame_num_offset >= h->prev_frame_num_offset) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "error in B-frame picture id\n"); yading@10: return -1; yading@10: } yading@10: } else { yading@10: h->prev_frame_num = h->frame_num; yading@10: h->frame_num = h->slice_num; yading@10: h->prev_frame_num_offset = h->frame_num - h->prev_frame_num; yading@10: yading@10: if (h->prev_frame_num_offset < 0) yading@10: h->prev_frame_num_offset += 256; yading@10: } yading@10: yading@10: for (m = 0; m < 2; m++) { yading@10: int i; yading@10: for (i = 0; i < 4; i++) { yading@10: int j; yading@10: for (j = -1; j < 4; j++) yading@10: h->ref_cache[m][scan8[0] + 8 * i + j] = 1; yading@10: if (i < 3) yading@10: h->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE; yading@10: } yading@10: } yading@10: yading@10: for (h->mb_y = 0; h->mb_y < h->mb_height; h->mb_y++) { yading@10: for (h->mb_x = 0; h->mb_x < h->mb_width; h->mb_x++) { yading@10: unsigned mb_type; yading@10: h->mb_xy = h->mb_x + h->mb_y * h->mb_stride; yading@10: yading@10: if ((get_bits_count(&h->gb) + 7) >= h->gb.size_in_bits && yading@10: ((get_bits_count(&h->gb) & 7) == 0 || yading@10: show_bits(&h->gb, -get_bits_count(&h->gb) & 7) == 0)) { yading@10: skip_bits(&h->gb, s->next_slice_index - get_bits_count(&h->gb)); yading@10: h->gb.size_in_bits = 8 * buf_size; yading@10: yading@10: if (svq3_decode_slice_header(avctx)) yading@10: return -1; yading@10: yading@10: /* TODO: support s->mb_skip_run */ yading@10: } yading@10: yading@10: mb_type = svq3_get_ue_golomb(&h->gb); yading@10: yading@10: if (h->pict_type == AV_PICTURE_TYPE_I) yading@10: mb_type += 8; yading@10: else if (h->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4) yading@10: mb_type += 4; yading@10: if (mb_type > 33 || svq3_decode_mb(s, mb_type)) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "error while decoding MB %d %d\n", h->mb_x, h->mb_y); yading@10: return -1; yading@10: } yading@10: yading@10: if (mb_type != 0 || h->cbp) yading@10: ff_h264_hl_decode_mb(h); yading@10: yading@10: if (h->pict_type != AV_PICTURE_TYPE_B && !h->low_delay) yading@10: h->cur_pic.mb_type[h->mb_x + h->mb_y * h->mb_stride] = yading@10: (h->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1; yading@10: } yading@10: yading@10: ff_draw_horiz_band(avctx, NULL, s->cur_pic, s->last_pic->f.data[0] ? s->last_pic : NULL, yading@10: 16 * h->mb_y, 16, h->picture_structure, 0, 0, yading@10: h->low_delay, h->mb_height * 16, h->mb_width * 16); yading@10: } yading@10: yading@10: left = buf_size*8 - get_bits_count(&h->gb); yading@10: yading@10: if (h->mb_y != h->mb_height || h->mb_x != h->mb_width) { yading@10: av_log(avctx, AV_LOG_INFO, "frame num %d incomplete pic x %d y %d left %d\n", avctx->frame_number, h->mb_y, h->mb_x, left); yading@10: //av_hex_dump(stderr, buf+buf_size-8, 8); yading@10: } yading@10: yading@10: if (left < 0) { yading@10: av_log(avctx, AV_LOG_ERROR, "frame num %d left %d\n", avctx->frame_number, left); yading@10: return -1; yading@10: } yading@10: yading@10: if (h->pict_type == AV_PICTURE_TYPE_B || h->low_delay) yading@10: ret = av_frame_ref(data, &s->cur_pic->f); yading@10: else if (s->last_pic->f.data[0]) yading@10: ret = av_frame_ref(data, &s->last_pic->f); yading@10: if (ret < 0) yading@10: return ret; yading@10: yading@10: /* Do not output the last pic after seeking. */ yading@10: if (s->last_pic->f.data[0] || h->low_delay) yading@10: *got_frame = 1; yading@10: yading@10: if (h->pict_type != AV_PICTURE_TYPE_B) { yading@10: FFSWAP(Picture*, s->cur_pic, s->next_pic); yading@10: } else { yading@10: av_frame_unref(&s->cur_pic->f); yading@10: } yading@10: yading@10: return buf_size; yading@10: } yading@10: yading@10: static int svq3_decode_end(AVCodecContext *avctx) yading@10: { yading@10: SVQ3Context *s = avctx->priv_data; yading@10: H264Context *h = &s->h; yading@10: yading@10: free_picture(avctx, s->cur_pic); yading@10: free_picture(avctx, s->next_pic); yading@10: free_picture(avctx, s->last_pic); yading@10: av_freep(&s->cur_pic); yading@10: av_freep(&s->next_pic); yading@10: av_freep(&s->last_pic); yading@10: yading@10: av_frame_unref(&h->cur_pic.f); yading@10: yading@10: ff_h264_free_context(h); yading@10: yading@10: av_freep(&s->buf); yading@10: s->buf_size = 0; yading@10: av_freep(&h->edge_emu_buffer); yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: AVCodec ff_svq3_decoder = { yading@10: .name = "svq3", yading@10: .type = AVMEDIA_TYPE_VIDEO, yading@10: .id = AV_CODEC_ID_SVQ3, yading@10: .priv_data_size = sizeof(SVQ3Context), yading@10: .init = svq3_decode_init, yading@10: .close = svq3_decode_end, yading@10: .decode = svq3_decode_frame, yading@10: .capabilities = CODEC_CAP_DRAW_HORIZ_BAND | yading@10: CODEC_CAP_DR1 | yading@10: CODEC_CAP_DELAY, yading@10: .long_name = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"), yading@10: .pix_fmts = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P, yading@10: AV_PIX_FMT_NONE}, yading@10: };