X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fsvq3.c;h=51657372ae8cf794989c04cf061d9a47faed2f35;hb=e807491fc6a336e4becc0cbc981274a8fde18aba;hp=316699c37c38147c5804ee592ef07a2a0950eccc;hpb=e7226984ac13aacb84eae77a372df8ff7685848f;p=ffmpeg diff --git a/libavcodec/svq3.c b/libavcodec/svq3.c index 316699c37c3..51657372ae8 100644 --- a/libavcodec/svq3.c +++ b/libavcodec/svq3.c @@ -46,13 +46,11 @@ #include "internal.h" #include "avcodec.h" #include "mpegutils.h" -#include "h264.h" - -#include "h264data.h" // FIXME FIXME FIXME - -#include "h264_mvpred.h" +#include "h264dec.h" +#include "h264data.h" #include "golomb.h" #include "hpeldsp.h" +#include "mathops.h" #include "rectangle.h" #include "tpeldsp.h" @@ -61,30 +59,89 @@ #endif #include "svq1.h" -#include "svq3.h" /** * @file * svq3 decoder. */ +typedef struct SVQ3Frame { + AVFrame *f; + + AVBufferRef *motion_val_buf[2]; + int16_t (*motion_val[2])[2]; + + AVBufferRef *mb_type_buf; + uint32_t *mb_type; + + + AVBufferRef *ref_index_buf[2]; + int8_t *ref_index[2]; +} SVQ3Frame; + typedef struct SVQ3Context { - H264Context h; + AVCodecContext *avctx; + + H264DSPContext h264dsp; + H264PredContext hpc; HpelDSPContext hdsp; TpelDSPContext tdsp; - H264Picture *cur_pic; - H264Picture *next_pic; - H264Picture *last_pic; + VideoDSPContext vdsp; + + SVQ3Frame *cur_pic; + SVQ3Frame *next_pic; + SVQ3Frame *last_pic; + GetBitContext gb; + GetBitContext gb_slice; + uint8_t *slice_buf; + int slice_size; int halfpel_flag; int thirdpel_flag; int unknown_flag; - int next_slice_index; uint32_t watermark_key; int adaptive_quant; int next_p_frame_damaged; int h_edge_pos; int v_edge_pos; int last_frame_output; + int slice_num; + int qscale; + int cbp; + int frame_num; + int frame_num_offset; + int prev_frame_num_offset; + int prev_frame_num; + + enum AVPictureType pict_type; + int low_delay; + + int mb_x, mb_y; + int mb_xy; + int mb_width, mb_height; + int mb_stride, mb_num; + int b_stride; + + uint32_t *mb2br_xy; + + int chroma_pred_mode; + int intra16x16_pred_mode; + + int8_t intra4x4_pred_mode_cache[5 * 8]; + int8_t (*intra4x4_pred_mode); + + unsigned int top_samples_available; + unsigned int topright_samples_available; + unsigned int left_samples_available; + + uint8_t *edge_emu_buffer; + + DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2]; + DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8]; + DECLARE_ALIGNED(16, int16_t, mb)[16 * 48 * 2]; + DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2]; + DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8]; + uint32_t dequant4_coeff[QP_MAX_NUM + 1][16]; + int block_offset[2 * (16 * 3)]; } SVQ3Context; #define FULLPEL_MODE 1 @@ -92,7 +149,7 @@ typedef struct SVQ3Context { #define THIRDPEL_MODE 3 #define PREDICT_MODE 4 -/* dual scan (from some older h264 draft) +/* dual scan (from some older H.264 draft) * o-->o-->o o * | /| * o o o / o @@ -159,7 +216,7 @@ static const uint32_t svq3_dequant_coeff[32] = { 61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533 }; -void ff_svq3_luma_dc_dequant_idct_c(int16_t *output, int16_t *input, int qp) +static void svq3_luma_dc_dequant_idct_c(int16_t *output, int16_t *input, int qp) { const int qmul = svq3_dequant_coeff[qp]; #define stride 16 @@ -194,8 +251,8 @@ void ff_svq3_luma_dc_dequant_idct_c(int16_t *output, int16_t *input, int qp) } #undef stride -void ff_svq3_add_idct_c(uint8_t *dst, int16_t *block, - int stride, int qp, int dc) +static void svq3_add_idct_c(uint8_t *dst, int16_t *block, + int stride, int qp, int dc) { const int qmul = svq3_dequant_coeff[qp]; int i; @@ -237,8 +294,9 @@ void ff_svq3_add_idct_c(uint8_t *dst, int16_t *block, static inline int svq3_decode_block(GetBitContext *gb, int16_t *block, int index, const int type) { - static const uint8_t *const scan_patterns[4] = - { luma_dc_zigzag_scan, zigzag_scan, svq3_scan, chroma_dc_scan }; + static const uint8_t *const scan_patterns[4] = { + luma_dc_zigzag_scan, ff_zigzag_scan, svq3_scan, ff_h264_chroma_dc_scan + }; int run, level, limit; unsigned vlc; @@ -246,7 +304,7 @@ static inline int svq3_decode_block(GetBitContext *gb, int16_t *block, const uint8_t *const scan = scan_patterns[type]; for (limit = (16 >> intra); index < 16; index = limit, limit += 8) { - for (; (vlc = svq3_get_ue_golomb(gb)) != 0; index++) { + for (; (vlc = get_interleaved_ue_golomb(gb)) != 0; index++) { int sign = (vlc & 1) ? 0 : -1; vlc = vlc + 1 >> 1; @@ -290,16 +348,88 @@ static inline int svq3_decode_block(GetBitContext *gb, int16_t *block, return 0; } +static av_always_inline int +svq3_fetch_diagonal_mv(const SVQ3Context *s, const int16_t **C, + int i, int list, int part_width) +{ + const int topright_ref = s->ref_cache[list][i - 8 + part_width]; + + if (topright_ref != PART_NOT_AVAILABLE) { + *C = s->mv_cache[list][i - 8 + part_width]; + return topright_ref; + } else { + *C = s->mv_cache[list][i - 8 - 1]; + return s->ref_cache[list][i - 8 - 1]; + } +} + +/** + * Get the predicted MV. + * @param n the block index + * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4) + * @param mx the x component of the predicted motion vector + * @param my the y component of the predicted motion vector + */ +static av_always_inline void svq3_pred_motion(const SVQ3Context *s, int n, + int part_width, int list, + int ref, int *const mx, int *const my) +{ + const int index8 = scan8[n]; + const int top_ref = s->ref_cache[list][index8 - 8]; + const int left_ref = s->ref_cache[list][index8 - 1]; + const int16_t *const A = s->mv_cache[list][index8 - 1]; + const int16_t *const B = s->mv_cache[list][index8 - 8]; + const int16_t *C; + int diagonal_ref, match_count; + +/* mv_cache + * B . . A T T T T + * U . . L . . , . + * U . . L . . . . + * U . . L . . , . + * . . . L . . . . + */ + + diagonal_ref = svq3_fetch_diagonal_mv(s, &C, index8, list, part_width); + match_count = (diagonal_ref == ref) + (top_ref == ref) + (left_ref == ref); + if (match_count > 1) { //most common + *mx = mid_pred(A[0], B[0], C[0]); + *my = mid_pred(A[1], B[1], C[1]); + } else if (match_count == 1) { + if (left_ref == ref) { + *mx = A[0]; + *my = A[1]; + } else if (top_ref == ref) { + *mx = B[0]; + *my = B[1]; + } else { + *mx = C[0]; + *my = C[1]; + } + } else { + if (top_ref == PART_NOT_AVAILABLE && + diagonal_ref == PART_NOT_AVAILABLE && + left_ref != PART_NOT_AVAILABLE) { + *mx = A[0]; + *my = A[1]; + } else { + *mx = mid_pred(A[0], B[0], C[0]); + *my = mid_pred(A[1], B[1], C[1]); + } + } +} + static inline void svq3_mc_dir_part(SVQ3Context *s, int x, int y, int width, int height, int mx, int my, int dxy, int thirdpel, int dir, int avg) { - H264Context *h = &s->h; - const H264Picture *pic = (dir == 0) ? s->last_pic : s->next_pic; + const SVQ3Frame *pic = (dir == 0) ? s->last_pic : s->next_pic; uint8_t *src, *dest; int i, emu = 0; int blocksize = 2 - (width >> 3); // 16->0, 8->1, 4->2 + int linesize = s->cur_pic->f->linesize[0]; + int uvlinesize = s->cur_pic->f->linesize[1]; mx += x; my += y; @@ -312,26 +442,26 @@ static inline void svq3_mc_dir_part(SVQ3Context *s, } /* form component predictions */ - dest = h->cur_pic.f.data[0] + x + y * h->linesize; - src = pic->f.data[0] + mx + my * h->linesize; + dest = s->cur_pic->f->data[0] + x + y * linesize; + src = pic->f->data[0] + mx + my * linesize; if (emu) { - h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src, - h->linesize, h->linesize, + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src, + linesize, linesize, width + 1, height + 1, mx, my, s->h_edge_pos, s->v_edge_pos); - src = h->edge_emu_buffer; + src = s->edge_emu_buffer; } if (thirdpel) (avg ? s->tdsp.avg_tpel_pixels_tab - : s->tdsp.put_tpel_pixels_tab)[dxy](dest, src, h->linesize, + : s->tdsp.put_tpel_pixels_tab)[dxy](dest, src, linesize, width, height); else (avg ? s->hdsp.avg_pixels_tab - : s->hdsp.put_pixels_tab)[blocksize][dxy](dest, src, h->linesize, + : s->hdsp.put_pixels_tab)[blocksize][dxy](dest, src, linesize, height); - if (!(h->flags & CODEC_FLAG_GRAY)) { + if (!(s->avctx->flags & AV_CODEC_FLAG_GRAY)) { mx = mx + (mx < (int) x) >> 1; my = my + (my < (int) y) >> 1; width = width >> 1; @@ -339,26 +469,26 @@ static inline void svq3_mc_dir_part(SVQ3Context *s, blocksize++; for (i = 1; i < 3; i++) { - dest = h->cur_pic.f.data[i] + (x >> 1) + (y >> 1) * h->uvlinesize; - src = pic->f.data[i] + mx + my * h->uvlinesize; + dest = s->cur_pic->f->data[i] + (x >> 1) + (y >> 1) * uvlinesize; + src = pic->f->data[i] + mx + my * uvlinesize; if (emu) { - h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src, - h->uvlinesize, h->uvlinesize, + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src, + uvlinesize, uvlinesize, width + 1, height + 1, mx, my, (s->h_edge_pos >> 1), s->v_edge_pos >> 1); - src = h->edge_emu_buffer; + src = s->edge_emu_buffer; } if (thirdpel) (avg ? s->tdsp.avg_tpel_pixels_tab : s->tdsp.put_tpel_pixels_tab)[dxy](dest, src, - h->uvlinesize, + uvlinesize, width, height); else (avg ? s->hdsp.avg_pixels_tab : s->hdsp.put_pixels_tab)[blocksize][dxy](dest, src, - h->uvlinesize, + uvlinesize, height); } } @@ -368,8 +498,6 @@ static inline int svq3_mc_dir(SVQ3Context *s, int size, int mode, int dir, int avg) { int i, j, k, mx, my, dx, dy, x, y; - H264Context *h = &s->h; - H264SliceContext *sl = &h->slice_ctx[0]; const int part_width = ((size & 5) == 4) ? 4 : 16 >> (size & 1); const int part_height = 16 >> ((unsigned)(size + 1) / 3); const int extra_width = (mode == PREDICT_MODE) ? -16 * 6 : 0; @@ -378,30 +506,30 @@ static inline int svq3_mc_dir(SVQ3Context *s, int size, int mode, for (i = 0; i < 16; i += part_height) for (j = 0; j < 16; j += part_width) { - const int b_xy = (4 * h->mb_x + (j >> 2)) + - (4 * h->mb_y + (i >> 2)) * h->b_stride; + const int b_xy = (4 * s->mb_x + (j >> 2)) + + (4 * s->mb_y + (i >> 2)) * s->b_stride; int dxy; - x = 16 * h->mb_x + j; - y = 16 * h->mb_y + i; + x = 16 * s->mb_x + j; + y = 16 * s->mb_y + i; k = (j >> 2 & 1) + (i >> 1 & 2) + (j >> 1 & 4) + (i & 8); if (mode != PREDICT_MODE) { - pred_motion(h, sl, k, part_width >> 2, dir, 1, &mx, &my); + svq3_pred_motion(s, k, part_width >> 2, dir, 1, &mx, &my); } else { mx = s->next_pic->motion_val[0][b_xy][0] << 1; my = s->next_pic->motion_val[0][b_xy][1] << 1; if (dir == 0) { - mx = mx * h->frame_num_offset / - h->prev_frame_num_offset + 1 >> 1; - my = my * h->frame_num_offset / - h->prev_frame_num_offset + 1 >> 1; + mx = mx * s->frame_num_offset / + s->prev_frame_num_offset + 1 >> 1; + my = my * s->frame_num_offset / + s->prev_frame_num_offset + 1 >> 1; } else { - mx = mx * (h->frame_num_offset - h->prev_frame_num_offset) / - h->prev_frame_num_offset + 1 >> 1; - my = my * (h->frame_num_offset - h->prev_frame_num_offset) / - h->prev_frame_num_offset + 1 >> 1; + mx = mx * (s->frame_num_offset - s->prev_frame_num_offset) / + s->prev_frame_num_offset + 1 >> 1; + my = my * (s->frame_num_offset - s->prev_frame_num_offset) / + s->prev_frame_num_offset + 1 >> 1; } } @@ -413,11 +541,11 @@ static inline int svq3_mc_dir(SVQ3Context *s, int size, int mode, if (mode == PREDICT_MODE) { dx = dy = 0; } else { - dy = svq3_get_se_golomb(&h->gb); - dx = svq3_get_se_golomb(&h->gb); + dy = get_interleaved_se_golomb(&s->gb_slice); + dx = get_interleaved_se_golomb(&s->gb_slice); if (dx == INVALID_VLC || dy == INVALID_VLC) { - av_log(h->avctx, AV_LOG_ERROR, "invalid MV vlc\n"); + av_log(s->avctx, AV_LOG_ERROR, "invalid MV vlc\n"); return -1; } } @@ -459,49 +587,149 @@ static inline int svq3_mc_dir(SVQ3Context *s, int size, int mode, int32_t mv = pack16to32(mx, my); if (part_height == 8 && i < 8) { - AV_WN32A(sl->mv_cache[dir][scan8[k] + 1 * 8], mv); + AV_WN32A(s->mv_cache[dir][scan8[k] + 1 * 8], mv); if (part_width == 8 && j < 8) - AV_WN32A(sl->mv_cache[dir][scan8[k] + 1 + 1 * 8], mv); + AV_WN32A(s->mv_cache[dir][scan8[k] + 1 + 1 * 8], mv); } if (part_width == 8 && j < 8) - AV_WN32A(sl->mv_cache[dir][scan8[k] + 1], mv); + AV_WN32A(s->mv_cache[dir][scan8[k] + 1], mv); if (part_width == 4 || part_height == 4) - AV_WN32A(sl->mv_cache[dir][scan8[k]], mv); + AV_WN32A(s->mv_cache[dir][scan8[k]], mv); } /* write back motion vectors */ - fill_rectangle(h->cur_pic.motion_val[dir][b_xy], - part_width >> 2, part_height >> 2, h->b_stride, + fill_rectangle(s->cur_pic->motion_val[dir][b_xy], + part_width >> 2, part_height >> 2, s->b_stride, pack16to32(mx, my), 4); } return 0; } +static av_always_inline void hl_decode_mb_idct_luma(SVQ3Context *s, + int mb_type, const int *block_offset, + int linesize, uint8_t *dest_y) +{ + int i; + if (!IS_INTRA4x4(mb_type)) { + for (i = 0; i < 16; i++) + if (s->non_zero_count_cache[scan8[i]] || s->mb[i * 16]) { + uint8_t *const ptr = dest_y + block_offset[i]; + svq3_add_idct_c(ptr, s->mb + i * 16, linesize, + s->qscale, IS_INTRA(mb_type) ? 1 : 0); + } + } +} + +static av_always_inline void hl_decode_mb_predict_luma(SVQ3Context *s, + int mb_type, + const int *block_offset, + int linesize, + uint8_t *dest_y) +{ + int i; + int qscale = s->qscale; + + if (IS_INTRA4x4(mb_type)) { + for (i = 0; i < 16; i++) { + uint8_t *const ptr = dest_y + block_offset[i]; + const int dir = s->intra4x4_pred_mode_cache[scan8[i]]; + + uint8_t *topright; + int nnz, tr; + if (dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED) { + const int topright_avail = (s->topright_samples_available << i) & 0x8000; + assert(s->mb_y || linesize <= block_offset[i]); + if (!topright_avail) { + tr = ptr[3 - linesize] * 0x01010101u; + topright = (uint8_t *)&tr; + } else + topright = ptr + 4 - linesize; + } else + topright = NULL; + + s->hpc.pred4x4[dir](ptr, topright, linesize); + nnz = s->non_zero_count_cache[scan8[i]]; + if (nnz) { + svq3_add_idct_c(ptr, s->mb + i * 16, linesize, qscale, 0); + } + } + } else { + s->hpc.pred16x16[s->intra16x16_pred_mode](dest_y, linesize); + svq3_luma_dc_dequant_idct_c(s->mb, s->mb_luma_dc[0], qscale); + } +} + +static void hl_decode_mb(SVQ3Context *s) +{ + const int mb_x = s->mb_x; + const int mb_y = s->mb_y; + const int mb_xy = s->mb_xy; + const int mb_type = s->cur_pic->mb_type[mb_xy]; + uint8_t *dest_y, *dest_cb, *dest_cr; + int linesize, uvlinesize; + int i, j; + const int *block_offset = &s->block_offset[0]; + const int block_h = 16 >> 1; + + linesize = s->cur_pic->f->linesize[0]; + uvlinesize = s->cur_pic->f->linesize[1]; + + dest_y = s->cur_pic->f->data[0] + (mb_x + mb_y * linesize) * 16; + dest_cb = s->cur_pic->f->data[1] + mb_x * 8 + mb_y * uvlinesize * block_h; + dest_cr = s->cur_pic->f->data[2] + mb_x * 8 + mb_y * uvlinesize * block_h; + + s->vdsp.prefetch(dest_y + (s->mb_x & 3) * 4 * linesize + 64, linesize, 4); + s->vdsp.prefetch(dest_cb + (s->mb_x & 7) * uvlinesize + 64, dest_cr - dest_cb, 2); + + if (IS_INTRA(mb_type)) { + s->hpc.pred8x8[s->chroma_pred_mode](dest_cb, uvlinesize); + s->hpc.pred8x8[s->chroma_pred_mode](dest_cr, uvlinesize); + + hl_decode_mb_predict_luma(s, mb_type, block_offset, linesize, dest_y); + } + + hl_decode_mb_idct_luma(s, mb_type, block_offset, linesize, dest_y); + + if (s->cbp & 0x30) { + uint8_t *dest[2] = { dest_cb, dest_cr }; + s->h264dsp.h264_chroma_dc_dequant_idct(s->mb + 16 * 16 * 1, + s->dequant4_coeff[4][0]); + s->h264dsp.h264_chroma_dc_dequant_idct(s->mb + 16 * 16 * 2, + s->dequant4_coeff[4][0]); + for (j = 1; j < 3; j++) { + for (i = j * 16; i < j * 16 + 4; i++) + if (s->non_zero_count_cache[scan8[i]] || s->mb[i * 16]) { + uint8_t *const ptr = dest[j - 1] + block_offset[i]; + svq3_add_idct_c(ptr, s->mb + i * 16, + uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2); + } + } + } +} + static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type) { - H264Context *h = &s->h; - H264SliceContext *sl = &h->slice_ctx[0]; int i, j, k, m, dir, mode; int cbp = 0; uint32_t vlc; int8_t *top, *left; - const int mb_xy = h->mb_xy; - const int b_xy = 4 * h->mb_x + 4 * h->mb_y * h->b_stride; + const int mb_xy = s->mb_xy; + const int b_xy = 4 * s->mb_x + 4 * s->mb_y * s->b_stride; - sl->top_samples_available = (h->mb_y == 0) ? 0x33FF : 0xFFFF; - sl->left_samples_available = (h->mb_x == 0) ? 0x5F5F : 0xFFFF; - sl->topright_samples_available = 0xFFFF; + s->top_samples_available = (s->mb_y == 0) ? 0x33FF : 0xFFFF; + s->left_samples_available = (s->mb_x == 0) ? 0x5F5F : 0xFFFF; + s->topright_samples_available = 0xFFFF; if (mb_type == 0) { /* SKIP */ - if (h->pict_type == AV_PICTURE_TYPE_P || + if (s->pict_type == AV_PICTURE_TYPE_P || s->next_pic->mb_type[mb_xy] == -1) { - svq3_mc_dir_part(s, 16 * h->mb_x, 16 * h->mb_y, 16, 16, + svq3_mc_dir_part(s, 16 * s->mb_x, 16 * s->mb_y, 16, 16, 0, 0, 0, 0, 0, 0); - if (h->pict_type == AV_PICTURE_TYPE_B) - svq3_mc_dir_part(s, 16 * h->mb_x, 16 * h->mb_y, 16, 16, + if (s->pict_type == AV_PICTURE_TYPE_B) + svq3_mc_dir_part(s, 16 * s->mb_x, 16 * s->mb_y, 16, 16, 0, 0, 0, 0, 1, 1); mb_type = MB_TYPE_SKIP; @@ -515,10 +743,10 @@ static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type) mb_type = MB_TYPE_16x16; } } else if (mb_type < 8) { /* INTER */ - if (s->thirdpel_flag && s->halfpel_flag == !get_bits1(&h->gb)) + if (s->thirdpel_flag && s->halfpel_flag == !get_bits1(&s->gb_slice)) mode = THIRDPEL_MODE; else if (s->halfpel_flag && - s->thirdpel_flag == !get_bits1(&h->gb)) + s->thirdpel_flag == !get_bits1(&s->gb_slice)) mode = HALFPEL_MODE; else mode = FULLPEL_MODE; @@ -533,46 +761,46 @@ static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type) */ for (m = 0; m < 2; m++) { - if (h->mb_x > 0 && sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - 1] + 6] != -1) { + if (s->mb_x > 0 && s->intra4x4_pred_mode[s->mb2br_xy[mb_xy - 1] + 6] != -1) { for (i = 0; i < 4; i++) - AV_COPY32(sl->mv_cache[m][scan8[0] - 1 + i * 8], - h->cur_pic.motion_val[m][b_xy - 1 + i * h->b_stride]); + AV_COPY32(s->mv_cache[m][scan8[0] - 1 + i * 8], + s->cur_pic->motion_val[m][b_xy - 1 + i * s->b_stride]); } else { for (i = 0; i < 4; i++) - AV_ZERO32(sl->mv_cache[m][scan8[0] - 1 + i * 8]); + AV_ZERO32(s->mv_cache[m][scan8[0] - 1 + i * 8]); } - if (h->mb_y > 0) { - memcpy(sl->mv_cache[m][scan8[0] - 1 * 8], - h->cur_pic.motion_val[m][b_xy - h->b_stride], + if (s->mb_y > 0) { + memcpy(s->mv_cache[m][scan8[0] - 1 * 8], + s->cur_pic->motion_val[m][b_xy - s->b_stride], 4 * 2 * sizeof(int16_t)); - memset(&sl->ref_cache[m][scan8[0] - 1 * 8], - (sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride]] == -1) ? PART_NOT_AVAILABLE : 1, 4); - - if (h->mb_x < h->mb_width - 1) { - AV_COPY32(sl->mv_cache[m][scan8[0] + 4 - 1 * 8], - h->cur_pic.motion_val[m][b_xy - h->b_stride + 4]); - sl->ref_cache[m][scan8[0] + 4 - 1 * 8] = - (sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride + 1] + 6] == -1 || - sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride]] == -1) ? PART_NOT_AVAILABLE : 1; + memset(&s->ref_cache[m][scan8[0] - 1 * 8], + (s->intra4x4_pred_mode[s->mb2br_xy[mb_xy - s->mb_stride]] == -1) ? PART_NOT_AVAILABLE : 1, 4); + + if (s->mb_x < s->mb_width - 1) { + AV_COPY32(s->mv_cache[m][scan8[0] + 4 - 1 * 8], + s->cur_pic->motion_val[m][b_xy - s->b_stride + 4]); + s->ref_cache[m][scan8[0] + 4 - 1 * 8] = + (s->intra4x4_pred_mode[s->mb2br_xy[mb_xy - s->mb_stride + 1] + 6] == -1 || + s->intra4x4_pred_mode[s->mb2br_xy[mb_xy - s->mb_stride]] == -1) ? PART_NOT_AVAILABLE : 1; } else - sl->ref_cache[m][scan8[0] + 4 - 1 * 8] = PART_NOT_AVAILABLE; - if (h->mb_x > 0) { - AV_COPY32(sl->mv_cache[m][scan8[0] - 1 - 1 * 8], - h->cur_pic.motion_val[m][b_xy - h->b_stride - 1]); - sl->ref_cache[m][scan8[0] - 1 - 1 * 8] = - (sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride - 1] + 3] == -1) ? PART_NOT_AVAILABLE : 1; + s->ref_cache[m][scan8[0] + 4 - 1 * 8] = PART_NOT_AVAILABLE; + if (s->mb_x > 0) { + AV_COPY32(s->mv_cache[m][scan8[0] - 1 - 1 * 8], + s->cur_pic->motion_val[m][b_xy - s->b_stride - 1]); + s->ref_cache[m][scan8[0] - 1 - 1 * 8] = + (s->intra4x4_pred_mode[s->mb2br_xy[mb_xy - s->mb_stride - 1] + 3] == -1) ? PART_NOT_AVAILABLE : 1; } else - sl->ref_cache[m][scan8[0] - 1 - 1 * 8] = PART_NOT_AVAILABLE; + s->ref_cache[m][scan8[0] - 1 - 1 * 8] = PART_NOT_AVAILABLE; } else - memset(&sl->ref_cache[m][scan8[0] - 1 * 8 - 1], + memset(&s->ref_cache[m][scan8[0] - 1 * 8 - 1], PART_NOT_AVAILABLE, 8); - if (h->pict_type != AV_PICTURE_TYPE_B) + if (s->pict_type != AV_PICTURE_TYPE_B) break; } /* decode motion vector(s) and form prediction(s) */ - if (h->pict_type == AV_PICTURE_TYPE_P) { + if (s->pict_type == AV_PICTURE_TYPE_P) { if (svq3_mc_dir(s, mb_type - 1, mode, 0, 0) < 0) return -1; } else { /* AV_PICTURE_TYPE_B */ @@ -581,7 +809,7 @@ static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type) return -1; } else { for (i = 0; i < 4; i++) - memset(h->cur_pic.motion_val[0][b_xy + i * h->b_stride], + memset(s->cur_pic->motion_val[0][b_xy + i * s->b_stride], 0, 4 * 2 * sizeof(int16_t)); } if (mb_type != 1) { @@ -589,128 +817,137 @@ static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type) return -1; } else { for (i = 0; i < 4; i++) - memset(h->cur_pic.motion_val[1][b_xy + i * h->b_stride], + memset(s->cur_pic->motion_val[1][b_xy + i * s->b_stride], 0, 4 * 2 * sizeof(int16_t)); } } mb_type = MB_TYPE_16x16; } else if (mb_type == 8 || mb_type == 33) { /* INTRA4x4 */ - memset(sl->intra4x4_pred_mode_cache, -1, 8 * 5 * sizeof(int8_t)); + int8_t *i4x4 = s->intra4x4_pred_mode + s->mb2br_xy[s->mb_xy]; + int8_t *i4x4_cache = s->intra4x4_pred_mode_cache; + + memset(s->intra4x4_pred_mode_cache, -1, 8 * 5 * sizeof(int8_t)); if (mb_type == 8) { - if (h->mb_x > 0) { + if (s->mb_x > 0) { for (i = 0; i < 4; i++) - sl->intra4x4_pred_mode_cache[scan8[0] - 1 + i * 8] = sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - 1] + 6 - i]; - if (sl->intra4x4_pred_mode_cache[scan8[0] - 1] == -1) - sl->left_samples_available = 0x5F5F; + s->intra4x4_pred_mode_cache[scan8[0] - 1 + i * 8] = s->intra4x4_pred_mode[s->mb2br_xy[mb_xy - 1] + 6 - i]; + if (s->intra4x4_pred_mode_cache[scan8[0] - 1] == -1) + s->left_samples_available = 0x5F5F; } - if (h->mb_y > 0) { - sl->intra4x4_pred_mode_cache[4 + 8 * 0] = sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride] + 0]; - sl->intra4x4_pred_mode_cache[5 + 8 * 0] = sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride] + 1]; - sl->intra4x4_pred_mode_cache[6 + 8 * 0] = sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride] + 2]; - sl->intra4x4_pred_mode_cache[7 + 8 * 0] = sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride] + 3]; - - if (sl->intra4x4_pred_mode_cache[4 + 8 * 0] == -1) - sl->top_samples_available = 0x33FF; + if (s->mb_y > 0) { + s->intra4x4_pred_mode_cache[4 + 8 * 0] = s->intra4x4_pred_mode[s->mb2br_xy[mb_xy - s->mb_stride] + 0]; + s->intra4x4_pred_mode_cache[5 + 8 * 0] = s->intra4x4_pred_mode[s->mb2br_xy[mb_xy - s->mb_stride] + 1]; + s->intra4x4_pred_mode_cache[6 + 8 * 0] = s->intra4x4_pred_mode[s->mb2br_xy[mb_xy - s->mb_stride] + 2]; + s->intra4x4_pred_mode_cache[7 + 8 * 0] = s->intra4x4_pred_mode[s->mb2br_xy[mb_xy - s->mb_stride] + 3]; + + if (s->intra4x4_pred_mode_cache[4 + 8 * 0] == -1) + s->top_samples_available = 0x33FF; } /* decode prediction codes for luma blocks */ for (i = 0; i < 16; i += 2) { - vlc = svq3_get_ue_golomb(&h->gb); + vlc = get_interleaved_ue_golomb(&s->gb_slice); if (vlc >= 25) { - av_log(h->avctx, AV_LOG_ERROR, + av_log(s->avctx, AV_LOG_ERROR, "luma prediction:%"PRIu32"\n", vlc); return -1; } - left = &sl->intra4x4_pred_mode_cache[scan8[i] - 1]; - top = &sl->intra4x4_pred_mode_cache[scan8[i] - 8]; + left = &s->intra4x4_pred_mode_cache[scan8[i] - 1]; + top = &s->intra4x4_pred_mode_cache[scan8[i] - 8]; left[1] = svq3_pred_1[top[0] + 1][left[0] + 1][svq3_pred_0[vlc][0]]; left[2] = svq3_pred_1[top[1] + 1][left[1] + 1][svq3_pred_0[vlc][1]]; if (left[1] == -1 || left[2] == -1) { - av_log(h->avctx, AV_LOG_ERROR, "weird prediction\n"); + av_log(s->avctx, AV_LOG_ERROR, "weird prediction\n"); return -1; } } } else { /* mb_type == 33, DC_128_PRED block type */ for (i = 0; i < 4; i++) - memset(&sl->intra4x4_pred_mode_cache[scan8[0] + 8 * i], DC_PRED, 4); + memset(&s->intra4x4_pred_mode_cache[scan8[0] + 8 * i], DC_PRED, 4); } - write_back_intra_pred_mode(h, sl); + AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4); + i4x4[4] = i4x4_cache[7 + 8 * 3]; + i4x4[5] = i4x4_cache[7 + 8 * 2]; + i4x4[6] = i4x4_cache[7 + 8 * 1]; if (mb_type == 8) { - ff_h264_check_intra4x4_pred_mode(h, sl); + ff_h264_check_intra4x4_pred_mode(s->intra4x4_pred_mode_cache, + s->avctx, s->top_samples_available, + s->left_samples_available); - sl->top_samples_available = (h->mb_y == 0) ? 0x33FF : 0xFFFF; - sl->left_samples_available = (h->mb_x == 0) ? 0x5F5F : 0xFFFF; + s->top_samples_available = (s->mb_y == 0) ? 0x33FF : 0xFFFF; + s->left_samples_available = (s->mb_x == 0) ? 0x5F5F : 0xFFFF; } else { for (i = 0; i < 4; i++) - memset(&sl->intra4x4_pred_mode_cache[scan8[0] + 8 * i], DC_128_PRED, 4); + memset(&s->intra4x4_pred_mode_cache[scan8[0] + 8 * i], DC_128_PRED, 4); - sl->top_samples_available = 0x33FF; - sl->left_samples_available = 0x5F5F; + s->top_samples_available = 0x33FF; + s->left_samples_available = 0x5F5F; } mb_type = MB_TYPE_INTRA4x4; } else { /* INTRA16x16 */ - dir = i_mb_type_info[mb_type - 8].pred_mode; + dir = ff_h264_i_mb_type_info[mb_type - 8].pred_mode; dir = (dir >> 1) ^ 3 * (dir & 1) ^ 1; - if ((sl->intra16x16_pred_mode = ff_h264_check_intra_pred_mode(h, sl, dir, 0)) < 0) { - av_log(h->avctx, AV_LOG_ERROR, "ff_h264_check_intra_pred_mode < 0\n"); - return sl->intra16x16_pred_mode; + if ((s->intra16x16_pred_mode = ff_h264_check_intra_pred_mode(s->avctx, s->top_samples_available, + s->left_samples_available, dir, 0)) < 0) { + av_log(s->avctx, AV_LOG_ERROR, "ff_h264_check_intra_pred_mode < 0\n"); + return s->intra16x16_pred_mode; } - cbp = i_mb_type_info[mb_type - 8].cbp; + cbp = ff_h264_i_mb_type_info[mb_type - 8].cbp; mb_type = MB_TYPE_INTRA16x16; } - if (!IS_INTER(mb_type) && h->pict_type != AV_PICTURE_TYPE_I) { + if (!IS_INTER(mb_type) && s->pict_type != AV_PICTURE_TYPE_I) { for (i = 0; i < 4; i++) - memset(h->cur_pic.motion_val[0][b_xy + i * h->b_stride], + memset(s->cur_pic->motion_val[0][b_xy + i * s->b_stride], 0, 4 * 2 * sizeof(int16_t)); - if (h->pict_type == AV_PICTURE_TYPE_B) { + if (s->pict_type == AV_PICTURE_TYPE_B) { for (i = 0; i < 4; i++) - memset(h->cur_pic.motion_val[1][b_xy + i * h->b_stride], + memset(s->cur_pic->motion_val[1][b_xy + i * s->b_stride], 0, 4 * 2 * sizeof(int16_t)); } } if (!IS_INTRA4x4(mb_type)) { - memset(sl->intra4x4_pred_mode + h->mb2br_xy[mb_xy], DC_PRED, 8); + memset(s->intra4x4_pred_mode + s->mb2br_xy[mb_xy], DC_PRED, 8); } - if (!IS_SKIP(mb_type) || h->pict_type == AV_PICTURE_TYPE_B) { - memset(sl->non_zero_count_cache + 8, 0, 14 * 8 * sizeof(uint8_t)); + if (!IS_SKIP(mb_type) || s->pict_type == AV_PICTURE_TYPE_B) { + memset(s->non_zero_count_cache + 8, 0, 14 * 8 * sizeof(uint8_t)); } if (!IS_INTRA16x16(mb_type) && - (!IS_SKIP(mb_type) || h->pict_type == AV_PICTURE_TYPE_B)) { - if ((vlc = svq3_get_ue_golomb(&h->gb)) >= 48) { - av_log(h->avctx, AV_LOG_ERROR, "cbp_vlc=%"PRIu32"\n", vlc); + (!IS_SKIP(mb_type) || s->pict_type == AV_PICTURE_TYPE_B)) { + if ((vlc = get_interleaved_ue_golomb(&s->gb_slice)) >= 48) { + av_log(s->avctx, AV_LOG_ERROR, "cbp_vlc=%"PRIu32"\n", vlc); return -1; } - cbp = IS_INTRA(mb_type) ? golomb_to_intra4x4_cbp[vlc] - : golomb_to_inter_cbp[vlc]; + cbp = IS_INTRA(mb_type) ? ff_h264_golomb_to_intra4x4_cbp[vlc] + : ff_h264_golomb_to_inter_cbp[vlc]; } if (IS_INTRA16x16(mb_type) || - (h->pict_type != AV_PICTURE_TYPE_I && s->adaptive_quant && cbp)) { - sl->qscale += svq3_get_se_golomb(&h->gb); + (s->pict_type != AV_PICTURE_TYPE_I && s->adaptive_quant && cbp)) { + s->qscale += get_interleaved_se_golomb(&s->gb_slice); - if (sl->qscale > 31u) { - av_log(h->avctx, AV_LOG_ERROR, "qscale:%d\n", sl->qscale); + if (s->qscale > 31u) { + av_log(s->avctx, AV_LOG_ERROR, "qscale:%d\n", s->qscale); return -1; } } if (IS_INTRA16x16(mb_type)) { - AV_ZERO128(sl->mb_luma_dc[0] + 0); - AV_ZERO128(sl->mb_luma_dc[0] + 8); - if (svq3_decode_block(&h->gb, sl->mb_luma_dc[0], 0, 1)) { - av_log(h->avctx, AV_LOG_ERROR, + AV_ZERO128(s->mb_luma_dc[0] + 0); + AV_ZERO128(s->mb_luma_dc[0] + 8); + if (svq3_decode_block(&s->gb_slice, s->mb_luma_dc[0], 0, 1)) { + av_log(s->avctx, AV_LOG_ERROR, "error while decoding intra luma dc\n"); return -1; } @@ -718,7 +955,7 @@ static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type) if (cbp) { const int index = IS_INTRA16x16(mb_type) ? 1 : 0; - const int type = ((sl->qscale < 24 && IS_INTRA4x4(mb_type)) ? 2 : 1); + const int type = ((s->qscale < 24 && IS_INTRA4x4(mb_type)) ? 2 : 1); for (i = 0; i < 4; i++) if ((cbp & (1 << i))) { @@ -726,10 +963,10 @@ static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type) k = index ? (1 * (j & 1) + 2 * (i & 1) + 2 * (j & 2) + 4 * (i & 2)) : (4 * i + j); - sl->non_zero_count_cache[scan8[k]] = 1; + s->non_zero_count_cache[scan8[k]] = 1; - if (svq3_decode_block(&h->gb, &sl->mb[16 * k], index, type)) { - av_log(h->avctx, AV_LOG_ERROR, + if (svq3_decode_block(&s->gb_slice, &s->mb[16 * k], index, type)) { + av_log(s->avctx, AV_LOG_ERROR, "error while decoding block\n"); return -1; } @@ -738,8 +975,8 @@ static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type) if ((cbp & 0x30)) { for (i = 1; i < 3; ++i) - if (svq3_decode_block(&h->gb, &sl->mb[16 * 16 * i], 0, 3)) { - av_log(h->avctx, AV_LOG_ERROR, + if (svq3_decode_block(&s->gb_slice, &s->mb[16 * 16 * i], 0, 3)) { + av_log(s->avctx, AV_LOG_ERROR, "error while decoding chroma dc block\n"); return -1; } @@ -748,10 +985,10 @@ static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type) for (i = 1; i < 3; i++) { for (j = 0; j < 4; j++) { k = 16 * i + j; - sl->non_zero_count_cache[scan8[k]] = 1; + s->non_zero_count_cache[scan8[k]] = 1; - if (svq3_decode_block(&h->gb, &sl->mb[16 * k], 1, 1)) { - av_log(h->avctx, AV_LOG_ERROR, + if (svq3_decode_block(&s->gb_slice, &s->mb[16 * k], 1, 1)) { + av_log(s->avctx, AV_LOG_ERROR, "error while decoding chroma ac block\n"); return -1; } @@ -761,11 +998,12 @@ static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type) } } - sl->cbp = cbp; - h->cur_pic.mb_type[mb_xy] = mb_type; + s->cbp = cbp; + s->cur_pic->mb_type[mb_xy] = mb_type; if (IS_INTRA(mb_type)) - sl->chroma_pred_mode = ff_h264_check_intra_pred_mode(h, sl, DC_PRED8x8, 1); + s->chroma_pred_mode = ff_h264_check_intra_pred_mode(s->avctx, s->top_samples_available, + s->left_samples_available, DC_PRED8x8, 1); return 0; } @@ -773,100 +1011,115 @@ static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type) static int svq3_decode_slice_header(AVCodecContext *avctx) { SVQ3Context *s = avctx->priv_data; - H264Context *h = &s->h; - H264SliceContext *sl = &h->slice_ctx[0]; - const int mb_xy = h->mb_xy; + const int mb_xy = s->mb_xy; int i, header; unsigned slice_id; - header = get_bits(&h->gb, 8); + header = get_bits(&s->gb, 8); if (((header & 0x9F) != 1 && (header & 0x9F) != 2) || (header & 0x60) == 0) { /* TODO: what? */ av_log(avctx, AV_LOG_ERROR, "unsupported slice header (%02X)\n", header); return -1; } else { + int slice_bits, slice_bytes, slice_length; int length = header >> 5 & 3; - s->next_slice_index = get_bits_count(&h->gb) + - 8 * show_bits(&h->gb, 8 * length) + - 8 * length; + slice_length = show_bits(&s->gb, 8 * length); + slice_bits = slice_length * 8; + slice_bytes = slice_length + length - 1; - if (s->next_slice_index > h->gb.size_in_bits) { + if (slice_bytes > get_bits_left(&s->gb)) { av_log(avctx, AV_LOG_ERROR, "slice after bitstream end\n"); return -1; } - h->gb.size_in_bits = s->next_slice_index - 8 * (length - 1); - skip_bits(&h->gb, 8); + skip_bits(&s->gb, 8); + + av_fast_malloc(&s->slice_buf, &s->slice_size, slice_bytes + AV_INPUT_BUFFER_PADDING_SIZE); + if (!s->slice_buf) + return AVERROR(ENOMEM); + + memcpy(s->slice_buf, s->gb.buffer + s->gb.index / 8, slice_bytes); + + init_get_bits(&s->gb_slice, s->slice_buf, slice_bits); if (s->watermark_key) { - uint32_t header = AV_RL32(&h->gb.buffer[(get_bits_count(&h->gb) >> 3) + 1]); - AV_WL32(&h->gb.buffer[(get_bits_count(&h->gb) >> 3) + 1], - header ^ s->watermark_key); + uint32_t header = AV_RL32(&s->gb_slice.buffer[1]); + AV_WL32(&s->gb_slice.buffer[1], header ^ s->watermark_key); } if (length > 0) { - memcpy((uint8_t *) &h->gb.buffer[get_bits_count(&h->gb) >> 3], - &h->gb.buffer[h->gb.size_in_bits >> 3], length - 1); + memcpy(s->slice_buf, &s->slice_buf[slice_length], length - 1); } - skip_bits_long(&h->gb, 0); + skip_bits_long(&s->gb, slice_bytes * 8); } - if ((slice_id = svq3_get_ue_golomb(&h->gb)) >= 3) { - av_log(h->avctx, AV_LOG_ERROR, "illegal slice type %u \n", slice_id); + if ((slice_id = get_interleaved_ue_golomb(&s->gb_slice)) >= 3) { + av_log(s->avctx, AV_LOG_ERROR, "illegal slice type %u \n", slice_id); return -1; } - sl->slice_type = golomb_to_pict_type[slice_id]; + s->pict_type = ff_h264_golomb_to_pict_type[slice_id]; if ((header & 0x9F) == 2) { - i = (h->mb_num < 64) ? 6 : (1 + av_log2(h->mb_num - 1)); - h->mb_skip_run = get_bits(&h->gb, i) - - (h->mb_y * h->mb_width + h->mb_x); + i = (s->mb_num < 64) ? 6 : (1 + av_log2(s->mb_num - 1)); + get_bits(&s->gb_slice, i); } else { - skip_bits1(&h->gb); - h->mb_skip_run = 0; + skip_bits1(&s->gb_slice); } - sl->slice_num = get_bits(&h->gb, 8); - sl->qscale = get_bits(&h->gb, 5); - s->adaptive_quant = get_bits1(&h->gb); + s->slice_num = get_bits(&s->gb_slice, 8); + s->qscale = get_bits(&s->gb_slice, 5); + s->adaptive_quant = get_bits1(&s->gb_slice); /* unknown fields */ - skip_bits1(&h->gb); + skip_bits1(&s->gb_slice); if (s->unknown_flag) - skip_bits1(&h->gb); + skip_bits1(&s->gb_slice); - skip_bits1(&h->gb); - skip_bits(&h->gb, 2); + skip_bits1(&s->gb_slice); + skip_bits(&s->gb_slice, 2); - while (get_bits1(&h->gb)) - skip_bits(&h->gb, 8); + while (get_bits1(&s->gb_slice)) + skip_bits(&s->gb_slice, 8); /* reset intra predictors and invalidate motion vector references */ - if (h->mb_x > 0) { - memset(sl->intra4x4_pred_mode + h->mb2br_xy[mb_xy - 1] + 3, + if (s->mb_x > 0) { + memset(s->intra4x4_pred_mode + s->mb2br_xy[mb_xy - 1] + 3, -1, 4 * sizeof(int8_t)); - memset(sl->intra4x4_pred_mode + h->mb2br_xy[mb_xy - h->mb_x], - -1, 8 * sizeof(int8_t) * h->mb_x); + memset(s->intra4x4_pred_mode + s->mb2br_xy[mb_xy - s->mb_x], + -1, 8 * sizeof(int8_t) * s->mb_x); } - if (h->mb_y > 0) { - memset(sl->intra4x4_pred_mode + h->mb2br_xy[mb_xy - h->mb_stride], - -1, 8 * sizeof(int8_t) * (h->mb_width - h->mb_x)); + if (s->mb_y > 0) { + memset(s->intra4x4_pred_mode + s->mb2br_xy[mb_xy - s->mb_stride], + -1, 8 * sizeof(int8_t) * (s->mb_width - s->mb_x)); - if (h->mb_x > 0) - sl->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride - 1] + 3] = -1; + if (s->mb_x > 0) + s->intra4x4_pred_mode[s->mb2br_xy[mb_xy - s->mb_stride - 1] + 3] = -1; } return 0; } +static void init_dequant4_coeff_table(SVQ3Context *s) +{ + int q, x; + const int max_qp = 51; + + for (q = 0; q < max_qp + 1; q++) { + int shift = ff_h264_quant_div6[q] + 2; + int idx = ff_h264_quant_rem6[q]; + for (x = 0; x < 16; x++) + s->dequant4_coeff[q][(x >> 2) | ((x << 2) & 0xF)] = + ((uint32_t)ff_h264_dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] * 16) << shift; + } +} + static av_cold int svq3_decode_init(AVCodecContext *avctx) { SVQ3Context *s = avctx->priv_data; - H264Context *h = &s->h; - int m; + int m, x, y; unsigned char *extradata; unsigned char *extradata_end; unsigned int size; @@ -882,21 +1135,23 @@ static av_cold int svq3_decode_init(AVCodecContext *avctx) return AVERROR(ENOMEM); } - if (ff_h264_decode_init(avctx) < 0) - return -1; + s->cur_pic->f = av_frame_alloc(); + s->last_pic->f = av_frame_alloc(); + s->next_pic->f = av_frame_alloc(); + if (!s->cur_pic->f || !s->last_pic->f || !s->next_pic->f) + return AVERROR(ENOMEM); + + ff_h264dsp_init(&s->h264dsp, 8, 1); + ff_h264_pred_init(&s->hpc, AV_CODEC_ID_SVQ3, 8, 1); + ff_videodsp_init(&s->vdsp, 8); ff_hpeldsp_init(&s->hdsp, avctx->flags); ff_tpeldsp_init(&s->tdsp); - h->flags = avctx->flags; - h->is_complex = 1; - h->picture_structure = PICT_FRAME; avctx->pix_fmt = AV_PIX_FMT_YUVJ420P; avctx->color_range = AVCOL_RANGE_JPEG; - h->slice_ctx[0].chroma_qp[0] = h->slice_ctx[0].chroma_qp[1] = 4; - h->chroma_x_shift = h->chroma_y_shift = 1; - + s->avctx = avctx; s->halfpel_flag = 1; s->thirdpel_flag = 1; s->unknown_flag = 0; @@ -970,7 +1225,7 @@ static av_cold int svq3_decode_init(AVCodecContext *avctx) skip_bits1(&gb); skip_bits1(&gb); - h->low_delay = get_bits1(&gb); + s->low_delay = get_bits1(&gb); /* unknown field */ skip_bits1(&gb); @@ -979,15 +1234,15 @@ static av_cold int svq3_decode_init(AVCodecContext *avctx) skip_bits(&gb, 8); s->unknown_flag = get_bits1(&gb); - avctx->has_b_frames = !h->low_delay; + avctx->has_b_frames = !s->low_delay; if (s->unknown_flag) { #if CONFIG_ZLIB - unsigned watermark_width = svq3_get_ue_golomb(&gb); - unsigned watermark_height = svq3_get_ue_golomb(&gb); - int u1 = svq3_get_ue_golomb(&gb); + unsigned watermark_width = get_interleaved_ue_golomb(&gb); + unsigned watermark_height = get_interleaved_ue_golomb(&gb); + int u1 = get_interleaved_ue_golomb(&gb); int u2 = get_bits(&gb, 8); int u3 = get_bits(&gb, 2); - int u4 = svq3_get_ue_golomb(&gb); + int u4 = get_interleaved_ue_golomb(&gb); unsigned long buf_len = watermark_width * watermark_height * 4; int offset = get_bits_count(&gb) + 7 >> 3; @@ -1023,25 +1278,36 @@ static av_cold int svq3_decode_init(AVCodecContext *avctx) } } - h->width = avctx->width; - h->height = avctx->height; - h->mb_width = (h->width + 15) / 16; - h->mb_height = (h->height + 15) / 16; - h->mb_stride = h->mb_width + 1; - h->mb_num = h->mb_width * h->mb_height; - h->b_stride = 4 * h->mb_width; - s->h_edge_pos = h->mb_width * 16; - s->v_edge_pos = h->mb_height * 16; - - if (ff_h264_alloc_tables(h) < 0) { - av_log(avctx, AV_LOG_ERROR, "svq3 memory allocation failed\n"); + s->mb_width = (avctx->width + 15) / 16; + s->mb_height = (avctx->height + 15) / 16; + s->mb_stride = s->mb_width + 1; + s->mb_num = s->mb_width * s->mb_height; + s->b_stride = 4 * s->mb_width; + s->h_edge_pos = s->mb_width * 16; + s->v_edge_pos = s->mb_height * 16; + + s->intra4x4_pred_mode = av_mallocz(s->mb_stride * 2 * 8); + if (!s->intra4x4_pred_mode) return AVERROR(ENOMEM); - } + + s->mb2br_xy = av_mallocz(s->mb_stride * (s->mb_height + 1) * + sizeof(*s->mb2br_xy)); + if (!s->mb2br_xy) + return AVERROR(ENOMEM); + + for (y = 0; y < s->mb_height; y++) + for (x = 0; x < s->mb_width; x++) { + const int mb_xy = x + y * s->mb_stride; + + s->mb2br_xy[mb_xy] = 8 * (mb_xy % (2 * s->mb_stride)); + } + + init_dequant4_coeff_table(s); return 0; } -static void free_picture(AVCodecContext *avctx, H264Picture *pic) +static void free_picture(AVCodecContext *avctx, SVQ3Frame *pic) { int i; for (i = 0; i < 2; i++) { @@ -1050,26 +1316,25 @@ static void free_picture(AVCodecContext *avctx, H264Picture *pic) } av_buffer_unref(&pic->mb_type_buf); - av_frame_unref(&pic->f); + av_frame_unref(pic->f); } -static int get_buffer(AVCodecContext *avctx, H264Picture *pic) +static int get_buffer(AVCodecContext *avctx, SVQ3Frame *pic) { SVQ3Context *s = avctx->priv_data; - H264Context *h = &s->h; - const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1; - const int mb_array_size = h->mb_stride * h->mb_height; - const int b4_stride = h->mb_width * 4 + 1; - const int b4_array_size = b4_stride * h->mb_height * 4; + const int big_mb_num = s->mb_stride * (s->mb_height + 1) + 1; + const int mb_array_size = s->mb_stride * s->mb_height; + const int b4_stride = s->mb_width * 4 + 1; + const int b4_array_size = b4_stride * s->mb_height * 4; int ret; if (!pic->motion_val_buf[0]) { int i; - pic->mb_type_buf = av_buffer_allocz((big_mb_num + h->mb_stride) * sizeof(uint32_t)); + pic->mb_type_buf = av_buffer_allocz((big_mb_num + s->mb_stride) * sizeof(uint32_t)); if (!pic->mb_type_buf) return AVERROR(ENOMEM); - pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1; + pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * s->mb_stride + 1; for (i = 0; i < 2; i++) { pic->motion_val_buf[i] = av_buffer_allocz(2 * (b4_array_size + 4) * sizeof(int16_t)); @@ -1083,22 +1348,19 @@ static int get_buffer(AVCodecContext *avctx, H264Picture *pic) pic->ref_index[i] = pic->ref_index_buf[i]->data; } } - pic->reference = !(h->pict_type == AV_PICTURE_TYPE_B); - ret = ff_get_buffer(avctx, &pic->f, - pic->reference ? AV_GET_BUFFER_FLAG_REF : 0); + ret = ff_get_buffer(avctx, pic->f, + (s->pict_type != AV_PICTURE_TYPE_B) ? + AV_GET_BUFFER_FLAG_REF : 0); if (ret < 0) goto fail; - if (!h->edge_emu_buffer) { - h->edge_emu_buffer = av_mallocz(pic->f.linesize[0] * 17); - if (!h->edge_emu_buffer) + if (!s->edge_emu_buffer) { + s->edge_emu_buffer = av_mallocz(pic->f->linesize[0] * 17); + if (!s->edge_emu_buffer) return AVERROR(ENOMEM); } - h->linesize = pic->f.linesize[0]; - h->uvlinesize = pic->f.linesize[1]; - return 0; fail: free_picture(avctx, pic); @@ -1110,15 +1372,13 @@ static int svq3_decode_frame(AVCodecContext *avctx, void *data, { const uint8_t *buf = avpkt->data; SVQ3Context *s = avctx->priv_data; - H264Context *h = &s->h; - H264SliceContext *sl = &h->slice_ctx[0]; int buf_size = avpkt->size; int ret, m, i; /* special case for last picture */ if (buf_size == 0) { - if (s->next_pic->f.data[0] && !h->low_delay && !s->last_frame_output) { - ret = av_frame_ref(data, &s->next_pic->f); + if (s->next_pic->f->data[0] && !s->low_delay && !s->last_frame_output) { + ret = av_frame_ref(data, s->next_pic->f); if (ret < 0) return ret; s->last_frame_output = 1; @@ -1127,108 +1387,101 @@ static int svq3_decode_frame(AVCodecContext *avctx, void *data, return 0; } - init_get_bits(&h->gb, buf, 8 * buf_size); + ret = init_get_bits(&s->gb, buf, 8 * buf_size); + if (ret < 0) + return ret; - h->mb_x = h->mb_y = h->mb_xy = 0; + s->mb_x = s->mb_y = s->mb_xy = 0; if (svq3_decode_slice_header(avctx)) return -1; - h->pict_type = sl->slice_type; + if (s->pict_type != AV_PICTURE_TYPE_B) + FFSWAP(SVQ3Frame*, s->next_pic, s->last_pic); - if (h->pict_type != AV_PICTURE_TYPE_B) - FFSWAP(H264Picture*, s->next_pic, s->last_pic); - - av_frame_unref(&s->cur_pic->f); + av_frame_unref(s->cur_pic->f); /* for skipping the frame */ - s->cur_pic->f.pict_type = h->pict_type; - s->cur_pic->f.key_frame = (h->pict_type == AV_PICTURE_TYPE_I); + s->cur_pic->f->pict_type = s->pict_type; + s->cur_pic->f->key_frame = (s->pict_type == AV_PICTURE_TYPE_I); ret = get_buffer(avctx, s->cur_pic); if (ret < 0) return ret; - h->cur_pic_ptr = s->cur_pic; - av_frame_unref(&h->cur_pic.f); - h->cur_pic = *s->cur_pic; - ret = av_frame_ref(&h->cur_pic.f, &s->cur_pic->f); - if (ret < 0) - return ret; - for (i = 0; i < 16; i++) { - h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3); - h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3); + s->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * s->cur_pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3); + s->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * s->cur_pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3); } for (i = 0; i < 16; i++) { - h->block_offset[16 + i] = - h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3); - h->block_offset[48 + 16 + i] = - h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3); + s->block_offset[16 + i] = + s->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * s->cur_pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3); + s->block_offset[48 + 16 + i] = + s->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * s->cur_pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3); } - if (h->pict_type != AV_PICTURE_TYPE_I) { - if (!s->last_pic->f.data[0]) { + if (s->pict_type != AV_PICTURE_TYPE_I) { + if (!s->last_pic->f->data[0]) { av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n"); ret = get_buffer(avctx, s->last_pic); if (ret < 0) return ret; - memset(s->last_pic->f.data[0], 0, avctx->height * s->last_pic->f.linesize[0]); - memset(s->last_pic->f.data[1], 0x80, (avctx->height / 2) * - s->last_pic->f.linesize[1]); - memset(s->last_pic->f.data[2], 0x80, (avctx->height / 2) * - s->last_pic->f.linesize[2]); + memset(s->last_pic->f->data[0], 0, avctx->height * s->last_pic->f->linesize[0]); + memset(s->last_pic->f->data[1], 0x80, (avctx->height / 2) * + s->last_pic->f->linesize[1]); + memset(s->last_pic->f->data[2], 0x80, (avctx->height / 2) * + s->last_pic->f->linesize[2]); } - if (h->pict_type == AV_PICTURE_TYPE_B && !s->next_pic->f.data[0]) { + if (s->pict_type == AV_PICTURE_TYPE_B && !s->next_pic->f->data[0]) { av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n"); ret = get_buffer(avctx, s->next_pic); if (ret < 0) return ret; - memset(s->next_pic->f.data[0], 0, avctx->height * s->next_pic->f.linesize[0]); - memset(s->next_pic->f.data[1], 0x80, (avctx->height / 2) * - s->next_pic->f.linesize[1]); - memset(s->next_pic->f.data[2], 0x80, (avctx->height / 2) * - s->next_pic->f.linesize[2]); + memset(s->next_pic->f->data[0], 0, avctx->height * s->next_pic->f->linesize[0]); + memset(s->next_pic->f->data[1], 0x80, (avctx->height / 2) * + s->next_pic->f->linesize[1]); + memset(s->next_pic->f->data[2], 0x80, (avctx->height / 2) * + s->next_pic->f->linesize[2]); } } if (avctx->debug & FF_DEBUG_PICT_INFO) - av_log(h->avctx, AV_LOG_DEBUG, + av_log(s->avctx, AV_LOG_DEBUG, "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n", - av_get_picture_type_char(h->pict_type), + av_get_picture_type_char(s->pict_type), s->halfpel_flag, s->thirdpel_flag, - s->adaptive_quant, h->slice_ctx[0].qscale, sl->slice_num); + s->adaptive_quant, s->qscale, s->slice_num); - if (avctx->skip_frame >= AVDISCARD_NONREF && h->pict_type == AV_PICTURE_TYPE_B || - avctx->skip_frame >= AVDISCARD_NONKEY && h->pict_type != AV_PICTURE_TYPE_I || + if (avctx->skip_frame >= AVDISCARD_NONREF && s->pict_type == AV_PICTURE_TYPE_B || + avctx->skip_frame >= AVDISCARD_NONKEY && s->pict_type != AV_PICTURE_TYPE_I || avctx->skip_frame >= AVDISCARD_ALL) return 0; if (s->next_p_frame_damaged) { - if (h->pict_type == AV_PICTURE_TYPE_B) + if (s->pict_type == AV_PICTURE_TYPE_B) return 0; else s->next_p_frame_damaged = 0; } - if (h->pict_type == AV_PICTURE_TYPE_B) { - h->frame_num_offset = sl->slice_num - h->prev_frame_num; + if (s->pict_type == AV_PICTURE_TYPE_B) { + s->frame_num_offset = s->slice_num - s->prev_frame_num; - if (h->frame_num_offset < 0) - h->frame_num_offset += 256; - if (h->frame_num_offset == 0 || - h->frame_num_offset >= h->prev_frame_num_offset) { - av_log(h->avctx, AV_LOG_ERROR, "error in B-frame picture id\n"); + if (s->frame_num_offset < 0) + s->frame_num_offset += 256; + if (s->frame_num_offset == 0 || + s->frame_num_offset >= s->prev_frame_num_offset) { + av_log(s->avctx, AV_LOG_ERROR, "error in B-frame picture id\n"); return -1; } } else { - h->prev_frame_num = h->frame_num; - h->frame_num = sl->slice_num; - h->prev_frame_num_offset = h->frame_num - h->prev_frame_num; + s->prev_frame_num = s->frame_num; + s->frame_num = s->slice_num; + s->prev_frame_num_offset = s->frame_num - s->prev_frame_num; - if (h->prev_frame_num_offset < 0) - h->prev_frame_num_offset += 256; + if (s->prev_frame_num_offset < 0) + s->prev_frame_num_offset += 256; } for (m = 0; m < 2; m++) { @@ -1236,70 +1489,68 @@ static int svq3_decode_frame(AVCodecContext *avctx, void *data, for (i = 0; i < 4; i++) { int j; for (j = -1; j < 4; j++) - sl->ref_cache[m][scan8[0] + 8 * i + j] = 1; + s->ref_cache[m][scan8[0] + 8 * i + j] = 1; if (i < 3) - sl->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE; + s->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE; } } - for (h->mb_y = 0; h->mb_y < h->mb_height; h->mb_y++) { - for (h->mb_x = 0; h->mb_x < h->mb_width; h->mb_x++) { + for (s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) { + for (s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++) { unsigned mb_type; - h->mb_xy = h->mb_x + h->mb_y * h->mb_stride; - - if ((get_bits_count(&h->gb) + 7) >= h->gb.size_in_bits && - ((get_bits_count(&h->gb) & 7) == 0 || - show_bits(&h->gb, -get_bits_count(&h->gb) & 7) == 0)) { - skip_bits(&h->gb, s->next_slice_index - get_bits_count(&h->gb)); - h->gb.size_in_bits = 8 * buf_size; + s->mb_xy = s->mb_x + s->mb_y * s->mb_stride; - if (svq3_decode_slice_header(avctx)) - return -1; + if ((get_bits_left(&s->gb_slice)) <= 7) { + if (((get_bits_count(&s->gb_slice) & 7) == 0 || + show_bits(&s->gb_slice, get_bits_left(&s->gb_slice) & 7) == 0)) { + if (svq3_decode_slice_header(avctx)) + return -1; + } /* TODO: support s->mb_skip_run */ } - mb_type = svq3_get_ue_golomb(&h->gb); + mb_type = get_interleaved_ue_golomb(&s->gb_slice); - if (h->pict_type == AV_PICTURE_TYPE_I) + if (s->pict_type == AV_PICTURE_TYPE_I) mb_type += 8; - else if (h->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4) + else if (s->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4) mb_type += 4; if (mb_type > 33 || svq3_decode_mb(s, mb_type)) { - av_log(h->avctx, AV_LOG_ERROR, - "error while decoding MB %d %d\n", h->mb_x, h->mb_y); + av_log(s->avctx, AV_LOG_ERROR, + "error while decoding MB %d %d\n", s->mb_x, s->mb_y); return -1; } if (mb_type != 0) - ff_h264_hl_decode_mb(h, &h->slice_ctx[0]); + hl_decode_mb(s); - if (h->pict_type != AV_PICTURE_TYPE_B && !h->low_delay) - h->cur_pic.mb_type[h->mb_x + h->mb_y * h->mb_stride] = - (h->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1; + if (s->pict_type != AV_PICTURE_TYPE_B && !s->low_delay) + s->cur_pic->mb_type[s->mb_x + s->mb_y * s->mb_stride] = + (s->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1; } - ff_draw_horiz_band(avctx, &s->cur_pic->f, - s->last_pic->f.data[0] ? &s->last_pic->f : NULL, - 16 * h->mb_y, 16, h->picture_structure, 0, - h->low_delay); + ff_draw_horiz_band(avctx, s->cur_pic->f, + s->last_pic->f->data[0] ? s->last_pic->f : NULL, + 16 * s->mb_y, 16, PICT_FRAME, 0, + s->low_delay); } - if (h->pict_type == AV_PICTURE_TYPE_B || h->low_delay) - ret = av_frame_ref(data, &s->cur_pic->f); - else if (s->last_pic->f.data[0]) - ret = av_frame_ref(data, &s->last_pic->f); + if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) + ret = av_frame_ref(data, s->cur_pic->f); + else if (s->last_pic->f->data[0]) + ret = av_frame_ref(data, s->last_pic->f); if (ret < 0) return ret; /* Do not output the last pic after seeking. */ - if (s->last_pic->f.data[0] || h->low_delay) + if (s->last_pic->f->data[0] || s->low_delay) *got_frame = 1; - if (h->pict_type != AV_PICTURE_TYPE_B) { - FFSWAP(H264Picture*, s->cur_pic, s->next_pic); + if (s->pict_type != AV_PICTURE_TYPE_B) { + FFSWAP(SVQ3Frame*, s->cur_pic, s->next_pic); } else { - av_frame_unref(&s->cur_pic->f); + av_frame_unref(s->cur_pic->f); } return buf_size; @@ -1308,18 +1559,20 @@ static int svq3_decode_frame(AVCodecContext *avctx, void *data, static av_cold int svq3_decode_end(AVCodecContext *avctx) { SVQ3Context *s = avctx->priv_data; - H264Context *h = &s->h; free_picture(avctx, s->cur_pic); free_picture(avctx, s->next_pic); free_picture(avctx, s->last_pic); + av_frame_free(&s->cur_pic->f); + av_frame_free(&s->next_pic->f); + av_frame_free(&s->last_pic->f); av_freep(&s->cur_pic); av_freep(&s->next_pic); av_freep(&s->last_pic); - - av_frame_unref(&h->cur_pic.f); - - ff_h264_free_context(h); + av_freep(&s->slice_buf); + av_freep(&s->intra4x4_pred_mode); + av_freep(&s->edge_emu_buffer); + av_freep(&s->mb2br_xy); return 0; } @@ -1333,9 +1586,9 @@ AVCodec ff_svq3_decoder = { .init = svq3_decode_init, .close = svq3_decode_end, .decode = svq3_decode_frame, - .capabilities = CODEC_CAP_DRAW_HORIZ_BAND | - CODEC_CAP_DR1 | - CODEC_CAP_DELAY, + .capabilities = AV_CODEC_CAP_DRAW_HORIZ_BAND | + AV_CODEC_CAP_DR1 | + AV_CODEC_CAP_DELAY, .pix_fmts = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_NONE}, };