yading@10: /* yading@10: * VP8 compatible video decoder yading@10: * yading@10: * Copyright (C) 2010 David Conrad yading@10: * Copyright (C) 2010 Ronald S. Bultje yading@10: * Copyright (C) 2010 Jason Garrett-Glaser yading@10: * Copyright (C) 2012 Daniel Kang 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: #include "libavutil/imgutils.h" yading@10: #include "avcodec.h" yading@10: #include "internal.h" yading@10: #include "vp8.h" yading@10: #include "vp8data.h" yading@10: #include "rectangle.h" yading@10: #include "thread.h" yading@10: yading@10: #if ARCH_ARM yading@10: # include "arm/vp8.h" yading@10: #endif yading@10: yading@10: static void free_buffers(VP8Context *s) yading@10: { yading@10: int i; yading@10: if (s->thread_data) yading@10: for (i = 0; i < MAX_THREADS; i++) { yading@10: #if HAVE_THREADS yading@10: pthread_cond_destroy(&s->thread_data[i].cond); yading@10: pthread_mutex_destroy(&s->thread_data[i].lock); yading@10: #endif yading@10: av_freep(&s->thread_data[i].filter_strength); yading@10: av_freep(&s->thread_data[i].edge_emu_buffer); yading@10: } yading@10: av_freep(&s->thread_data); yading@10: av_freep(&s->macroblocks_base); yading@10: av_freep(&s->intra4x4_pred_mode_top); yading@10: av_freep(&s->top_nnz); yading@10: av_freep(&s->top_border); yading@10: yading@10: s->macroblocks = NULL; yading@10: } yading@10: yading@10: static int vp8_alloc_frame(VP8Context *s, VP8Frame *f, int ref) yading@10: { yading@10: int ret; yading@10: if ((ret = ff_thread_get_buffer(s->avctx, &f->tf, yading@10: ref ? AV_GET_BUFFER_FLAG_REF : 0)) < 0) yading@10: return ret; yading@10: if (!(f->seg_map = av_buffer_allocz(s->mb_width * s->mb_height))) { yading@10: ff_thread_release_buffer(s->avctx, &f->tf); yading@10: return AVERROR(ENOMEM); yading@10: } yading@10: return 0; yading@10: } yading@10: yading@10: static void vp8_release_frame(VP8Context *s, VP8Frame *f) yading@10: { yading@10: av_buffer_unref(&f->seg_map); yading@10: ff_thread_release_buffer(s->avctx, &f->tf); yading@10: } yading@10: yading@10: static int vp8_ref_frame(VP8Context *s, VP8Frame *dst, VP8Frame *src) yading@10: { yading@10: int ret; yading@10: yading@10: vp8_release_frame(s, dst); yading@10: yading@10: if ((ret = ff_thread_ref_frame(&dst->tf, &src->tf)) < 0) yading@10: return ret; yading@10: if (src->seg_map && yading@10: !(dst->seg_map = av_buffer_ref(src->seg_map))) { yading@10: vp8_release_frame(s, dst); yading@10: return AVERROR(ENOMEM); yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: yading@10: static void vp8_decode_flush_impl(AVCodecContext *avctx, int free_mem) yading@10: { yading@10: VP8Context *s = avctx->priv_data; yading@10: int i; yading@10: yading@10: for (i = 0; i < FF_ARRAY_ELEMS(s->frames); i++) yading@10: vp8_release_frame(s, &s->frames[i]); yading@10: memset(s->framep, 0, sizeof(s->framep)); yading@10: yading@10: if (free_mem) yading@10: free_buffers(s); yading@10: } yading@10: yading@10: static void vp8_decode_flush(AVCodecContext *avctx) yading@10: { yading@10: vp8_decode_flush_impl(avctx, 0); yading@10: } yading@10: yading@10: static int update_dimensions(VP8Context *s, int width, int height) yading@10: { yading@10: AVCodecContext *avctx = s->avctx; yading@10: int i; yading@10: yading@10: if (width != s->avctx->width || ((width+15)/16 != s->mb_width || (height+15)/16 != s->mb_height) && s->macroblocks_base || yading@10: height != s->avctx->height) { yading@10: if (av_image_check_size(width, height, 0, s->avctx)) yading@10: return AVERROR_INVALIDDATA; yading@10: yading@10: vp8_decode_flush_impl(s->avctx, 1); yading@10: yading@10: avcodec_set_dimensions(s->avctx, width, height); yading@10: } yading@10: yading@10: s->mb_width = (s->avctx->coded_width +15) / 16; yading@10: s->mb_height = (s->avctx->coded_height+15) / 16; yading@10: yading@10: s->mb_layout = (avctx->active_thread_type == FF_THREAD_SLICE) && (FFMIN(s->num_coeff_partitions, avctx->thread_count) > 1); yading@10: if (!s->mb_layout) { // Frame threading and one thread yading@10: s->macroblocks_base = av_mallocz((s->mb_width+s->mb_height*2+1)*sizeof(*s->macroblocks)); yading@10: s->intra4x4_pred_mode_top = av_mallocz(s->mb_width*4); yading@10: } yading@10: else // Sliced threading yading@10: s->macroblocks_base = av_mallocz((s->mb_width+2)*(s->mb_height+2)*sizeof(*s->macroblocks)); yading@10: s->top_nnz = av_mallocz(s->mb_width*sizeof(*s->top_nnz)); yading@10: s->top_border = av_mallocz((s->mb_width+1)*sizeof(*s->top_border)); yading@10: s->thread_data = av_mallocz(MAX_THREADS*sizeof(VP8ThreadData)); yading@10: yading@10: for (i = 0; i < MAX_THREADS; i++) { yading@10: s->thread_data[i].filter_strength = av_mallocz(s->mb_width*sizeof(*s->thread_data[0].filter_strength)); yading@10: #if HAVE_THREADS yading@10: pthread_mutex_init(&s->thread_data[i].lock, NULL); yading@10: pthread_cond_init(&s->thread_data[i].cond, NULL); yading@10: #endif yading@10: } yading@10: yading@10: if (!s->macroblocks_base || !s->top_nnz || !s->top_border || yading@10: (!s->intra4x4_pred_mode_top && !s->mb_layout)) yading@10: return AVERROR(ENOMEM); yading@10: yading@10: s->macroblocks = s->macroblocks_base + 1; yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static void parse_segment_info(VP8Context *s) yading@10: { yading@10: VP56RangeCoder *c = &s->c; yading@10: int i; yading@10: yading@10: s->segmentation.update_map = vp8_rac_get(c); yading@10: yading@10: if (vp8_rac_get(c)) { // update segment feature data yading@10: s->segmentation.absolute_vals = vp8_rac_get(c); yading@10: yading@10: for (i = 0; i < 4; i++) yading@10: s->segmentation.base_quant[i] = vp8_rac_get_sint(c, 7); yading@10: yading@10: for (i = 0; i < 4; i++) yading@10: s->segmentation.filter_level[i] = vp8_rac_get_sint(c, 6); yading@10: } yading@10: if (s->segmentation.update_map) yading@10: for (i = 0; i < 3; i++) yading@10: s->prob->segmentid[i] = vp8_rac_get(c) ? vp8_rac_get_uint(c, 8) : 255; yading@10: } yading@10: yading@10: static void update_lf_deltas(VP8Context *s) yading@10: { yading@10: VP56RangeCoder *c = &s->c; yading@10: int i; yading@10: yading@10: for (i = 0; i < 4; i++) { yading@10: if (vp8_rac_get(c)) { yading@10: s->lf_delta.ref[i] = vp8_rac_get_uint(c, 6); yading@10: yading@10: if (vp8_rac_get(c)) yading@10: s->lf_delta.ref[i] = -s->lf_delta.ref[i]; yading@10: } yading@10: } yading@10: yading@10: for (i = MODE_I4x4; i <= VP8_MVMODE_SPLIT; i++) { yading@10: if (vp8_rac_get(c)) { yading@10: s->lf_delta.mode[i] = vp8_rac_get_uint(c, 6); yading@10: yading@10: if (vp8_rac_get(c)) yading@10: s->lf_delta.mode[i] = -s->lf_delta.mode[i]; yading@10: } yading@10: } yading@10: } yading@10: yading@10: static int setup_partitions(VP8Context *s, const uint8_t *buf, int buf_size) yading@10: { yading@10: const uint8_t *sizes = buf; yading@10: int i; yading@10: yading@10: s->num_coeff_partitions = 1 << vp8_rac_get_uint(&s->c, 2); yading@10: yading@10: buf += 3*(s->num_coeff_partitions-1); yading@10: buf_size -= 3*(s->num_coeff_partitions-1); yading@10: if (buf_size < 0) yading@10: return -1; yading@10: yading@10: for (i = 0; i < s->num_coeff_partitions-1; i++) { yading@10: int size = AV_RL24(sizes + 3*i); yading@10: if (buf_size - size < 0) yading@10: return -1; yading@10: yading@10: ff_vp56_init_range_decoder(&s->coeff_partition[i], buf, size); yading@10: buf += size; yading@10: buf_size -= size; yading@10: } yading@10: ff_vp56_init_range_decoder(&s->coeff_partition[i], buf, buf_size); yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static void get_quants(VP8Context *s) yading@10: { yading@10: VP56RangeCoder *c = &s->c; yading@10: int i, base_qi; yading@10: yading@10: int yac_qi = vp8_rac_get_uint(c, 7); yading@10: int ydc_delta = vp8_rac_get_sint(c, 4); yading@10: int y2dc_delta = vp8_rac_get_sint(c, 4); yading@10: int y2ac_delta = vp8_rac_get_sint(c, 4); yading@10: int uvdc_delta = vp8_rac_get_sint(c, 4); yading@10: int uvac_delta = vp8_rac_get_sint(c, 4); yading@10: yading@10: for (i = 0; i < 4; i++) { yading@10: if (s->segmentation.enabled) { yading@10: base_qi = s->segmentation.base_quant[i]; yading@10: if (!s->segmentation.absolute_vals) yading@10: base_qi += yac_qi; yading@10: } else yading@10: base_qi = yac_qi; yading@10: yading@10: s->qmat[i].luma_qmul[0] = vp8_dc_qlookup[av_clip_uintp2(base_qi + ydc_delta , 7)]; yading@10: s->qmat[i].luma_qmul[1] = vp8_ac_qlookup[av_clip_uintp2(base_qi , 7)]; yading@10: s->qmat[i].luma_dc_qmul[0] = 2 * vp8_dc_qlookup[av_clip_uintp2(base_qi + y2dc_delta, 7)]; yading@10: /* 101581>>16 is equivalent to 155/100 */ yading@10: s->qmat[i].luma_dc_qmul[1] = (101581 * vp8_ac_qlookup[av_clip_uintp2(base_qi + y2ac_delta, 7)]) >> 16; yading@10: s->qmat[i].chroma_qmul[0] = vp8_dc_qlookup[av_clip_uintp2(base_qi + uvdc_delta, 7)]; yading@10: s->qmat[i].chroma_qmul[1] = vp8_ac_qlookup[av_clip_uintp2(base_qi + uvac_delta, 7)]; yading@10: yading@10: s->qmat[i].luma_dc_qmul[1] = FFMAX(s->qmat[i].luma_dc_qmul[1], 8); yading@10: s->qmat[i].chroma_qmul[0] = FFMIN(s->qmat[i].chroma_qmul[0], 132); yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * Determine which buffers golden and altref should be updated with after this frame. yading@10: * The spec isn't clear here, so I'm going by my understanding of what libvpx does yading@10: * yading@10: * Intra frames update all 3 references yading@10: * Inter frames update VP56_FRAME_PREVIOUS if the update_last flag is set yading@10: * If the update (golden|altref) flag is set, it's updated with the current frame yading@10: * if update_last is set, and VP56_FRAME_PREVIOUS otherwise. yading@10: * If the flag is not set, the number read means: yading@10: * 0: no update yading@10: * 1: VP56_FRAME_PREVIOUS yading@10: * 2: update golden with altref, or update altref with golden yading@10: */ yading@10: static VP56Frame ref_to_update(VP8Context *s, int update, VP56Frame ref) yading@10: { yading@10: VP56RangeCoder *c = &s->c; yading@10: yading@10: if (update) yading@10: return VP56_FRAME_CURRENT; yading@10: yading@10: switch (vp8_rac_get_uint(c, 2)) { yading@10: case 1: yading@10: return VP56_FRAME_PREVIOUS; yading@10: case 2: yading@10: return (ref == VP56_FRAME_GOLDEN) ? VP56_FRAME_GOLDEN2 : VP56_FRAME_GOLDEN; yading@10: } yading@10: return VP56_FRAME_NONE; yading@10: } yading@10: yading@10: static void update_refs(VP8Context *s) yading@10: { yading@10: VP56RangeCoder *c = &s->c; yading@10: yading@10: int update_golden = vp8_rac_get(c); yading@10: int update_altref = vp8_rac_get(c); yading@10: yading@10: s->update_golden = ref_to_update(s, update_golden, VP56_FRAME_GOLDEN); yading@10: s->update_altref = ref_to_update(s, update_altref, VP56_FRAME_GOLDEN2); yading@10: } yading@10: yading@10: static int decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size) yading@10: { yading@10: VP56RangeCoder *c = &s->c; yading@10: int header_size, hscale, vscale, i, j, k, l, m, ret; yading@10: int width = s->avctx->width; yading@10: int height = s->avctx->height; yading@10: yading@10: s->keyframe = !(buf[0] & 1); yading@10: s->profile = (buf[0]>>1) & 7; yading@10: s->invisible = !(buf[0] & 0x10); yading@10: header_size = AV_RL24(buf) >> 5; yading@10: buf += 3; yading@10: buf_size -= 3; yading@10: yading@10: if (s->profile > 3) yading@10: av_log(s->avctx, AV_LOG_WARNING, "Unknown profile %d\n", s->profile); yading@10: yading@10: if (!s->profile) yading@10: memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_epel_pixels_tab, sizeof(s->put_pixels_tab)); yading@10: else // profile 1-3 use bilinear, 4+ aren't defined so whatever yading@10: memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_bilinear_pixels_tab, sizeof(s->put_pixels_tab)); yading@10: yading@10: if (header_size > buf_size - 7*s->keyframe) { yading@10: av_log(s->avctx, AV_LOG_ERROR, "Header size larger than data provided\n"); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: yading@10: if (s->keyframe) { yading@10: if (AV_RL24(buf) != 0x2a019d) { yading@10: av_log(s->avctx, AV_LOG_ERROR, "Invalid start code 0x%x\n", AV_RL24(buf)); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: width = AV_RL16(buf+3) & 0x3fff; yading@10: height = AV_RL16(buf+5) & 0x3fff; yading@10: hscale = buf[4] >> 6; yading@10: vscale = buf[6] >> 6; yading@10: buf += 7; yading@10: buf_size -= 7; yading@10: yading@10: if (hscale || vscale) yading@10: avpriv_request_sample(s->avctx, "Upscaling"); yading@10: yading@10: s->update_golden = s->update_altref = VP56_FRAME_CURRENT; yading@10: for (i = 0; i < 4; i++) yading@10: for (j = 0; j < 16; j++) yading@10: memcpy(s->prob->token[i][j], vp8_token_default_probs[i][vp8_coeff_band[j]], yading@10: sizeof(s->prob->token[i][j])); yading@10: memcpy(s->prob->pred16x16, vp8_pred16x16_prob_inter, sizeof(s->prob->pred16x16)); yading@10: memcpy(s->prob->pred8x8c , vp8_pred8x8c_prob_inter , sizeof(s->prob->pred8x8c)); yading@10: memcpy(s->prob->mvc , vp8_mv_default_prob , sizeof(s->prob->mvc)); yading@10: memset(&s->segmentation, 0, sizeof(s->segmentation)); yading@10: memset(&s->lf_delta, 0, sizeof(s->lf_delta)); yading@10: } yading@10: yading@10: ff_vp56_init_range_decoder(c, buf, header_size); yading@10: buf += header_size; yading@10: buf_size -= header_size; yading@10: yading@10: if (s->keyframe) { yading@10: if (vp8_rac_get(c)) yading@10: av_log(s->avctx, AV_LOG_WARNING, "Unspecified colorspace\n"); yading@10: vp8_rac_get(c); // whether we can skip clamping in dsp functions yading@10: } yading@10: yading@10: if ((s->segmentation.enabled = vp8_rac_get(c))) yading@10: parse_segment_info(s); yading@10: else yading@10: s->segmentation.update_map = 0; // FIXME: move this to some init function? yading@10: yading@10: s->filter.simple = vp8_rac_get(c); yading@10: s->filter.level = vp8_rac_get_uint(c, 6); yading@10: s->filter.sharpness = vp8_rac_get_uint(c, 3); yading@10: yading@10: if ((s->lf_delta.enabled = vp8_rac_get(c))) yading@10: if (vp8_rac_get(c)) yading@10: update_lf_deltas(s); yading@10: yading@10: if (setup_partitions(s, buf, buf_size)) { yading@10: av_log(s->avctx, AV_LOG_ERROR, "Invalid partitions\n"); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: yading@10: if (!s->macroblocks_base || /* first frame */ yading@10: width != s->avctx->width || height != s->avctx->height || (width+15)/16 != s->mb_width || (height+15)/16 != s->mb_height) { yading@10: if ((ret = update_dimensions(s, width, height)) < 0) yading@10: return ret; yading@10: } yading@10: yading@10: get_quants(s); yading@10: yading@10: if (!s->keyframe) { yading@10: update_refs(s); yading@10: s->sign_bias[VP56_FRAME_GOLDEN] = vp8_rac_get(c); yading@10: s->sign_bias[VP56_FRAME_GOLDEN2 /* altref */] = vp8_rac_get(c); yading@10: } yading@10: yading@10: // if we aren't saving this frame's probabilities for future frames, yading@10: // make a copy of the current probabilities yading@10: if (!(s->update_probabilities = vp8_rac_get(c))) yading@10: s->prob[1] = s->prob[0]; yading@10: yading@10: s->update_last = s->keyframe || vp8_rac_get(c); yading@10: yading@10: for (i = 0; i < 4; i++) yading@10: for (j = 0; j < 8; j++) yading@10: for (k = 0; k < 3; k++) yading@10: for (l = 0; l < NUM_DCT_TOKENS-1; l++) yading@10: if (vp56_rac_get_prob_branchy(c, vp8_token_update_probs[i][j][k][l])) { yading@10: int prob = vp8_rac_get_uint(c, 8); yading@10: for (m = 0; vp8_coeff_band_indexes[j][m] >= 0; m++) yading@10: s->prob->token[i][vp8_coeff_band_indexes[j][m]][k][l] = prob; yading@10: } yading@10: yading@10: if ((s->mbskip_enabled = vp8_rac_get(c))) yading@10: s->prob->mbskip = vp8_rac_get_uint(c, 8); yading@10: yading@10: if (!s->keyframe) { yading@10: s->prob->intra = vp8_rac_get_uint(c, 8); yading@10: s->prob->last = vp8_rac_get_uint(c, 8); yading@10: s->prob->golden = vp8_rac_get_uint(c, 8); yading@10: yading@10: if (vp8_rac_get(c)) yading@10: for (i = 0; i < 4; i++) yading@10: s->prob->pred16x16[i] = vp8_rac_get_uint(c, 8); yading@10: if (vp8_rac_get(c)) yading@10: for (i = 0; i < 3; i++) yading@10: s->prob->pred8x8c[i] = vp8_rac_get_uint(c, 8); yading@10: yading@10: // 17.2 MV probability update yading@10: for (i = 0; i < 2; i++) yading@10: for (j = 0; j < 19; j++) yading@10: if (vp56_rac_get_prob_branchy(c, vp8_mv_update_prob[i][j])) yading@10: s->prob->mvc[i][j] = vp8_rac_get_nn(c); yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static av_always_inline void clamp_mv(VP8Context *s, VP56mv *dst, const VP56mv *src) yading@10: { yading@10: dst->x = av_clip(src->x, s->mv_min.x, s->mv_max.x); yading@10: dst->y = av_clip(src->y, s->mv_min.y, s->mv_max.y); yading@10: } yading@10: yading@10: /** yading@10: * Motion vector coding, 17.1. yading@10: */ yading@10: static int read_mv_component(VP56RangeCoder *c, const uint8_t *p) yading@10: { yading@10: int bit, x = 0; yading@10: yading@10: if (vp56_rac_get_prob_branchy(c, p[0])) { yading@10: int i; yading@10: yading@10: for (i = 0; i < 3; i++) yading@10: x += vp56_rac_get_prob(c, p[9 + i]) << i; yading@10: for (i = 9; i > 3; i--) yading@10: x += vp56_rac_get_prob(c, p[9 + i]) << i; yading@10: if (!(x & 0xFFF0) || vp56_rac_get_prob(c, p[12])) yading@10: x += 8; yading@10: } else { yading@10: // small_mvtree yading@10: const uint8_t *ps = p+2; yading@10: bit = vp56_rac_get_prob(c, *ps); yading@10: ps += 1 + 3*bit; yading@10: x += 4*bit; yading@10: bit = vp56_rac_get_prob(c, *ps); yading@10: ps += 1 + bit; yading@10: x += 2*bit; yading@10: x += vp56_rac_get_prob(c, *ps); yading@10: } yading@10: yading@10: return (x && vp56_rac_get_prob(c, p[1])) ? -x : x; yading@10: } yading@10: yading@10: static av_always_inline yading@10: const uint8_t *get_submv_prob(uint32_t left, uint32_t top) yading@10: { yading@10: if (left == top) yading@10: return vp8_submv_prob[4-!!left]; yading@10: if (!top) yading@10: return vp8_submv_prob[2]; yading@10: return vp8_submv_prob[1-!!left]; yading@10: } yading@10: yading@10: /** yading@10: * Split motion vector prediction, 16.4. yading@10: * @returns the number of motion vectors parsed (2, 4 or 16) yading@10: */ yading@10: static av_always_inline yading@10: int decode_splitmvs(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb, int layout) yading@10: { yading@10: int part_idx; yading@10: int n, num; yading@10: VP8Macroblock *top_mb; yading@10: VP8Macroblock *left_mb = &mb[-1]; yading@10: const uint8_t *mbsplits_left = vp8_mbsplits[left_mb->partitioning], yading@10: *mbsplits_top, yading@10: *mbsplits_cur, *firstidx; yading@10: VP56mv *top_mv; yading@10: VP56mv *left_mv = left_mb->bmv; yading@10: VP56mv *cur_mv = mb->bmv; yading@10: yading@10: if (!layout) // layout is inlined, s->mb_layout is not yading@10: top_mb = &mb[2]; yading@10: else yading@10: top_mb = &mb[-s->mb_width-1]; yading@10: mbsplits_top = vp8_mbsplits[top_mb->partitioning]; yading@10: top_mv = top_mb->bmv; yading@10: yading@10: if (vp56_rac_get_prob_branchy(c, vp8_mbsplit_prob[0])) { yading@10: if (vp56_rac_get_prob_branchy(c, vp8_mbsplit_prob[1])) { yading@10: part_idx = VP8_SPLITMVMODE_16x8 + vp56_rac_get_prob(c, vp8_mbsplit_prob[2]); yading@10: } else { yading@10: part_idx = VP8_SPLITMVMODE_8x8; yading@10: } yading@10: } else { yading@10: part_idx = VP8_SPLITMVMODE_4x4; yading@10: } yading@10: yading@10: num = vp8_mbsplit_count[part_idx]; yading@10: mbsplits_cur = vp8_mbsplits[part_idx], yading@10: firstidx = vp8_mbfirstidx[part_idx]; yading@10: mb->partitioning = part_idx; yading@10: yading@10: for (n = 0; n < num; n++) { yading@10: int k = firstidx[n]; yading@10: uint32_t left, above; yading@10: const uint8_t *submv_prob; yading@10: yading@10: if (!(k & 3)) yading@10: left = AV_RN32A(&left_mv[mbsplits_left[k + 3]]); yading@10: else yading@10: left = AV_RN32A(&cur_mv[mbsplits_cur[k - 1]]); yading@10: if (k <= 3) yading@10: above = AV_RN32A(&top_mv[mbsplits_top[k + 12]]); yading@10: else yading@10: above = AV_RN32A(&cur_mv[mbsplits_cur[k - 4]]); yading@10: yading@10: submv_prob = get_submv_prob(left, above); yading@10: yading@10: if (vp56_rac_get_prob_branchy(c, submv_prob[0])) { yading@10: if (vp56_rac_get_prob_branchy(c, submv_prob[1])) { yading@10: if (vp56_rac_get_prob_branchy(c, submv_prob[2])) { yading@10: mb->bmv[n].y = mb->mv.y + read_mv_component(c, s->prob->mvc[0]); yading@10: mb->bmv[n].x = mb->mv.x + read_mv_component(c, s->prob->mvc[1]); yading@10: } else { yading@10: AV_ZERO32(&mb->bmv[n]); yading@10: } yading@10: } else { yading@10: AV_WN32A(&mb->bmv[n], above); yading@10: } yading@10: } else { yading@10: AV_WN32A(&mb->bmv[n], left); yading@10: } yading@10: } yading@10: yading@10: return num; yading@10: } yading@10: yading@10: static av_always_inline yading@10: void decode_mvs(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y, int layout) yading@10: { yading@10: VP8Macroblock *mb_edge[3] = { 0 /* top */, yading@10: mb - 1 /* left */, yading@10: 0 /* top-left */ }; yading@10: enum { CNT_ZERO, CNT_NEAREST, CNT_NEAR, CNT_SPLITMV }; yading@10: enum { VP8_EDGE_TOP, VP8_EDGE_LEFT, VP8_EDGE_TOPLEFT }; yading@10: int idx = CNT_ZERO; yading@10: int cur_sign_bias = s->sign_bias[mb->ref_frame]; yading@10: int8_t *sign_bias = s->sign_bias; yading@10: VP56mv near_mv[4]; yading@10: uint8_t cnt[4] = { 0 }; yading@10: VP56RangeCoder *c = &s->c; yading@10: yading@10: if (!layout) { // layout is inlined (s->mb_layout is not) yading@10: mb_edge[0] = mb + 2; yading@10: mb_edge[2] = mb + 1; yading@10: } yading@10: else { yading@10: mb_edge[0] = mb - s->mb_width-1; yading@10: mb_edge[2] = mb - s->mb_width-2; yading@10: } yading@10: yading@10: AV_ZERO32(&near_mv[0]); yading@10: AV_ZERO32(&near_mv[1]); yading@10: AV_ZERO32(&near_mv[2]); yading@10: yading@10: /* Process MB on top, left and top-left */ yading@10: #define MV_EDGE_CHECK(n)\ yading@10: {\ yading@10: VP8Macroblock *edge = mb_edge[n];\ yading@10: int edge_ref = edge->ref_frame;\ yading@10: if (edge_ref != VP56_FRAME_CURRENT) {\ yading@10: uint32_t mv = AV_RN32A(&edge->mv);\ yading@10: if (mv) {\ yading@10: if (cur_sign_bias != sign_bias[edge_ref]) {\ yading@10: /* SWAR negate of the values in mv. */\ yading@10: mv = ~mv;\ yading@10: mv = ((mv&0x7fff7fff) + 0x00010001) ^ (mv&0x80008000);\ yading@10: }\ yading@10: if (!n || mv != AV_RN32A(&near_mv[idx]))\ yading@10: AV_WN32A(&near_mv[++idx], mv);\ yading@10: cnt[idx] += 1 + (n != 2);\ yading@10: } else\ yading@10: cnt[CNT_ZERO] += 1 + (n != 2);\ yading@10: }\ yading@10: } yading@10: yading@10: MV_EDGE_CHECK(0) yading@10: MV_EDGE_CHECK(1) yading@10: MV_EDGE_CHECK(2) yading@10: yading@10: mb->partitioning = VP8_SPLITMVMODE_NONE; yading@10: if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_ZERO]][0])) { yading@10: mb->mode = VP8_MVMODE_MV; yading@10: yading@10: /* If we have three distinct MVs, merge first and last if they're the same */ yading@10: if (cnt[CNT_SPLITMV] && AV_RN32A(&near_mv[1 + VP8_EDGE_TOP]) == AV_RN32A(&near_mv[1 + VP8_EDGE_TOPLEFT])) yading@10: cnt[CNT_NEAREST] += 1; yading@10: yading@10: /* Swap near and nearest if necessary */ yading@10: if (cnt[CNT_NEAR] > cnt[CNT_NEAREST]) { yading@10: FFSWAP(uint8_t, cnt[CNT_NEAREST], cnt[CNT_NEAR]); yading@10: FFSWAP( VP56mv, near_mv[CNT_NEAREST], near_mv[CNT_NEAR]); yading@10: } yading@10: yading@10: if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_NEAREST]][1])) { yading@10: if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_NEAR]][2])) { yading@10: yading@10: /* Choose the best mv out of 0,0 and the nearest mv */ yading@10: clamp_mv(s, &mb->mv, &near_mv[CNT_ZERO + (cnt[CNT_NEAREST] >= cnt[CNT_ZERO])]); yading@10: cnt[CNT_SPLITMV] = ((mb_edge[VP8_EDGE_LEFT]->mode == VP8_MVMODE_SPLIT) + yading@10: (mb_edge[VP8_EDGE_TOP]->mode == VP8_MVMODE_SPLIT)) * 2 + yading@10: (mb_edge[VP8_EDGE_TOPLEFT]->mode == VP8_MVMODE_SPLIT); yading@10: yading@10: if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_SPLITMV]][3])) { yading@10: mb->mode = VP8_MVMODE_SPLIT; yading@10: mb->mv = mb->bmv[decode_splitmvs(s, c, mb, layout) - 1]; yading@10: } else { yading@10: mb->mv.y += read_mv_component(c, s->prob->mvc[0]); yading@10: mb->mv.x += read_mv_component(c, s->prob->mvc[1]); yading@10: mb->bmv[0] = mb->mv; yading@10: } yading@10: } else { yading@10: clamp_mv(s, &mb->mv, &near_mv[CNT_NEAR]); yading@10: mb->bmv[0] = mb->mv; yading@10: } yading@10: } else { yading@10: clamp_mv(s, &mb->mv, &near_mv[CNT_NEAREST]); yading@10: mb->bmv[0] = mb->mv; yading@10: } yading@10: } else { yading@10: mb->mode = VP8_MVMODE_ZERO; yading@10: AV_ZERO32(&mb->mv); yading@10: mb->bmv[0] = mb->mv; yading@10: } yading@10: } yading@10: yading@10: static av_always_inline yading@10: void decode_intra4x4_modes(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb, yading@10: int mb_x, int keyframe, int layout) yading@10: { yading@10: uint8_t *intra4x4 = mb->intra4x4_pred_mode_mb; yading@10: yading@10: if (layout == 1) { yading@10: VP8Macroblock *mb_top = mb - s->mb_width - 1; yading@10: memcpy(mb->intra4x4_pred_mode_top, mb_top->intra4x4_pred_mode_top, 4); yading@10: } yading@10: if (keyframe) { yading@10: int x, y; yading@10: uint8_t* top; yading@10: uint8_t* const left = s->intra4x4_pred_mode_left; yading@10: if (layout == 1) yading@10: top = mb->intra4x4_pred_mode_top; yading@10: else yading@10: top = s->intra4x4_pred_mode_top + 4 * mb_x; yading@10: for (y = 0; y < 4; y++) { yading@10: for (x = 0; x < 4; x++) { yading@10: const uint8_t *ctx; yading@10: ctx = vp8_pred4x4_prob_intra[top[x]][left[y]]; yading@10: *intra4x4 = vp8_rac_get_tree(c, vp8_pred4x4_tree, ctx); yading@10: left[y] = top[x] = *intra4x4; yading@10: intra4x4++; yading@10: } yading@10: } yading@10: } else { yading@10: int i; yading@10: for (i = 0; i < 16; i++) yading@10: intra4x4[i] = vp8_rac_get_tree(c, vp8_pred4x4_tree, vp8_pred4x4_prob_inter); yading@10: } yading@10: } yading@10: yading@10: static av_always_inline yading@10: void decode_mb_mode(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y, yading@10: uint8_t *segment, uint8_t *ref, int layout) yading@10: { yading@10: VP56RangeCoder *c = &s->c; yading@10: yading@10: if (s->segmentation.update_map) { yading@10: int bit = vp56_rac_get_prob(c, s->prob->segmentid[0]); yading@10: *segment = vp56_rac_get_prob(c, s->prob->segmentid[1+bit]) + 2*bit; yading@10: } else if (s->segmentation.enabled) yading@10: *segment = ref ? *ref : *segment; yading@10: mb->segment = *segment; yading@10: yading@10: mb->skip = s->mbskip_enabled ? vp56_rac_get_prob(c, s->prob->mbskip) : 0; yading@10: yading@10: if (s->keyframe) { yading@10: mb->mode = vp8_rac_get_tree(c, vp8_pred16x16_tree_intra, vp8_pred16x16_prob_intra); yading@10: yading@10: if (mb->mode == MODE_I4x4) { yading@10: decode_intra4x4_modes(s, c, mb, mb_x, 1, layout); yading@10: } else { yading@10: const uint32_t modes = vp8_pred4x4_mode[mb->mode] * 0x01010101u; yading@10: if (s->mb_layout == 1) yading@10: AV_WN32A(mb->intra4x4_pred_mode_top, modes); yading@10: else yading@10: AV_WN32A(s->intra4x4_pred_mode_top + 4 * mb_x, modes); yading@10: AV_WN32A( s->intra4x4_pred_mode_left, modes); yading@10: } yading@10: yading@10: mb->chroma_pred_mode = vp8_rac_get_tree(c, vp8_pred8x8c_tree, vp8_pred8x8c_prob_intra); yading@10: mb->ref_frame = VP56_FRAME_CURRENT; yading@10: } else if (vp56_rac_get_prob_branchy(c, s->prob->intra)) { yading@10: // inter MB, 16.2 yading@10: if (vp56_rac_get_prob_branchy(c, s->prob->last)) yading@10: mb->ref_frame = vp56_rac_get_prob(c, s->prob->golden) ? yading@10: VP56_FRAME_GOLDEN2 /* altref */ : VP56_FRAME_GOLDEN; yading@10: else yading@10: mb->ref_frame = VP56_FRAME_PREVIOUS; yading@10: s->ref_count[mb->ref_frame-1]++; yading@10: yading@10: // motion vectors, 16.3 yading@10: decode_mvs(s, mb, mb_x, mb_y, layout); yading@10: } else { yading@10: // intra MB, 16.1 yading@10: mb->mode = vp8_rac_get_tree(c, vp8_pred16x16_tree_inter, s->prob->pred16x16); yading@10: yading@10: if (mb->mode == MODE_I4x4) yading@10: decode_intra4x4_modes(s, c, mb, mb_x, 0, layout); yading@10: yading@10: mb->chroma_pred_mode = vp8_rac_get_tree(c, vp8_pred8x8c_tree, s->prob->pred8x8c); yading@10: mb->ref_frame = VP56_FRAME_CURRENT; yading@10: mb->partitioning = VP8_SPLITMVMODE_NONE; yading@10: AV_ZERO32(&mb->bmv[0]); yading@10: } yading@10: } yading@10: yading@10: #ifndef decode_block_coeffs_internal yading@10: /** yading@10: * @param r arithmetic bitstream reader context yading@10: * @param block destination for block coefficients yading@10: * @param probs probabilities to use when reading trees from the bitstream yading@10: * @param i initial coeff index, 0 unless a separate DC block is coded yading@10: * @param qmul array holding the dc/ac dequant factor at position 0/1 yading@10: * @return 0 if no coeffs were decoded yading@10: * otherwise, the index of the last coeff decoded plus one yading@10: */ yading@10: static int decode_block_coeffs_internal(VP56RangeCoder *r, int16_t block[16], yading@10: uint8_t probs[16][3][NUM_DCT_TOKENS-1], yading@10: int i, uint8_t *token_prob, int16_t qmul[2]) yading@10: { yading@10: VP56RangeCoder c = *r; yading@10: goto skip_eob; yading@10: do { yading@10: int coeff; yading@10: if (!vp56_rac_get_prob_branchy(&c, token_prob[0])) // DCT_EOB yading@10: break; yading@10: yading@10: skip_eob: yading@10: if (!vp56_rac_get_prob_branchy(&c, token_prob[1])) { // DCT_0 yading@10: if (++i == 16) yading@10: break; // invalid input; blocks should end with EOB yading@10: token_prob = probs[i][0]; yading@10: goto skip_eob; yading@10: } yading@10: yading@10: if (!vp56_rac_get_prob_branchy(&c, token_prob[2])) { // DCT_1 yading@10: coeff = 1; yading@10: token_prob = probs[i+1][1]; yading@10: } else { yading@10: if (!vp56_rac_get_prob_branchy(&c, token_prob[3])) { // DCT 2,3,4 yading@10: coeff = vp56_rac_get_prob_branchy(&c, token_prob[4]); yading@10: if (coeff) yading@10: coeff += vp56_rac_get_prob(&c, token_prob[5]); yading@10: coeff += 2; yading@10: } else { yading@10: // DCT_CAT* yading@10: if (!vp56_rac_get_prob_branchy(&c, token_prob[6])) { yading@10: if (!vp56_rac_get_prob_branchy(&c, token_prob[7])) { // DCT_CAT1 yading@10: coeff = 5 + vp56_rac_get_prob(&c, vp8_dct_cat1_prob[0]); yading@10: } else { // DCT_CAT2 yading@10: coeff = 7; yading@10: coeff += vp56_rac_get_prob(&c, vp8_dct_cat2_prob[0]) << 1; yading@10: coeff += vp56_rac_get_prob(&c, vp8_dct_cat2_prob[1]); yading@10: } yading@10: } else { // DCT_CAT3 and up yading@10: int a = vp56_rac_get_prob(&c, token_prob[8]); yading@10: int b = vp56_rac_get_prob(&c, token_prob[9+a]); yading@10: int cat = (a<<1) + b; yading@10: coeff = 3 + (8<segment; yading@10: int block_dc = 0; yading@10: yading@10: if (mb->mode != MODE_I4x4 && mb->mode != VP8_MVMODE_SPLIT) { yading@10: nnz_pred = t_nnz[8] + l_nnz[8]; yading@10: yading@10: // decode DC values and do hadamard yading@10: nnz = decode_block_coeffs(c, td->block_dc, s->prob->token[1], 0, nnz_pred, yading@10: s->qmat[segment].luma_dc_qmul); yading@10: l_nnz[8] = t_nnz[8] = !!nnz; yading@10: if (nnz) { yading@10: nnz_total += nnz; yading@10: block_dc = 1; yading@10: if (nnz == 1) yading@10: s->vp8dsp.vp8_luma_dc_wht_dc(td->block, td->block_dc); yading@10: else yading@10: s->vp8dsp.vp8_luma_dc_wht(td->block, td->block_dc); yading@10: } yading@10: luma_start = 1; yading@10: luma_ctx = 0; yading@10: } yading@10: yading@10: // luma blocks yading@10: for (y = 0; y < 4; y++) yading@10: for (x = 0; x < 4; x++) { yading@10: nnz_pred = l_nnz[y] + t_nnz[x]; yading@10: nnz = decode_block_coeffs(c, td->block[y][x], s->prob->token[luma_ctx], luma_start, yading@10: nnz_pred, s->qmat[segment].luma_qmul); yading@10: // nnz+block_dc may be one more than the actual last index, but we don't care yading@10: td->non_zero_count_cache[y][x] = nnz + block_dc; yading@10: t_nnz[x] = l_nnz[y] = !!nnz; yading@10: nnz_total += nnz; yading@10: } yading@10: yading@10: // chroma blocks yading@10: // TODO: what to do about dimensions? 2nd dim for luma is x, yading@10: // but for chroma it's (y<<1)|x yading@10: for (i = 4; i < 6; i++) yading@10: for (y = 0; y < 2; y++) yading@10: for (x = 0; x < 2; x++) { yading@10: nnz_pred = l_nnz[i+2*y] + t_nnz[i+2*x]; yading@10: nnz = decode_block_coeffs(c, td->block[i][(y<<1)+x], s->prob->token[2], 0, yading@10: nnz_pred, s->qmat[segment].chroma_qmul); yading@10: td->non_zero_count_cache[i][(y<<1)+x] = nnz; yading@10: t_nnz[i+2*x] = l_nnz[i+2*y] = !!nnz; yading@10: nnz_total += nnz; yading@10: } yading@10: yading@10: // if there were no coded coeffs despite the macroblock not being marked skip, yading@10: // we MUST not do the inner loop filter and should not do IDCT yading@10: // Since skip isn't used for bitstream prediction, just manually set it. yading@10: if (!nnz_total) yading@10: mb->skip = 1; yading@10: } yading@10: yading@10: static av_always_inline yading@10: void backup_mb_border(uint8_t *top_border, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, yading@10: int linesize, int uvlinesize, int simple) yading@10: { yading@10: AV_COPY128(top_border, src_y + 15*linesize); yading@10: if (!simple) { yading@10: AV_COPY64(top_border+16, src_cb + 7*uvlinesize); yading@10: AV_COPY64(top_border+24, src_cr + 7*uvlinesize); yading@10: } yading@10: } yading@10: yading@10: static av_always_inline yading@10: void xchg_mb_border(uint8_t *top_border, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, yading@10: int linesize, int uvlinesize, int mb_x, int mb_y, int mb_width, yading@10: int simple, int xchg) yading@10: { yading@10: uint8_t *top_border_m1 = top_border-32; // for TL prediction yading@10: src_y -= linesize; yading@10: src_cb -= uvlinesize; yading@10: src_cr -= uvlinesize; yading@10: yading@10: #define XCHG(a,b,xchg) do { \ yading@10: if (xchg) AV_SWAP64(b,a); \ yading@10: else AV_COPY64(b,a); \ yading@10: } while (0) yading@10: yading@10: XCHG(top_border_m1+8, src_y-8, xchg); yading@10: XCHG(top_border, src_y, xchg); yading@10: XCHG(top_border+8, src_y+8, 1); yading@10: if (mb_x < mb_width-1) yading@10: XCHG(top_border+32, src_y+16, 1); yading@10: yading@10: // only copy chroma for normal loop filter yading@10: // or to initialize the top row to 127 yading@10: if (!simple || !mb_y) { yading@10: XCHG(top_border_m1+16, src_cb-8, xchg); yading@10: XCHG(top_border_m1+24, src_cr-8, xchg); yading@10: XCHG(top_border+16, src_cb, 1); yading@10: XCHG(top_border+24, src_cr, 1); yading@10: } yading@10: } yading@10: yading@10: static av_always_inline yading@10: int check_dc_pred8x8_mode(int mode, int mb_x, int mb_y) yading@10: { yading@10: if (!mb_x) { yading@10: return mb_y ? TOP_DC_PRED8x8 : DC_128_PRED8x8; yading@10: } else { yading@10: return mb_y ? mode : LEFT_DC_PRED8x8; yading@10: } yading@10: } yading@10: yading@10: static av_always_inline yading@10: int check_tm_pred8x8_mode(int mode, int mb_x, int mb_y) yading@10: { yading@10: if (!mb_x) { yading@10: return mb_y ? VERT_PRED8x8 : DC_129_PRED8x8; yading@10: } else { yading@10: return mb_y ? mode : HOR_PRED8x8; yading@10: } yading@10: } yading@10: yading@10: static av_always_inline yading@10: int check_intra_pred8x8_mode(int mode, int mb_x, int mb_y) yading@10: { yading@10: if (mode == DC_PRED8x8) { yading@10: return check_dc_pred8x8_mode(mode, mb_x, mb_y); yading@10: } else { yading@10: return mode; yading@10: } yading@10: } yading@10: yading@10: static av_always_inline yading@10: int check_intra_pred8x8_mode_emuedge(int mode, int mb_x, int mb_y) yading@10: { yading@10: switch (mode) { yading@10: case DC_PRED8x8: yading@10: return check_dc_pred8x8_mode(mode, mb_x, mb_y); yading@10: case VERT_PRED8x8: yading@10: return !mb_y ? DC_127_PRED8x8 : mode; yading@10: case HOR_PRED8x8: yading@10: return !mb_x ? DC_129_PRED8x8 : mode; yading@10: case PLANE_PRED8x8 /*TM*/: yading@10: return check_tm_pred8x8_mode(mode, mb_x, mb_y); yading@10: } yading@10: return mode; yading@10: } yading@10: yading@10: static av_always_inline yading@10: int check_tm_pred4x4_mode(int mode, int mb_x, int mb_y) yading@10: { yading@10: if (!mb_x) { yading@10: return mb_y ? VERT_VP8_PRED : DC_129_PRED; yading@10: } else { yading@10: return mb_y ? mode : HOR_VP8_PRED; yading@10: } yading@10: } yading@10: yading@10: static av_always_inline yading@10: int check_intra_pred4x4_mode_emuedge(int mode, int mb_x, int mb_y, int *copy_buf) yading@10: { yading@10: switch (mode) { yading@10: case VERT_PRED: yading@10: if (!mb_x && mb_y) { yading@10: *copy_buf = 1; yading@10: return mode; yading@10: } yading@10: /* fall-through */ yading@10: case DIAG_DOWN_LEFT_PRED: yading@10: case VERT_LEFT_PRED: yading@10: return !mb_y ? DC_127_PRED : mode; yading@10: case HOR_PRED: yading@10: if (!mb_y) { yading@10: *copy_buf = 1; yading@10: return mode; yading@10: } yading@10: /* fall-through */ yading@10: case HOR_UP_PRED: yading@10: return !mb_x ? DC_129_PRED : mode; yading@10: case TM_VP8_PRED: yading@10: return check_tm_pred4x4_mode(mode, mb_x, mb_y); yading@10: case DC_PRED: // 4x4 DC doesn't use the same "H.264-style" exceptions as 16x16/8x8 DC yading@10: case DIAG_DOWN_RIGHT_PRED: yading@10: case VERT_RIGHT_PRED: yading@10: case HOR_DOWN_PRED: yading@10: if (!mb_y || !mb_x) yading@10: *copy_buf = 1; yading@10: return mode; yading@10: } yading@10: return mode; yading@10: } yading@10: yading@10: static av_always_inline yading@10: void intra_predict(VP8Context *s, VP8ThreadData *td, uint8_t *dst[3], yading@10: VP8Macroblock *mb, int mb_x, int mb_y) yading@10: { yading@10: AVCodecContext *avctx = s->avctx; yading@10: int x, y, mode, nnz; yading@10: uint32_t tr; yading@10: yading@10: // for the first row, we need to run xchg_mb_border to init the top edge to 127 yading@10: // otherwise, skip it if we aren't going to deblock yading@10: if (!(avctx->flags & CODEC_FLAG_EMU_EDGE && !mb_y) && (s->deblock_filter || !mb_y) && td->thread_nr == 0) yading@10: xchg_mb_border(s->top_border[mb_x+1], dst[0], dst[1], dst[2], yading@10: s->linesize, s->uvlinesize, mb_x, mb_y, s->mb_width, yading@10: s->filter.simple, 1); yading@10: yading@10: if (mb->mode < MODE_I4x4) { yading@10: if (avctx->flags & CODEC_FLAG_EMU_EDGE) { // tested yading@10: mode = check_intra_pred8x8_mode_emuedge(mb->mode, mb_x, mb_y); yading@10: } else { yading@10: mode = check_intra_pred8x8_mode(mb->mode, mb_x, mb_y); yading@10: } yading@10: s->hpc.pred16x16[mode](dst[0], s->linesize); yading@10: } else { yading@10: uint8_t *ptr = dst[0]; yading@10: uint8_t *intra4x4 = mb->intra4x4_pred_mode_mb; yading@10: uint8_t tr_top[4] = { 127, 127, 127, 127 }; yading@10: yading@10: // all blocks on the right edge of the macroblock use bottom edge yading@10: // the top macroblock for their topright edge yading@10: uint8_t *tr_right = ptr - s->linesize + 16; yading@10: yading@10: // if we're on the right edge of the frame, said edge is extended yading@10: // from the top macroblock yading@10: if (!(!mb_y && avctx->flags & CODEC_FLAG_EMU_EDGE) && yading@10: mb_x == s->mb_width-1) { yading@10: tr = tr_right[-1]*0x01010101u; yading@10: tr_right = (uint8_t *)&tr; yading@10: } yading@10: yading@10: if (mb->skip) yading@10: AV_ZERO128(td->non_zero_count_cache); yading@10: yading@10: for (y = 0; y < 4; y++) { yading@10: uint8_t *topright = ptr + 4 - s->linesize; yading@10: for (x = 0; x < 4; x++) { yading@10: int copy = 0, linesize = s->linesize; yading@10: uint8_t *dst = ptr+4*x; yading@10: DECLARE_ALIGNED(4, uint8_t, copy_dst)[5*8]; yading@10: yading@10: if ((y == 0 || x == 3) && mb_y == 0 && avctx->flags & CODEC_FLAG_EMU_EDGE) { yading@10: topright = tr_top; yading@10: } else if (x == 3) yading@10: topright = tr_right; yading@10: yading@10: if (avctx->flags & CODEC_FLAG_EMU_EDGE) { // mb_x+x or mb_y+y is a hack but works yading@10: mode = check_intra_pred4x4_mode_emuedge(intra4x4[x], mb_x + x, mb_y + y, ©); yading@10: if (copy) { yading@10: dst = copy_dst + 12; yading@10: linesize = 8; yading@10: if (!(mb_y + y)) { yading@10: copy_dst[3] = 127U; yading@10: AV_WN32A(copy_dst+4, 127U * 0x01010101U); yading@10: } else { yading@10: AV_COPY32(copy_dst+4, ptr+4*x-s->linesize); yading@10: if (!(mb_x + x)) { yading@10: copy_dst[3] = 129U; yading@10: } else { yading@10: copy_dst[3] = ptr[4*x-s->linesize-1]; yading@10: } yading@10: } yading@10: if (!(mb_x + x)) { yading@10: copy_dst[11] = yading@10: copy_dst[19] = yading@10: copy_dst[27] = yading@10: copy_dst[35] = 129U; yading@10: } else { yading@10: copy_dst[11] = ptr[4*x -1]; yading@10: copy_dst[19] = ptr[4*x+s->linesize -1]; yading@10: copy_dst[27] = ptr[4*x+s->linesize*2-1]; yading@10: copy_dst[35] = ptr[4*x+s->linesize*3-1]; yading@10: } yading@10: } yading@10: } else { yading@10: mode = intra4x4[x]; yading@10: } yading@10: s->hpc.pred4x4[mode](dst, topright, linesize); yading@10: if (copy) { yading@10: AV_COPY32(ptr+4*x , copy_dst+12); yading@10: AV_COPY32(ptr+4*x+s->linesize , copy_dst+20); yading@10: AV_COPY32(ptr+4*x+s->linesize*2, copy_dst+28); yading@10: AV_COPY32(ptr+4*x+s->linesize*3, copy_dst+36); yading@10: } yading@10: yading@10: nnz = td->non_zero_count_cache[y][x]; yading@10: if (nnz) { yading@10: if (nnz == 1) yading@10: s->vp8dsp.vp8_idct_dc_add(ptr+4*x, td->block[y][x], s->linesize); yading@10: else yading@10: s->vp8dsp.vp8_idct_add(ptr+4*x, td->block[y][x], s->linesize); yading@10: } yading@10: topright += 4; yading@10: } yading@10: yading@10: ptr += 4*s->linesize; yading@10: intra4x4 += 4; yading@10: } yading@10: } yading@10: yading@10: if (avctx->flags & CODEC_FLAG_EMU_EDGE) { yading@10: mode = check_intra_pred8x8_mode_emuedge(mb->chroma_pred_mode, mb_x, mb_y); yading@10: } else { yading@10: mode = check_intra_pred8x8_mode(mb->chroma_pred_mode, mb_x, mb_y); yading@10: } yading@10: s->hpc.pred8x8[mode](dst[1], s->uvlinesize); yading@10: s->hpc.pred8x8[mode](dst[2], s->uvlinesize); yading@10: yading@10: if (!(avctx->flags & CODEC_FLAG_EMU_EDGE && !mb_y) && (s->deblock_filter || !mb_y) && td->thread_nr == 0) yading@10: xchg_mb_border(s->top_border[mb_x+1], dst[0], dst[1], dst[2], yading@10: s->linesize, s->uvlinesize, mb_x, mb_y, s->mb_width, yading@10: s->filter.simple, 0); yading@10: } yading@10: yading@10: static const uint8_t subpel_idx[3][8] = { yading@10: { 0, 1, 2, 1, 2, 1, 2, 1 }, // nr. of left extra pixels, yading@10: // also function pointer index yading@10: { 0, 3, 5, 3, 5, 3, 5, 3 }, // nr. of extra pixels required yading@10: { 0, 2, 3, 2, 3, 2, 3, 2 }, // nr. of right extra pixels yading@10: }; yading@10: yading@10: /** yading@10: * luma MC function yading@10: * yading@10: * @param s VP8 decoding context yading@10: * @param dst target buffer for block data at block position yading@10: * @param ref reference picture buffer at origin (0, 0) yading@10: * @param mv motion vector (relative to block position) to get pixel data from yading@10: * @param x_off horizontal position of block from origin (0, 0) yading@10: * @param y_off vertical position of block from origin (0, 0) yading@10: * @param block_w width of block (16, 8 or 4) yading@10: * @param block_h height of block (always same as block_w) yading@10: * @param width width of src/dst plane data yading@10: * @param height height of src/dst plane data yading@10: * @param linesize size of a single line of plane data, including padding yading@10: * @param mc_func motion compensation function pointers (bilinear or sixtap MC) yading@10: */ yading@10: static av_always_inline yading@10: void vp8_mc_luma(VP8Context *s, VP8ThreadData *td, uint8_t *dst, yading@10: ThreadFrame *ref, const VP56mv *mv, yading@10: int x_off, int y_off, int block_w, int block_h, yading@10: int width, int height, int linesize, yading@10: vp8_mc_func mc_func[3][3]) yading@10: { yading@10: uint8_t *src = ref->f->data[0]; yading@10: yading@10: if (AV_RN32A(mv)) { yading@10: yading@10: int mx = (mv->x << 1)&7, mx_idx = subpel_idx[0][mx]; yading@10: int my = (mv->y << 1)&7, my_idx = subpel_idx[0][my]; yading@10: yading@10: x_off += mv->x >> 2; yading@10: y_off += mv->y >> 2; yading@10: yading@10: // edge emulation yading@10: ff_thread_await_progress(ref, (3 + y_off + block_h + subpel_idx[2][my]) >> 4, 0); yading@10: src += y_off * linesize + x_off; yading@10: if (x_off < mx_idx || x_off >= width - block_w - subpel_idx[2][mx] || yading@10: y_off < my_idx || y_off >= height - block_h - subpel_idx[2][my]) { yading@10: s->vdsp.emulated_edge_mc(td->edge_emu_buffer, src - my_idx * linesize - mx_idx, linesize, yading@10: block_w + subpel_idx[1][mx], block_h + subpel_idx[1][my], yading@10: x_off - mx_idx, y_off - my_idx, width, height); yading@10: src = td->edge_emu_buffer + mx_idx + linesize * my_idx; yading@10: } yading@10: mc_func[my_idx][mx_idx](dst, linesize, src, linesize, block_h, mx, my); yading@10: } else { yading@10: ff_thread_await_progress(ref, (3 + y_off + block_h) >> 4, 0); yading@10: mc_func[0][0](dst, linesize, src + y_off * linesize + x_off, linesize, block_h, 0, 0); yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * chroma MC function yading@10: * yading@10: * @param s VP8 decoding context yading@10: * @param dst1 target buffer for block data at block position (U plane) yading@10: * @param dst2 target buffer for block data at block position (V plane) yading@10: * @param ref reference picture buffer at origin (0, 0) yading@10: * @param mv motion vector (relative to block position) to get pixel data from yading@10: * @param x_off horizontal position of block from origin (0, 0) yading@10: * @param y_off vertical position of block from origin (0, 0) yading@10: * @param block_w width of block (16, 8 or 4) yading@10: * @param block_h height of block (always same as block_w) yading@10: * @param width width of src/dst plane data yading@10: * @param height height of src/dst plane data yading@10: * @param linesize size of a single line of plane data, including padding yading@10: * @param mc_func motion compensation function pointers (bilinear or sixtap MC) yading@10: */ yading@10: static av_always_inline yading@10: void vp8_mc_chroma(VP8Context *s, VP8ThreadData *td, uint8_t *dst1, uint8_t *dst2, yading@10: ThreadFrame *ref, const VP56mv *mv, int x_off, int y_off, yading@10: int block_w, int block_h, int width, int height, int linesize, yading@10: vp8_mc_func mc_func[3][3]) yading@10: { yading@10: uint8_t *src1 = ref->f->data[1], *src2 = ref->f->data[2]; yading@10: yading@10: if (AV_RN32A(mv)) { yading@10: int mx = mv->x&7, mx_idx = subpel_idx[0][mx]; yading@10: int my = mv->y&7, my_idx = subpel_idx[0][my]; yading@10: yading@10: x_off += mv->x >> 3; yading@10: y_off += mv->y >> 3; yading@10: yading@10: // edge emulation yading@10: src1 += y_off * linesize + x_off; yading@10: src2 += y_off * linesize + x_off; yading@10: ff_thread_await_progress(ref, (3 + y_off + block_h + subpel_idx[2][my]) >> 3, 0); yading@10: if (x_off < mx_idx || x_off >= width - block_w - subpel_idx[2][mx] || yading@10: y_off < my_idx || y_off >= height - block_h - subpel_idx[2][my]) { yading@10: s->vdsp.emulated_edge_mc(td->edge_emu_buffer, src1 - my_idx * linesize - mx_idx, linesize, yading@10: block_w + subpel_idx[1][mx], block_h + subpel_idx[1][my], yading@10: x_off - mx_idx, y_off - my_idx, width, height); yading@10: src1 = td->edge_emu_buffer + mx_idx + linesize * my_idx; yading@10: mc_func[my_idx][mx_idx](dst1, linesize, src1, linesize, block_h, mx, my); yading@10: yading@10: s->vdsp.emulated_edge_mc(td->edge_emu_buffer, src2 - my_idx * linesize - mx_idx, linesize, yading@10: block_w + subpel_idx[1][mx], block_h + subpel_idx[1][my], yading@10: x_off - mx_idx, y_off - my_idx, width, height); yading@10: src2 = td->edge_emu_buffer + mx_idx + linesize * my_idx; yading@10: mc_func[my_idx][mx_idx](dst2, linesize, src2, linesize, block_h, mx, my); yading@10: } else { yading@10: mc_func[my_idx][mx_idx](dst1, linesize, src1, linesize, block_h, mx, my); yading@10: mc_func[my_idx][mx_idx](dst2, linesize, src2, linesize, block_h, mx, my); yading@10: } yading@10: } else { yading@10: ff_thread_await_progress(ref, (3 + y_off + block_h) >> 3, 0); yading@10: mc_func[0][0](dst1, linesize, src1 + y_off * linesize + x_off, linesize, block_h, 0, 0); yading@10: mc_func[0][0](dst2, linesize, src2 + y_off * linesize + x_off, linesize, block_h, 0, 0); yading@10: } yading@10: } yading@10: yading@10: static av_always_inline yading@10: void vp8_mc_part(VP8Context *s, VP8ThreadData *td, uint8_t *dst[3], yading@10: ThreadFrame *ref_frame, int x_off, int y_off, yading@10: int bx_off, int by_off, yading@10: int block_w, int block_h, yading@10: int width, int height, VP56mv *mv) yading@10: { yading@10: VP56mv uvmv = *mv; yading@10: yading@10: /* Y */ yading@10: vp8_mc_luma(s, td, dst[0] + by_off * s->linesize + bx_off, yading@10: ref_frame, mv, x_off + bx_off, y_off + by_off, yading@10: block_w, block_h, width, height, s->linesize, yading@10: s->put_pixels_tab[block_w == 8]); yading@10: yading@10: /* U/V */ yading@10: if (s->profile == 3) { yading@10: uvmv.x &= ~7; yading@10: uvmv.y &= ~7; yading@10: } yading@10: x_off >>= 1; y_off >>= 1; yading@10: bx_off >>= 1; by_off >>= 1; yading@10: width >>= 1; height >>= 1; yading@10: block_w >>= 1; block_h >>= 1; yading@10: vp8_mc_chroma(s, td, dst[1] + by_off * s->uvlinesize + bx_off, yading@10: dst[2] + by_off * s->uvlinesize + bx_off, ref_frame, yading@10: &uvmv, x_off + bx_off, y_off + by_off, yading@10: block_w, block_h, width, height, s->uvlinesize, yading@10: s->put_pixels_tab[1 + (block_w == 4)]); yading@10: } yading@10: yading@10: /* Fetch pixels for estimated mv 4 macroblocks ahead. yading@10: * Optimized for 64-byte cache lines. Inspired by ffh264 prefetch_motion. */ yading@10: static av_always_inline void prefetch_motion(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y, int mb_xy, int ref) yading@10: { yading@10: /* Don't prefetch refs that haven't been used very often this frame. */ yading@10: if (s->ref_count[ref-1] > (mb_xy >> 5)) { yading@10: int x_off = mb_x << 4, y_off = mb_y << 4; yading@10: int mx = (mb->mv.x>>2) + x_off + 8; yading@10: int my = (mb->mv.y>>2) + y_off; yading@10: uint8_t **src= s->framep[ref]->tf.f->data; yading@10: int off= mx + (my + (mb_x&3)*4)*s->linesize + 64; yading@10: /* For threading, a ff_thread_await_progress here might be useful, but yading@10: * it actually slows down the decoder. Since a bad prefetch doesn't yading@10: * generate bad decoder output, we don't run it here. */ yading@10: s->vdsp.prefetch(src[0]+off, s->linesize, 4); yading@10: off= (mx>>1) + ((my>>1) + (mb_x&7))*s->uvlinesize + 64; yading@10: s->vdsp.prefetch(src[1]+off, src[2]-src[1], 2); yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * Apply motion vectors to prediction buffer, chapter 18. yading@10: */ yading@10: static av_always_inline yading@10: void inter_predict(VP8Context *s, VP8ThreadData *td, uint8_t *dst[3], yading@10: VP8Macroblock *mb, int mb_x, int mb_y) yading@10: { yading@10: int x_off = mb_x << 4, y_off = mb_y << 4; yading@10: int width = 16*s->mb_width, height = 16*s->mb_height; yading@10: ThreadFrame *ref = &s->framep[mb->ref_frame]->tf; yading@10: VP56mv *bmv = mb->bmv; yading@10: yading@10: switch (mb->partitioning) { yading@10: case VP8_SPLITMVMODE_NONE: yading@10: vp8_mc_part(s, td, dst, ref, x_off, y_off, yading@10: 0, 0, 16, 16, width, height, &mb->mv); yading@10: break; yading@10: case VP8_SPLITMVMODE_4x4: { yading@10: int x, y; yading@10: VP56mv uvmv; yading@10: yading@10: /* Y */ yading@10: for (y = 0; y < 4; y++) { yading@10: for (x = 0; x < 4; x++) { yading@10: vp8_mc_luma(s, td, dst[0] + 4*y*s->linesize + x*4, yading@10: ref, &bmv[4*y + x], yading@10: 4*x + x_off, 4*y + y_off, 4, 4, yading@10: width, height, s->linesize, yading@10: s->put_pixels_tab[2]); yading@10: } yading@10: } yading@10: yading@10: /* U/V */ yading@10: x_off >>= 1; y_off >>= 1; width >>= 1; height >>= 1; yading@10: for (y = 0; y < 2; y++) { yading@10: for (x = 0; x < 2; x++) { yading@10: uvmv.x = mb->bmv[ 2*y * 4 + 2*x ].x + yading@10: mb->bmv[ 2*y * 4 + 2*x+1].x + yading@10: mb->bmv[(2*y+1) * 4 + 2*x ].x + yading@10: mb->bmv[(2*y+1) * 4 + 2*x+1].x; yading@10: uvmv.y = mb->bmv[ 2*y * 4 + 2*x ].y + yading@10: mb->bmv[ 2*y * 4 + 2*x+1].y + yading@10: mb->bmv[(2*y+1) * 4 + 2*x ].y + yading@10: mb->bmv[(2*y+1) * 4 + 2*x+1].y; yading@10: uvmv.x = (uvmv.x + 2 + (uvmv.x >> (INT_BIT-1))) >> 2; yading@10: uvmv.y = (uvmv.y + 2 + (uvmv.y >> (INT_BIT-1))) >> 2; yading@10: if (s->profile == 3) { yading@10: uvmv.x &= ~7; yading@10: uvmv.y &= ~7; yading@10: } yading@10: vp8_mc_chroma(s, td, dst[1] + 4*y*s->uvlinesize + x*4, yading@10: dst[2] + 4*y*s->uvlinesize + x*4, ref, &uvmv, yading@10: 4*x + x_off, 4*y + y_off, 4, 4, yading@10: width, height, s->uvlinesize, yading@10: s->put_pixels_tab[2]); yading@10: } yading@10: } yading@10: break; yading@10: } yading@10: case VP8_SPLITMVMODE_16x8: yading@10: vp8_mc_part(s, td, dst, ref, x_off, y_off, yading@10: 0, 0, 16, 8, width, height, &bmv[0]); yading@10: vp8_mc_part(s, td, dst, ref, x_off, y_off, yading@10: 0, 8, 16, 8, width, height, &bmv[1]); yading@10: break; yading@10: case VP8_SPLITMVMODE_8x16: yading@10: vp8_mc_part(s, td, dst, ref, x_off, y_off, yading@10: 0, 0, 8, 16, width, height, &bmv[0]); yading@10: vp8_mc_part(s, td, dst, ref, x_off, y_off, yading@10: 8, 0, 8, 16, width, height, &bmv[1]); yading@10: break; yading@10: case VP8_SPLITMVMODE_8x8: yading@10: vp8_mc_part(s, td, dst, ref, x_off, y_off, yading@10: 0, 0, 8, 8, width, height, &bmv[0]); yading@10: vp8_mc_part(s, td, dst, ref, x_off, y_off, yading@10: 8, 0, 8, 8, width, height, &bmv[1]); yading@10: vp8_mc_part(s, td, dst, ref, x_off, y_off, yading@10: 0, 8, 8, 8, width, height, &bmv[2]); yading@10: vp8_mc_part(s, td, dst, ref, x_off, y_off, yading@10: 8, 8, 8, 8, width, height, &bmv[3]); yading@10: break; yading@10: } yading@10: } yading@10: yading@10: static av_always_inline void idct_mb(VP8Context *s, VP8ThreadData *td, yading@10: uint8_t *dst[3], VP8Macroblock *mb) yading@10: { yading@10: int x, y, ch; yading@10: yading@10: if (mb->mode != MODE_I4x4) { yading@10: uint8_t *y_dst = dst[0]; yading@10: for (y = 0; y < 4; y++) { yading@10: uint32_t nnz4 = AV_RL32(td->non_zero_count_cache[y]); yading@10: if (nnz4) { yading@10: if (nnz4&~0x01010101) { yading@10: for (x = 0; x < 4; x++) { yading@10: if ((uint8_t)nnz4 == 1) yading@10: s->vp8dsp.vp8_idct_dc_add(y_dst+4*x, td->block[y][x], s->linesize); yading@10: else if((uint8_t)nnz4 > 1) yading@10: s->vp8dsp.vp8_idct_add(y_dst+4*x, td->block[y][x], s->linesize); yading@10: nnz4 >>= 8; yading@10: if (!nnz4) yading@10: break; yading@10: } yading@10: } else { yading@10: s->vp8dsp.vp8_idct_dc_add4y(y_dst, td->block[y], s->linesize); yading@10: } yading@10: } yading@10: y_dst += 4*s->linesize; yading@10: } yading@10: } yading@10: yading@10: for (ch = 0; ch < 2; ch++) { yading@10: uint32_t nnz4 = AV_RL32(td->non_zero_count_cache[4+ch]); yading@10: if (nnz4) { yading@10: uint8_t *ch_dst = dst[1+ch]; yading@10: if (nnz4&~0x01010101) { yading@10: for (y = 0; y < 2; y++) { yading@10: for (x = 0; x < 2; x++) { yading@10: if ((uint8_t)nnz4 == 1) yading@10: s->vp8dsp.vp8_idct_dc_add(ch_dst+4*x, td->block[4+ch][(y<<1)+x], s->uvlinesize); yading@10: else if((uint8_t)nnz4 > 1) yading@10: s->vp8dsp.vp8_idct_add(ch_dst+4*x, td->block[4+ch][(y<<1)+x], s->uvlinesize); yading@10: nnz4 >>= 8; yading@10: if (!nnz4) yading@10: goto chroma_idct_end; yading@10: } yading@10: ch_dst += 4*s->uvlinesize; yading@10: } yading@10: } else { yading@10: s->vp8dsp.vp8_idct_dc_add4uv(ch_dst, td->block[4+ch], s->uvlinesize); yading@10: } yading@10: } yading@10: chroma_idct_end: ; yading@10: } yading@10: } yading@10: yading@10: static av_always_inline void filter_level_for_mb(VP8Context *s, VP8Macroblock *mb, VP8FilterStrength *f ) yading@10: { yading@10: int interior_limit, filter_level; yading@10: yading@10: if (s->segmentation.enabled) { yading@10: filter_level = s->segmentation.filter_level[mb->segment]; yading@10: if (!s->segmentation.absolute_vals) yading@10: filter_level += s->filter.level; yading@10: } else yading@10: filter_level = s->filter.level; yading@10: yading@10: if (s->lf_delta.enabled) { yading@10: filter_level += s->lf_delta.ref[mb->ref_frame]; yading@10: filter_level += s->lf_delta.mode[mb->mode]; yading@10: } yading@10: yading@10: filter_level = av_clip_uintp2(filter_level, 6); yading@10: yading@10: interior_limit = filter_level; yading@10: if (s->filter.sharpness) { yading@10: interior_limit >>= (s->filter.sharpness + 3) >> 2; yading@10: interior_limit = FFMIN(interior_limit, 9 - s->filter.sharpness); yading@10: } yading@10: interior_limit = FFMAX(interior_limit, 1); yading@10: yading@10: f->filter_level = filter_level; yading@10: f->inner_limit = interior_limit; yading@10: f->inner_filter = !mb->skip || mb->mode == MODE_I4x4 || mb->mode == VP8_MVMODE_SPLIT; yading@10: } yading@10: yading@10: static av_always_inline void filter_mb(VP8Context *s, uint8_t *dst[3], VP8FilterStrength *f, int mb_x, int mb_y) yading@10: { yading@10: int mbedge_lim, bedge_lim, hev_thresh; yading@10: int filter_level = f->filter_level; yading@10: int inner_limit = f->inner_limit; yading@10: int inner_filter = f->inner_filter; yading@10: int linesize = s->linesize; yading@10: int uvlinesize = s->uvlinesize; yading@10: static const uint8_t hev_thresh_lut[2][64] = { yading@10: { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, yading@10: 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, yading@10: 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, yading@10: 3, 3, 3, 3 }, yading@10: { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, yading@10: 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, yading@10: 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, yading@10: 2, 2, 2, 2 } yading@10: }; yading@10: yading@10: if (!filter_level) yading@10: return; yading@10: yading@10: bedge_lim = 2*filter_level + inner_limit; yading@10: mbedge_lim = bedge_lim + 4; yading@10: yading@10: hev_thresh = hev_thresh_lut[s->keyframe][filter_level]; yading@10: yading@10: if (mb_x) { yading@10: s->vp8dsp.vp8_h_loop_filter16y(dst[0], linesize, yading@10: mbedge_lim, inner_limit, hev_thresh); yading@10: s->vp8dsp.vp8_h_loop_filter8uv(dst[1], dst[2], uvlinesize, yading@10: mbedge_lim, inner_limit, hev_thresh); yading@10: } yading@10: yading@10: if (inner_filter) { yading@10: s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0]+ 4, linesize, bedge_lim, yading@10: inner_limit, hev_thresh); yading@10: s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0]+ 8, linesize, bedge_lim, yading@10: inner_limit, hev_thresh); yading@10: s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0]+12, linesize, bedge_lim, yading@10: inner_limit, hev_thresh); yading@10: s->vp8dsp.vp8_h_loop_filter8uv_inner(dst[1] + 4, dst[2] + 4, yading@10: uvlinesize, bedge_lim, yading@10: inner_limit, hev_thresh); yading@10: } yading@10: yading@10: if (mb_y) { yading@10: s->vp8dsp.vp8_v_loop_filter16y(dst[0], linesize, yading@10: mbedge_lim, inner_limit, hev_thresh); yading@10: s->vp8dsp.vp8_v_loop_filter8uv(dst[1], dst[2], uvlinesize, yading@10: mbedge_lim, inner_limit, hev_thresh); yading@10: } yading@10: yading@10: if (inner_filter) { yading@10: s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0]+ 4*linesize, yading@10: linesize, bedge_lim, yading@10: inner_limit, hev_thresh); yading@10: s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0]+ 8*linesize, yading@10: linesize, bedge_lim, yading@10: inner_limit, hev_thresh); yading@10: s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0]+12*linesize, yading@10: linesize, bedge_lim, yading@10: inner_limit, hev_thresh); yading@10: s->vp8dsp.vp8_v_loop_filter8uv_inner(dst[1] + 4 * uvlinesize, yading@10: dst[2] + 4 * uvlinesize, yading@10: uvlinesize, bedge_lim, yading@10: inner_limit, hev_thresh); yading@10: } yading@10: } yading@10: yading@10: static av_always_inline void filter_mb_simple(VP8Context *s, uint8_t *dst, VP8FilterStrength *f, int mb_x, int mb_y) yading@10: { yading@10: int mbedge_lim, bedge_lim; yading@10: int filter_level = f->filter_level; yading@10: int inner_limit = f->inner_limit; yading@10: int inner_filter = f->inner_filter; yading@10: int linesize = s->linesize; yading@10: yading@10: if (!filter_level) yading@10: return; yading@10: yading@10: bedge_lim = 2*filter_level + inner_limit; yading@10: mbedge_lim = bedge_lim + 4; yading@10: yading@10: if (mb_x) yading@10: s->vp8dsp.vp8_h_loop_filter_simple(dst, linesize, mbedge_lim); yading@10: if (inner_filter) { yading@10: s->vp8dsp.vp8_h_loop_filter_simple(dst+ 4, linesize, bedge_lim); yading@10: s->vp8dsp.vp8_h_loop_filter_simple(dst+ 8, linesize, bedge_lim); yading@10: s->vp8dsp.vp8_h_loop_filter_simple(dst+12, linesize, bedge_lim); yading@10: } yading@10: yading@10: if (mb_y) yading@10: s->vp8dsp.vp8_v_loop_filter_simple(dst, linesize, mbedge_lim); yading@10: if (inner_filter) { yading@10: s->vp8dsp.vp8_v_loop_filter_simple(dst+ 4*linesize, linesize, bedge_lim); yading@10: s->vp8dsp.vp8_v_loop_filter_simple(dst+ 8*linesize, linesize, bedge_lim); yading@10: s->vp8dsp.vp8_v_loop_filter_simple(dst+12*linesize, linesize, bedge_lim); yading@10: } yading@10: } yading@10: yading@10: #define MARGIN (16 << 2) yading@10: static void vp8_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *curframe, yading@10: VP8Frame *prev_frame) yading@10: { yading@10: VP8Context *s = avctx->priv_data; yading@10: int mb_x, mb_y; yading@10: yading@10: s->mv_min.y = -MARGIN; yading@10: s->mv_max.y = ((s->mb_height - 1) << 6) + MARGIN; yading@10: for (mb_y = 0; mb_y < s->mb_height; mb_y++) { yading@10: VP8Macroblock *mb = s->macroblocks_base + ((s->mb_width+1)*(mb_y + 1) + 1); yading@10: int mb_xy = mb_y*s->mb_width; yading@10: yading@10: AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED*0x01010101); yading@10: yading@10: s->mv_min.x = -MARGIN; yading@10: s->mv_max.x = ((s->mb_width - 1) << 6) + MARGIN; yading@10: for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) { yading@10: if (mb_y == 0) yading@10: AV_WN32A((mb-s->mb_width-1)->intra4x4_pred_mode_top, DC_PRED*0x01010101); yading@10: decode_mb_mode(s, mb, mb_x, mb_y, curframe->seg_map->data + mb_xy, yading@10: prev_frame && prev_frame->seg_map ? yading@10: prev_frame->seg_map->data + mb_xy : NULL, 1); yading@10: s->mv_min.x -= 64; yading@10: s->mv_max.x -= 64; yading@10: } yading@10: s->mv_min.y -= 64; yading@10: s->mv_max.y -= 64; yading@10: } yading@10: } yading@10: yading@10: #if HAVE_THREADS yading@10: #define check_thread_pos(td, otd, mb_x_check, mb_y_check)\ yading@10: do {\ yading@10: int tmp = (mb_y_check << 16) | (mb_x_check & 0xFFFF);\ yading@10: if (otd->thread_mb_pos < tmp) {\ yading@10: pthread_mutex_lock(&otd->lock);\ yading@10: td->wait_mb_pos = tmp;\ yading@10: do {\ yading@10: if (otd->thread_mb_pos >= tmp)\ yading@10: break;\ yading@10: pthread_cond_wait(&otd->cond, &otd->lock);\ yading@10: } while (1);\ yading@10: td->wait_mb_pos = INT_MAX;\ yading@10: pthread_mutex_unlock(&otd->lock);\ yading@10: }\ yading@10: } while(0); yading@10: yading@10: #define update_pos(td, mb_y, mb_x)\ yading@10: do {\ yading@10: int pos = (mb_y << 16) | (mb_x & 0xFFFF);\ yading@10: int sliced_threading = (avctx->active_thread_type == FF_THREAD_SLICE) && (num_jobs > 1);\ yading@10: int is_null = (next_td == NULL) || (prev_td == NULL);\ yading@10: int pos_check = (is_null) ? 1 :\ yading@10: (next_td != td && pos >= next_td->wait_mb_pos) ||\ yading@10: (prev_td != td && pos >= prev_td->wait_mb_pos);\ yading@10: td->thread_mb_pos = pos;\ yading@10: if (sliced_threading && pos_check) {\ yading@10: pthread_mutex_lock(&td->lock);\ yading@10: pthread_cond_broadcast(&td->cond);\ yading@10: pthread_mutex_unlock(&td->lock);\ yading@10: }\ yading@10: } while(0); yading@10: #else yading@10: #define check_thread_pos(td, otd, mb_x_check, mb_y_check) yading@10: #define update_pos(td, mb_y, mb_x) yading@10: #endif yading@10: yading@10: static void vp8_decode_mb_row_no_filter(AVCodecContext *avctx, void *tdata, yading@10: int jobnr, int threadnr) yading@10: { yading@10: VP8Context *s = avctx->priv_data; yading@10: VP8ThreadData *prev_td, *next_td, *td = &s->thread_data[threadnr]; yading@10: int mb_y = td->thread_mb_pos>>16; yading@10: int i, y, mb_x, mb_xy = mb_y*s->mb_width; yading@10: int num_jobs = s->num_jobs; yading@10: VP8Frame *curframe = s->curframe, *prev_frame = s->prev_frame; yading@10: VP56RangeCoder *c = &s->coeff_partition[mb_y & (s->num_coeff_partitions-1)]; yading@10: VP8Macroblock *mb; yading@10: uint8_t *dst[3] = { yading@10: curframe->tf.f->data[0] + 16*mb_y*s->linesize, yading@10: curframe->tf.f->data[1] + 8*mb_y*s->uvlinesize, yading@10: curframe->tf.f->data[2] + 8*mb_y*s->uvlinesize yading@10: }; yading@10: if (mb_y == 0) prev_td = td; yading@10: else prev_td = &s->thread_data[(jobnr + num_jobs - 1)%num_jobs]; yading@10: if (mb_y == s->mb_height-1) next_td = td; yading@10: else next_td = &s->thread_data[(jobnr + 1)%num_jobs]; yading@10: if (s->mb_layout == 1) yading@10: mb = s->macroblocks_base + ((s->mb_width+1)*(mb_y + 1) + 1); yading@10: else { yading@10: mb = s->macroblocks + (s->mb_height - mb_y - 1)*2; yading@10: memset(mb - 1, 0, sizeof(*mb)); // zero left macroblock yading@10: AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED*0x01010101); yading@10: } yading@10: yading@10: memset(td->left_nnz, 0, sizeof(td->left_nnz)); yading@10: // left edge of 129 for intra prediction yading@10: if (!(avctx->flags & CODEC_FLAG_EMU_EDGE)) { yading@10: for (i = 0; i < 3; i++) yading@10: for (y = 0; y < 16>>!!i; y++) yading@10: dst[i][y*curframe->tf.f->linesize[i]-1] = 129; yading@10: if (mb_y == 1) { yading@10: s->top_border[0][15] = s->top_border[0][23] = s->top_border[0][31] = 129; yading@10: } yading@10: } yading@10: yading@10: s->mv_min.x = -MARGIN; yading@10: s->mv_max.x = ((s->mb_width - 1) << 6) + MARGIN; yading@10: yading@10: for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) { yading@10: // Wait for previous thread to read mb_x+2, and reach mb_y-1. yading@10: if (prev_td != td) { yading@10: if (threadnr != 0) { yading@10: check_thread_pos(td, prev_td, mb_x+1, mb_y-1); yading@10: } else { yading@10: check_thread_pos(td, prev_td, (s->mb_width+3) + (mb_x+1), mb_y-1); yading@10: } yading@10: } yading@10: yading@10: s->vdsp.prefetch(dst[0] + (mb_x&3)*4*s->linesize + 64, s->linesize, 4); yading@10: s->vdsp.prefetch(dst[1] + (mb_x&7)*s->uvlinesize + 64, dst[2] - dst[1], 2); yading@10: yading@10: if (!s->mb_layout) yading@10: decode_mb_mode(s, mb, mb_x, mb_y, curframe->seg_map->data + mb_xy, yading@10: prev_frame && prev_frame->seg_map ? yading@10: prev_frame->seg_map->data + mb_xy : NULL, 0); yading@10: yading@10: prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_PREVIOUS); yading@10: yading@10: if (!mb->skip) yading@10: decode_mb_coeffs(s, td, c, mb, s->top_nnz[mb_x], td->left_nnz); yading@10: yading@10: if (mb->mode <= MODE_I4x4) yading@10: intra_predict(s, td, dst, mb, mb_x, mb_y); yading@10: else yading@10: inter_predict(s, td, dst, mb, mb_x, mb_y); yading@10: yading@10: prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_GOLDEN); yading@10: yading@10: if (!mb->skip) { yading@10: idct_mb(s, td, dst, mb); yading@10: } else { yading@10: AV_ZERO64(td->left_nnz); yading@10: AV_WN64(s->top_nnz[mb_x], 0); // array of 9, so unaligned yading@10: yading@10: // Reset DC block predictors if they would exist if the mb had coefficients yading@10: if (mb->mode != MODE_I4x4 && mb->mode != VP8_MVMODE_SPLIT) { yading@10: td->left_nnz[8] = 0; yading@10: s->top_nnz[mb_x][8] = 0; yading@10: } yading@10: } yading@10: yading@10: if (s->deblock_filter) yading@10: filter_level_for_mb(s, mb, &td->filter_strength[mb_x]); yading@10: yading@10: if (s->deblock_filter && num_jobs != 1 && threadnr == num_jobs-1) { yading@10: if (s->filter.simple) yading@10: backup_mb_border(s->top_border[mb_x+1], dst[0], NULL, NULL, s->linesize, 0, 1); yading@10: else yading@10: backup_mb_border(s->top_border[mb_x+1], dst[0], dst[1], dst[2], s->linesize, s->uvlinesize, 0); yading@10: } yading@10: yading@10: prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_GOLDEN2); yading@10: yading@10: dst[0] += 16; yading@10: dst[1] += 8; yading@10: dst[2] += 8; yading@10: s->mv_min.x -= 64; yading@10: s->mv_max.x -= 64; yading@10: yading@10: if (mb_x == s->mb_width+1) { yading@10: update_pos(td, mb_y, s->mb_width+3); yading@10: } else { yading@10: update_pos(td, mb_y, mb_x); yading@10: } yading@10: } yading@10: } yading@10: yading@10: static void vp8_filter_mb_row(AVCodecContext *avctx, void *tdata, yading@10: int jobnr, int threadnr) yading@10: { yading@10: VP8Context *s = avctx->priv_data; yading@10: VP8ThreadData *td = &s->thread_data[threadnr]; yading@10: int mb_x, mb_y = td->thread_mb_pos>>16, num_jobs = s->num_jobs; yading@10: AVFrame *curframe = s->curframe->tf.f; yading@10: VP8Macroblock *mb; yading@10: VP8ThreadData *prev_td, *next_td; yading@10: uint8_t *dst[3] = { yading@10: curframe->data[0] + 16*mb_y*s->linesize, yading@10: curframe->data[1] + 8*mb_y*s->uvlinesize, yading@10: curframe->data[2] + 8*mb_y*s->uvlinesize yading@10: }; yading@10: yading@10: if (s->mb_layout == 1) yading@10: mb = s->macroblocks_base + ((s->mb_width+1)*(mb_y + 1) + 1); yading@10: else yading@10: mb = s->macroblocks + (s->mb_height - mb_y - 1)*2; yading@10: yading@10: if (mb_y == 0) prev_td = td; yading@10: else prev_td = &s->thread_data[(jobnr + num_jobs - 1)%num_jobs]; yading@10: if (mb_y == s->mb_height-1) next_td = td; yading@10: else next_td = &s->thread_data[(jobnr + 1)%num_jobs]; yading@10: yading@10: for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb++) { yading@10: VP8FilterStrength *f = &td->filter_strength[mb_x]; yading@10: if (prev_td != td) { yading@10: check_thread_pos(td, prev_td, (mb_x+1) + (s->mb_width+3), mb_y-1); yading@10: } yading@10: if (next_td != td) yading@10: if (next_td != &s->thread_data[0]) { yading@10: check_thread_pos(td, next_td, mb_x+1, mb_y+1); yading@10: } yading@10: yading@10: if (num_jobs == 1) { yading@10: if (s->filter.simple) yading@10: backup_mb_border(s->top_border[mb_x+1], dst[0], NULL, NULL, s->linesize, 0, 1); yading@10: else yading@10: backup_mb_border(s->top_border[mb_x+1], dst[0], dst[1], dst[2], s->linesize, s->uvlinesize, 0); yading@10: } yading@10: yading@10: if (s->filter.simple) yading@10: filter_mb_simple(s, dst[0], f, mb_x, mb_y); yading@10: else yading@10: filter_mb(s, dst, f, mb_x, mb_y); yading@10: dst[0] += 16; yading@10: dst[1] += 8; yading@10: dst[2] += 8; yading@10: yading@10: update_pos(td, mb_y, (s->mb_width+3) + mb_x); yading@10: } yading@10: } yading@10: yading@10: static int vp8_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata, yading@10: int jobnr, int threadnr) yading@10: { yading@10: VP8Context *s = avctx->priv_data; yading@10: VP8ThreadData *td = &s->thread_data[jobnr]; yading@10: VP8ThreadData *next_td = NULL, *prev_td = NULL; yading@10: VP8Frame *curframe = s->curframe; yading@10: int mb_y, num_jobs = s->num_jobs; yading@10: td->thread_nr = threadnr; yading@10: for (mb_y = jobnr; mb_y < s->mb_height; mb_y += num_jobs) { yading@10: if (mb_y >= s->mb_height) break; yading@10: td->thread_mb_pos = mb_y<<16; yading@10: vp8_decode_mb_row_no_filter(avctx, tdata, jobnr, threadnr); yading@10: if (s->deblock_filter) yading@10: vp8_filter_mb_row(avctx, tdata, jobnr, threadnr); yading@10: update_pos(td, mb_y, INT_MAX & 0xFFFF); yading@10: yading@10: s->mv_min.y -= 64; yading@10: s->mv_max.y -= 64; yading@10: yading@10: if (avctx->active_thread_type == FF_THREAD_FRAME) yading@10: ff_thread_report_progress(&curframe->tf, mb_y, 0); yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static int vp8_decode_frame(AVCodecContext *avctx, void *data, int *got_frame, yading@10: AVPacket *avpkt) yading@10: { yading@10: VP8Context *s = avctx->priv_data; yading@10: int ret, i, referenced, num_jobs; yading@10: enum AVDiscard skip_thresh; yading@10: VP8Frame *av_uninit(curframe), *prev_frame; yading@10: yading@10: if ((ret = decode_frame_header(s, avpkt->data, avpkt->size)) < 0) yading@10: goto err; yading@10: yading@10: prev_frame = s->framep[VP56_FRAME_CURRENT]; yading@10: yading@10: referenced = s->update_last || s->update_golden == VP56_FRAME_CURRENT yading@10: || s->update_altref == VP56_FRAME_CURRENT; yading@10: yading@10: skip_thresh = !referenced ? AVDISCARD_NONREF : yading@10: !s->keyframe ? AVDISCARD_NONKEY : AVDISCARD_ALL; yading@10: yading@10: if (avctx->skip_frame >= skip_thresh) { yading@10: s->invisible = 1; yading@10: memcpy(&s->next_framep[0], &s->framep[0], sizeof(s->framep[0]) * 4); yading@10: goto skip_decode; yading@10: } yading@10: s->deblock_filter = s->filter.level && avctx->skip_loop_filter < skip_thresh; yading@10: yading@10: // release no longer referenced frames yading@10: for (i = 0; i < 5; i++) yading@10: if (s->frames[i].tf.f->data[0] && yading@10: &s->frames[i] != prev_frame && yading@10: &s->frames[i] != s->framep[VP56_FRAME_PREVIOUS] && yading@10: &s->frames[i] != s->framep[VP56_FRAME_GOLDEN] && yading@10: &s->frames[i] != s->framep[VP56_FRAME_GOLDEN2]) yading@10: vp8_release_frame(s, &s->frames[i]); yading@10: yading@10: // find a free buffer yading@10: for (i = 0; i < 5; i++) yading@10: if (&s->frames[i] != prev_frame && yading@10: &s->frames[i] != s->framep[VP56_FRAME_PREVIOUS] && yading@10: &s->frames[i] != s->framep[VP56_FRAME_GOLDEN] && yading@10: &s->frames[i] != s->framep[VP56_FRAME_GOLDEN2]) { yading@10: curframe = s->framep[VP56_FRAME_CURRENT] = &s->frames[i]; yading@10: break; yading@10: } yading@10: if (i == 5) { yading@10: av_log(avctx, AV_LOG_FATAL, "Ran out of free frames!\n"); yading@10: abort(); yading@10: } yading@10: if (curframe->tf.f->data[0]) yading@10: vp8_release_frame(s, curframe); yading@10: yading@10: // Given that arithmetic probabilities are updated every frame, it's quite likely yading@10: // that the values we have on a random interframe are complete junk if we didn't yading@10: // start decode on a keyframe. So just don't display anything rather than junk. yading@10: if (!s->keyframe && (!s->framep[VP56_FRAME_PREVIOUS] || yading@10: !s->framep[VP56_FRAME_GOLDEN] || yading@10: !s->framep[VP56_FRAME_GOLDEN2])) { yading@10: av_log(avctx, AV_LOG_WARNING, "Discarding interframe without a prior keyframe!\n"); yading@10: ret = AVERROR_INVALIDDATA; yading@10: goto err; yading@10: } yading@10: yading@10: curframe->tf.f->key_frame = s->keyframe; yading@10: curframe->tf.f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P; yading@10: if ((ret = vp8_alloc_frame(s, curframe, referenced)) < 0) yading@10: goto err; yading@10: yading@10: // check if golden and altref are swapped yading@10: if (s->update_altref != VP56_FRAME_NONE) { yading@10: s->next_framep[VP56_FRAME_GOLDEN2] = s->framep[s->update_altref]; yading@10: } else { yading@10: s->next_framep[VP56_FRAME_GOLDEN2] = s->framep[VP56_FRAME_GOLDEN2]; yading@10: } yading@10: if (s->update_golden != VP56_FRAME_NONE) { yading@10: s->next_framep[VP56_FRAME_GOLDEN] = s->framep[s->update_golden]; yading@10: } else { yading@10: s->next_framep[VP56_FRAME_GOLDEN] = s->framep[VP56_FRAME_GOLDEN]; yading@10: } yading@10: if (s->update_last) { yading@10: s->next_framep[VP56_FRAME_PREVIOUS] = curframe; yading@10: } else { yading@10: s->next_framep[VP56_FRAME_PREVIOUS] = s->framep[VP56_FRAME_PREVIOUS]; yading@10: } yading@10: s->next_framep[VP56_FRAME_CURRENT] = curframe; yading@10: yading@10: ff_thread_finish_setup(avctx); yading@10: yading@10: s->linesize = curframe->tf.f->linesize[0]; yading@10: s->uvlinesize = curframe->tf.f->linesize[1]; yading@10: yading@10: if (!s->thread_data[0].edge_emu_buffer) yading@10: for (i = 0; i < MAX_THREADS; i++) yading@10: s->thread_data[i].edge_emu_buffer = av_malloc(21*s->linesize); yading@10: yading@10: memset(s->top_nnz, 0, s->mb_width*sizeof(*s->top_nnz)); yading@10: /* Zero macroblock structures for top/top-left prediction from outside the frame. */ yading@10: if (!s->mb_layout) yading@10: memset(s->macroblocks + s->mb_height*2 - 1, 0, (s->mb_width+1)*sizeof(*s->macroblocks)); yading@10: if (!s->mb_layout && s->keyframe) yading@10: memset(s->intra4x4_pred_mode_top, DC_PRED, s->mb_width*4); yading@10: yading@10: // top edge of 127 for intra prediction yading@10: if (!(avctx->flags & CODEC_FLAG_EMU_EDGE)) { yading@10: s->top_border[0][15] = s->top_border[0][23] = 127; yading@10: s->top_border[0][31] = 127; yading@10: memset(s->top_border[1], 127, s->mb_width*sizeof(*s->top_border)); yading@10: } yading@10: memset(s->ref_count, 0, sizeof(s->ref_count)); yading@10: yading@10: yading@10: // Make sure the previous frame has read its segmentation map, yading@10: // if we re-use the same map. yading@10: if (prev_frame && s->segmentation.enabled && !s->segmentation.update_map) yading@10: ff_thread_await_progress(&prev_frame->tf, 1, 0); yading@10: yading@10: if (s->mb_layout == 1) yading@10: vp8_decode_mv_mb_modes(avctx, curframe, prev_frame); yading@10: yading@10: if (avctx->active_thread_type == FF_THREAD_FRAME) yading@10: num_jobs = 1; yading@10: else yading@10: num_jobs = FFMIN(s->num_coeff_partitions, avctx->thread_count); yading@10: s->num_jobs = num_jobs; yading@10: s->curframe = curframe; yading@10: s->prev_frame = prev_frame; yading@10: s->mv_min.y = -MARGIN; yading@10: s->mv_max.y = ((s->mb_height - 1) << 6) + MARGIN; yading@10: for (i = 0; i < MAX_THREADS; i++) { yading@10: s->thread_data[i].thread_mb_pos = 0; yading@10: s->thread_data[i].wait_mb_pos = INT_MAX; yading@10: } yading@10: avctx->execute2(avctx, vp8_decode_mb_row_sliced, s->thread_data, NULL, num_jobs); yading@10: yading@10: ff_thread_report_progress(&curframe->tf, INT_MAX, 0); yading@10: memcpy(&s->framep[0], &s->next_framep[0], sizeof(s->framep[0]) * 4); yading@10: yading@10: skip_decode: yading@10: // if future frames don't use the updated probabilities, yading@10: // reset them to the values we saved yading@10: if (!s->update_probabilities) yading@10: s->prob[0] = s->prob[1]; yading@10: yading@10: if (!s->invisible) { yading@10: if ((ret = av_frame_ref(data, curframe->tf.f)) < 0) yading@10: return ret; yading@10: *got_frame = 1; yading@10: } yading@10: yading@10: return avpkt->size; yading@10: err: yading@10: memcpy(&s->next_framep[0], &s->framep[0], sizeof(s->framep[0]) * 4); yading@10: return ret; yading@10: } yading@10: yading@10: static av_cold int vp8_decode_free(AVCodecContext *avctx) yading@10: { yading@10: VP8Context *s = avctx->priv_data; yading@10: int i; yading@10: yading@10: vp8_decode_flush_impl(avctx, 1); yading@10: for (i = 0; i < FF_ARRAY_ELEMS(s->frames); i++) yading@10: av_frame_free(&s->frames[i].tf.f); yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static av_cold int vp8_init_frames(VP8Context *s) yading@10: { yading@10: int i; yading@10: for (i = 0; i < FF_ARRAY_ELEMS(s->frames); i++) { yading@10: s->frames[i].tf.f = av_frame_alloc(); yading@10: if (!s->frames[i].tf.f) yading@10: return AVERROR(ENOMEM); yading@10: } yading@10: return 0; yading@10: } yading@10: yading@10: static av_cold int vp8_decode_init(AVCodecContext *avctx) yading@10: { yading@10: VP8Context *s = avctx->priv_data; yading@10: int ret; yading@10: yading@10: s->avctx = avctx; yading@10: avctx->pix_fmt = AV_PIX_FMT_YUV420P; yading@10: avctx->internal->allocate_progress = 1; yading@10: yading@10: ff_videodsp_init(&s->vdsp, 8); yading@10: ff_h264_pred_init(&s->hpc, AV_CODEC_ID_VP8, 8, 1); yading@10: ff_vp8dsp_init(&s->vp8dsp); yading@10: yading@10: if ((ret = vp8_init_frames(s)) < 0) { yading@10: vp8_decode_free(avctx); yading@10: return ret; yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static av_cold int vp8_decode_init_thread_copy(AVCodecContext *avctx) yading@10: { yading@10: VP8Context *s = avctx->priv_data; yading@10: int ret; yading@10: yading@10: s->avctx = avctx; yading@10: yading@10: if ((ret = vp8_init_frames(s)) < 0) { yading@10: vp8_decode_free(avctx); yading@10: return ret; yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: #define REBASE(pic) \ yading@10: pic ? pic - &s_src->frames[0] + &s->frames[0] : NULL yading@10: yading@10: static int vp8_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src) yading@10: { yading@10: VP8Context *s = dst->priv_data, *s_src = src->priv_data; yading@10: int i; yading@10: yading@10: if (s->macroblocks_base && yading@10: (s_src->mb_width != s->mb_width || s_src->mb_height != s->mb_height)) { yading@10: free_buffers(s); yading@10: s->mb_width = s_src->mb_width; yading@10: s->mb_height = s_src->mb_height; yading@10: } yading@10: yading@10: s->prob[0] = s_src->prob[!s_src->update_probabilities]; yading@10: s->segmentation = s_src->segmentation; yading@10: s->lf_delta = s_src->lf_delta; yading@10: memcpy(s->sign_bias, s_src->sign_bias, sizeof(s->sign_bias)); yading@10: yading@10: for (i = 0; i < FF_ARRAY_ELEMS(s_src->frames); i++) { yading@10: if (s_src->frames[i].tf.f->data[0]) { yading@10: int ret = vp8_ref_frame(s, &s->frames[i], &s_src->frames[i]); yading@10: if (ret < 0) yading@10: return ret; yading@10: } yading@10: } yading@10: yading@10: s->framep[0] = REBASE(s_src->next_framep[0]); yading@10: s->framep[1] = REBASE(s_src->next_framep[1]); yading@10: s->framep[2] = REBASE(s_src->next_framep[2]); yading@10: s->framep[3] = REBASE(s_src->next_framep[3]); yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static unsigned apply_padding(unsigned size) { return size + (size & 1); } yading@10: yading@10: static int webp_decode_frame(AVCodecContext *avctx, void *data, int *data_size, yading@10: AVPacket *avpkt) yading@10: { yading@10: const uint8_t *buf = avpkt->data; yading@10: int buf_size = avpkt->size; yading@10: AVPacket pkt = *avpkt; yading@10: yading@10: if (buf_size >= 16 yading@10: && AV_RL32(buf ) == AV_RL32("RIFF") yading@10: && AV_RL32(buf+ 8) == AV_RL32("WEBP")) { yading@10: unsigned riff_size = apply_padding(AV_RL32(buf+4)) + 8; yading@10: buf += 12; // Skip over main header yading@10: buf_size -= 12; yading@10: if (buf_size < 8 || riff_size < 8) { yading@10: av_log(avctx, AV_LOG_ERROR, "Incomplete header.\n"); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: if (AV_RL32(buf) == AV_RL32("VP8L")) { yading@10: av_log(avctx, AV_LOG_ERROR, "Unsupported WebP lossless format.\n"); yading@10: return AVERROR_PATCHWELCOME; yading@10: } yading@10: if (AV_RL32(buf) == AV_RL32("VP8X") && AV_RL32(buf+4) < (unsigned)buf_size) { yading@10: unsigned size = apply_padding(AV_RL32(buf+4) + 8); yading@10: buf += size; yading@10: buf_size -= size; yading@10: } yading@10: if (buf_size >= 8 yading@10: && AV_RL32(buf) == AV_RL32("ALPH") && AV_RL32(buf+4) < (unsigned)buf_size) { yading@10: unsigned size = apply_padding(AV_RL32(buf+4) + 8); yading@10: buf += size; yading@10: buf_size -= size; yading@10: av_log(avctx, AV_LOG_WARNING, "Skipping alpha plane\n"); yading@10: } yading@10: if (buf_size >= 8 && AV_RL32(buf) == AV_RL32("VP8 ")) { yading@10: buf += 8; yading@10: buf_size -= 8; yading@10: } yading@10: } yading@10: pkt.data = buf; yading@10: pkt.size = buf_size; yading@10: yading@10: return vp8_decode_frame(avctx, data, data_size, &pkt); yading@10: } yading@10: yading@10: AVCodec ff_vp8_decoder = { yading@10: .name = "vp8", yading@10: .type = AVMEDIA_TYPE_VIDEO, yading@10: .id = AV_CODEC_ID_VP8, yading@10: .priv_data_size = sizeof(VP8Context), yading@10: .init = vp8_decode_init, yading@10: .close = vp8_decode_free, yading@10: .decode = vp8_decode_frame, yading@10: .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS | CODEC_CAP_SLICE_THREADS, yading@10: .flush = vp8_decode_flush, yading@10: .long_name = NULL_IF_CONFIG_SMALL("On2 VP8"), yading@10: .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp8_decode_init_thread_copy), yading@10: .update_thread_context = ONLY_IF_THREADS_ENABLED(vp8_decode_update_thread_context), yading@10: }; yading@10: yading@10: AVCodec ff_webp_decoder = { yading@10: .name = "webp", yading@10: .type = AVMEDIA_TYPE_VIDEO, yading@10: .id = AV_CODEC_ID_WEBP, yading@10: .priv_data_size = sizeof(VP8Context), yading@10: .init = vp8_decode_init, yading@10: .close = vp8_decode_free, yading@10: .decode = webp_decode_frame, yading@10: .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS | CODEC_CAP_SLICE_THREADS, yading@10: .flush = vp8_decode_flush, yading@10: .long_name = NULL_IF_CONFIG_SMALL("WebP"), yading@10: .init_thread_copy = ONLY_IF_THREADS_ENABLED(vp8_decode_init_thread_copy), yading@10: .update_thread_context = ONLY_IF_THREADS_ENABLED(vp8_decode_update_thread_context), yading@10: };