This is following Libav layout to ease merges.
OBJS-$(CONFIG_VP8_CUVID_DECODER) += cuvid.o
OBJS-$(CONFIG_VP8_MEDIACODEC_DECODER) += mediacodecdec.o
OBJS-$(CONFIG_VP8_VAAPI_ENCODER) += vaapi_encode_vp8.o
-OBJS-$(CONFIG_VP9_DECODER) += vp9.o vp9dsp.o vp56rac.o vp9dsp_8bpp.o \
- vp9dsp_10bpp.o vp9dsp_12bpp.o
+OBJS-$(CONFIG_VP9_DECODER) += vp9.o vp9data.o vp9dsp.o \
+ vp9block.o vp9prob.o vp9mvs.o vp56rac.o \
+ vp9dsp_8bpp.o vp9dsp_10bpp.o vp9dsp_12bpp.o
OBJS-$(CONFIG_VP9_CUVID_DECODER) += cuvid.o
OBJS-$(CONFIG_VP9_MEDIACODEC_DECODER) += mediacodecdec.o
OBJS-$(CONFIG_VPLAYER_DECODER) += textdec.o ass.o
#ifndef AVCODEC_AARCH64_VP9DSP_INIT_H
#define AVCODEC_AARCH64_VP9DSP_INIT_H
-#include "libavcodec/vp9dsp.h"
+#include "libavcodec/vp9.h"
void ff_vp9dsp_init_10bpp_aarch64(VP9DSPContext *dsp);
void ff_vp9dsp_init_12bpp_aarch64(VP9DSPContext *dsp);
#include "libavutil/attributes.h"
#include "libavutil/aarch64/cpu.h"
-#include "libavcodec/vp9dsp.h"
+#include "libavcodec/vp9.h"
#include "vp9dsp_init.h"
#define declare_fpel(type, sz) \
#ifndef AVCODEC_ARM_VP9DSP_INIT_H
#define AVCODEC_ARM_VP9DSP_INIT_H
-#include "libavcodec/vp9dsp.h"
+#include "libavcodec/vp9.h"
void ff_vp9dsp_init_10bpp_arm(VP9DSPContext *dsp);
void ff_vp9dsp_init_12bpp_arm(VP9DSPContext *dsp);
#include "libavutil/attributes.h"
#include "libavutil/arm/cpu.h"
-#include "libavcodec/vp9dsp.h"
+#include "libavcodec/vp9.h"
#include "vp9dsp_init.h"
#define declare_fpel(type, sz) \
*/
#include <string.h>
-#include "libavcodec/vp9dsp.h"
+#include "libavcodec/vp9.h"
#include "libavutil/mips/generic_macros_msa.h"
#include "vp9dsp_mips.h"
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
-#include "libavcodec/vp9dsp.h"
+#include "libavcodec/vp9.h"
#include "libavutil/mips/generic_macros_msa.h"
#include "vp9dsp_mips.h"
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
-#include "libavcodec/vp9dsp.h"
+#include "libavcodec/vp9.h"
#include "libavutil/mips/generic_macros_msa.h"
#include "vp9dsp_mips.h"
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
-#include "libavcodec/vp9dsp.h"
+#include "libavcodec/vp9.h"
#include "libavutil/mips/generic_macros_msa.h"
#include "vp9dsp_mips.h"
#include "config.h"
#include "libavutil/common.h"
-#include "libavcodec/vp9dsp.h"
+#include "libavcodec/vp9.h"
#include "vp9dsp_mips.h"
#if HAVE_MSA
#include "vp56.h"
#include "vp9.h"
#include "vp9data.h"
-#include "vp9dsp.h"
+#include "vp9.h"
#include "libavutil/avassert.h"
#include "libavutil/pixdesc.h"
#define VP9_SYNCCODE 0x498342
-struct VP9Filter {
- uint8_t level[8 * 8];
- uint8_t /* bit=col */ mask[2 /* 0=y, 1=uv */][2 /* 0=col, 1=row */]
- [8 /* rows */][4 /* 0=16, 1=8, 2=4, 3=inner4 */];
-};
-
-typedef struct VP9Block {
- uint8_t seg_id, intra, comp, ref[2], mode[4], uvmode, skip;
- enum FilterMode filter;
- VP56mv mv[4 /* b_idx */][2 /* ref */];
- enum BlockSize bs;
- enum TxfmMode tx, uvtx;
- enum BlockLevel bl;
- enum BlockPartition bp;
-} VP9Block;
-
-typedef struct VP9Context {
- VP9SharedContext s;
-
- VP9DSPContext dsp;
- VideoDSPContext vdsp;
- GetBitContext gb;
- VP56RangeCoder c;
- VP56RangeCoder *c_b;
- unsigned c_b_size;
- VP9Block *b_base, *b;
- int pass;
- int row, row7, col, col7;
- uint8_t *dst[3];
- ptrdiff_t y_stride, uv_stride;
-
- uint8_t ss_h, ss_v;
- uint8_t last_bpp, bpp_index, bytesperpixel;
- uint8_t last_keyframe;
- // sb_cols/rows, rows/cols and last_fmt are used for allocating all internal
- // arrays, and are thus per-thread. w/h and gf_fmt are synced between threads
- // and are therefore per-stream. pix_fmt represents the value in the header
- // of the currently processed frame.
- int w, h;
- enum AVPixelFormat pix_fmt, last_fmt, gf_fmt;
- unsigned sb_cols, sb_rows, rows, cols;
- ThreadFrame next_refs[8];
-
- struct {
- uint8_t lim_lut[64];
- uint8_t mblim_lut[64];
- } filter_lut;
- unsigned tile_row_start, tile_row_end, tile_col_start, tile_col_end;
- struct {
- prob_context p;
- uint8_t coef[4][2][2][6][6][3];
- } prob_ctx[4];
- struct {
- prob_context p;
- uint8_t coef[4][2][2][6][6][11];
- } prob;
- struct {
- unsigned y_mode[4][10];
- unsigned uv_mode[10][10];
- unsigned filter[4][3];
- unsigned mv_mode[7][4];
- unsigned intra[4][2];
- unsigned comp[5][2];
- unsigned single_ref[5][2][2];
- unsigned comp_ref[5][2];
- unsigned tx32p[2][4];
- unsigned tx16p[2][3];
- unsigned tx8p[2][2];
- unsigned skip[3][2];
- unsigned mv_joint[4];
- struct {
- unsigned sign[2];
- unsigned classes[11];
- unsigned class0[2];
- unsigned bits[10][2];
- unsigned class0_fp[2][4];
- unsigned fp[4];
- unsigned class0_hp[2];
- unsigned hp[2];
- } mv_comp[2];
- unsigned partition[4][4][4];
- unsigned coef[4][2][2][6][6][3];
- unsigned eob[4][2][2][6][6][2];
- } counts;
-
- // contextual (left/above) cache
- DECLARE_ALIGNED(16, uint8_t, left_y_nnz_ctx)[16];
- DECLARE_ALIGNED(16, uint8_t, left_mode_ctx)[16];
- DECLARE_ALIGNED(16, VP56mv, left_mv_ctx)[16][2];
- DECLARE_ALIGNED(16, uint8_t, left_uv_nnz_ctx)[2][16];
- DECLARE_ALIGNED(8, uint8_t, left_partition_ctx)[8];
- DECLARE_ALIGNED(8, uint8_t, left_skip_ctx)[8];
- DECLARE_ALIGNED(8, uint8_t, left_txfm_ctx)[8];
- DECLARE_ALIGNED(8, uint8_t, left_segpred_ctx)[8];
- DECLARE_ALIGNED(8, uint8_t, left_intra_ctx)[8];
- DECLARE_ALIGNED(8, uint8_t, left_comp_ctx)[8];
- DECLARE_ALIGNED(8, uint8_t, left_ref_ctx)[8];
- DECLARE_ALIGNED(8, uint8_t, left_filter_ctx)[8];
- uint8_t *above_partition_ctx;
- uint8_t *above_mode_ctx;
- // FIXME maybe merge some of the below in a flags field?
- uint8_t *above_y_nnz_ctx;
- uint8_t *above_uv_nnz_ctx[2];
- uint8_t *above_skip_ctx; // 1bit
- uint8_t *above_txfm_ctx; // 2bit
- uint8_t *above_segpred_ctx; // 1bit
- uint8_t *above_intra_ctx; // 1bit
- uint8_t *above_comp_ctx; // 1bit
- uint8_t *above_ref_ctx; // 2bit
- uint8_t *above_filter_ctx;
- VP56mv (*above_mv_ctx)[2];
-
- // whole-frame cache
- uint8_t *intra_pred_data[3];
- struct VP9Filter *lflvl;
- DECLARE_ALIGNED(32, uint8_t, edge_emu_buffer)[135 * 144 * 2];
-
- // block reconstruction intermediates
- int block_alloc_using_2pass;
- int16_t *block_base, *block, *uvblock_base[2], *uvblock[2];
- uint8_t *eob_base, *uveob_base[2], *eob, *uveob[2];
- struct { int x, y; } min_mv, max_mv;
- DECLARE_ALIGNED(32, uint8_t, tmp_y)[64 * 64 * 2];
- DECLARE_ALIGNED(32, uint8_t, tmp_uv)[2][64 * 64 * 2];
- uint16_t mvscale[3][2];
- uint8_t mvstep[3][2];
-} VP9Context;
-
-static const uint8_t bwh_tab[2][N_BS_SIZES][2] = {
- {
- { 16, 16 }, { 16, 8 }, { 8, 16 }, { 8, 8 }, { 8, 4 }, { 4, 8 },
- { 4, 4 }, { 4, 2 }, { 2, 4 }, { 2, 2 }, { 2, 1 }, { 1, 2 }, { 1, 1 },
- }, {
- { 8, 8 }, { 8, 4 }, { 4, 8 }, { 4, 4 }, { 4, 2 }, { 2, 4 },
- { 2, 2 }, { 2, 1 }, { 1, 2 }, { 1, 1 }, { 1, 1 }, { 1, 1 }, { 1, 1 },
- }
-};
-
static void vp9_unref_frame(AVCodecContext *ctx, VP9Frame *f)
{
ff_thread_release_buffer(ctx, &f->tf);
quvac = av_clip_uintp2(qyac + s->s.h.uvac_qdelta, 8);
qyac = av_clip_uintp2(qyac, 8);
- s->s.h.segmentation.feat[i].qmul[0][0] = vp9_dc_qlookup[s->bpp_index][qydc];
- s->s.h.segmentation.feat[i].qmul[0][1] = vp9_ac_qlookup[s->bpp_index][qyac];
- s->s.h.segmentation.feat[i].qmul[1][0] = vp9_dc_qlookup[s->bpp_index][quvdc];
- s->s.h.segmentation.feat[i].qmul[1][1] = vp9_ac_qlookup[s->bpp_index][quvac];
+ s->s.h.segmentation.feat[i].qmul[0][0] = ff_vp9_dc_qlookup[s->bpp_index][qydc];
+ s->s.h.segmentation.feat[i].qmul[0][1] = ff_vp9_ac_qlookup[s->bpp_index][qyac];
+ s->s.h.segmentation.feat[i].qmul[1][0] = ff_vp9_dc_qlookup[s->bpp_index][quvdc];
+ s->s.h.segmentation.feat[i].qmul[1][1] = ff_vp9_ac_qlookup[s->bpp_index][quvac];
sh = s->s.h.filter.level >= 32;
if (s->s.h.segmentation.enabled && s->s.h.segmentation.feat[i].lf_enabled) {
if (s->s.h.keyframe || s->s.h.errorres || (s->s.h.intraonly && s->s.h.resetctx == 3)) {
s->prob_ctx[0].p = s->prob_ctx[1].p = s->prob_ctx[2].p =
- s->prob_ctx[3].p = vp9_default_probs;
- memcpy(s->prob_ctx[0].coef, vp9_default_coef_probs,
- sizeof(vp9_default_coef_probs));
- memcpy(s->prob_ctx[1].coef, vp9_default_coef_probs,
- sizeof(vp9_default_coef_probs));
- memcpy(s->prob_ctx[2].coef, vp9_default_coef_probs,
- sizeof(vp9_default_coef_probs));
- memcpy(s->prob_ctx[3].coef, vp9_default_coef_probs,
- sizeof(vp9_default_coef_probs));
+ s->prob_ctx[3].p = ff_vp9_default_probs;
+ memcpy(s->prob_ctx[0].coef, ff_vp9_default_coef_probs,
+ sizeof(ff_vp9_default_coef_probs));
+ memcpy(s->prob_ctx[1].coef, ff_vp9_default_coef_probs,
+ sizeof(ff_vp9_default_coef_probs));
+ memcpy(s->prob_ctx[2].coef, ff_vp9_default_coef_probs,
+ sizeof(ff_vp9_default_coef_probs));
+ memcpy(s->prob_ctx[3].coef, ff_vp9_default_coef_probs,
+ sizeof(ff_vp9_default_coef_probs));
} else if (s->s.h.intraonly && s->s.h.resetctx == 2) {
- s->prob_ctx[c].p = vp9_default_probs;
- memcpy(s->prob_ctx[c].coef, vp9_default_coef_probs,
- sizeof(vp9_default_coef_probs));
+ s->prob_ctx[c].p = ff_vp9_default_probs;
+ memcpy(s->prob_ctx[c].coef, ff_vp9_default_coef_probs,
+ sizeof(ff_vp9_default_coef_probs));
}
// next 16 bits is size of the rest of the header (arith-coded)
return (data2 - data) + size2;
}
-static av_always_inline void clamp_mv(VP56mv *dst, const VP56mv *src,
- VP9Context *s)
-{
- dst->x = av_clip(src->x, s->min_mv.x, s->max_mv.x);
- dst->y = av_clip(src->y, s->min_mv.y, s->max_mv.y);
-}
-
-static void find_ref_mvs(VP9Context *s,
- VP56mv *pmv, int ref, int z, int idx, int sb)
+static void decode_sb(AVCodecContext *ctx, int row, int col, struct VP9Filter *lflvl,
+ ptrdiff_t yoff, ptrdiff_t uvoff, enum BlockLevel bl)
{
- static const int8_t mv_ref_blk_off[N_BS_SIZES][8][2] = {
- [BS_64x64] = {{ 3, -1 }, { -1, 3 }, { 4, -1 }, { -1, 4 },
- { -1, -1 }, { 0, -1 }, { -1, 0 }, { 6, -1 }},
- [BS_64x32] = {{ 0, -1 }, { -1, 0 }, { 4, -1 }, { -1, 2 },
- { -1, -1 }, { 0, -3 }, { -3, 0 }, { 2, -1 }},
- [BS_32x64] = {{ -1, 0 }, { 0, -1 }, { -1, 4 }, { 2, -1 },
- { -1, -1 }, { -3, 0 }, { 0, -3 }, { -1, 2 }},
- [BS_32x32] = {{ 1, -1 }, { -1, 1 }, { 2, -1 }, { -1, 2 },
- { -1, -1 }, { 0, -3 }, { -3, 0 }, { -3, -3 }},
- [BS_32x16] = {{ 0, -1 }, { -1, 0 }, { 2, -1 }, { -1, -1 },
- { -1, 1 }, { 0, -3 }, { -3, 0 }, { -3, -3 }},
- [BS_16x32] = {{ -1, 0 }, { 0, -1 }, { -1, 2 }, { -1, -1 },
- { 1, -1 }, { -3, 0 }, { 0, -3 }, { -3, -3 }},
- [BS_16x16] = {{ 0, -1 }, { -1, 0 }, { 1, -1 }, { -1, 1 },
- { -1, -1 }, { 0, -3 }, { -3, 0 }, { -3, -3 }},
- [BS_16x8] = {{ 0, -1 }, { -1, 0 }, { 1, -1 }, { -1, -1 },
- { 0, -2 }, { -2, 0 }, { -2, -1 }, { -1, -2 }},
- [BS_8x16] = {{ -1, 0 }, { 0, -1 }, { -1, 1 }, { -1, -1 },
- { -2, 0 }, { 0, -2 }, { -1, -2 }, { -2, -1 }},
- [BS_8x8] = {{ 0, -1 }, { -1, 0 }, { -1, -1 }, { 0, -2 },
- { -2, 0 }, { -1, -2 }, { -2, -1 }, { -2, -2 }},
- [BS_8x4] = {{ 0, -1 }, { -1, 0 }, { -1, -1 }, { 0, -2 },
- { -2, 0 }, { -1, -2 }, { -2, -1 }, { -2, -2 }},
- [BS_4x8] = {{ 0, -1 }, { -1, 0 }, { -1, -1 }, { 0, -2 },
- { -2, 0 }, { -1, -2 }, { -2, -1 }, { -2, -2 }},
- [BS_4x4] = {{ 0, -1 }, { -1, 0 }, { -1, -1 }, { 0, -2 },
- { -2, 0 }, { -1, -2 }, { -2, -1 }, { -2, -2 }},
- };
- VP9Block *b = s->b;
- int row = s->row, col = s->col, row7 = s->row7;
- const int8_t (*p)[2] = mv_ref_blk_off[b->bs];
-#define INVALID_MV 0x80008000U
- uint32_t mem = INVALID_MV, mem_sub8x8 = INVALID_MV;
- int i;
-
-#define RETURN_DIRECT_MV(mv) \
- do { \
- uint32_t m = AV_RN32A(&mv); \
- if (!idx) { \
- AV_WN32A(pmv, m); \
- return; \
- } else if (mem == INVALID_MV) { \
- mem = m; \
- } else if (m != mem) { \
- AV_WN32A(pmv, m); \
- return; \
- } \
- } while (0)
-
- if (sb >= 0) {
- if (sb == 2 || sb == 1) {
- RETURN_DIRECT_MV(b->mv[0][z]);
- } else if (sb == 3) {
- RETURN_DIRECT_MV(b->mv[2][z]);
- RETURN_DIRECT_MV(b->mv[1][z]);
- RETURN_DIRECT_MV(b->mv[0][z]);
- }
+ VP9Context *s = ctx->priv_data;
+ int c = ((s->above_partition_ctx[col] >> (3 - bl)) & 1) |
+ (((s->left_partition_ctx[row & 0x7] >> (3 - bl)) & 1) << 1);
+ const uint8_t *p = s->s.h.keyframe || s->s.h.intraonly ? ff_vp9_default_kf_partition_probs[bl][c] :
+ s->prob.p.partition[bl][c];
+ enum BlockPartition bp;
+ ptrdiff_t hbs = 4 >> bl;
+ AVFrame *f = s->s.frames[CUR_FRAME].tf.f;
+ ptrdiff_t y_stride = f->linesize[0], uv_stride = f->linesize[1];
+ int bytesperpixel = s->bytesperpixel;
-#define RETURN_MV(mv) \
- do { \
- if (sb > 0) { \
- VP56mv tmp; \
- uint32_t m; \
- av_assert2(idx == 1); \
- av_assert2(mem != INVALID_MV); \
- if (mem_sub8x8 == INVALID_MV) { \
- clamp_mv(&tmp, &mv, s); \
- m = AV_RN32A(&tmp); \
- if (m != mem) { \
- AV_WN32A(pmv, m); \
- return; \
- } \
- mem_sub8x8 = AV_RN32A(&mv); \
- } else if (mem_sub8x8 != AV_RN32A(&mv)) { \
- clamp_mv(&tmp, &mv, s); \
- m = AV_RN32A(&tmp); \
- if (m != mem) { \
- AV_WN32A(pmv, m); \
- } else { \
- /* BUG I'm pretty sure this isn't the intention */ \
- AV_WN32A(pmv, 0); \
- } \
- return; \
- } \
- } else { \
- uint32_t m = AV_RN32A(&mv); \
- if (!idx) { \
- clamp_mv(pmv, &mv, s); \
- return; \
- } else if (mem == INVALID_MV) { \
- mem = m; \
- } else if (m != mem) { \
- clamp_mv(pmv, &mv, s); \
- return; \
- } \
- } \
- } while (0)
-
- if (row > 0) {
- struct VP9mvrefPair *mv = &s->s.frames[CUR_FRAME].mv[(row - 1) * s->sb_cols * 8 + col];
- if (mv->ref[0] == ref) {
- RETURN_MV(s->above_mv_ctx[2 * col + (sb & 1)][0]);
- } else if (mv->ref[1] == ref) {
- RETURN_MV(s->above_mv_ctx[2 * col + (sb & 1)][1]);
+ if (bl == BL_8X8) {
+ bp = vp8_rac_get_tree(&s->c, ff_vp9_partition_tree, p);
+ ff_vp9_decode_block(ctx, row, col, lflvl, yoff, uvoff, bl, bp);
+ } else if (col + hbs < s->cols) { // FIXME why not <=?
+ if (row + hbs < s->rows) { // FIXME why not <=?
+ bp = vp8_rac_get_tree(&s->c, ff_vp9_partition_tree, p);
+ switch (bp) {
+ case PARTITION_NONE:
+ ff_vp9_decode_block(ctx, row, col, lflvl, yoff, uvoff, bl, bp);
+ break;
+ case PARTITION_H:
+ ff_vp9_decode_block(ctx, row, col, lflvl, yoff, uvoff, bl, bp);
+ yoff += hbs * 8 * y_stride;
+ uvoff += hbs * 8 * uv_stride >> s->ss_v;
+ ff_vp9_decode_block(ctx, row + hbs, col, lflvl, yoff, uvoff, bl, bp);
+ break;
+ case PARTITION_V:
+ ff_vp9_decode_block(ctx, row, col, lflvl, yoff, uvoff, bl, bp);
+ yoff += hbs * 8 * bytesperpixel;
+ uvoff += hbs * 8 * bytesperpixel >> s->ss_h;
+ ff_vp9_decode_block(ctx, row, col + hbs, lflvl, yoff, uvoff, bl, bp);
+ break;
+ case PARTITION_SPLIT:
+ decode_sb(ctx, row, col, lflvl, yoff, uvoff, bl + 1);
+ decode_sb(ctx, row, col + hbs, lflvl,
+ yoff + 8 * hbs * bytesperpixel,
+ uvoff + (8 * hbs * bytesperpixel >> s->ss_h), bl + 1);
+ yoff += hbs * 8 * y_stride;
+ uvoff += hbs * 8 * uv_stride >> s->ss_v;
+ decode_sb(ctx, row + hbs, col, lflvl, yoff, uvoff, bl + 1);
+ decode_sb(ctx, row + hbs, col + hbs, lflvl,
+ yoff + 8 * hbs * bytesperpixel,
+ uvoff + (8 * hbs * bytesperpixel >> s->ss_h), bl + 1);
+ break;
+ default:
+ av_assert0(0);
}
+ } else if (vp56_rac_get_prob_branchy(&s->c, p[1])) {
+ bp = PARTITION_SPLIT;
+ decode_sb(ctx, row, col, lflvl, yoff, uvoff, bl + 1);
+ decode_sb(ctx, row, col + hbs, lflvl,
+ yoff + 8 * hbs * bytesperpixel,
+ uvoff + (8 * hbs * bytesperpixel >> s->ss_h), bl + 1);
+ } else {
+ bp = PARTITION_H;
+ ff_vp9_decode_block(ctx, row, col, lflvl, yoff, uvoff, bl, bp);
}
- if (col > s->tile_col_start) {
- struct VP9mvrefPair *mv = &s->s.frames[CUR_FRAME].mv[row * s->sb_cols * 8 + col - 1];
- if (mv->ref[0] == ref) {
- RETURN_MV(s->left_mv_ctx[2 * row7 + (sb >> 1)][0]);
- } else if (mv->ref[1] == ref) {
- RETURN_MV(s->left_mv_ctx[2 * row7 + (sb >> 1)][1]);
- }
+ } else if (row + hbs < s->rows) { // FIXME why not <=?
+ if (vp56_rac_get_prob_branchy(&s->c, p[2])) {
+ bp = PARTITION_SPLIT;
+ decode_sb(ctx, row, col, lflvl, yoff, uvoff, bl + 1);
+ yoff += hbs * 8 * y_stride;
+ uvoff += hbs * 8 * uv_stride >> s->ss_v;
+ decode_sb(ctx, row + hbs, col, lflvl, yoff, uvoff, bl + 1);
+ } else {
+ bp = PARTITION_V;
+ ff_vp9_decode_block(ctx, row, col, lflvl, yoff, uvoff, bl, bp);
}
- i = 2;
} else {
- i = 0;
+ bp = PARTITION_SPLIT;
+ decode_sb(ctx, row, col, lflvl, yoff, uvoff, bl + 1);
}
+ s->counts.partition[bl][c][bp]++;
+}
- // previously coded MVs in this neighbourhood, using same reference frame
- for (; i < 8; i++) {
- int c = p[i][0] + col, r = p[i][1] + row;
-
- if (c >= s->tile_col_start && c < s->cols && r >= 0 && r < s->rows) {
- struct VP9mvrefPair *mv = &s->s.frames[CUR_FRAME].mv[r * s->sb_cols * 8 + c];
+static void decode_sb_mem(AVCodecContext *ctx, int row, int col, struct VP9Filter *lflvl,
+ ptrdiff_t yoff, ptrdiff_t uvoff, enum BlockLevel bl)
+{
+ VP9Context *s = ctx->priv_data;
+ VP9Block *b = s->b;
+ ptrdiff_t hbs = 4 >> bl;
+ AVFrame *f = s->s.frames[CUR_FRAME].tf.f;
+ ptrdiff_t y_stride = f->linesize[0], uv_stride = f->linesize[1];
+ int bytesperpixel = s->bytesperpixel;
- if (mv->ref[0] == ref) {
- RETURN_MV(mv->mv[0]);
- } else if (mv->ref[1] == ref) {
- RETURN_MV(mv->mv[1]);
+ if (bl == BL_8X8) {
+ av_assert2(b->bl == BL_8X8);
+ ff_vp9_decode_block(ctx, row, col, lflvl, yoff, uvoff, b->bl, b->bp);
+ } else if (s->b->bl == bl) {
+ ff_vp9_decode_block(ctx, row, col, lflvl, yoff, uvoff, b->bl, b->bp);
+ if (b->bp == PARTITION_H && row + hbs < s->rows) {
+ yoff += hbs * 8 * y_stride;
+ uvoff += hbs * 8 * uv_stride >> s->ss_v;
+ ff_vp9_decode_block(ctx, row + hbs, col, lflvl, yoff, uvoff, b->bl, b->bp);
+ } else if (b->bp == PARTITION_V && col + hbs < s->cols) {
+ yoff += hbs * 8 * bytesperpixel;
+ uvoff += hbs * 8 * bytesperpixel >> s->ss_h;
+ ff_vp9_decode_block(ctx, row, col + hbs, lflvl, yoff, uvoff, b->bl, b->bp);
+ }
+ } else {
+ decode_sb_mem(ctx, row, col, lflvl, yoff, uvoff, bl + 1);
+ if (col + hbs < s->cols) { // FIXME why not <=?
+ if (row + hbs < s->rows) {
+ decode_sb_mem(ctx, row, col + hbs, lflvl, yoff + 8 * hbs * bytesperpixel,
+ uvoff + (8 * hbs * bytesperpixel >> s->ss_h), bl + 1);
+ yoff += hbs * 8 * y_stride;
+ uvoff += hbs * 8 * uv_stride >> s->ss_v;
+ decode_sb_mem(ctx, row + hbs, col, lflvl, yoff, uvoff, bl + 1);
+ decode_sb_mem(ctx, row + hbs, col + hbs, lflvl,
+ yoff + 8 * hbs * bytesperpixel,
+ uvoff + (8 * hbs * bytesperpixel >> s->ss_h), bl + 1);
+ } else {
+ yoff += hbs * 8 * bytesperpixel;
+ uvoff += hbs * 8 * bytesperpixel >> s->ss_h;
+ decode_sb_mem(ctx, row, col + hbs, lflvl, yoff, uvoff, bl + 1);
}
+ } else if (row + hbs < s->rows) {
+ yoff += hbs * 8 * y_stride;
+ uvoff += hbs * 8 * uv_stride >> s->ss_v;
+ decode_sb_mem(ctx, row + hbs, col, lflvl, yoff, uvoff, bl + 1);
}
}
+}
- // MV at this position in previous frame, using same reference frame
- if (s->s.h.use_last_frame_mvs) {
- struct VP9mvrefPair *mv = &s->s.frames[REF_FRAME_MVPAIR].mv[row * s->sb_cols * 8 + col];
+static av_always_inline void filter_plane_cols(VP9Context *s, int col, int ss_h, int ss_v,
+ uint8_t *lvl, uint8_t (*mask)[4],
+ uint8_t *dst, ptrdiff_t ls)
+{
+ int y, x, bytesperpixel = s->bytesperpixel;
- if (!s->s.frames[REF_FRAME_MVPAIR].uses_2pass)
- ff_thread_await_progress(&s->s.frames[REF_FRAME_MVPAIR].tf, row >> 3, 0);
- if (mv->ref[0] == ref) {
- RETURN_MV(mv->mv[0]);
- } else if (mv->ref[1] == ref) {
- RETURN_MV(mv->mv[1]);
- }
- }
+ // filter edges between columns (e.g. block1 | block2)
+ for (y = 0; y < 8; y += 2 << ss_v, dst += 16 * ls, lvl += 16 << ss_v) {
+ uint8_t *ptr = dst, *l = lvl, *hmask1 = mask[y], *hmask2 = mask[y + 1 + ss_v];
+ unsigned hm1 = hmask1[0] | hmask1[1] | hmask1[2], hm13 = hmask1[3];
+ unsigned hm2 = hmask2[1] | hmask2[2], hm23 = hmask2[3];
+ unsigned hm = hm1 | hm2 | hm13 | hm23;
-#define RETURN_SCALE_MV(mv, scale) \
- do { \
- if (scale) { \
- VP56mv mv_temp = { -mv.x, -mv.y }; \
- RETURN_MV(mv_temp); \
- } else { \
- RETURN_MV(mv); \
- } \
- } while (0)
-
- // previously coded MVs in this neighbourhood, using different reference frame
- for (i = 0; i < 8; i++) {
- int c = p[i][0] + col, r = p[i][1] + row;
+ for (x = 1; hm & ~(x - 1); x <<= 1, ptr += 8 * bytesperpixel >> ss_h) {
+ if (col || x > 1) {
+ if (hm1 & x) {
+ int L = *l, H = L >> 4;
+ int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
- if (c >= s->tile_col_start && c < s->cols && r >= 0 && r < s->rows) {
- struct VP9mvrefPair *mv = &s->s.frames[CUR_FRAME].mv[r * s->sb_cols * 8 + c];
+ if (hmask1[0] & x) {
+ if (hmask2[0] & x) {
+ av_assert2(l[8 << ss_v] == L);
+ s->dsp.loop_filter_16[0](ptr, ls, E, I, H);
+ } else {
+ s->dsp.loop_filter_8[2][0](ptr, ls, E, I, H);
+ }
+ } else if (hm2 & x) {
+ L = l[8 << ss_v];
+ H |= (L >> 4) << 8;
+ E |= s->filter_lut.mblim_lut[L] << 8;
+ I |= s->filter_lut.lim_lut[L] << 8;
+ s->dsp.loop_filter_mix2[!!(hmask1[1] & x)]
+ [!!(hmask2[1] & x)]
+ [0](ptr, ls, E, I, H);
+ } else {
+ s->dsp.loop_filter_8[!!(hmask1[1] & x)]
+ [0](ptr, ls, E, I, H);
+ }
+ } else if (hm2 & x) {
+ int L = l[8 << ss_v], H = L >> 4;
+ int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
- if (mv->ref[0] != ref && mv->ref[0] >= 0) {
- RETURN_SCALE_MV(mv->mv[0], s->s.h.signbias[mv->ref[0]] != s->s.h.signbias[ref]);
- }
- if (mv->ref[1] != ref && mv->ref[1] >= 0 &&
- // BUG - libvpx has this condition regardless of whether
- // we used the first ref MV and pre-scaling
- AV_RN32A(&mv->mv[0]) != AV_RN32A(&mv->mv[1])) {
- RETURN_SCALE_MV(mv->mv[1], s->s.h.signbias[mv->ref[1]] != s->s.h.signbias[ref]);
+ s->dsp.loop_filter_8[!!(hmask2[1] & x)]
+ [0](ptr + 8 * ls, ls, E, I, H);
+ }
}
- }
- }
+ if (ss_h) {
+ if (x & 0xAA)
+ l += 2;
+ } else {
+ if (hm13 & x) {
+ int L = *l, H = L >> 4;
+ int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
- // MV at this position in previous frame, using different reference frame
- if (s->s.h.use_last_frame_mvs) {
- struct VP9mvrefPair *mv = &s->s.frames[REF_FRAME_MVPAIR].mv[row * s->sb_cols * 8 + col];
+ if (hm23 & x) {
+ L = l[8 << ss_v];
+ H |= (L >> 4) << 8;
+ E |= s->filter_lut.mblim_lut[L] << 8;
+ I |= s->filter_lut.lim_lut[L] << 8;
+ s->dsp.loop_filter_mix2[0][0][0](ptr + 4 * bytesperpixel, ls, E, I, H);
+ } else {
+ s->dsp.loop_filter_8[0][0](ptr + 4 * bytesperpixel, ls, E, I, H);
+ }
+ } else if (hm23 & x) {
+ int L = l[8 << ss_v], H = L >> 4;
+ int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
- // no need to await_progress, because we already did that above
- if (mv->ref[0] != ref && mv->ref[0] >= 0) {
- RETURN_SCALE_MV(mv->mv[0], s->s.h.signbias[mv->ref[0]] != s->s.h.signbias[ref]);
- }
- if (mv->ref[1] != ref && mv->ref[1] >= 0 &&
- // BUG - libvpx has this condition regardless of whether
- // we used the first ref MV and pre-scaling
- AV_RN32A(&mv->mv[0]) != AV_RN32A(&mv->mv[1])) {
- RETURN_SCALE_MV(mv->mv[1], s->s.h.signbias[mv->ref[1]] != s->s.h.signbias[ref]);
+ s->dsp.loop_filter_8[0][0](ptr + 8 * ls + 4 * bytesperpixel, ls, E, I, H);
+ }
+ l++;
+ }
}
}
-
- AV_ZERO32(pmv);
- clamp_mv(pmv, pmv, s);
-#undef INVALID_MV
-#undef RETURN_MV
-#undef RETURN_SCALE_MV
}
-static av_always_inline int read_mv_component(VP9Context *s, int idx, int hp)
+static av_always_inline void filter_plane_rows(VP9Context *s, int row, int ss_h, int ss_v,
+ uint8_t *lvl, uint8_t (*mask)[4],
+ uint8_t *dst, ptrdiff_t ls)
{
- int bit, sign = vp56_rac_get_prob(&s->c, s->prob.p.mv_comp[idx].sign);
- int n, c = vp8_rac_get_tree(&s->c, vp9_mv_class_tree,
- s->prob.p.mv_comp[idx].classes);
-
- s->counts.mv_comp[idx].sign[sign]++;
- s->counts.mv_comp[idx].classes[c]++;
- if (c) {
- int m;
-
- for (n = 0, m = 0; m < c; m++) {
- bit = vp56_rac_get_prob(&s->c, s->prob.p.mv_comp[idx].bits[m]);
- n |= bit << m;
- s->counts.mv_comp[idx].bits[m][bit]++;
- }
- n <<= 3;
- bit = vp8_rac_get_tree(&s->c, vp9_mv_fp_tree, s->prob.p.mv_comp[idx].fp);
- n |= bit << 1;
- s->counts.mv_comp[idx].fp[bit]++;
- if (hp) {
- bit = vp56_rac_get_prob(&s->c, s->prob.p.mv_comp[idx].hp);
- s->counts.mv_comp[idx].hp[bit]++;
- n |= bit;
- } else {
- n |= 1;
- // bug in libvpx - we count for bw entropy purposes even if the
- // bit wasn't coded
- s->counts.mv_comp[idx].hp[1]++;
- }
- n += 8 << c;
- } else {
- n = vp56_rac_get_prob(&s->c, s->prob.p.mv_comp[idx].class0);
- s->counts.mv_comp[idx].class0[n]++;
- bit = vp8_rac_get_tree(&s->c, vp9_mv_fp_tree,
- s->prob.p.mv_comp[idx].class0_fp[n]);
- s->counts.mv_comp[idx].class0_fp[n][bit]++;
- n = (n << 3) | (bit << 1);
- if (hp) {
- bit = vp56_rac_get_prob(&s->c, s->prob.p.mv_comp[idx].class0_hp);
- s->counts.mv_comp[idx].class0_hp[bit]++;
- n |= bit;
- } else {
- n |= 1;
- // bug in libvpx - we count for bw entropy purposes even if the
- // bit wasn't coded
- s->counts.mv_comp[idx].class0_hp[1]++;
- }
- }
+ int y, x, bytesperpixel = s->bytesperpixel;
- return sign ? -(n + 1) : (n + 1);
-}
+ // block1
+ // filter edges between rows (e.g. ------)
+ // block2
+ for (y = 0; y < 8; y++, dst += 8 * ls >> ss_v) {
+ uint8_t *ptr = dst, *l = lvl, *vmask = mask[y];
+ unsigned vm = vmask[0] | vmask[1] | vmask[2], vm3 = vmask[3];
-static void fill_mv(VP9Context *s,
- VP56mv *mv, int mode, int sb)
-{
- VP9Block *b = s->b;
+ for (x = 1; vm & ~(x - 1); x <<= (2 << ss_h), ptr += 16 * bytesperpixel, l += 2 << ss_h) {
+ if (row || y) {
+ if (vm & x) {
+ int L = *l, H = L >> 4;
+ int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
- if (mode == ZEROMV) {
- AV_ZERO64(mv);
- } else {
- int hp;
-
- // FIXME cache this value and reuse for other subblocks
- find_ref_mvs(s, &mv[0], b->ref[0], 0, mode == NEARMV,
- mode == NEWMV ? -1 : sb);
- // FIXME maybe move this code into find_ref_mvs()
- if ((mode == NEWMV || sb == -1) &&
- !(hp = s->s.h.highprecisionmvs && abs(mv[0].x) < 64 && abs(mv[0].y) < 64)) {
- if (mv[0].y & 1) {
- if (mv[0].y < 0)
- mv[0].y++;
- else
- mv[0].y--;
- }
- if (mv[0].x & 1) {
- if (mv[0].x < 0)
- mv[0].x++;
- else
- mv[0].x--;
- }
- }
- if (mode == NEWMV) {
- enum MVJoint j = vp8_rac_get_tree(&s->c, vp9_mv_joint_tree,
- s->prob.p.mv_joint);
-
- s->counts.mv_joint[j]++;
- if (j >= MV_JOINT_V)
- mv[0].y += read_mv_component(s, 0, hp);
- if (j & 1)
- mv[0].x += read_mv_component(s, 1, hp);
- }
+ if (vmask[0] & x) {
+ if (vmask[0] & (x << (1 + ss_h))) {
+ av_assert2(l[1 + ss_h] == L);
+ s->dsp.loop_filter_16[1](ptr, ls, E, I, H);
+ } else {
+ s->dsp.loop_filter_8[2][1](ptr, ls, E, I, H);
+ }
+ } else if (vm & (x << (1 + ss_h))) {
+ L = l[1 + ss_h];
+ H |= (L >> 4) << 8;
+ E |= s->filter_lut.mblim_lut[L] << 8;
+ I |= s->filter_lut.lim_lut[L] << 8;
+ s->dsp.loop_filter_mix2[!!(vmask[1] & x)]
+ [!!(vmask[1] & (x << (1 + ss_h)))]
+ [1](ptr, ls, E, I, H);
+ } else {
+ s->dsp.loop_filter_8[!!(vmask[1] & x)]
+ [1](ptr, ls, E, I, H);
+ }
+ } else if (vm & (x << (1 + ss_h))) {
+ int L = l[1 + ss_h], H = L >> 4;
+ int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
- if (b->comp) {
- // FIXME cache this value and reuse for other subblocks
- find_ref_mvs(s, &mv[1], b->ref[1], 1, mode == NEARMV,
- mode == NEWMV ? -1 : sb);
- if ((mode == NEWMV || sb == -1) &&
- !(hp = s->s.h.highprecisionmvs && abs(mv[1].x) < 64 && abs(mv[1].y) < 64)) {
- if (mv[1].y & 1) {
- if (mv[1].y < 0)
- mv[1].y++;
- else
- mv[1].y--;
- }
- if (mv[1].x & 1) {
- if (mv[1].x < 0)
- mv[1].x++;
- else
- mv[1].x--;
+ s->dsp.loop_filter_8[!!(vmask[1] & (x << (1 + ss_h)))]
+ [1](ptr + 8 * bytesperpixel, ls, E, I, H);
}
}
- if (mode == NEWMV) {
- enum MVJoint j = vp8_rac_get_tree(&s->c, vp9_mv_joint_tree,
- s->prob.p.mv_joint);
-
- s->counts.mv_joint[j]++;
- if (j >= MV_JOINT_V)
- mv[1].y += read_mv_component(s, 0, hp);
- if (j & 1)
- mv[1].x += read_mv_component(s, 1, hp);
- }
- }
- }
-}
-
-static av_always_inline void setctx_2d(uint8_t *ptr, int w, int h,
- ptrdiff_t stride, int v)
-{
- switch (w) {
- case 1:
- do {
- *ptr = v;
- ptr += stride;
- } while (--h);
- break;
- case 2: {
- int v16 = v * 0x0101;
- do {
- AV_WN16A(ptr, v16);
- ptr += stride;
- } while (--h);
- break;
- }
- case 4: {
- uint32_t v32 = v * 0x01010101;
- do {
- AV_WN32A(ptr, v32);
- ptr += stride;
- } while (--h);
- break;
- }
- case 8: {
-#if HAVE_FAST_64BIT
- uint64_t v64 = v * 0x0101010101010101ULL;
- do {
- AV_WN64A(ptr, v64);
- ptr += stride;
- } while (--h);
-#else
- uint32_t v32 = v * 0x01010101;
- do {
- AV_WN32A(ptr, v32);
- AV_WN32A(ptr + 4, v32);
- ptr += stride;
- } while (--h);
-#endif
- break;
- }
- }
-}
-
-static void decode_mode(AVCodecContext *ctx)
-{
- static const uint8_t left_ctx[N_BS_SIZES] = {
- 0x0, 0x8, 0x0, 0x8, 0xc, 0x8, 0xc, 0xe, 0xc, 0xe, 0xf, 0xe, 0xf
- };
- static const uint8_t above_ctx[N_BS_SIZES] = {
- 0x0, 0x0, 0x8, 0x8, 0x8, 0xc, 0xc, 0xc, 0xe, 0xe, 0xe, 0xf, 0xf
- };
- static const uint8_t max_tx_for_bl_bp[N_BS_SIZES] = {
- TX_32X32, TX_32X32, TX_32X32, TX_32X32, TX_16X16, TX_16X16,
- TX_16X16, TX_8X8, TX_8X8, TX_8X8, TX_4X4, TX_4X4, TX_4X4
- };
- VP9Context *s = ctx->priv_data;
- VP9Block *b = s->b;
- int row = s->row, col = s->col, row7 = s->row7;
- enum TxfmMode max_tx = max_tx_for_bl_bp[b->bs];
- int bw4 = bwh_tab[1][b->bs][0], w4 = FFMIN(s->cols - col, bw4);
- int bh4 = bwh_tab[1][b->bs][1], h4 = FFMIN(s->rows - row, bh4), y;
- int have_a = row > 0, have_l = col > s->tile_col_start;
- int vref, filter_id;
-
- if (!s->s.h.segmentation.enabled) {
- b->seg_id = 0;
- } else if (s->s.h.keyframe || s->s.h.intraonly) {
- b->seg_id = !s->s.h.segmentation.update_map ? 0 :
- vp8_rac_get_tree(&s->c, vp9_segmentation_tree, s->s.h.segmentation.prob);
- } else if (!s->s.h.segmentation.update_map ||
- (s->s.h.segmentation.temporal &&
- vp56_rac_get_prob_branchy(&s->c,
- s->s.h.segmentation.pred_prob[s->above_segpred_ctx[col] +
- s->left_segpred_ctx[row7]]))) {
- if (!s->s.h.errorres && s->s.frames[REF_FRAME_SEGMAP].segmentation_map) {
- int pred = 8, x;
- uint8_t *refsegmap = s->s.frames[REF_FRAME_SEGMAP].segmentation_map;
-
- if (!s->s.frames[REF_FRAME_SEGMAP].uses_2pass)
- ff_thread_await_progress(&s->s.frames[REF_FRAME_SEGMAP].tf, row >> 3, 0);
- for (y = 0; y < h4; y++) {
- int idx_base = (y + row) * 8 * s->sb_cols + col;
- for (x = 0; x < w4; x++)
- pred = FFMIN(pred, refsegmap[idx_base + x]);
- }
- av_assert1(pred < 8);
- b->seg_id = pred;
- } else {
- b->seg_id = 0;
- }
-
- memset(&s->above_segpred_ctx[col], 1, w4);
- memset(&s->left_segpred_ctx[row7], 1, h4);
- } else {
- b->seg_id = vp8_rac_get_tree(&s->c, vp9_segmentation_tree,
- s->s.h.segmentation.prob);
-
- memset(&s->above_segpred_ctx[col], 0, w4);
- memset(&s->left_segpred_ctx[row7], 0, h4);
- }
- if (s->s.h.segmentation.enabled &&
- (s->s.h.segmentation.update_map || s->s.h.keyframe || s->s.h.intraonly)) {
- setctx_2d(&s->s.frames[CUR_FRAME].segmentation_map[row * 8 * s->sb_cols + col],
- bw4, bh4, 8 * s->sb_cols, b->seg_id);
- }
-
- b->skip = s->s.h.segmentation.enabled &&
- s->s.h.segmentation.feat[b->seg_id].skip_enabled;
- if (!b->skip) {
- int c = s->left_skip_ctx[row7] + s->above_skip_ctx[col];
- b->skip = vp56_rac_get_prob(&s->c, s->prob.p.skip[c]);
- s->counts.skip[c][b->skip]++;
- }
-
- if (s->s.h.keyframe || s->s.h.intraonly) {
- b->intra = 1;
- } else if (s->s.h.segmentation.enabled && s->s.h.segmentation.feat[b->seg_id].ref_enabled) {
- b->intra = !s->s.h.segmentation.feat[b->seg_id].ref_val;
- } else {
- int c, bit;
-
- if (have_a && have_l) {
- c = s->above_intra_ctx[col] + s->left_intra_ctx[row7];
- c += (c == 2);
- } else {
- c = have_a ? 2 * s->above_intra_ctx[col] :
- have_l ? 2 * s->left_intra_ctx[row7] : 0;
- }
- bit = vp56_rac_get_prob(&s->c, s->prob.p.intra[c]);
- s->counts.intra[c][bit]++;
- b->intra = !bit;
- }
-
- if ((b->intra || !b->skip) && s->s.h.txfmmode == TX_SWITCHABLE) {
- int c;
- if (have_a) {
- if (have_l) {
- c = (s->above_skip_ctx[col] ? max_tx :
- s->above_txfm_ctx[col]) +
- (s->left_skip_ctx[row7] ? max_tx :
- s->left_txfm_ctx[row7]) > max_tx;
- } else {
- c = s->above_skip_ctx[col] ? 1 :
- (s->above_txfm_ctx[col] * 2 > max_tx);
- }
- } else if (have_l) {
- c = s->left_skip_ctx[row7] ? 1 :
- (s->left_txfm_ctx[row7] * 2 > max_tx);
- } else {
- c = 1;
- }
- switch (max_tx) {
- case TX_32X32:
- b->tx = vp56_rac_get_prob(&s->c, s->prob.p.tx32p[c][0]);
- if (b->tx) {
- b->tx += vp56_rac_get_prob(&s->c, s->prob.p.tx32p[c][1]);
- if (b->tx == 2)
- b->tx += vp56_rac_get_prob(&s->c, s->prob.p.tx32p[c][2]);
- }
- s->counts.tx32p[c][b->tx]++;
- break;
- case TX_16X16:
- b->tx = vp56_rac_get_prob(&s->c, s->prob.p.tx16p[c][0]);
- if (b->tx)
- b->tx += vp56_rac_get_prob(&s->c, s->prob.p.tx16p[c][1]);
- s->counts.tx16p[c][b->tx]++;
- break;
- case TX_8X8:
- b->tx = vp56_rac_get_prob(&s->c, s->prob.p.tx8p[c]);
- s->counts.tx8p[c][b->tx]++;
- break;
- case TX_4X4:
- b->tx = TX_4X4;
- break;
- }
- } else {
- b->tx = FFMIN(max_tx, s->s.h.txfmmode);
- }
-
- if (s->s.h.keyframe || s->s.h.intraonly) {
- uint8_t *a = &s->above_mode_ctx[col * 2];
- uint8_t *l = &s->left_mode_ctx[(row7) << 1];
-
- b->comp = 0;
- if (b->bs > BS_8x8) {
- // FIXME the memory storage intermediates here aren't really
- // necessary, they're just there to make the code slightly
- // simpler for now
- b->mode[0] = a[0] = vp8_rac_get_tree(&s->c, vp9_intramode_tree,
- vp9_default_kf_ymode_probs[a[0]][l[0]]);
- if (b->bs != BS_8x4) {
- b->mode[1] = vp8_rac_get_tree(&s->c, vp9_intramode_tree,
- vp9_default_kf_ymode_probs[a[1]][b->mode[0]]);
- l[0] = a[1] = b->mode[1];
- } else {
- l[0] = a[1] = b->mode[1] = b->mode[0];
- }
- if (b->bs != BS_4x8) {
- b->mode[2] = a[0] = vp8_rac_get_tree(&s->c, vp9_intramode_tree,
- vp9_default_kf_ymode_probs[a[0]][l[1]]);
- if (b->bs != BS_8x4) {
- b->mode[3] = vp8_rac_get_tree(&s->c, vp9_intramode_tree,
- vp9_default_kf_ymode_probs[a[1]][b->mode[2]]);
- l[1] = a[1] = b->mode[3];
- } else {
- l[1] = a[1] = b->mode[3] = b->mode[2];
- }
- } else {
- b->mode[2] = b->mode[0];
- l[1] = a[1] = b->mode[3] = b->mode[1];
- }
- } else {
- b->mode[0] = vp8_rac_get_tree(&s->c, vp9_intramode_tree,
- vp9_default_kf_ymode_probs[*a][*l]);
- b->mode[3] = b->mode[2] = b->mode[1] = b->mode[0];
- // FIXME this can probably be optimized
- memset(a, b->mode[0], bwh_tab[0][b->bs][0]);
- memset(l, b->mode[0], bwh_tab[0][b->bs][1]);
- }
- b->uvmode = vp8_rac_get_tree(&s->c, vp9_intramode_tree,
- vp9_default_kf_uvmode_probs[b->mode[3]]);
- } else if (b->intra) {
- b->comp = 0;
- if (b->bs > BS_8x8) {
- b->mode[0] = vp8_rac_get_tree(&s->c, vp9_intramode_tree,
- s->prob.p.y_mode[0]);
- s->counts.y_mode[0][b->mode[0]]++;
- if (b->bs != BS_8x4) {
- b->mode[1] = vp8_rac_get_tree(&s->c, vp9_intramode_tree,
- s->prob.p.y_mode[0]);
- s->counts.y_mode[0][b->mode[1]]++;
- } else {
- b->mode[1] = b->mode[0];
- }
- if (b->bs != BS_4x8) {
- b->mode[2] = vp8_rac_get_tree(&s->c, vp9_intramode_tree,
- s->prob.p.y_mode[0]);
- s->counts.y_mode[0][b->mode[2]]++;
- if (b->bs != BS_8x4) {
- b->mode[3] = vp8_rac_get_tree(&s->c, vp9_intramode_tree,
- s->prob.p.y_mode[0]);
- s->counts.y_mode[0][b->mode[3]]++;
- } else {
- b->mode[3] = b->mode[2];
- }
- } else {
- b->mode[2] = b->mode[0];
- b->mode[3] = b->mode[1];
- }
- } else {
- static const uint8_t size_group[10] = {
- 3, 3, 3, 3, 2, 2, 2, 1, 1, 1
- };
- int sz = size_group[b->bs];
-
- b->mode[0] = vp8_rac_get_tree(&s->c, vp9_intramode_tree,
- s->prob.p.y_mode[sz]);
- b->mode[1] = b->mode[2] = b->mode[3] = b->mode[0];
- s->counts.y_mode[sz][b->mode[3]]++;
- }
- b->uvmode = vp8_rac_get_tree(&s->c, vp9_intramode_tree,
- s->prob.p.uv_mode[b->mode[3]]);
- s->counts.uv_mode[b->mode[3]][b->uvmode]++;
- } else {
- static const uint8_t inter_mode_ctx_lut[14][14] = {
- { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
- { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
- { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
- { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
- { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
- { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
- { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
- { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
- { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
- { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
- { 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 2, 2, 1, 3 },
- { 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 2, 2, 1, 3 },
- { 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 1, 1, 0, 3 },
- { 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 3, 3, 3, 4 },
- };
-
- if (s->s.h.segmentation.enabled && s->s.h.segmentation.feat[b->seg_id].ref_enabled) {
- av_assert2(s->s.h.segmentation.feat[b->seg_id].ref_val != 0);
- b->comp = 0;
- b->ref[0] = s->s.h.segmentation.feat[b->seg_id].ref_val - 1;
- } else {
- // read comp_pred flag
- if (s->s.h.comppredmode != PRED_SWITCHABLE) {
- b->comp = s->s.h.comppredmode == PRED_COMPREF;
- } else {
- int c;
-
- // FIXME add intra as ref=0xff (or -1) to make these easier?
- if (have_a) {
- if (have_l) {
- if (s->above_comp_ctx[col] && s->left_comp_ctx[row7]) {
- c = 4;
- } else if (s->above_comp_ctx[col]) {
- c = 2 + (s->left_intra_ctx[row7] ||
- s->left_ref_ctx[row7] == s->s.h.fixcompref);
- } else if (s->left_comp_ctx[row7]) {
- c = 2 + (s->above_intra_ctx[col] ||
- s->above_ref_ctx[col] == s->s.h.fixcompref);
- } else {
- c = (!s->above_intra_ctx[col] &&
- s->above_ref_ctx[col] == s->s.h.fixcompref) ^
- (!s->left_intra_ctx[row7] &&
- s->left_ref_ctx[row & 7] == s->s.h.fixcompref);
- }
- } else {
- c = s->above_comp_ctx[col] ? 3 :
- (!s->above_intra_ctx[col] && s->above_ref_ctx[col] == s->s.h.fixcompref);
- }
- } else if (have_l) {
- c = s->left_comp_ctx[row7] ? 3 :
- (!s->left_intra_ctx[row7] && s->left_ref_ctx[row7] == s->s.h.fixcompref);
- } else {
- c = 1;
- }
- b->comp = vp56_rac_get_prob(&s->c, s->prob.p.comp[c]);
- s->counts.comp[c][b->comp]++;
- }
-
- // read actual references
- // FIXME probably cache a few variables here to prevent repetitive
- // memory accesses below
- if (b->comp) /* two references */ {
- int fix_idx = s->s.h.signbias[s->s.h.fixcompref], var_idx = !fix_idx, c, bit;
-
- b->ref[fix_idx] = s->s.h.fixcompref;
- // FIXME can this codeblob be replaced by some sort of LUT?
- if (have_a) {
- if (have_l) {
- if (s->above_intra_ctx[col]) {
- if (s->left_intra_ctx[row7]) {
- c = 2;
- } else {
- c = 1 + 2 * (s->left_ref_ctx[row7] != s->s.h.varcompref[1]);
- }
- } else if (s->left_intra_ctx[row7]) {
- c = 1 + 2 * (s->above_ref_ctx[col] != s->s.h.varcompref[1]);
- } else {
- int refl = s->left_ref_ctx[row7], refa = s->above_ref_ctx[col];
-
- if (refl == refa && refa == s->s.h.varcompref[1]) {
- c = 0;
- } else if (!s->left_comp_ctx[row7] && !s->above_comp_ctx[col]) {
- if ((refa == s->s.h.fixcompref && refl == s->s.h.varcompref[0]) ||
- (refl == s->s.h.fixcompref && refa == s->s.h.varcompref[0])) {
- c = 4;
- } else {
- c = (refa == refl) ? 3 : 1;
- }
- } else if (!s->left_comp_ctx[row7]) {
- if (refa == s->s.h.varcompref[1] && refl != s->s.h.varcompref[1]) {
- c = 1;
- } else {
- c = (refl == s->s.h.varcompref[1] &&
- refa != s->s.h.varcompref[1]) ? 2 : 4;
- }
- } else if (!s->above_comp_ctx[col]) {
- if (refl == s->s.h.varcompref[1] && refa != s->s.h.varcompref[1]) {
- c = 1;
- } else {
- c = (refa == s->s.h.varcompref[1] &&
- refl != s->s.h.varcompref[1]) ? 2 : 4;
- }
- } else {
- c = (refl == refa) ? 4 : 2;
- }
- }
- } else {
- if (s->above_intra_ctx[col]) {
- c = 2;
- } else if (s->above_comp_ctx[col]) {
- c = 4 * (s->above_ref_ctx[col] != s->s.h.varcompref[1]);
- } else {
- c = 3 * (s->above_ref_ctx[col] != s->s.h.varcompref[1]);
- }
- }
- } else if (have_l) {
- if (s->left_intra_ctx[row7]) {
- c = 2;
- } else if (s->left_comp_ctx[row7]) {
- c = 4 * (s->left_ref_ctx[row7] != s->s.h.varcompref[1]);
- } else {
- c = 3 * (s->left_ref_ctx[row7] != s->s.h.varcompref[1]);
- }
- } else {
- c = 2;
- }
- bit = vp56_rac_get_prob(&s->c, s->prob.p.comp_ref[c]);
- b->ref[var_idx] = s->s.h.varcompref[bit];
- s->counts.comp_ref[c][bit]++;
- } else /* single reference */ {
- int bit, c;
-
- if (have_a && !s->above_intra_ctx[col]) {
- if (have_l && !s->left_intra_ctx[row7]) {
- if (s->left_comp_ctx[row7]) {
- if (s->above_comp_ctx[col]) {
- c = 1 + (!s->s.h.fixcompref || !s->left_ref_ctx[row7] ||
- !s->above_ref_ctx[col]);
- } else {
- c = (3 * !s->above_ref_ctx[col]) +
- (!s->s.h.fixcompref || !s->left_ref_ctx[row7]);
- }
- } else if (s->above_comp_ctx[col]) {
- c = (3 * !s->left_ref_ctx[row7]) +
- (!s->s.h.fixcompref || !s->above_ref_ctx[col]);
- } else {
- c = 2 * !s->left_ref_ctx[row7] + 2 * !s->above_ref_ctx[col];
- }
- } else if (s->above_intra_ctx[col]) {
- c = 2;
- } else if (s->above_comp_ctx[col]) {
- c = 1 + (!s->s.h.fixcompref || !s->above_ref_ctx[col]);
- } else {
- c = 4 * (!s->above_ref_ctx[col]);
- }
- } else if (have_l && !s->left_intra_ctx[row7]) {
- if (s->left_intra_ctx[row7]) {
- c = 2;
- } else if (s->left_comp_ctx[row7]) {
- c = 1 + (!s->s.h.fixcompref || !s->left_ref_ctx[row7]);
- } else {
- c = 4 * (!s->left_ref_ctx[row7]);
- }
- } else {
- c = 2;
- }
- bit = vp56_rac_get_prob(&s->c, s->prob.p.single_ref[c][0]);
- s->counts.single_ref[c][0][bit]++;
- if (!bit) {
- b->ref[0] = 0;
- } else {
- // FIXME can this codeblob be replaced by some sort of LUT?
- if (have_a) {
- if (have_l) {
- if (s->left_intra_ctx[row7]) {
- if (s->above_intra_ctx[col]) {
- c = 2;
- } else if (s->above_comp_ctx[col]) {
- c = 1 + 2 * (s->s.h.fixcompref == 1 ||
- s->above_ref_ctx[col] == 1);
- } else if (!s->above_ref_ctx[col]) {
- c = 3;
- } else {
- c = 4 * (s->above_ref_ctx[col] == 1);
- }
- } else if (s->above_intra_ctx[col]) {
- if (s->left_intra_ctx[row7]) {
- c = 2;
- } else if (s->left_comp_ctx[row7]) {
- c = 1 + 2 * (s->s.h.fixcompref == 1 ||
- s->left_ref_ctx[row7] == 1);
- } else if (!s->left_ref_ctx[row7]) {
- c = 3;
- } else {
- c = 4 * (s->left_ref_ctx[row7] == 1);
- }
- } else if (s->above_comp_ctx[col]) {
- if (s->left_comp_ctx[row7]) {
- if (s->left_ref_ctx[row7] == s->above_ref_ctx[col]) {
- c = 3 * (s->s.h.fixcompref == 1 ||
- s->left_ref_ctx[row7] == 1);
- } else {
- c = 2;
- }
- } else if (!s->left_ref_ctx[row7]) {
- c = 1 + 2 * (s->s.h.fixcompref == 1 ||
- s->above_ref_ctx[col] == 1);
- } else {
- c = 3 * (s->left_ref_ctx[row7] == 1) +
- (s->s.h.fixcompref == 1 || s->above_ref_ctx[col] == 1);
- }
- } else if (s->left_comp_ctx[row7]) {
- if (!s->above_ref_ctx[col]) {
- c = 1 + 2 * (s->s.h.fixcompref == 1 ||
- s->left_ref_ctx[row7] == 1);
- } else {
- c = 3 * (s->above_ref_ctx[col] == 1) +
- (s->s.h.fixcompref == 1 || s->left_ref_ctx[row7] == 1);
- }
- } else if (!s->above_ref_ctx[col]) {
- if (!s->left_ref_ctx[row7]) {
- c = 3;
- } else {
- c = 4 * (s->left_ref_ctx[row7] == 1);
- }
- } else if (!s->left_ref_ctx[row7]) {
- c = 4 * (s->above_ref_ctx[col] == 1);
- } else {
- c = 2 * (s->left_ref_ctx[row7] == 1) +
- 2 * (s->above_ref_ctx[col] == 1);
- }
- } else {
- if (s->above_intra_ctx[col] ||
- (!s->above_comp_ctx[col] && !s->above_ref_ctx[col])) {
- c = 2;
- } else if (s->above_comp_ctx[col]) {
- c = 3 * (s->s.h.fixcompref == 1 || s->above_ref_ctx[col] == 1);
- } else {
- c = 4 * (s->above_ref_ctx[col] == 1);
- }
- }
- } else if (have_l) {
- if (s->left_intra_ctx[row7] ||
- (!s->left_comp_ctx[row7] && !s->left_ref_ctx[row7])) {
- c = 2;
- } else if (s->left_comp_ctx[row7]) {
- c = 3 * (s->s.h.fixcompref == 1 || s->left_ref_ctx[row7] == 1);
- } else {
- c = 4 * (s->left_ref_ctx[row7] == 1);
- }
- } else {
- c = 2;
- }
- bit = vp56_rac_get_prob(&s->c, s->prob.p.single_ref[c][1]);
- s->counts.single_ref[c][1][bit]++;
- b->ref[0] = 1 + bit;
- }
- }
- }
-
- if (b->bs <= BS_8x8) {
- if (s->s.h.segmentation.enabled && s->s.h.segmentation.feat[b->seg_id].skip_enabled) {
- b->mode[0] = b->mode[1] = b->mode[2] = b->mode[3] = ZEROMV;
- } else {
- static const uint8_t off[10] = {
- 3, 0, 0, 1, 0, 0, 0, 0, 0, 0
- };
-
- // FIXME this needs to use the LUT tables from find_ref_mvs
- // because not all are -1,0/0,-1
- int c = inter_mode_ctx_lut[s->above_mode_ctx[col + off[b->bs]]]
- [s->left_mode_ctx[row7 + off[b->bs]]];
-
- b->mode[0] = vp8_rac_get_tree(&s->c, vp9_inter_mode_tree,
- s->prob.p.mv_mode[c]);
- b->mode[1] = b->mode[2] = b->mode[3] = b->mode[0];
- s->counts.mv_mode[c][b->mode[0] - 10]++;
- }
- }
-
- if (s->s.h.filtermode == FILTER_SWITCHABLE) {
- int c;
-
- if (have_a && s->above_mode_ctx[col] >= NEARESTMV) {
- if (have_l && s->left_mode_ctx[row7] >= NEARESTMV) {
- c = s->above_filter_ctx[col] == s->left_filter_ctx[row7] ?
- s->left_filter_ctx[row7] : 3;
- } else {
- c = s->above_filter_ctx[col];
- }
- } else if (have_l && s->left_mode_ctx[row7] >= NEARESTMV) {
- c = s->left_filter_ctx[row7];
- } else {
- c = 3;
- }
-
- filter_id = vp8_rac_get_tree(&s->c, vp9_filter_tree,
- s->prob.p.filter[c]);
- s->counts.filter[c][filter_id]++;
- b->filter = vp9_filter_lut[filter_id];
- } else {
- b->filter = s->s.h.filtermode;
- }
-
- if (b->bs > BS_8x8) {
- int c = inter_mode_ctx_lut[s->above_mode_ctx[col]][s->left_mode_ctx[row7]];
-
- b->mode[0] = vp8_rac_get_tree(&s->c, vp9_inter_mode_tree,
- s->prob.p.mv_mode[c]);
- s->counts.mv_mode[c][b->mode[0] - 10]++;
- fill_mv(s, b->mv[0], b->mode[0], 0);
-
- if (b->bs != BS_8x4) {
- b->mode[1] = vp8_rac_get_tree(&s->c, vp9_inter_mode_tree,
- s->prob.p.mv_mode[c]);
- s->counts.mv_mode[c][b->mode[1] - 10]++;
- fill_mv(s, b->mv[1], b->mode[1], 1);
- } else {
- b->mode[1] = b->mode[0];
- AV_COPY32(&b->mv[1][0], &b->mv[0][0]);
- AV_COPY32(&b->mv[1][1], &b->mv[0][1]);
- }
-
- if (b->bs != BS_4x8) {
- b->mode[2] = vp8_rac_get_tree(&s->c, vp9_inter_mode_tree,
- s->prob.p.mv_mode[c]);
- s->counts.mv_mode[c][b->mode[2] - 10]++;
- fill_mv(s, b->mv[2], b->mode[2], 2);
-
- if (b->bs != BS_8x4) {
- b->mode[3] = vp8_rac_get_tree(&s->c, vp9_inter_mode_tree,
- s->prob.p.mv_mode[c]);
- s->counts.mv_mode[c][b->mode[3] - 10]++;
- fill_mv(s, b->mv[3], b->mode[3], 3);
- } else {
- b->mode[3] = b->mode[2];
- AV_COPY32(&b->mv[3][0], &b->mv[2][0]);
- AV_COPY32(&b->mv[3][1], &b->mv[2][1]);
- }
- } else {
- b->mode[2] = b->mode[0];
- AV_COPY32(&b->mv[2][0], &b->mv[0][0]);
- AV_COPY32(&b->mv[2][1], &b->mv[0][1]);
- b->mode[3] = b->mode[1];
- AV_COPY32(&b->mv[3][0], &b->mv[1][0]);
- AV_COPY32(&b->mv[3][1], &b->mv[1][1]);
- }
- } else {
- fill_mv(s, b->mv[0], b->mode[0], -1);
- AV_COPY32(&b->mv[1][0], &b->mv[0][0]);
- AV_COPY32(&b->mv[2][0], &b->mv[0][0]);
- AV_COPY32(&b->mv[3][0], &b->mv[0][0]);
- AV_COPY32(&b->mv[1][1], &b->mv[0][1]);
- AV_COPY32(&b->mv[2][1], &b->mv[0][1]);
- AV_COPY32(&b->mv[3][1], &b->mv[0][1]);
- }
-
- vref = b->ref[b->comp ? s->s.h.signbias[s->s.h.varcompref[0]] : 0];
- }
-
-#if HAVE_FAST_64BIT
-#define SPLAT_CTX(var, val, n) \
- switch (n) { \
- case 1: var = val; break; \
- case 2: AV_WN16A(&var, val * 0x0101); break; \
- case 4: AV_WN32A(&var, val * 0x01010101); break; \
- case 8: AV_WN64A(&var, val * 0x0101010101010101ULL); break; \
- case 16: { \
- uint64_t v64 = val * 0x0101010101010101ULL; \
- AV_WN64A( &var, v64); \
- AV_WN64A(&((uint8_t *) &var)[8], v64); \
- break; \
- } \
- }
-#else
-#define SPLAT_CTX(var, val, n) \
- switch (n) { \
- case 1: var = val; break; \
- case 2: AV_WN16A(&var, val * 0x0101); break; \
- case 4: AV_WN32A(&var, val * 0x01010101); break; \
- case 8: { \
- uint32_t v32 = val * 0x01010101; \
- AV_WN32A( &var, v32); \
- AV_WN32A(&((uint8_t *) &var)[4], v32); \
- break; \
- } \
- case 16: { \
- uint32_t v32 = val * 0x01010101; \
- AV_WN32A( &var, v32); \
- AV_WN32A(&((uint8_t *) &var)[4], v32); \
- AV_WN32A(&((uint8_t *) &var)[8], v32); \
- AV_WN32A(&((uint8_t *) &var)[12], v32); \
- break; \
- } \
- }
-#endif
-
- switch (bwh_tab[1][b->bs][0]) {
-#define SET_CTXS(dir, off, n) \
- do { \
- SPLAT_CTX(s->dir##_skip_ctx[off], b->skip, n); \
- SPLAT_CTX(s->dir##_txfm_ctx[off], b->tx, n); \
- SPLAT_CTX(s->dir##_partition_ctx[off], dir##_ctx[b->bs], n); \
- if (!s->s.h.keyframe && !s->s.h.intraonly) { \
- SPLAT_CTX(s->dir##_intra_ctx[off], b->intra, n); \
- SPLAT_CTX(s->dir##_comp_ctx[off], b->comp, n); \
- SPLAT_CTX(s->dir##_mode_ctx[off], b->mode[3], n); \
- if (!b->intra) { \
- SPLAT_CTX(s->dir##_ref_ctx[off], vref, n); \
- if (s->s.h.filtermode == FILTER_SWITCHABLE) { \
- SPLAT_CTX(s->dir##_filter_ctx[off], filter_id, n); \
- } \
- } \
- } \
- } while (0)
- case 1: SET_CTXS(above, col, 1); break;
- case 2: SET_CTXS(above, col, 2); break;
- case 4: SET_CTXS(above, col, 4); break;
- case 8: SET_CTXS(above, col, 8); break;
- }
- switch (bwh_tab[1][b->bs][1]) {
- case 1: SET_CTXS(left, row7, 1); break;
- case 2: SET_CTXS(left, row7, 2); break;
- case 4: SET_CTXS(left, row7, 4); break;
- case 8: SET_CTXS(left, row7, 8); break;
- }
-#undef SPLAT_CTX
-#undef SET_CTXS
-
- if (!s->s.h.keyframe && !s->s.h.intraonly) {
- if (b->bs > BS_8x8) {
- int mv0 = AV_RN32A(&b->mv[3][0]), mv1 = AV_RN32A(&b->mv[3][1]);
-
- AV_COPY32(&s->left_mv_ctx[row7 * 2 + 0][0], &b->mv[1][0]);
- AV_COPY32(&s->left_mv_ctx[row7 * 2 + 0][1], &b->mv[1][1]);
- AV_WN32A(&s->left_mv_ctx[row7 * 2 + 1][0], mv0);
- AV_WN32A(&s->left_mv_ctx[row7 * 2 + 1][1], mv1);
- AV_COPY32(&s->above_mv_ctx[col * 2 + 0][0], &b->mv[2][0]);
- AV_COPY32(&s->above_mv_ctx[col * 2 + 0][1], &b->mv[2][1]);
- AV_WN32A(&s->above_mv_ctx[col * 2 + 1][0], mv0);
- AV_WN32A(&s->above_mv_ctx[col * 2 + 1][1], mv1);
- } else {
- int n, mv0 = AV_RN32A(&b->mv[3][0]), mv1 = AV_RN32A(&b->mv[3][1]);
-
- for (n = 0; n < w4 * 2; n++) {
- AV_WN32A(&s->above_mv_ctx[col * 2 + n][0], mv0);
- AV_WN32A(&s->above_mv_ctx[col * 2 + n][1], mv1);
- }
- for (n = 0; n < h4 * 2; n++) {
- AV_WN32A(&s->left_mv_ctx[row7 * 2 + n][0], mv0);
- AV_WN32A(&s->left_mv_ctx[row7 * 2 + n][1], mv1);
- }
- }
- }
-
- // FIXME kinda ugly
- for (y = 0; y < h4; y++) {
- int x, o = (row + y) * s->sb_cols * 8 + col;
- struct VP9mvrefPair *mv = &s->s.frames[CUR_FRAME].mv[o];
-
- if (b->intra) {
- for (x = 0; x < w4; x++) {
- mv[x].ref[0] =
- mv[x].ref[1] = -1;
- }
- } else if (b->comp) {
- for (x = 0; x < w4; x++) {
- mv[x].ref[0] = b->ref[0];
- mv[x].ref[1] = b->ref[1];
- AV_COPY32(&mv[x].mv[0], &b->mv[3][0]);
- AV_COPY32(&mv[x].mv[1], &b->mv[3][1]);
- }
- } else {
- for (x = 0; x < w4; x++) {
- mv[x].ref[0] = b->ref[0];
- mv[x].ref[1] = -1;
- AV_COPY32(&mv[x].mv[0], &b->mv[3][0]);
- }
- }
- }
-}
-
-// FIXME merge cnt/eob arguments?
-static av_always_inline int
-decode_coeffs_b_generic(VP56RangeCoder *c, int16_t *coef, int n_coeffs,
- int is_tx32x32, int is8bitsperpixel, int bpp, unsigned (*cnt)[6][3],
- unsigned (*eob)[6][2], uint8_t (*p)[6][11],
- int nnz, const int16_t *scan, const int16_t (*nb)[2],
- const int16_t *band_counts, const int16_t *qmul)
-{
- int i = 0, band = 0, band_left = band_counts[band];
- uint8_t *tp = p[0][nnz];
- uint8_t cache[1024];
-
- do {
- int val, rc;
-
- val = vp56_rac_get_prob_branchy(c, tp[0]); // eob
- eob[band][nnz][val]++;
- if (!val)
- break;
-
- skip_eob:
- if (!vp56_rac_get_prob_branchy(c, tp[1])) { // zero
- cnt[band][nnz][0]++;
- if (!--band_left)
- band_left = band_counts[++band];
- cache[scan[i]] = 0;
- nnz = (1 + cache[nb[i][0]] + cache[nb[i][1]]) >> 1;
- tp = p[band][nnz];
- if (++i == n_coeffs)
- break; //invalid input; blocks should end with EOB
- goto skip_eob;
- }
-
- rc = scan[i];
- if (!vp56_rac_get_prob_branchy(c, tp[2])) { // one
- cnt[band][nnz][1]++;
- val = 1;
- cache[rc] = 1;
- } else {
- // fill in p[3-10] (model fill) - only once per frame for each pos
- if (!tp[3])
- memcpy(&tp[3], vp9_model_pareto8[tp[2]], 8);
-
- cnt[band][nnz][2]++;
- if (!vp56_rac_get_prob_branchy(c, tp[3])) { // 2, 3, 4
- if (!vp56_rac_get_prob_branchy(c, tp[4])) {
- cache[rc] = val = 2;
- } else {
- val = 3 + vp56_rac_get_prob(c, tp[5]);
- cache[rc] = 3;
- }
- } else if (!vp56_rac_get_prob_branchy(c, tp[6])) { // cat1/2
- cache[rc] = 4;
- if (!vp56_rac_get_prob_branchy(c, tp[7])) {
- val = 5 + vp56_rac_get_prob(c, 159);
- } else {
- val = 7 + (vp56_rac_get_prob(c, 165) << 1);
- val += vp56_rac_get_prob(c, 145);
- }
- } else { // cat 3-6
- cache[rc] = 5;
- if (!vp56_rac_get_prob_branchy(c, tp[8])) {
- if (!vp56_rac_get_prob_branchy(c, tp[9])) {
- val = 11 + (vp56_rac_get_prob(c, 173) << 2);
- val += (vp56_rac_get_prob(c, 148) << 1);
- val += vp56_rac_get_prob(c, 140);
- } else {
- val = 19 + (vp56_rac_get_prob(c, 176) << 3);
- val += (vp56_rac_get_prob(c, 155) << 2);
- val += (vp56_rac_get_prob(c, 140) << 1);
- val += vp56_rac_get_prob(c, 135);
- }
- } else if (!vp56_rac_get_prob_branchy(c, tp[10])) {
- val = 35 + (vp56_rac_get_prob(c, 180) << 4);
- val += (vp56_rac_get_prob(c, 157) << 3);
- val += (vp56_rac_get_prob(c, 141) << 2);
- val += (vp56_rac_get_prob(c, 134) << 1);
- val += vp56_rac_get_prob(c, 130);
- } else {
- val = 67;
- if (!is8bitsperpixel) {
- if (bpp == 12) {
- val += vp56_rac_get_prob(c, 255) << 17;
- val += vp56_rac_get_prob(c, 255) << 16;
- }
- val += (vp56_rac_get_prob(c, 255) << 15);
- val += (vp56_rac_get_prob(c, 255) << 14);
- }
- val += (vp56_rac_get_prob(c, 254) << 13);
- val += (vp56_rac_get_prob(c, 254) << 12);
- val += (vp56_rac_get_prob(c, 254) << 11);
- val += (vp56_rac_get_prob(c, 252) << 10);
- val += (vp56_rac_get_prob(c, 249) << 9);
- val += (vp56_rac_get_prob(c, 243) << 8);
- val += (vp56_rac_get_prob(c, 230) << 7);
- val += (vp56_rac_get_prob(c, 196) << 6);
- val += (vp56_rac_get_prob(c, 177) << 5);
- val += (vp56_rac_get_prob(c, 153) << 4);
- val += (vp56_rac_get_prob(c, 140) << 3);
- val += (vp56_rac_get_prob(c, 133) << 2);
- val += (vp56_rac_get_prob(c, 130) << 1);
- val += vp56_rac_get_prob(c, 129);
- }
- }
- }
-#define STORE_COEF(c, i, v) do { \
- if (is8bitsperpixel) { \
- c[i] = v; \
- } else { \
- AV_WN32A(&c[i * 2], v); \
- } \
-} while (0)
- if (!--band_left)
- band_left = band_counts[++band];
- if (is_tx32x32)
- STORE_COEF(coef, rc, ((vp8_rac_get(c) ? -val : val) * qmul[!!i]) / 2);
- else
- STORE_COEF(coef, rc, (vp8_rac_get(c) ? -val : val) * qmul[!!i]);
- nnz = (1 + cache[nb[i][0]] + cache[nb[i][1]]) >> 1;
- tp = p[band][nnz];
- } while (++i < n_coeffs);
-
- return i;
-}
-
-static int decode_coeffs_b_8bpp(VP9Context *s, int16_t *coef, int n_coeffs,
- unsigned (*cnt)[6][3], unsigned (*eob)[6][2],
- uint8_t (*p)[6][11], int nnz, const int16_t *scan,
- const int16_t (*nb)[2], const int16_t *band_counts,
- const int16_t *qmul)
-{
- return decode_coeffs_b_generic(&s->c, coef, n_coeffs, 0, 1, 8, cnt, eob, p,
- nnz, scan, nb, band_counts, qmul);
-}
-
-static int decode_coeffs_b32_8bpp(VP9Context *s, int16_t *coef, int n_coeffs,
- unsigned (*cnt)[6][3], unsigned (*eob)[6][2],
- uint8_t (*p)[6][11], int nnz, const int16_t *scan,
- const int16_t (*nb)[2], const int16_t *band_counts,
- const int16_t *qmul)
-{
- return decode_coeffs_b_generic(&s->c, coef, n_coeffs, 1, 1, 8, cnt, eob, p,
- nnz, scan, nb, band_counts, qmul);
-}
-
-static int decode_coeffs_b_16bpp(VP9Context *s, int16_t *coef, int n_coeffs,
- unsigned (*cnt)[6][3], unsigned (*eob)[6][2],
- uint8_t (*p)[6][11], int nnz, const int16_t *scan,
- const int16_t (*nb)[2], const int16_t *band_counts,
- const int16_t *qmul)
-{
- return decode_coeffs_b_generic(&s->c, coef, n_coeffs, 0, 0, s->s.h.bpp, cnt, eob, p,
- nnz, scan, nb, band_counts, qmul);
-}
-
-static int decode_coeffs_b32_16bpp(VP9Context *s, int16_t *coef, int n_coeffs,
- unsigned (*cnt)[6][3], unsigned (*eob)[6][2],
- uint8_t (*p)[6][11], int nnz, const int16_t *scan,
- const int16_t (*nb)[2], const int16_t *band_counts,
- const int16_t *qmul)
-{
- return decode_coeffs_b_generic(&s->c, coef, n_coeffs, 1, 0, s->s.h.bpp, cnt, eob, p,
- nnz, scan, nb, band_counts, qmul);
-}
-
-static av_always_inline int decode_coeffs(AVCodecContext *ctx, int is8bitsperpixel)
-{
- VP9Context *s = ctx->priv_data;
- VP9Block *b = s->b;
- int row = s->row, col = s->col;
- uint8_t (*p)[6][11] = s->prob.coef[b->tx][0 /* y */][!b->intra];
- unsigned (*c)[6][3] = s->counts.coef[b->tx][0 /* y */][!b->intra];
- unsigned (*e)[6][2] = s->counts.eob[b->tx][0 /* y */][!b->intra];
- int w4 = bwh_tab[1][b->bs][0] << 1, h4 = bwh_tab[1][b->bs][1] << 1;
- int end_x = FFMIN(2 * (s->cols - col), w4);
- int end_y = FFMIN(2 * (s->rows - row), h4);
- int n, pl, x, y, res;
- int16_t (*qmul)[2] = s->s.h.segmentation.feat[b->seg_id].qmul;
- int tx = 4 * s->s.h.lossless + b->tx;
- const int16_t * const *yscans = vp9_scans[tx];
- const int16_t (* const *ynbs)[2] = vp9_scans_nb[tx];
- const int16_t *uvscan = vp9_scans[b->uvtx][DCT_DCT];
- const int16_t (*uvnb)[2] = vp9_scans_nb[b->uvtx][DCT_DCT];
- uint8_t *a = &s->above_y_nnz_ctx[col * 2];
- uint8_t *l = &s->left_y_nnz_ctx[(row & 7) << 1];
- static const int16_t band_counts[4][8] = {
- { 1, 2, 3, 4, 3, 16 - 13 },
- { 1, 2, 3, 4, 11, 64 - 21 },
- { 1, 2, 3, 4, 11, 256 - 21 },
- { 1, 2, 3, 4, 11, 1024 - 21 },
- };
- const int16_t *y_band_counts = band_counts[b->tx];
- const int16_t *uv_band_counts = band_counts[b->uvtx];
- int bytesperpixel = is8bitsperpixel ? 1 : 2;
- int total_coeff = 0;
-
-#define MERGE(la, end, step, rd) \
- for (n = 0; n < end; n += step) \
- la[n] = !!rd(&la[n])
-#define MERGE_CTX(step, rd) \
- do { \
- MERGE(l, end_y, step, rd); \
- MERGE(a, end_x, step, rd); \
- } while (0)
-
-#define DECODE_Y_COEF_LOOP(step, mode_index, v) \
- for (n = 0, y = 0; y < end_y; y += step) { \
- for (x = 0; x < end_x; x += step, n += step * step) { \
- enum TxfmType txtp = vp9_intra_txfm_type[b->mode[mode_index]]; \
- res = (is8bitsperpixel ? decode_coeffs_b##v##_8bpp : decode_coeffs_b##v##_16bpp) \
- (s, s->block + 16 * n * bytesperpixel, 16 * step * step, \
- c, e, p, a[x] + l[y], yscans[txtp], \
- ynbs[txtp], y_band_counts, qmul[0]); \
- a[x] = l[y] = !!res; \
- total_coeff |= !!res; \
- if (step >= 4) { \
- AV_WN16A(&s->eob[n], res); \
- } else { \
- s->eob[n] = res; \
- } \
- } \
- }
-
-#define SPLAT(la, end, step, cond) \
- if (step == 2) { \
- for (n = 1; n < end; n += step) \
- la[n] = la[n - 1]; \
- } else if (step == 4) { \
- if (cond) { \
- for (n = 0; n < end; n += step) \
- AV_WN32A(&la[n], la[n] * 0x01010101); \
- } else { \
- for (n = 0; n < end; n += step) \
- memset(&la[n + 1], la[n], FFMIN(end - n - 1, 3)); \
- } \
- } else /* step == 8 */ { \
- if (cond) { \
- if (HAVE_FAST_64BIT) { \
- for (n = 0; n < end; n += step) \
- AV_WN64A(&la[n], la[n] * 0x0101010101010101ULL); \
- } else { \
- for (n = 0; n < end; n += step) { \
- uint32_t v32 = la[n] * 0x01010101; \
- AV_WN32A(&la[n], v32); \
- AV_WN32A(&la[n + 4], v32); \
- } \
- } \
- } else { \
- for (n = 0; n < end; n += step) \
- memset(&la[n + 1], la[n], FFMIN(end - n - 1, 7)); \
- } \
- }
-#define SPLAT_CTX(step) \
- do { \
- SPLAT(a, end_x, step, end_x == w4); \
- SPLAT(l, end_y, step, end_y == h4); \
- } while (0)
-
- /* y tokens */
- switch (b->tx) {
- case TX_4X4:
- DECODE_Y_COEF_LOOP(1, b->bs > BS_8x8 ? n : 0,);
- break;
- case TX_8X8:
- MERGE_CTX(2, AV_RN16A);
- DECODE_Y_COEF_LOOP(2, 0,);
- SPLAT_CTX(2);
- break;
- case TX_16X16:
- MERGE_CTX(4, AV_RN32A);
- DECODE_Y_COEF_LOOP(4, 0,);
- SPLAT_CTX(4);
- break;
- case TX_32X32:
- MERGE_CTX(8, AV_RN64A);
- DECODE_Y_COEF_LOOP(8, 0, 32);
- SPLAT_CTX(8);
- break;
- }
-
-#define DECODE_UV_COEF_LOOP(step, v) \
- for (n = 0, y = 0; y < end_y; y += step) { \
- for (x = 0; x < end_x; x += step, n += step * step) { \
- res = (is8bitsperpixel ? decode_coeffs_b##v##_8bpp : decode_coeffs_b##v##_16bpp) \
- (s, s->uvblock[pl] + 16 * n * bytesperpixel, \
- 16 * step * step, c, e, p, a[x] + l[y], \
- uvscan, uvnb, uv_band_counts, qmul[1]); \
- a[x] = l[y] = !!res; \
- total_coeff |= !!res; \
- if (step >= 4) { \
- AV_WN16A(&s->uveob[pl][n], res); \
- } else { \
- s->uveob[pl][n] = res; \
- } \
- } \
- }
-
- p = s->prob.coef[b->uvtx][1 /* uv */][!b->intra];
- c = s->counts.coef[b->uvtx][1 /* uv */][!b->intra];
- e = s->counts.eob[b->uvtx][1 /* uv */][!b->intra];
- w4 >>= s->ss_h;
- end_x >>= s->ss_h;
- h4 >>= s->ss_v;
- end_y >>= s->ss_v;
- for (pl = 0; pl < 2; pl++) {
- a = &s->above_uv_nnz_ctx[pl][col << !s->ss_h];
- l = &s->left_uv_nnz_ctx[pl][(row & 7) << !s->ss_v];
- switch (b->uvtx) {
- case TX_4X4:
- DECODE_UV_COEF_LOOP(1,);
- break;
- case TX_8X8:
- MERGE_CTX(2, AV_RN16A);
- DECODE_UV_COEF_LOOP(2,);
- SPLAT_CTX(2);
- break;
- case TX_16X16:
- MERGE_CTX(4, AV_RN32A);
- DECODE_UV_COEF_LOOP(4,);
- SPLAT_CTX(4);
- break;
- case TX_32X32:
- MERGE_CTX(8, AV_RN64A);
- DECODE_UV_COEF_LOOP(8, 32);
- SPLAT_CTX(8);
- break;
- }
- }
-
- return total_coeff;
-}
-
-static int decode_coeffs_8bpp(AVCodecContext *ctx)
-{
- return decode_coeffs(ctx, 1);
-}
-
-static int decode_coeffs_16bpp(AVCodecContext *ctx)
-{
- return decode_coeffs(ctx, 0);
-}
-
-static av_always_inline int check_intra_mode(VP9Context *s, int mode, uint8_t **a,
- uint8_t *dst_edge, ptrdiff_t stride_edge,
- uint8_t *dst_inner, ptrdiff_t stride_inner,
- uint8_t *l, int col, int x, int w,
- int row, int y, enum TxfmMode tx,
- int p, int ss_h, int ss_v, int bytesperpixel)
-{
- int have_top = row > 0 || y > 0;
- int have_left = col > s->tile_col_start || x > 0;
- int have_right = x < w - 1;
- int bpp = s->s.h.bpp;
- static const uint8_t mode_conv[10][2 /* have_left */][2 /* have_top */] = {
- [VERT_PRED] = { { DC_127_PRED, VERT_PRED },
- { DC_127_PRED, VERT_PRED } },
- [HOR_PRED] = { { DC_129_PRED, DC_129_PRED },
- { HOR_PRED, HOR_PRED } },
- [DC_PRED] = { { DC_128_PRED, TOP_DC_PRED },
- { LEFT_DC_PRED, DC_PRED } },
- [DIAG_DOWN_LEFT_PRED] = { { DC_127_PRED, DIAG_DOWN_LEFT_PRED },
- { DC_127_PRED, DIAG_DOWN_LEFT_PRED } },
- [DIAG_DOWN_RIGHT_PRED] = { { DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_RIGHT_PRED },
- { DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_RIGHT_PRED } },
- [VERT_RIGHT_PRED] = { { VERT_RIGHT_PRED, VERT_RIGHT_PRED },
- { VERT_RIGHT_PRED, VERT_RIGHT_PRED } },
- [HOR_DOWN_PRED] = { { HOR_DOWN_PRED, HOR_DOWN_PRED },
- { HOR_DOWN_PRED, HOR_DOWN_PRED } },
- [VERT_LEFT_PRED] = { { DC_127_PRED, VERT_LEFT_PRED },
- { DC_127_PRED, VERT_LEFT_PRED } },
- [HOR_UP_PRED] = { { DC_129_PRED, DC_129_PRED },
- { HOR_UP_PRED, HOR_UP_PRED } },
- [TM_VP8_PRED] = { { DC_129_PRED, VERT_PRED },
- { HOR_PRED, TM_VP8_PRED } },
- };
- static const struct {
- uint8_t needs_left:1;
- uint8_t needs_top:1;
- uint8_t needs_topleft:1;
- uint8_t needs_topright:1;
- uint8_t invert_left:1;
- } edges[N_INTRA_PRED_MODES] = {
- [VERT_PRED] = { .needs_top = 1 },
- [HOR_PRED] = { .needs_left = 1 },
- [DC_PRED] = { .needs_top = 1, .needs_left = 1 },
- [DIAG_DOWN_LEFT_PRED] = { .needs_top = 1, .needs_topright = 1 },
- [DIAG_DOWN_RIGHT_PRED] = { .needs_left = 1, .needs_top = 1, .needs_topleft = 1 },
- [VERT_RIGHT_PRED] = { .needs_left = 1, .needs_top = 1, .needs_topleft = 1 },
- [HOR_DOWN_PRED] = { .needs_left = 1, .needs_top = 1, .needs_topleft = 1 },
- [VERT_LEFT_PRED] = { .needs_top = 1, .needs_topright = 1 },
- [HOR_UP_PRED] = { .needs_left = 1, .invert_left = 1 },
- [TM_VP8_PRED] = { .needs_left = 1, .needs_top = 1, .needs_topleft = 1 },
- [LEFT_DC_PRED] = { .needs_left = 1 },
- [TOP_DC_PRED] = { .needs_top = 1 },
- [DC_128_PRED] = { 0 },
- [DC_127_PRED] = { 0 },
- [DC_129_PRED] = { 0 }
- };
-
- av_assert2(mode >= 0 && mode < 10);
- mode = mode_conv[mode][have_left][have_top];
- if (edges[mode].needs_top) {
- uint8_t *top, *topleft;
- int n_px_need = 4 << tx, n_px_have = (((s->cols - col) << !ss_h) - x) * 4;
- int n_px_need_tr = 0;
-
- if (tx == TX_4X4 && edges[mode].needs_topright && have_right)
- n_px_need_tr = 4;
-
- // if top of sb64-row, use s->intra_pred_data[] instead of
- // dst[-stride] for intra prediction (it contains pre- instead of
- // post-loopfilter data)
- if (have_top) {
- top = !(row & 7) && !y ?
- s->intra_pred_data[p] + (col * (8 >> ss_h) + x * 4) * bytesperpixel :
- y == 0 ? &dst_edge[-stride_edge] : &dst_inner[-stride_inner];
- if (have_left)
- topleft = !(row & 7) && !y ?
- s->intra_pred_data[p] + (col * (8 >> ss_h) + x * 4) * bytesperpixel :
- y == 0 || x == 0 ? &dst_edge[-stride_edge] :
- &dst_inner[-stride_inner];
- }
-
- if (have_top &&
- (!edges[mode].needs_topleft || (have_left && top == topleft)) &&
- (tx != TX_4X4 || !edges[mode].needs_topright || have_right) &&
- n_px_need + n_px_need_tr <= n_px_have) {
- *a = top;
- } else {
- if (have_top) {
- if (n_px_need <= n_px_have) {
- memcpy(*a, top, n_px_need * bytesperpixel);
- } else {
-#define memset_bpp(c, i1, v, i2, num) do { \
- if (bytesperpixel == 1) { \
- memset(&(c)[(i1)], (v)[(i2)], (num)); \
- } else { \
- int n, val = AV_RN16A(&(v)[(i2) * 2]); \
- for (n = 0; n < (num); n++) { \
- AV_WN16A(&(c)[((i1) + n) * 2], val); \
- } \
- } \
-} while (0)
- memcpy(*a, top, n_px_have * bytesperpixel);
- memset_bpp(*a, n_px_have, (*a), n_px_have - 1, n_px_need - n_px_have);
- }
- } else {
-#define memset_val(c, val, num) do { \
- if (bytesperpixel == 1) { \
- memset((c), (val), (num)); \
- } else { \
- int n; \
- for (n = 0; n < (num); n++) { \
- AV_WN16A(&(c)[n * 2], (val)); \
- } \
- } \
-} while (0)
- memset_val(*a, (128 << (bpp - 8)) - 1, n_px_need);
- }
- if (edges[mode].needs_topleft) {
- if (have_left && have_top) {
-#define assign_bpp(c, i1, v, i2) do { \
- if (bytesperpixel == 1) { \
- (c)[(i1)] = (v)[(i2)]; \
- } else { \
- AV_COPY16(&(c)[(i1) * 2], &(v)[(i2) * 2]); \
- } \
-} while (0)
- assign_bpp(*a, -1, topleft, -1);
- } else {
-#define assign_val(c, i, v) do { \
- if (bytesperpixel == 1) { \
- (c)[(i)] = (v); \
- } else { \
- AV_WN16A(&(c)[(i) * 2], (v)); \
- } \
-} while (0)
- assign_val((*a), -1, (128 << (bpp - 8)) + (have_top ? +1 : -1));
- }
- }
- if (tx == TX_4X4 && edges[mode].needs_topright) {
- if (have_top && have_right &&
- n_px_need + n_px_need_tr <= n_px_have) {
- memcpy(&(*a)[4 * bytesperpixel], &top[4 * bytesperpixel], 4 * bytesperpixel);
- } else {
- memset_bpp(*a, 4, *a, 3, 4);
- }
- }
- }
- }
- if (edges[mode].needs_left) {
- if (have_left) {
- int n_px_need = 4 << tx, i, n_px_have = (((s->rows - row) << !ss_v) - y) * 4;
- uint8_t *dst = x == 0 ? dst_edge : dst_inner;
- ptrdiff_t stride = x == 0 ? stride_edge : stride_inner;
-
- if (edges[mode].invert_left) {
- if (n_px_need <= n_px_have) {
- for (i = 0; i < n_px_need; i++)
- assign_bpp(l, i, &dst[i * stride], -1);
- } else {
- for (i = 0; i < n_px_have; i++)
- assign_bpp(l, i, &dst[i * stride], -1);
- memset_bpp(l, n_px_have, l, n_px_have - 1, n_px_need - n_px_have);
- }
- } else {
- if (n_px_need <= n_px_have) {
- for (i = 0; i < n_px_need; i++)
- assign_bpp(l, n_px_need - 1 - i, &dst[i * stride], -1);
- } else {
- for (i = 0; i < n_px_have; i++)
- assign_bpp(l, n_px_need - 1 - i, &dst[i * stride], -1);
- memset_bpp(l, 0, l, n_px_need - n_px_have, n_px_need - n_px_have);
- }
- }
- } else {
- memset_val(l, (128 << (bpp - 8)) + 1, 4 << tx);
- }
- }
-
- return mode;
-}
-
-static av_always_inline void intra_recon(AVCodecContext *ctx, ptrdiff_t y_off,
- ptrdiff_t uv_off, int bytesperpixel)
-{
- VP9Context *s = ctx->priv_data;
- VP9Block *b = s->b;
- int row = s->row, col = s->col;
- int w4 = bwh_tab[1][b->bs][0] << 1, step1d = 1 << b->tx, n;
- int h4 = bwh_tab[1][b->bs][1] << 1, x, y, step = 1 << (b->tx * 2);
- int end_x = FFMIN(2 * (s->cols - col), w4);
- int end_y = FFMIN(2 * (s->rows - row), h4);
- int tx = 4 * s->s.h.lossless + b->tx, uvtx = b->uvtx + 4 * s->s.h.lossless;
- int uvstep1d = 1 << b->uvtx, p;
- uint8_t *dst = s->dst[0], *dst_r = s->s.frames[CUR_FRAME].tf.f->data[0] + y_off;
- LOCAL_ALIGNED_32(uint8_t, a_buf, [96]);
- LOCAL_ALIGNED_32(uint8_t, l, [64]);
-
- for (n = 0, y = 0; y < end_y; y += step1d) {
- uint8_t *ptr = dst, *ptr_r = dst_r;
- for (x = 0; x < end_x; x += step1d, ptr += 4 * step1d * bytesperpixel,
- ptr_r += 4 * step1d * bytesperpixel, n += step) {
- int mode = b->mode[b->bs > BS_8x8 && b->tx == TX_4X4 ?
- y * 2 + x : 0];
- uint8_t *a = &a_buf[32];
- enum TxfmType txtp = vp9_intra_txfm_type[mode];
- int eob = b->skip ? 0 : b->tx > TX_8X8 ? AV_RN16A(&s->eob[n]) : s->eob[n];
-
- mode = check_intra_mode(s, mode, &a, ptr_r,
- s->s.frames[CUR_FRAME].tf.f->linesize[0],
- ptr, s->y_stride, l,
- col, x, w4, row, y, b->tx, 0, 0, 0, bytesperpixel);
- s->dsp.intra_pred[b->tx][mode](ptr, s->y_stride, l, a);
- if (eob)
- s->dsp.itxfm_add[tx][txtp](ptr, s->y_stride,
- s->block + 16 * n * bytesperpixel, eob);
- }
- dst_r += 4 * step1d * s->s.frames[CUR_FRAME].tf.f->linesize[0];
- dst += 4 * step1d * s->y_stride;
- }
-
- // U/V
- w4 >>= s->ss_h;
- end_x >>= s->ss_h;
- end_y >>= s->ss_v;
- step = 1 << (b->uvtx * 2);
- for (p = 0; p < 2; p++) {
- dst = s->dst[1 + p];
- dst_r = s->s.frames[CUR_FRAME].tf.f->data[1 + p] + uv_off;
- for (n = 0, y = 0; y < end_y; y += uvstep1d) {
- uint8_t *ptr = dst, *ptr_r = dst_r;
- for (x = 0; x < end_x; x += uvstep1d, ptr += 4 * uvstep1d * bytesperpixel,
- ptr_r += 4 * uvstep1d * bytesperpixel, n += step) {
- int mode = b->uvmode;
- uint8_t *a = &a_buf[32];
- int eob = b->skip ? 0 : b->uvtx > TX_8X8 ? AV_RN16A(&s->uveob[p][n]) : s->uveob[p][n];
-
- mode = check_intra_mode(s, mode, &a, ptr_r,
- s->s.frames[CUR_FRAME].tf.f->linesize[1],
- ptr, s->uv_stride, l, col, x, w4, row, y,
- b->uvtx, p + 1, s->ss_h, s->ss_v, bytesperpixel);
- s->dsp.intra_pred[b->uvtx][mode](ptr, s->uv_stride, l, a);
- if (eob)
- s->dsp.itxfm_add[uvtx][DCT_DCT](ptr, s->uv_stride,
- s->uvblock[p] + 16 * n * bytesperpixel, eob);
- }
- dst_r += 4 * uvstep1d * s->s.frames[CUR_FRAME].tf.f->linesize[1];
- dst += 4 * uvstep1d * s->uv_stride;
- }
- }
-}
-
-static void intra_recon_8bpp(AVCodecContext *ctx, ptrdiff_t y_off, ptrdiff_t uv_off)
-{
- intra_recon(ctx, y_off, uv_off, 1);
-}
-
-static void intra_recon_16bpp(AVCodecContext *ctx, ptrdiff_t y_off, ptrdiff_t uv_off)
-{
- intra_recon(ctx, y_off, uv_off, 2);
-}
-
-static av_always_inline void mc_luma_unscaled(VP9Context *s, vp9_mc_func (*mc)[2],
- uint8_t *dst, ptrdiff_t dst_stride,
- const uint8_t *ref, ptrdiff_t ref_stride,
- ThreadFrame *ref_frame,
- ptrdiff_t y, ptrdiff_t x, const VP56mv *mv,
- int bw, int bh, int w, int h, int bytesperpixel)
-{
- int mx = mv->x, my = mv->y, th;
-
- y += my >> 3;
- x += mx >> 3;
- ref += y * ref_stride + x * bytesperpixel;
- mx &= 7;
- my &= 7;
- // FIXME bilinear filter only needs 0/1 pixels, not 3/4
- // we use +7 because the last 7 pixels of each sbrow can be changed in
- // the longest loopfilter of the next sbrow
- th = (y + bh + 4 * !!my + 7) >> 6;
- ff_thread_await_progress(ref_frame, FFMAX(th, 0), 0);
- // The arm/aarch64 _hv filters read one more row than what actually is
- // needed, so switch to emulated edge one pixel sooner vertically
- // (!!my * 5) than horizontally (!!mx * 4).
- if (x < !!mx * 3 || y < !!my * 3 ||
- x + !!mx * 4 > w - bw || y + !!my * 5 > h - bh) {
- s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
- ref - !!my * 3 * ref_stride - !!mx * 3 * bytesperpixel,
- 160, ref_stride,
- bw + !!mx * 7, bh + !!my * 7,
- x - !!mx * 3, y - !!my * 3, w, h);
- ref = s->edge_emu_buffer + !!my * 3 * 160 + !!mx * 3 * bytesperpixel;
- ref_stride = 160;
- }
- mc[!!mx][!!my](dst, dst_stride, ref, ref_stride, bh, mx << 1, my << 1);
-}
-
-static av_always_inline void mc_chroma_unscaled(VP9Context *s, vp9_mc_func (*mc)[2],
- uint8_t *dst_u, uint8_t *dst_v,
- ptrdiff_t dst_stride,
- const uint8_t *ref_u, ptrdiff_t src_stride_u,
- const uint8_t *ref_v, ptrdiff_t src_stride_v,
- ThreadFrame *ref_frame,
- ptrdiff_t y, ptrdiff_t x, const VP56mv *mv,
- int bw, int bh, int w, int h, int bytesperpixel)
-{
- int mx = mv->x * (1 << !s->ss_h), my = mv->y * (1 << !s->ss_v), th;
-
- y += my >> 4;
- x += mx >> 4;
- ref_u += y * src_stride_u + x * bytesperpixel;
- ref_v += y * src_stride_v + x * bytesperpixel;
- mx &= 15;
- my &= 15;
- // FIXME bilinear filter only needs 0/1 pixels, not 3/4
- // we use +7 because the last 7 pixels of each sbrow can be changed in
- // the longest loopfilter of the next sbrow
- th = (y + bh + 4 * !!my + 7) >> (6 - s->ss_v);
- ff_thread_await_progress(ref_frame, FFMAX(th, 0), 0);
- // The arm/aarch64 _hv filters read one more row than what actually is
- // needed, so switch to emulated edge one pixel sooner vertically
- // (!!my * 5) than horizontally (!!mx * 4).
- if (x < !!mx * 3 || y < !!my * 3 ||
- x + !!mx * 4 > w - bw || y + !!my * 5 > h - bh) {
- s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
- ref_u - !!my * 3 * src_stride_u - !!mx * 3 * bytesperpixel,
- 160, src_stride_u,
- bw + !!mx * 7, bh + !!my * 7,
- x - !!mx * 3, y - !!my * 3, w, h);
- ref_u = s->edge_emu_buffer + !!my * 3 * 160 + !!mx * 3 * bytesperpixel;
- mc[!!mx][!!my](dst_u, dst_stride, ref_u, 160, bh, mx, my);
-
- s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
- ref_v - !!my * 3 * src_stride_v - !!mx * 3 * bytesperpixel,
- 160, src_stride_v,
- bw + !!mx * 7, bh + !!my * 7,
- x - !!mx * 3, y - !!my * 3, w, h);
- ref_v = s->edge_emu_buffer + !!my * 3 * 160 + !!mx * 3 * bytesperpixel;
- mc[!!mx][!!my](dst_v, dst_stride, ref_v, 160, bh, mx, my);
- } else {
- mc[!!mx][!!my](dst_u, dst_stride, ref_u, src_stride_u, bh, mx, my);
- mc[!!mx][!!my](dst_v, dst_stride, ref_v, src_stride_v, bh, mx, my);
- }
-}
-
-#define mc_luma_dir(s, mc, dst, dst_ls, src, src_ls, tref, row, col, mv, \
- px, py, pw, ph, bw, bh, w, h, i) \
- mc_luma_unscaled(s, s->dsp.mc, dst, dst_ls, src, src_ls, tref, row, col, \
- mv, bw, bh, w, h, bytesperpixel)
-#define mc_chroma_dir(s, mc, dstu, dstv, dst_ls, srcu, srcu_ls, srcv, srcv_ls, tref, \
- row, col, mv, px, py, pw, ph, bw, bh, w, h, i) \
- mc_chroma_unscaled(s, s->dsp.mc, dstu, dstv, dst_ls, srcu, srcu_ls, srcv, srcv_ls, tref, \
- row, col, mv, bw, bh, w, h, bytesperpixel)
-#define SCALED 0
-#define FN(x) x##_8bpp
-#define BYTES_PER_PIXEL 1
-#include "vp9_mc_template.c"
-#undef FN
-#undef BYTES_PER_PIXEL
-#define FN(x) x##_16bpp
-#define BYTES_PER_PIXEL 2
-#include "vp9_mc_template.c"
-#undef mc_luma_dir
-#undef mc_chroma_dir
-#undef FN
-#undef BYTES_PER_PIXEL
-#undef SCALED
-
-static av_always_inline void mc_luma_scaled(VP9Context *s, vp9_scaled_mc_func smc,
- vp9_mc_func (*mc)[2],
- uint8_t *dst, ptrdiff_t dst_stride,
- const uint8_t *ref, ptrdiff_t ref_stride,
- ThreadFrame *ref_frame,
- ptrdiff_t y, ptrdiff_t x, const VP56mv *in_mv,
- int px, int py, int pw, int ph,
- int bw, int bh, int w, int h, int bytesperpixel,
- const uint16_t *scale, const uint8_t *step)
-{
- if (s->s.frames[CUR_FRAME].tf.f->width == ref_frame->f->width &&
- s->s.frames[CUR_FRAME].tf.f->height == ref_frame->f->height) {
- mc_luma_unscaled(s, mc, dst, dst_stride, ref, ref_stride, ref_frame,
- y, x, in_mv, bw, bh, w, h, bytesperpixel);
- } else {
-#define scale_mv(n, dim) (((int64_t)(n) * scale[dim]) >> 14)
- int mx, my;
- int refbw_m1, refbh_m1;
- int th;
- VP56mv mv;
-
- mv.x = av_clip(in_mv->x, -(x + pw - px + 4) * 8, (s->cols * 8 - x + px + 3) * 8);
- mv.y = av_clip(in_mv->y, -(y + ph - py + 4) * 8, (s->rows * 8 - y + py + 3) * 8);
- // BUG libvpx seems to scale the two components separately. This introduces
- // rounding errors but we have to reproduce them to be exactly compatible
- // with the output from libvpx...
- mx = scale_mv(mv.x * 2, 0) + scale_mv(x * 16, 0);
- my = scale_mv(mv.y * 2, 1) + scale_mv(y * 16, 1);
-
- y = my >> 4;
- x = mx >> 4;
- ref += y * ref_stride + x * bytesperpixel;
- mx &= 15;
- my &= 15;
- refbw_m1 = ((bw - 1) * step[0] + mx) >> 4;
- refbh_m1 = ((bh - 1) * step[1] + my) >> 4;
- // FIXME bilinear filter only needs 0/1 pixels, not 3/4
- // we use +7 because the last 7 pixels of each sbrow can be changed in
- // the longest loopfilter of the next sbrow
- th = (y + refbh_m1 + 4 + 7) >> 6;
- ff_thread_await_progress(ref_frame, FFMAX(th, 0), 0);
- // The arm/aarch64 _hv filters read one more row than what actually is
- // needed, so switch to emulated edge one pixel sooner vertically
- // (y + 5 >= h - refbh_m1) than horizontally (x + 4 >= w - refbw_m1).
- if (x < 3 || y < 3 || x + 4 >= w - refbw_m1 || y + 5 >= h - refbh_m1) {
- s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
- ref - 3 * ref_stride - 3 * bytesperpixel,
- 288, ref_stride,
- refbw_m1 + 8, refbh_m1 + 8,
- x - 3, y - 3, w, h);
- ref = s->edge_emu_buffer + 3 * 288 + 3 * bytesperpixel;
- ref_stride = 288;
- }
- smc(dst, dst_stride, ref, ref_stride, bh, mx, my, step[0], step[1]);
- }
-}
-
-static av_always_inline void mc_chroma_scaled(VP9Context *s, vp9_scaled_mc_func smc,
- vp9_mc_func (*mc)[2],
- uint8_t *dst_u, uint8_t *dst_v,
- ptrdiff_t dst_stride,
- const uint8_t *ref_u, ptrdiff_t src_stride_u,
- const uint8_t *ref_v, ptrdiff_t src_stride_v,
- ThreadFrame *ref_frame,
- ptrdiff_t y, ptrdiff_t x, const VP56mv *in_mv,
- int px, int py, int pw, int ph,
- int bw, int bh, int w, int h, int bytesperpixel,
- const uint16_t *scale, const uint8_t *step)
-{
- if (s->s.frames[CUR_FRAME].tf.f->width == ref_frame->f->width &&
- s->s.frames[CUR_FRAME].tf.f->height == ref_frame->f->height) {
- mc_chroma_unscaled(s, mc, dst_u, dst_v, dst_stride, ref_u, src_stride_u,
- ref_v, src_stride_v, ref_frame,
- y, x, in_mv, bw, bh, w, h, bytesperpixel);
- } else {
- int mx, my;
- int refbw_m1, refbh_m1;
- int th;
- VP56mv mv;
-
- if (s->ss_h) {
- // BUG https://code.google.com/p/webm/issues/detail?id=820
- mv.x = av_clip(in_mv->x, -(x + pw - px + 4) * 16, (s->cols * 4 - x + px + 3) * 16);
- mx = scale_mv(mv.x, 0) + (scale_mv(x * 16, 0) & ~15) + (scale_mv(x * 32, 0) & 15);
- } else {
- mv.x = av_clip(in_mv->x, -(x + pw - px + 4) * 8, (s->cols * 8 - x + px + 3) * 8);
- mx = scale_mv(mv.x * 2, 0) + scale_mv(x * 16, 0);
- }
- if (s->ss_v) {
- // BUG https://code.google.com/p/webm/issues/detail?id=820
- mv.y = av_clip(in_mv->y, -(y + ph - py + 4) * 16, (s->rows * 4 - y + py + 3) * 16);
- my = scale_mv(mv.y, 1) + (scale_mv(y * 16, 1) & ~15) + (scale_mv(y * 32, 1) & 15);
- } else {
- mv.y = av_clip(in_mv->y, -(y + ph - py + 4) * 8, (s->rows * 8 - y + py + 3) * 8);
- my = scale_mv(mv.y * 2, 1) + scale_mv(y * 16, 1);
- }
-#undef scale_mv
- y = my >> 4;
- x = mx >> 4;
- ref_u += y * src_stride_u + x * bytesperpixel;
- ref_v += y * src_stride_v + x * bytesperpixel;
- mx &= 15;
- my &= 15;
- refbw_m1 = ((bw - 1) * step[0] + mx) >> 4;
- refbh_m1 = ((bh - 1) * step[1] + my) >> 4;
- // FIXME bilinear filter only needs 0/1 pixels, not 3/4
- // we use +7 because the last 7 pixels of each sbrow can be changed in
- // the longest loopfilter of the next sbrow
- th = (y + refbh_m1 + 4 + 7) >> (6 - s->ss_v);
- ff_thread_await_progress(ref_frame, FFMAX(th, 0), 0);
- // The arm/aarch64 _hv filters read one more row than what actually is
- // needed, so switch to emulated edge one pixel sooner vertically
- // (y + 5 >= h - refbh_m1) than horizontally (x + 4 >= w - refbw_m1).
- if (x < 3 || y < 3 || x + 4 >= w - refbw_m1 || y + 5 >= h - refbh_m1) {
- s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
- ref_u - 3 * src_stride_u - 3 * bytesperpixel,
- 288, src_stride_u,
- refbw_m1 + 8, refbh_m1 + 8,
- x - 3, y - 3, w, h);
- ref_u = s->edge_emu_buffer + 3 * 288 + 3 * bytesperpixel;
- smc(dst_u, dst_stride, ref_u, 288, bh, mx, my, step[0], step[1]);
-
- s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
- ref_v - 3 * src_stride_v - 3 * bytesperpixel,
- 288, src_stride_v,
- refbw_m1 + 8, refbh_m1 + 8,
- x - 3, y - 3, w, h);
- ref_v = s->edge_emu_buffer + 3 * 288 + 3 * bytesperpixel;
- smc(dst_v, dst_stride, ref_v, 288, bh, mx, my, step[0], step[1]);
- } else {
- smc(dst_u, dst_stride, ref_u, src_stride_u, bh, mx, my, step[0], step[1]);
- smc(dst_v, dst_stride, ref_v, src_stride_v, bh, mx, my, step[0], step[1]);
- }
- }
-}
-
-#define mc_luma_dir(s, mc, dst, dst_ls, src, src_ls, tref, row, col, mv, \
- px, py, pw, ph, bw, bh, w, h, i) \
- mc_luma_scaled(s, s->dsp.s##mc, s->dsp.mc, dst, dst_ls, src, src_ls, tref, row, col, \
- mv, px, py, pw, ph, bw, bh, w, h, bytesperpixel, \
- s->mvscale[b->ref[i]], s->mvstep[b->ref[i]])
-#define mc_chroma_dir(s, mc, dstu, dstv, dst_ls, srcu, srcu_ls, srcv, srcv_ls, tref, \
- row, col, mv, px, py, pw, ph, bw, bh, w, h, i) \
- mc_chroma_scaled(s, s->dsp.s##mc, s->dsp.mc, dstu, dstv, dst_ls, srcu, srcu_ls, srcv, srcv_ls, tref, \
- row, col, mv, px, py, pw, ph, bw, bh, w, h, bytesperpixel, \
- s->mvscale[b->ref[i]], s->mvstep[b->ref[i]])
-#define SCALED 1
-#define FN(x) x##_scaled_8bpp
-#define BYTES_PER_PIXEL 1
-#include "vp9_mc_template.c"
-#undef FN
-#undef BYTES_PER_PIXEL
-#define FN(x) x##_scaled_16bpp
-#define BYTES_PER_PIXEL 2
-#include "vp9_mc_template.c"
-#undef mc_luma_dir
-#undef mc_chroma_dir
-#undef FN
-#undef BYTES_PER_PIXEL
-#undef SCALED
-
-static av_always_inline void inter_recon(AVCodecContext *ctx, int bytesperpixel)
-{
- VP9Context *s = ctx->priv_data;
- VP9Block *b = s->b;
- int row = s->row, col = s->col;
-
- if (s->mvscale[b->ref[0]][0] || (b->comp && s->mvscale[b->ref[1]][0])) {
- if (bytesperpixel == 1) {
- inter_pred_scaled_8bpp(ctx);
- } else {
- inter_pred_scaled_16bpp(ctx);
- }
- } else {
- if (bytesperpixel == 1) {
- inter_pred_8bpp(ctx);
- } else {
- inter_pred_16bpp(ctx);
- }
- }
- if (!b->skip) {
- /* mostly copied intra_recon() */
-
- int w4 = bwh_tab[1][b->bs][0] << 1, step1d = 1 << b->tx, n;
- int h4 = bwh_tab[1][b->bs][1] << 1, x, y, step = 1 << (b->tx * 2);
- int end_x = FFMIN(2 * (s->cols - col), w4);
- int end_y = FFMIN(2 * (s->rows - row), h4);
- int tx = 4 * s->s.h.lossless + b->tx, uvtx = b->uvtx + 4 * s->s.h.lossless;
- int uvstep1d = 1 << b->uvtx, p;
- uint8_t *dst = s->dst[0];
-
- // y itxfm add
- for (n = 0, y = 0; y < end_y; y += step1d) {
- uint8_t *ptr = dst;
- for (x = 0; x < end_x; x += step1d,
- ptr += 4 * step1d * bytesperpixel, n += step) {
- int eob = b->tx > TX_8X8 ? AV_RN16A(&s->eob[n]) : s->eob[n];
-
- if (eob)
- s->dsp.itxfm_add[tx][DCT_DCT](ptr, s->y_stride,
- s->block + 16 * n * bytesperpixel, eob);
- }
- dst += 4 * s->y_stride * step1d;
- }
-
- // uv itxfm add
- end_x >>= s->ss_h;
- end_y >>= s->ss_v;
- step = 1 << (b->uvtx * 2);
- for (p = 0; p < 2; p++) {
- dst = s->dst[p + 1];
- for (n = 0, y = 0; y < end_y; y += uvstep1d) {
- uint8_t *ptr = dst;
- for (x = 0; x < end_x; x += uvstep1d,
- ptr += 4 * uvstep1d * bytesperpixel, n += step) {
- int eob = b->uvtx > TX_8X8 ? AV_RN16A(&s->uveob[p][n]) : s->uveob[p][n];
-
- if (eob)
- s->dsp.itxfm_add[uvtx][DCT_DCT](ptr, s->uv_stride,
- s->uvblock[p] + 16 * n * bytesperpixel, eob);
- }
- dst += 4 * uvstep1d * s->uv_stride;
- }
- }
- }
-}
-
-static void inter_recon_8bpp(AVCodecContext *ctx)
-{
- inter_recon(ctx, 1);
-}
-
-static void inter_recon_16bpp(AVCodecContext *ctx)
-{
- inter_recon(ctx, 2);
-}
-
-static av_always_inline void mask_edges(uint8_t (*mask)[8][4], int ss_h, int ss_v,
- int row_and_7, int col_and_7,
- int w, int h, int col_end, int row_end,
- enum TxfmMode tx, int skip_inter)
-{
- static const unsigned wide_filter_col_mask[2] = { 0x11, 0x01 };
- static const unsigned wide_filter_row_mask[2] = { 0x03, 0x07 };
-
- // FIXME I'm pretty sure all loops can be replaced by a single LUT if
- // we make VP9Filter.mask uint64_t (i.e. row/col all single variable)
- // and make the LUT 5-indexed (bl, bp, is_uv, tx and row/col), and then
- // use row_and_7/col_and_7 as shifts (1*col_and_7+8*row_and_7)
-
- // the intended behaviour of the vp9 loopfilter is to work on 8-pixel
- // edges. This means that for UV, we work on two subsampled blocks at
- // a time, and we only use the topleft block's mode information to set
- // things like block strength. Thus, for any block size smaller than
- // 16x16, ignore the odd portion of the block.
- if (tx == TX_4X4 && (ss_v | ss_h)) {
- if (h == ss_v) {
- if (row_and_7 & 1)
- return;
- if (!row_end)
- h += 1;
- }
- if (w == ss_h) {
- if (col_and_7 & 1)
- return;
- if (!col_end)
- w += 1;
- }
- }
-
- if (tx == TX_4X4 && !skip_inter) {
- int t = 1 << col_and_7, m_col = (t << w) - t, y;
- // on 32-px edges, use the 8-px wide loopfilter; else, use 4-px wide
- int m_row_8 = m_col & wide_filter_col_mask[ss_h], m_row_4 = m_col - m_row_8;
-
- for (y = row_and_7; y < h + row_and_7; y++) {
- int col_mask_id = 2 - !(y & wide_filter_row_mask[ss_v]);
-
- mask[0][y][1] |= m_row_8;
- mask[0][y][2] |= m_row_4;
- // for odd lines, if the odd col is not being filtered,
- // skip odd row also:
- // .---. <-- a
- // | |
- // |___| <-- b
- // ^ ^
- // c d
- //
- // if a/c are even row/col and b/d are odd, and d is skipped,
- // e.g. right edge of size-66x66.webm, then skip b also (bug)
- if ((ss_h & ss_v) && (col_end & 1) && (y & 1)) {
- mask[1][y][col_mask_id] |= (t << (w - 1)) - t;
- } else {
- mask[1][y][col_mask_id] |= m_col;
- }
- if (!ss_h)
- mask[0][y][3] |= m_col;
- if (!ss_v) {
- if (ss_h && (col_end & 1))
- mask[1][y][3] |= (t << (w - 1)) - t;
- else
- mask[1][y][3] |= m_col;
- }
- }
- } else {
- int y, t = 1 << col_and_7, m_col = (t << w) - t;
-
- if (!skip_inter) {
- int mask_id = (tx == TX_8X8);
- static const unsigned masks[4] = { 0xff, 0x55, 0x11, 0x01 };
- int l2 = tx + ss_h - 1, step1d;
- int m_row = m_col & masks[l2];
-
- // at odd UV col/row edges tx16/tx32 loopfilter edges, force
- // 8wd loopfilter to prevent going off the visible edge.
- if (ss_h && tx > TX_8X8 && (w ^ (w - 1)) == 1) {
- int m_row_16 = ((t << (w - 1)) - t) & masks[l2];
- int m_row_8 = m_row - m_row_16;
-
- for (y = row_and_7; y < h + row_and_7; y++) {
- mask[0][y][0] |= m_row_16;
- mask[0][y][1] |= m_row_8;
- }
- } else {
- for (y = row_and_7; y < h + row_and_7; y++)
- mask[0][y][mask_id] |= m_row;
- }
-
- l2 = tx + ss_v - 1;
- step1d = 1 << l2;
- if (ss_v && tx > TX_8X8 && (h ^ (h - 1)) == 1) {
- for (y = row_and_7; y < h + row_and_7 - 1; y += step1d)
- mask[1][y][0] |= m_col;
- if (y - row_and_7 == h - 1)
- mask[1][y][1] |= m_col;
- } else {
- for (y = row_and_7; y < h + row_and_7; y += step1d)
- mask[1][y][mask_id] |= m_col;
- }
- } else if (tx != TX_4X4) {
- int mask_id;
-
- mask_id = (tx == TX_8X8) || (h == ss_v);
- mask[1][row_and_7][mask_id] |= m_col;
- mask_id = (tx == TX_8X8) || (w == ss_h);
- for (y = row_and_7; y < h + row_and_7; y++)
- mask[0][y][mask_id] |= t;
- } else {
- int t8 = t & wide_filter_col_mask[ss_h], t4 = t - t8;
-
- for (y = row_and_7; y < h + row_and_7; y++) {
- mask[0][y][2] |= t4;
- mask[0][y][1] |= t8;
- }
- mask[1][row_and_7][2 - !(row_and_7 & wide_filter_row_mask[ss_v])] |= m_col;
- }
- }
-}
-
-static void decode_b(AVCodecContext *ctx, int row, int col,
- struct VP9Filter *lflvl, ptrdiff_t yoff, ptrdiff_t uvoff,
- enum BlockLevel bl, enum BlockPartition bp)
-{
- VP9Context *s = ctx->priv_data;
- VP9Block *b = s->b;
- enum BlockSize bs = bl * 3 + bp;
- int bytesperpixel = s->bytesperpixel;
- int w4 = bwh_tab[1][bs][0], h4 = bwh_tab[1][bs][1], lvl;
- int emu[2];
- AVFrame *f = s->s.frames[CUR_FRAME].tf.f;
-
- s->row = row;
- s->row7 = row & 7;
- s->col = col;
- s->col7 = col & 7;
- s->min_mv.x = -(128 + col * 64);
- s->min_mv.y = -(128 + row * 64);
- s->max_mv.x = 128 + (s->cols - col - w4) * 64;
- s->max_mv.y = 128 + (s->rows - row - h4) * 64;
- if (s->pass < 2) {
- b->bs = bs;
- b->bl = bl;
- b->bp = bp;
- decode_mode(ctx);
- b->uvtx = b->tx - ((s->ss_h && w4 * 2 == (1 << b->tx)) ||
- (s->ss_v && h4 * 2 == (1 << b->tx)));
-
- if (!b->skip) {
- int has_coeffs;
-
- if (bytesperpixel == 1) {
- has_coeffs = decode_coeffs_8bpp(ctx);
- } else {
- has_coeffs = decode_coeffs_16bpp(ctx);
- }
- if (!has_coeffs && b->bs <= BS_8x8 && !b->intra) {
- b->skip = 1;
- memset(&s->above_skip_ctx[col], 1, w4);
- memset(&s->left_skip_ctx[s->row7], 1, h4);
- }
- } else {
- int row7 = s->row7;
-
-#define SPLAT_ZERO_CTX(v, n) \
- switch (n) { \
- case 1: v = 0; break; \
- case 2: AV_ZERO16(&v); break; \
- case 4: AV_ZERO32(&v); break; \
- case 8: AV_ZERO64(&v); break; \
- case 16: AV_ZERO128(&v); break; \
- }
-#define SPLAT_ZERO_YUV(dir, var, off, n, dir2) \
- do { \
- SPLAT_ZERO_CTX(s->dir##_y_##var[off * 2], n * 2); \
- if (s->ss_##dir2) { \
- SPLAT_ZERO_CTX(s->dir##_uv_##var[0][off], n); \
- SPLAT_ZERO_CTX(s->dir##_uv_##var[1][off], n); \
- } else { \
- SPLAT_ZERO_CTX(s->dir##_uv_##var[0][off * 2], n * 2); \
- SPLAT_ZERO_CTX(s->dir##_uv_##var[1][off * 2], n * 2); \
- } \
- } while (0)
-
- switch (w4) {
- case 1: SPLAT_ZERO_YUV(above, nnz_ctx, col, 1, h); break;
- case 2: SPLAT_ZERO_YUV(above, nnz_ctx, col, 2, h); break;
- case 4: SPLAT_ZERO_YUV(above, nnz_ctx, col, 4, h); break;
- case 8: SPLAT_ZERO_YUV(above, nnz_ctx, col, 8, h); break;
- }
- switch (h4) {
- case 1: SPLAT_ZERO_YUV(left, nnz_ctx, row7, 1, v); break;
- case 2: SPLAT_ZERO_YUV(left, nnz_ctx, row7, 2, v); break;
- case 4: SPLAT_ZERO_YUV(left, nnz_ctx, row7, 4, v); break;
- case 8: SPLAT_ZERO_YUV(left, nnz_ctx, row7, 8, v); break;
- }
- }
-
- if (s->pass == 1) {
- s->b++;
- s->block += w4 * h4 * 64 * bytesperpixel;
- s->uvblock[0] += w4 * h4 * 64 * bytesperpixel >> (s->ss_h + s->ss_v);
- s->uvblock[1] += w4 * h4 * 64 * bytesperpixel >> (s->ss_h + s->ss_v);
- s->eob += 4 * w4 * h4;
- s->uveob[0] += 4 * w4 * h4 >> (s->ss_h + s->ss_v);
- s->uveob[1] += 4 * w4 * h4 >> (s->ss_h + s->ss_v);
-
- return;
- }
- }
-
- // emulated overhangs if the stride of the target buffer can't hold. This
- // makes it possible to support emu-edge and so on even if we have large block
- // overhangs
- emu[0] = (col + w4) * 8 * bytesperpixel > f->linesize[0] ||
- (row + h4) > s->rows;
- emu[1] = ((col + w4) * 8 >> s->ss_h) * bytesperpixel > f->linesize[1] ||
- (row + h4) > s->rows;
- if (emu[0]) {
- s->dst[0] = s->tmp_y;
- s->y_stride = 128;
- } else {
- s->dst[0] = f->data[0] + yoff;
- s->y_stride = f->linesize[0];
- }
- if (emu[1]) {
- s->dst[1] = s->tmp_uv[0];
- s->dst[2] = s->tmp_uv[1];
- s->uv_stride = 128;
- } else {
- s->dst[1] = f->data[1] + uvoff;
- s->dst[2] = f->data[2] + uvoff;
- s->uv_stride = f->linesize[1];
- }
- if (b->intra) {
- if (s->s.h.bpp > 8) {
- intra_recon_16bpp(ctx, yoff, uvoff);
- } else {
- intra_recon_8bpp(ctx, yoff, uvoff);
- }
- } else {
- if (s->s.h.bpp > 8) {
- inter_recon_16bpp(ctx);
- } else {
- inter_recon_8bpp(ctx);
- }
- }
- if (emu[0]) {
- int w = FFMIN(s->cols - col, w4) * 8, h = FFMIN(s->rows - row, h4) * 8, n, o = 0;
-
- for (n = 0; o < w; n++) {
- int bw = 64 >> n;
-
- av_assert2(n <= 4);
- if (w & bw) {
- s->dsp.mc[n][0][0][0][0](f->data[0] + yoff + o * bytesperpixel, f->linesize[0],
- s->tmp_y + o * bytesperpixel, 128, h, 0, 0);
- o += bw;
- }
- }
- }
- if (emu[1]) {
- int w = FFMIN(s->cols - col, w4) * 8 >> s->ss_h;
- int h = FFMIN(s->rows - row, h4) * 8 >> s->ss_v, n, o = 0;
-
- for (n = s->ss_h; o < w; n++) {
- int bw = 64 >> n;
-
- av_assert2(n <= 4);
- if (w & bw) {
- s->dsp.mc[n][0][0][0][0](f->data[1] + uvoff + o * bytesperpixel, f->linesize[1],
- s->tmp_uv[0] + o * bytesperpixel, 128, h, 0, 0);
- s->dsp.mc[n][0][0][0][0](f->data[2] + uvoff + o * bytesperpixel, f->linesize[2],
- s->tmp_uv[1] + o * bytesperpixel, 128, h, 0, 0);
- o += bw;
- }
- }
- }
-
- // pick filter level and find edges to apply filter to
- if (s->s.h.filter.level &&
- (lvl = s->s.h.segmentation.feat[b->seg_id].lflvl[b->intra ? 0 : b->ref[0] + 1]
- [b->mode[3] != ZEROMV]) > 0) {
- int x_end = FFMIN(s->cols - col, w4), y_end = FFMIN(s->rows - row, h4);
- int skip_inter = !b->intra && b->skip, col7 = s->col7, row7 = s->row7;
-
- setctx_2d(&lflvl->level[row7 * 8 + col7], w4, h4, 8, lvl);
- mask_edges(lflvl->mask[0], 0, 0, row7, col7, x_end, y_end, 0, 0, b->tx, skip_inter);
- if (s->ss_h || s->ss_v)
- mask_edges(lflvl->mask[1], s->ss_h, s->ss_v, row7, col7, x_end, y_end,
- s->cols & 1 && col + w4 >= s->cols ? s->cols & 7 : 0,
- s->rows & 1 && row + h4 >= s->rows ? s->rows & 7 : 0,
- b->uvtx, skip_inter);
-
- if (!s->filter_lut.lim_lut[lvl]) {
- int sharp = s->s.h.filter.sharpness;
- int limit = lvl;
-
- if (sharp > 0) {
- limit >>= (sharp + 3) >> 2;
- limit = FFMIN(limit, 9 - sharp);
- }
- limit = FFMAX(limit, 1);
-
- s->filter_lut.lim_lut[lvl] = limit;
- s->filter_lut.mblim_lut[lvl] = 2 * (lvl + 2) + limit;
- }
- }
-
- if (s->pass == 2) {
- s->b++;
- s->block += w4 * h4 * 64 * bytesperpixel;
- s->uvblock[0] += w4 * h4 * 64 * bytesperpixel >> (s->ss_v + s->ss_h);
- s->uvblock[1] += w4 * h4 * 64 * bytesperpixel >> (s->ss_v + s->ss_h);
- s->eob += 4 * w4 * h4;
- s->uveob[0] += 4 * w4 * h4 >> (s->ss_v + s->ss_h);
- s->uveob[1] += 4 * w4 * h4 >> (s->ss_v + s->ss_h);
- }
-}
-
-static void decode_sb(AVCodecContext *ctx, int row, int col, struct VP9Filter *lflvl,
- ptrdiff_t yoff, ptrdiff_t uvoff, enum BlockLevel bl)
-{
- VP9Context *s = ctx->priv_data;
- int c = ((s->above_partition_ctx[col] >> (3 - bl)) & 1) |
- (((s->left_partition_ctx[row & 0x7] >> (3 - bl)) & 1) << 1);
- const uint8_t *p = s->s.h.keyframe || s->s.h.intraonly ? vp9_default_kf_partition_probs[bl][c] :
- s->prob.p.partition[bl][c];
- enum BlockPartition bp;
- ptrdiff_t hbs = 4 >> bl;
- AVFrame *f = s->s.frames[CUR_FRAME].tf.f;
- ptrdiff_t y_stride = f->linesize[0], uv_stride = f->linesize[1];
- int bytesperpixel = s->bytesperpixel;
-
- if (bl == BL_8X8) {
- bp = vp8_rac_get_tree(&s->c, vp9_partition_tree, p);
- decode_b(ctx, row, col, lflvl, yoff, uvoff, bl, bp);
- } else if (col + hbs < s->cols) { // FIXME why not <=?
- if (row + hbs < s->rows) { // FIXME why not <=?
- bp = vp8_rac_get_tree(&s->c, vp9_partition_tree, p);
- switch (bp) {
- case PARTITION_NONE:
- decode_b(ctx, row, col, lflvl, yoff, uvoff, bl, bp);
- break;
- case PARTITION_H:
- decode_b(ctx, row, col, lflvl, yoff, uvoff, bl, bp);
- yoff += hbs * 8 * y_stride;
- uvoff += hbs * 8 * uv_stride >> s->ss_v;
- decode_b(ctx, row + hbs, col, lflvl, yoff, uvoff, bl, bp);
- break;
- case PARTITION_V:
- decode_b(ctx, row, col, lflvl, yoff, uvoff, bl, bp);
- yoff += hbs * 8 * bytesperpixel;
- uvoff += hbs * 8 * bytesperpixel >> s->ss_h;
- decode_b(ctx, row, col + hbs, lflvl, yoff, uvoff, bl, bp);
- break;
- case PARTITION_SPLIT:
- decode_sb(ctx, row, col, lflvl, yoff, uvoff, bl + 1);
- decode_sb(ctx, row, col + hbs, lflvl,
- yoff + 8 * hbs * bytesperpixel,
- uvoff + (8 * hbs * bytesperpixel >> s->ss_h), bl + 1);
- yoff += hbs * 8 * y_stride;
- uvoff += hbs * 8 * uv_stride >> s->ss_v;
- decode_sb(ctx, row + hbs, col, lflvl, yoff, uvoff, bl + 1);
- decode_sb(ctx, row + hbs, col + hbs, lflvl,
- yoff + 8 * hbs * bytesperpixel,
- uvoff + (8 * hbs * bytesperpixel >> s->ss_h), bl + 1);
- break;
- default:
- av_assert0(0);
- }
- } else if (vp56_rac_get_prob_branchy(&s->c, p[1])) {
- bp = PARTITION_SPLIT;
- decode_sb(ctx, row, col, lflvl, yoff, uvoff, bl + 1);
- decode_sb(ctx, row, col + hbs, lflvl,
- yoff + 8 * hbs * bytesperpixel,
- uvoff + (8 * hbs * bytesperpixel >> s->ss_h), bl + 1);
- } else {
- bp = PARTITION_H;
- decode_b(ctx, row, col, lflvl, yoff, uvoff, bl, bp);
- }
- } else if (row + hbs < s->rows) { // FIXME why not <=?
- if (vp56_rac_get_prob_branchy(&s->c, p[2])) {
- bp = PARTITION_SPLIT;
- decode_sb(ctx, row, col, lflvl, yoff, uvoff, bl + 1);
- yoff += hbs * 8 * y_stride;
- uvoff += hbs * 8 * uv_stride >> s->ss_v;
- decode_sb(ctx, row + hbs, col, lflvl, yoff, uvoff, bl + 1);
- } else {
- bp = PARTITION_V;
- decode_b(ctx, row, col, lflvl, yoff, uvoff, bl, bp);
- }
- } else {
- bp = PARTITION_SPLIT;
- decode_sb(ctx, row, col, lflvl, yoff, uvoff, bl + 1);
- }
- s->counts.partition[bl][c][bp]++;
-}
-
-static void decode_sb_mem(AVCodecContext *ctx, int row, int col, struct VP9Filter *lflvl,
- ptrdiff_t yoff, ptrdiff_t uvoff, enum BlockLevel bl)
-{
- VP9Context *s = ctx->priv_data;
- VP9Block *b = s->b;
- ptrdiff_t hbs = 4 >> bl;
- AVFrame *f = s->s.frames[CUR_FRAME].tf.f;
- ptrdiff_t y_stride = f->linesize[0], uv_stride = f->linesize[1];
- int bytesperpixel = s->bytesperpixel;
-
- if (bl == BL_8X8) {
- av_assert2(b->bl == BL_8X8);
- decode_b(ctx, row, col, lflvl, yoff, uvoff, b->bl, b->bp);
- } else if (s->b->bl == bl) {
- decode_b(ctx, row, col, lflvl, yoff, uvoff, b->bl, b->bp);
- if (b->bp == PARTITION_H && row + hbs < s->rows) {
- yoff += hbs * 8 * y_stride;
- uvoff += hbs * 8 * uv_stride >> s->ss_v;
- decode_b(ctx, row + hbs, col, lflvl, yoff, uvoff, b->bl, b->bp);
- } else if (b->bp == PARTITION_V && col + hbs < s->cols) {
- yoff += hbs * 8 * bytesperpixel;
- uvoff += hbs * 8 * bytesperpixel >> s->ss_h;
- decode_b(ctx, row, col + hbs, lflvl, yoff, uvoff, b->bl, b->bp);
- }
- } else {
- decode_sb_mem(ctx, row, col, lflvl, yoff, uvoff, bl + 1);
- if (col + hbs < s->cols) { // FIXME why not <=?
- if (row + hbs < s->rows) {
- decode_sb_mem(ctx, row, col + hbs, lflvl, yoff + 8 * hbs * bytesperpixel,
- uvoff + (8 * hbs * bytesperpixel >> s->ss_h), bl + 1);
- yoff += hbs * 8 * y_stride;
- uvoff += hbs * 8 * uv_stride >> s->ss_v;
- decode_sb_mem(ctx, row + hbs, col, lflvl, yoff, uvoff, bl + 1);
- decode_sb_mem(ctx, row + hbs, col + hbs, lflvl,
- yoff + 8 * hbs * bytesperpixel,
- uvoff + (8 * hbs * bytesperpixel >> s->ss_h), bl + 1);
- } else {
- yoff += hbs * 8 * bytesperpixel;
- uvoff += hbs * 8 * bytesperpixel >> s->ss_h;
- decode_sb_mem(ctx, row, col + hbs, lflvl, yoff, uvoff, bl + 1);
- }
- } else if (row + hbs < s->rows) {
- yoff += hbs * 8 * y_stride;
- uvoff += hbs * 8 * uv_stride >> s->ss_v;
- decode_sb_mem(ctx, row + hbs, col, lflvl, yoff, uvoff, bl + 1);
- }
- }
-}
-
-static av_always_inline void filter_plane_cols(VP9Context *s, int col, int ss_h, int ss_v,
- uint8_t *lvl, uint8_t (*mask)[4],
- uint8_t *dst, ptrdiff_t ls)
-{
- int y, x, bytesperpixel = s->bytesperpixel;
-
- // filter edges between columns (e.g. block1 | block2)
- for (y = 0; y < 8; y += 2 << ss_v, dst += 16 * ls, lvl += 16 << ss_v) {
- uint8_t *ptr = dst, *l = lvl, *hmask1 = mask[y], *hmask2 = mask[y + 1 + ss_v];
- unsigned hm1 = hmask1[0] | hmask1[1] | hmask1[2], hm13 = hmask1[3];
- unsigned hm2 = hmask2[1] | hmask2[2], hm23 = hmask2[3];
- unsigned hm = hm1 | hm2 | hm13 | hm23;
-
- for (x = 1; hm & ~(x - 1); x <<= 1, ptr += 8 * bytesperpixel >> ss_h) {
- if (col || x > 1) {
- if (hm1 & x) {
- int L = *l, H = L >> 4;
- int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
-
- if (hmask1[0] & x) {
- if (hmask2[0] & x) {
- av_assert2(l[8 << ss_v] == L);
- s->dsp.loop_filter_16[0](ptr, ls, E, I, H);
- } else {
- s->dsp.loop_filter_8[2][0](ptr, ls, E, I, H);
- }
- } else if (hm2 & x) {
- L = l[8 << ss_v];
- H |= (L >> 4) << 8;
- E |= s->filter_lut.mblim_lut[L] << 8;
- I |= s->filter_lut.lim_lut[L] << 8;
- s->dsp.loop_filter_mix2[!!(hmask1[1] & x)]
- [!!(hmask2[1] & x)]
- [0](ptr, ls, E, I, H);
- } else {
- s->dsp.loop_filter_8[!!(hmask1[1] & x)]
- [0](ptr, ls, E, I, H);
- }
- } else if (hm2 & x) {
- int L = l[8 << ss_v], H = L >> 4;
- int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
-
- s->dsp.loop_filter_8[!!(hmask2[1] & x)]
- [0](ptr + 8 * ls, ls, E, I, H);
- }
- }
- if (ss_h) {
- if (x & 0xAA)
- l += 2;
- } else {
- if (hm13 & x) {
- int L = *l, H = L >> 4;
- int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
-
- if (hm23 & x) {
- L = l[8 << ss_v];
- H |= (L >> 4) << 8;
- E |= s->filter_lut.mblim_lut[L] << 8;
- I |= s->filter_lut.lim_lut[L] << 8;
- s->dsp.loop_filter_mix2[0][0][0](ptr + 4 * bytesperpixel, ls, E, I, H);
- } else {
- s->dsp.loop_filter_8[0][0](ptr + 4 * bytesperpixel, ls, E, I, H);
- }
- } else if (hm23 & x) {
- int L = l[8 << ss_v], H = L >> 4;
- int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
-
- s->dsp.loop_filter_8[0][0](ptr + 8 * ls + 4 * bytesperpixel, ls, E, I, H);
- }
- l++;
- }
- }
- }
-}
-
-static av_always_inline void filter_plane_rows(VP9Context *s, int row, int ss_h, int ss_v,
- uint8_t *lvl, uint8_t (*mask)[4],
- uint8_t *dst, ptrdiff_t ls)
-{
- int y, x, bytesperpixel = s->bytesperpixel;
-
- // block1
- // filter edges between rows (e.g. ------)
- // block2
- for (y = 0; y < 8; y++, dst += 8 * ls >> ss_v) {
- uint8_t *ptr = dst, *l = lvl, *vmask = mask[y];
- unsigned vm = vmask[0] | vmask[1] | vmask[2], vm3 = vmask[3];
-
- for (x = 1; vm & ~(x - 1); x <<= (2 << ss_h), ptr += 16 * bytesperpixel, l += 2 << ss_h) {
- if (row || y) {
- if (vm & x) {
- int L = *l, H = L >> 4;
- int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
-
- if (vmask[0] & x) {
- if (vmask[0] & (x << (1 + ss_h))) {
- av_assert2(l[1 + ss_h] == L);
- s->dsp.loop_filter_16[1](ptr, ls, E, I, H);
- } else {
- s->dsp.loop_filter_8[2][1](ptr, ls, E, I, H);
- }
- } else if (vm & (x << (1 + ss_h))) {
- L = l[1 + ss_h];
- H |= (L >> 4) << 8;
- E |= s->filter_lut.mblim_lut[L] << 8;
- I |= s->filter_lut.lim_lut[L] << 8;
- s->dsp.loop_filter_mix2[!!(vmask[1] & x)]
- [!!(vmask[1] & (x << (1 + ss_h)))]
- [1](ptr, ls, E, I, H);
- } else {
- s->dsp.loop_filter_8[!!(vmask[1] & x)]
- [1](ptr, ls, E, I, H);
- }
- } else if (vm & (x << (1 + ss_h))) {
- int L = l[1 + ss_h], H = L >> 4;
- int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
-
- s->dsp.loop_filter_8[!!(vmask[1] & (x << (1 + ss_h)))]
- [1](ptr + 8 * bytesperpixel, ls, E, I, H);
- }
- }
- if (!ss_v) {
- if (vm3 & x) {
- int L = *l, H = L >> 4;
- int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
+ if (!ss_v) {
+ if (vm3 & x) {
+ int L = *l, H = L >> 4;
+ int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
if (vm3 & (x << (1 + ss_h))) {
L = l[1 + ss_h];
*end = FFMIN(sb_end, n) << 3;
}
-static av_always_inline void adapt_prob(uint8_t *p, unsigned ct0, unsigned ct1,
- int max_count, int update_factor)
-{
- unsigned ct = ct0 + ct1, p2, p1;
-
- if (!ct)
- return;
-
- update_factor = FASTDIV(update_factor * FFMIN(ct, max_count), max_count);
- p1 = *p;
- p2 = ((((int64_t) ct0) << 8) + (ct >> 1)) / ct;
- p2 = av_clip(p2, 1, 255);
-
- // (p1 * (256 - update_factor) + p2 * update_factor + 128) >> 8
- *p = p1 + (((p2 - p1) * update_factor + 128) >> 8);
-}
-
-static void adapt_probs(VP9Context *s)
-{
- int i, j, k, l, m;
- prob_context *p = &s->prob_ctx[s->s.h.framectxid].p;
- int uf = (s->s.h.keyframe || s->s.h.intraonly || !s->last_keyframe) ? 112 : 128;
-
- // coefficients
- for (i = 0; i < 4; i++)
- for (j = 0; j < 2; j++)
- for (k = 0; k < 2; k++)
- for (l = 0; l < 6; l++)
- for (m = 0; m < 6; m++) {
- uint8_t *pp = s->prob_ctx[s->s.h.framectxid].coef[i][j][k][l][m];
- unsigned *e = s->counts.eob[i][j][k][l][m];
- unsigned *c = s->counts.coef[i][j][k][l][m];
-
- if (l == 0 && m >= 3) // dc only has 3 pt
- break;
-
- adapt_prob(&pp[0], e[0], e[1], 24, uf);
- adapt_prob(&pp[1], c[0], c[1] + c[2], 24, uf);
- adapt_prob(&pp[2], c[1], c[2], 24, uf);
- }
-
- if (s->s.h.keyframe || s->s.h.intraonly) {
- memcpy(p->skip, s->prob.p.skip, sizeof(p->skip));
- memcpy(p->tx32p, s->prob.p.tx32p, sizeof(p->tx32p));
- memcpy(p->tx16p, s->prob.p.tx16p, sizeof(p->tx16p));
- memcpy(p->tx8p, s->prob.p.tx8p, sizeof(p->tx8p));
- return;
- }
-
- // skip flag
- for (i = 0; i < 3; i++)
- adapt_prob(&p->skip[i], s->counts.skip[i][0], s->counts.skip[i][1], 20, 128);
-
- // intra/inter flag
- for (i = 0; i < 4; i++)
- adapt_prob(&p->intra[i], s->counts.intra[i][0], s->counts.intra[i][1], 20, 128);
-
- // comppred flag
- if (s->s.h.comppredmode == PRED_SWITCHABLE) {
- for (i = 0; i < 5; i++)
- adapt_prob(&p->comp[i], s->counts.comp[i][0], s->counts.comp[i][1], 20, 128);
- }
-
- // reference frames
- if (s->s.h.comppredmode != PRED_SINGLEREF) {
- for (i = 0; i < 5; i++)
- adapt_prob(&p->comp_ref[i], s->counts.comp_ref[i][0],
- s->counts.comp_ref[i][1], 20, 128);
- }
-
- if (s->s.h.comppredmode != PRED_COMPREF) {
- for (i = 0; i < 5; i++) {
- uint8_t *pp = p->single_ref[i];
- unsigned (*c)[2] = s->counts.single_ref[i];
-
- adapt_prob(&pp[0], c[0][0], c[0][1], 20, 128);
- adapt_prob(&pp[1], c[1][0], c[1][1], 20, 128);
- }
- }
-
- // block partitioning
- for (i = 0; i < 4; i++)
- for (j = 0; j < 4; j++) {
- uint8_t *pp = p->partition[i][j];
- unsigned *c = s->counts.partition[i][j];
-
- adapt_prob(&pp[0], c[0], c[1] + c[2] + c[3], 20, 128);
- adapt_prob(&pp[1], c[1], c[2] + c[3], 20, 128);
- adapt_prob(&pp[2], c[2], c[3], 20, 128);
- }
-
- // tx size
- if (s->s.h.txfmmode == TX_SWITCHABLE) {
- for (i = 0; i < 2; i++) {
- unsigned *c16 = s->counts.tx16p[i], *c32 = s->counts.tx32p[i];
-
- adapt_prob(&p->tx8p[i], s->counts.tx8p[i][0], s->counts.tx8p[i][1], 20, 128);
- adapt_prob(&p->tx16p[i][0], c16[0], c16[1] + c16[2], 20, 128);
- adapt_prob(&p->tx16p[i][1], c16[1], c16[2], 20, 128);
- adapt_prob(&p->tx32p[i][0], c32[0], c32[1] + c32[2] + c32[3], 20, 128);
- adapt_prob(&p->tx32p[i][1], c32[1], c32[2] + c32[3], 20, 128);
- adapt_prob(&p->tx32p[i][2], c32[2], c32[3], 20, 128);
- }
- }
-
- // interpolation filter
- if (s->s.h.filtermode == FILTER_SWITCHABLE) {
- for (i = 0; i < 4; i++) {
- uint8_t *pp = p->filter[i];
- unsigned *c = s->counts.filter[i];
-
- adapt_prob(&pp[0], c[0], c[1] + c[2], 20, 128);
- adapt_prob(&pp[1], c[1], c[2], 20, 128);
- }
- }
-
- // inter modes
- for (i = 0; i < 7; i++) {
- uint8_t *pp = p->mv_mode[i];
- unsigned *c = s->counts.mv_mode[i];
-
- adapt_prob(&pp[0], c[2], c[1] + c[0] + c[3], 20, 128);
- adapt_prob(&pp[1], c[0], c[1] + c[3], 20, 128);
- adapt_prob(&pp[2], c[1], c[3], 20, 128);
- }
-
- // mv joints
- {
- uint8_t *pp = p->mv_joint;
- unsigned *c = s->counts.mv_joint;
-
- adapt_prob(&pp[0], c[0], c[1] + c[2] + c[3], 20, 128);
- adapt_prob(&pp[1], c[1], c[2] + c[3], 20, 128);
- adapt_prob(&pp[2], c[2], c[3], 20, 128);
- }
-
- // mv components
- for (i = 0; i < 2; i++) {
- uint8_t *pp;
- unsigned *c, (*c2)[2], sum;
-
- adapt_prob(&p->mv_comp[i].sign, s->counts.mv_comp[i].sign[0],
- s->counts.mv_comp[i].sign[1], 20, 128);
-
- pp = p->mv_comp[i].classes;
- c = s->counts.mv_comp[i].classes;
- sum = c[1] + c[2] + c[3] + c[4] + c[5] + c[6] + c[7] + c[8] + c[9] + c[10];
- adapt_prob(&pp[0], c[0], sum, 20, 128);
- sum -= c[1];
- adapt_prob(&pp[1], c[1], sum, 20, 128);
- sum -= c[2] + c[3];
- adapt_prob(&pp[2], c[2] + c[3], sum, 20, 128);
- adapt_prob(&pp[3], c[2], c[3], 20, 128);
- sum -= c[4] + c[5];
- adapt_prob(&pp[4], c[4] + c[5], sum, 20, 128);
- adapt_prob(&pp[5], c[4], c[5], 20, 128);
- sum -= c[6];
- adapt_prob(&pp[6], c[6], sum, 20, 128);
- adapt_prob(&pp[7], c[7] + c[8], c[9] + c[10], 20, 128);
- adapt_prob(&pp[8], c[7], c[8], 20, 128);
- adapt_prob(&pp[9], c[9], c[10], 20, 128);
-
- adapt_prob(&p->mv_comp[i].class0, s->counts.mv_comp[i].class0[0],
- s->counts.mv_comp[i].class0[1], 20, 128);
- pp = p->mv_comp[i].bits;
- c2 = s->counts.mv_comp[i].bits;
- for (j = 0; j < 10; j++)
- adapt_prob(&pp[j], c2[j][0], c2[j][1], 20, 128);
-
- for (j = 0; j < 2; j++) {
- pp = p->mv_comp[i].class0_fp[j];
- c = s->counts.mv_comp[i].class0_fp[j];
- adapt_prob(&pp[0], c[0], c[1] + c[2] + c[3], 20, 128);
- adapt_prob(&pp[1], c[1], c[2] + c[3], 20, 128);
- adapt_prob(&pp[2], c[2], c[3], 20, 128);
- }
- pp = p->mv_comp[i].fp;
- c = s->counts.mv_comp[i].fp;
- adapt_prob(&pp[0], c[0], c[1] + c[2] + c[3], 20, 128);
- adapt_prob(&pp[1], c[1], c[2] + c[3], 20, 128);
- adapt_prob(&pp[2], c[2], c[3], 20, 128);
-
- if (s->s.h.highprecisionmvs) {
- adapt_prob(&p->mv_comp[i].class0_hp, s->counts.mv_comp[i].class0_hp[0],
- s->counts.mv_comp[i].class0_hp[1], 20, 128);
- adapt_prob(&p->mv_comp[i].hp, s->counts.mv_comp[i].hp[0],
- s->counts.mv_comp[i].hp[1], 20, 128);
- }
- }
-
- // y intra modes
- for (i = 0; i < 4; i++) {
- uint8_t *pp = p->y_mode[i];
- unsigned *c = s->counts.y_mode[i], sum, s2;
-
- sum = c[0] + c[1] + c[3] + c[4] + c[5] + c[6] + c[7] + c[8] + c[9];
- adapt_prob(&pp[0], c[DC_PRED], sum, 20, 128);
- sum -= c[TM_VP8_PRED];
- adapt_prob(&pp[1], c[TM_VP8_PRED], sum, 20, 128);
- sum -= c[VERT_PRED];
- adapt_prob(&pp[2], c[VERT_PRED], sum, 20, 128);
- s2 = c[HOR_PRED] + c[DIAG_DOWN_RIGHT_PRED] + c[VERT_RIGHT_PRED];
- sum -= s2;
- adapt_prob(&pp[3], s2, sum, 20, 128);
- s2 -= c[HOR_PRED];
- adapt_prob(&pp[4], c[HOR_PRED], s2, 20, 128);
- adapt_prob(&pp[5], c[DIAG_DOWN_RIGHT_PRED], c[VERT_RIGHT_PRED], 20, 128);
- sum -= c[DIAG_DOWN_LEFT_PRED];
- adapt_prob(&pp[6], c[DIAG_DOWN_LEFT_PRED], sum, 20, 128);
- sum -= c[VERT_LEFT_PRED];
- adapt_prob(&pp[7], c[VERT_LEFT_PRED], sum, 20, 128);
- adapt_prob(&pp[8], c[HOR_DOWN_PRED], c[HOR_UP_PRED], 20, 128);
- }
-
- // uv intra modes
- for (i = 0; i < 10; i++) {
- uint8_t *pp = p->uv_mode[i];
- unsigned *c = s->counts.uv_mode[i], sum, s2;
-
- sum = c[0] + c[1] + c[3] + c[4] + c[5] + c[6] + c[7] + c[8] + c[9];
- adapt_prob(&pp[0], c[DC_PRED], sum, 20, 128);
- sum -= c[TM_VP8_PRED];
- adapt_prob(&pp[1], c[TM_VP8_PRED], sum, 20, 128);
- sum -= c[VERT_PRED];
- adapt_prob(&pp[2], c[VERT_PRED], sum, 20, 128);
- s2 = c[HOR_PRED] + c[DIAG_DOWN_RIGHT_PRED] + c[VERT_RIGHT_PRED];
- sum -= s2;
- adapt_prob(&pp[3], s2, sum, 20, 128);
- s2 -= c[HOR_PRED];
- adapt_prob(&pp[4], c[HOR_PRED], s2, 20, 128);
- adapt_prob(&pp[5], c[DIAG_DOWN_RIGHT_PRED], c[VERT_RIGHT_PRED], 20, 128);
- sum -= c[DIAG_DOWN_LEFT_PRED];
- adapt_prob(&pp[6], c[DIAG_DOWN_LEFT_PRED], sum, 20, 128);
- sum -= c[VERT_LEFT_PRED];
- adapt_prob(&pp[7], c[VERT_LEFT_PRED], sum, 20, 128);
- adapt_prob(&pp[8], c[HOR_DOWN_PRED], c[HOR_UP_PRED], 20, 128);
- }
-}
-
static void free_buffers(VP9Context *s)
{
av_freep(&s->intra_pred_data[0]);
}
if (s->pass < 2 && s->s.h.refreshctx && !s->s.h.parallelmode) {
- adapt_probs(s);
+ ff_vp9_adapt_probs(s);
ff_thread_finish_setup(ctx);
}
} while (s->pass++ == 1);
#ifndef AVCODEC_VP9_H
#define AVCODEC_VP9_H
+#include <stddef.h>
#include <stdint.h>
+#include "libavutil/buffer.h"
+#include "libavutil/internal.h"
+
+#include "avcodec.h"
#include "thread.h"
#include "vp56.h"
PRED_SWITCHABLE,
};
+enum MVJoint {
+ MV_JOINT_ZERO,
+ MV_JOINT_H,
+ MV_JOINT_V,
+ MV_JOINT_HV,
+};
+
+typedef struct ProbContext {
+ uint8_t y_mode[4][9];
+ uint8_t uv_mode[10][9];
+ uint8_t filter[4][2];
+ uint8_t mv_mode[7][3];
+ uint8_t intra[4];
+ uint8_t comp[5];
+ uint8_t single_ref[5][2];
+ uint8_t comp_ref[5];
+ uint8_t tx32p[2][3];
+ uint8_t tx16p[2][2];
+ uint8_t tx8p[2];
+ uint8_t skip[3];
+ uint8_t mv_joint[3];
+ struct {
+ uint8_t sign;
+ uint8_t classes[10];
+ uint8_t class0;
+ uint8_t bits[10];
+ uint8_t class0_fp[2][3];
+ uint8_t fp[3];
+ uint8_t class0_hp;
+ uint8_t hp;
+ } mv_comp[2];
+ uint8_t partition[4][4][3];
+} ProbContext;
+
+typedef void (*vp9_mc_func)(uint8_t *dst, ptrdiff_t dst_stride,
+ const uint8_t *ref, ptrdiff_t ref_stride,
+ int h, int mx, int my);
+typedef void (*vp9_scaled_mc_func)(uint8_t *dst, ptrdiff_t dst_stride,
+ const uint8_t *ref, ptrdiff_t ref_stride,
+ int h, int mx, int my, int dx, int dy);
+
+typedef struct VP9DSPContext {
+ /*
+ * dimension 1: 0=4x4, 1=8x8, 2=16x16, 3=32x32
+ * dimension 2: intra prediction modes
+ *
+ * dst/left/top is aligned by transform-size (i.e. 4, 8, 16 or 32 pixels)
+ * stride is aligned by 16 pixels
+ * top[-1] is top/left; top[4,7] is top-right for 4x4
+ */
+ // FIXME(rbultje) maybe replace left/top pointers with HAVE_TOP/
+ // HAVE_LEFT/HAVE_TOPRIGHT flags instead, and then handle it in-place?
+ // also needs to fit in with what H.264/VP8/etc do
+ void (*intra_pred[N_TXFM_SIZES][N_INTRA_PRED_MODES])(uint8_t *dst,
+ ptrdiff_t stride,
+ const uint8_t *left,
+ const uint8_t *top);
+
+ /*
+ * dimension 1: 0=4x4, 1=8x8, 2=16x16, 3=32x32, 4=lossless (3-4=dct only)
+ * dimension 2: 0=dct/dct, 1=dct/adst, 2=adst/dct, 3=adst/adst
+ *
+ * dst is aligned by transform-size (i.e. 4, 8, 16 or 32 pixels)
+ * stride is aligned by 16 pixels
+ * block is 16-byte aligned
+ * eob indicates the position (+1) of the last non-zero coefficient,
+ * in scan-order. This can be used to write faster versions, e.g. a
+ * dc-only 4x4/8x8/16x16/32x32, or a 4x4-only (eob<10) 8x8/16x16/32x32,
+ * etc.
+ */
+ // FIXME also write idct_add_block() versions for whole (inter) pred
+ // blocks, so we can do 2 4x4s at once
+ void (*itxfm_add[N_TXFM_SIZES + 1][N_TXFM_TYPES])(uint8_t *dst,
+ ptrdiff_t stride,
+ int16_t *block, int eob);
+
+ /*
+ * dimension 1: width of filter (0=4, 1=8, 2=16)
+ * dimension 2: 0=col-edge filter (h), 1=row-edge filter (v)
+ *
+ * dst/stride are aligned by 8
+ */
+ void (*loop_filter_8[3][2])(uint8_t *dst, ptrdiff_t stride,
+ int mb_lim, int lim, int hev_thr);
+
+ /*
+ * dimension 1: 0=col-edge filter (h), 1=row-edge filter (v)
+ *
+ * The width of filter is assumed to be 16; dst/stride are aligned by 16
+ */
+ void (*loop_filter_16[2])(uint8_t *dst, ptrdiff_t stride,
+ int mb_lim, int lim, int hev_thr);
+
+ /*
+ * dimension 1/2: width of filter (0=4, 1=8) for each filter half
+ * dimension 3: 0=col-edge filter (h), 1=row-edge filter (v)
+ *
+ * dst/stride are aligned by operation size
+ * this basically calls loop_filter[d1][d3][0](), followed by
+ * loop_filter[d2][d3][0]() on the next 8 pixels
+ * mb_lim/lim/hev_thr contain two values in the lowest two bytes of the
+ * integer.
+ */
+ // FIXME perhaps a mix4 that operates on 32px (for AVX2)
+ void (*loop_filter_mix2[2][2][2])(uint8_t *dst, ptrdiff_t stride,
+ int mb_lim, int lim, int hev_thr);
+
+ /*
+ * dimension 1: hsize (0: 64, 1: 32, 2: 16, 3: 8, 4: 4)
+ * dimension 2: filter type (0: smooth, 1: regular, 2: sharp, 3: bilin)
+ * dimension 3: averaging type (0: put, 1: avg)
+ * dimension 4: x subpel interpolation (0: none, 1: 8tap/bilin)
+ * dimension 5: y subpel interpolation (0: none, 1: 8tap/bilin)
+ *
+ * dst/stride are aligned by hsize
+ */
+ vp9_mc_func mc[5][4][2][2][2];
+
+ /*
+ * for scalable MC, first 3 dimensions identical to above, the other two
+ * don't exist since it changes per stepsize.
+ */
+ vp9_scaled_mc_func smc[5][4][2];
+} VP9DSPContext;
+
+
struct VP9mvrefPair {
VP56mv mv[2];
int8_t ref[2];
};
+struct VP9Filter {
+ uint8_t level[8 * 8];
+ uint8_t /* bit=col */ mask[2 /* 0=y, 1=uv */][2 /* 0=col, 1=row */]
+ [8 /* rows */][4 /* 0=16, 1=8, 2=4, 3=inner4 */];
+};
+
typedef struct VP9Frame {
ThreadFrame tf;
AVBufferRef *extradata;
VP9Frame frames[3];
} VP9SharedContext;
+typedef struct VP9Block {
+ uint8_t seg_id, intra, comp, ref[2], mode[4], uvmode, skip;
+ enum FilterMode filter;
+ VP56mv mv[4 /* b_idx */][2 /* ref */];
+ enum BlockSize bs;
+ enum TxfmMode tx, uvtx;
+ enum BlockLevel bl;
+ enum BlockPartition bp;
+} VP9Block;
+
+typedef struct VP9Context {
+ VP9SharedContext s;
+
+ VP9DSPContext dsp;
+ VideoDSPContext vdsp;
+ GetBitContext gb;
+ VP56RangeCoder c;
+ VP56RangeCoder *c_b;
+ unsigned c_b_size;
+ VP9Block *b_base, *b;
+ int pass;
+ int row, row7, col, col7;
+ uint8_t *dst[3];
+ ptrdiff_t y_stride, uv_stride;
+
+ uint8_t ss_h, ss_v;
+ uint8_t last_bpp, bpp_index, bytesperpixel;
+ uint8_t last_keyframe;
+ // sb_cols/rows, rows/cols and last_fmt are used for allocating all internal
+ // arrays, and are thus per-thread. w/h and gf_fmt are synced between threads
+ // and are therefore per-stream. pix_fmt represents the value in the header
+ // of the currently processed frame.
+ int w, h;
+ enum AVPixelFormat pix_fmt, last_fmt, gf_fmt;
+ unsigned sb_cols, sb_rows, rows, cols;
+ ThreadFrame next_refs[8];
+
+ struct {
+ uint8_t lim_lut[64];
+ uint8_t mblim_lut[64];
+ } filter_lut;
+ unsigned tile_row_start, tile_row_end, tile_col_start, tile_col_end;
+ struct {
+ ProbContext p;
+ uint8_t coef[4][2][2][6][6][3];
+ } prob_ctx[4];
+ struct {
+ ProbContext p;
+ uint8_t coef[4][2][2][6][6][11];
+ } prob;
+ struct {
+ unsigned y_mode[4][10];
+ unsigned uv_mode[10][10];
+ unsigned filter[4][3];
+ unsigned mv_mode[7][4];
+ unsigned intra[4][2];
+ unsigned comp[5][2];
+ unsigned single_ref[5][2][2];
+ unsigned comp_ref[5][2];
+ unsigned tx32p[2][4];
+ unsigned tx16p[2][3];
+ unsigned tx8p[2][2];
+ unsigned skip[3][2];
+ unsigned mv_joint[4];
+ struct {
+ unsigned sign[2];
+ unsigned classes[11];
+ unsigned class0[2];
+ unsigned bits[10][2];
+ unsigned class0_fp[2][4];
+ unsigned fp[4];
+ unsigned class0_hp[2];
+ unsigned hp[2];
+ } mv_comp[2];
+ unsigned partition[4][4][4];
+ unsigned coef[4][2][2][6][6][3];
+ unsigned eob[4][2][2][6][6][2];
+ } counts;
+
+ // contextual (left/above) cache
+ DECLARE_ALIGNED(16, uint8_t, left_y_nnz_ctx)[16];
+ DECLARE_ALIGNED(16, uint8_t, left_mode_ctx)[16];
+ DECLARE_ALIGNED(16, VP56mv, left_mv_ctx)[16][2];
+ DECLARE_ALIGNED(16, uint8_t, left_uv_nnz_ctx)[2][16];
+ DECLARE_ALIGNED(8, uint8_t, left_partition_ctx)[8];
+ DECLARE_ALIGNED(8, uint8_t, left_skip_ctx)[8];
+ DECLARE_ALIGNED(8, uint8_t, left_txfm_ctx)[8];
+ DECLARE_ALIGNED(8, uint8_t, left_segpred_ctx)[8];
+ DECLARE_ALIGNED(8, uint8_t, left_intra_ctx)[8];
+ DECLARE_ALIGNED(8, uint8_t, left_comp_ctx)[8];
+ DECLARE_ALIGNED(8, uint8_t, left_ref_ctx)[8];
+ DECLARE_ALIGNED(8, uint8_t, left_filter_ctx)[8];
+ uint8_t *above_partition_ctx;
+ uint8_t *above_mode_ctx;
+ // FIXME maybe merge some of the below in a flags field?
+ uint8_t *above_y_nnz_ctx;
+ uint8_t *above_uv_nnz_ctx[2];
+ uint8_t *above_skip_ctx; // 1bit
+ uint8_t *above_txfm_ctx; // 2bit
+ uint8_t *above_segpred_ctx; // 1bit
+ uint8_t *above_intra_ctx; // 1bit
+ uint8_t *above_comp_ctx; // 1bit
+ uint8_t *above_ref_ctx; // 2bit
+ uint8_t *above_filter_ctx;
+ VP56mv (*above_mv_ctx)[2];
+
+ // whole-frame cache
+ uint8_t *intra_pred_data[3];
+ struct VP9Filter *lflvl;
+ DECLARE_ALIGNED(32, uint8_t, edge_emu_buffer)[135 * 144 * 2];
+
+ // block reconstruction intermediates
+ int block_alloc_using_2pass;
+ int16_t *block_base, *block, *uvblock_base[2], *uvblock[2];
+ uint8_t *eob_base, *uveob_base[2], *eob, *uveob[2];
+ struct { int x, y; } min_mv, max_mv;
+ DECLARE_ALIGNED(32, uint8_t, tmp_y)[64 * 64 * 2];
+ DECLARE_ALIGNED(32, uint8_t, tmp_uv)[2][64 * 64 * 2];
+ uint16_t mvscale[3][2];
+ uint8_t mvstep[3][2];
+} VP9Context;
+
+extern const int16_t ff_vp9_subpel_filters[3][16][8];
+
+void ff_vp9dsp_init(VP9DSPContext *dsp, int bpp, int bitexact);
+
+void ff_vp9dsp_init_8(VP9DSPContext *dsp);
+void ff_vp9dsp_init_10(VP9DSPContext *dsp);
+void ff_vp9dsp_init_12(VP9DSPContext *dsp);
+
+void ff_vp9dsp_init_aarch64(VP9DSPContext *dsp, int bpp);
+void ff_vp9dsp_init_arm(VP9DSPContext *dsp, int bpp);
+void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp, int bitexact);
+void ff_vp9dsp_init_mips(VP9DSPContext *dsp, int bpp);
+
+void ff_vp9_fill_mv(VP9Context *s, VP56mv *mv, int mode, int sb);
+
+void ff_vp9_adapt_probs(VP9Context *s);
+
+void ff_vp9_decode_block(AVCodecContext *ctx, int row, int col,
+ struct VP9Filter *lflvl, ptrdiff_t yoff, ptrdiff_t uvoff,
+ enum BlockLevel bl, enum BlockPartition bp);
+
#endif /* AVCODEC_VP9_H */
--- /dev/null
+/*
+ * VP9 compatible video decoder
+ *
+ * Copyright (C) 2013 Ronald S. Bultje <rsbultje gmail com>
+ * Copyright (C) 2013 Clément Bœsch <u pkh me>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/avassert.h"
+
+#include "avcodec.h"
+#include "internal.h"
+#include "videodsp.h"
+#include "vp56.h"
+#include "vp9.h"
+#include "vp9data.h"
+
+static const uint8_t bwh_tab[2][N_BS_SIZES][2] = {
+ {
+ { 16, 16 }, { 16, 8 }, { 8, 16 }, { 8, 8 }, { 8, 4 }, { 4, 8 },
+ { 4, 4 }, { 4, 2 }, { 2, 4 }, { 2, 2 }, { 2, 1 }, { 1, 2 }, { 1, 1 },
+ }, {
+ { 8, 8 }, { 8, 4 }, { 4, 8 }, { 4, 4 }, { 4, 2 }, { 2, 4 },
+ { 2, 2 }, { 2, 1 }, { 1, 2 }, { 1, 1 }, { 1, 1 }, { 1, 1 }, { 1, 1 },
+ }
+};
+
+static av_always_inline void setctx_2d(uint8_t *ptr, int w, int h,
+ ptrdiff_t stride, int v)
+{
+ switch (w) {
+ case 1:
+ do {
+ *ptr = v;
+ ptr += stride;
+ } while (--h);
+ break;
+ case 2: {
+ int v16 = v * 0x0101;
+ do {
+ AV_WN16A(ptr, v16);
+ ptr += stride;
+ } while (--h);
+ break;
+ }
+ case 4: {
+ uint32_t v32 = v * 0x01010101;
+ do {
+ AV_WN32A(ptr, v32);
+ ptr += stride;
+ } while (--h);
+ break;
+ }
+ case 8: {
+#if HAVE_FAST_64BIT
+ uint64_t v64 = v * 0x0101010101010101ULL;
+ do {
+ AV_WN64A(ptr, v64);
+ ptr += stride;
+ } while (--h);
+#else
+ uint32_t v32 = v * 0x01010101;
+ do {
+ AV_WN32A(ptr, v32);
+ AV_WN32A(ptr + 4, v32);
+ ptr += stride;
+ } while (--h);
+#endif
+ break;
+ }
+ }
+}
+
+static void decode_mode(AVCodecContext *ctx)
+{
+ static const uint8_t left_ctx[N_BS_SIZES] = {
+ 0x0, 0x8, 0x0, 0x8, 0xc, 0x8, 0xc, 0xe, 0xc, 0xe, 0xf, 0xe, 0xf
+ };
+ static const uint8_t above_ctx[N_BS_SIZES] = {
+ 0x0, 0x0, 0x8, 0x8, 0x8, 0xc, 0xc, 0xc, 0xe, 0xe, 0xe, 0xf, 0xf
+ };
+ static const uint8_t max_tx_for_bl_bp[N_BS_SIZES] = {
+ TX_32X32, TX_32X32, TX_32X32, TX_32X32, TX_16X16, TX_16X16,
+ TX_16X16, TX_8X8, TX_8X8, TX_8X8, TX_4X4, TX_4X4, TX_4X4
+ };
+ VP9Context *s = ctx->priv_data;
+ VP9Block *b = s->b;
+ int row = s->row, col = s->col, row7 = s->row7;
+ enum TxfmMode max_tx = max_tx_for_bl_bp[b->bs];
+ int bw4 = bwh_tab[1][b->bs][0], w4 = FFMIN(s->cols - col, bw4);
+ int bh4 = bwh_tab[1][b->bs][1], h4 = FFMIN(s->rows - row, bh4), y;
+ int have_a = row > 0, have_l = col > s->tile_col_start;
+ int vref, filter_id;
+
+ if (!s->s.h.segmentation.enabled) {
+ b->seg_id = 0;
+ } else if (s->s.h.keyframe || s->s.h.intraonly) {
+ b->seg_id = !s->s.h.segmentation.update_map ? 0 :
+ vp8_rac_get_tree(&s->c, ff_vp9_segmentation_tree, s->s.h.segmentation.prob);
+ } else if (!s->s.h.segmentation.update_map ||
+ (s->s.h.segmentation.temporal &&
+ vp56_rac_get_prob_branchy(&s->c,
+ s->s.h.segmentation.pred_prob[s->above_segpred_ctx[col] +
+ s->left_segpred_ctx[row7]]))) {
+ if (!s->s.h.errorres && s->s.frames[REF_FRAME_SEGMAP].segmentation_map) {
+ int pred = 8, x;
+ uint8_t *refsegmap = s->s.frames[REF_FRAME_SEGMAP].segmentation_map;
+
+ if (!s->s.frames[REF_FRAME_SEGMAP].uses_2pass)
+ ff_thread_await_progress(&s->s.frames[REF_FRAME_SEGMAP].tf, row >> 3, 0);
+ for (y = 0; y < h4; y++) {
+ int idx_base = (y + row) * 8 * s->sb_cols + col;
+ for (x = 0; x < w4; x++)
+ pred = FFMIN(pred, refsegmap[idx_base + x]);
+ }
+ av_assert1(pred < 8);
+ b->seg_id = pred;
+ } else {
+ b->seg_id = 0;
+ }
+
+ memset(&s->above_segpred_ctx[col], 1, w4);
+ memset(&s->left_segpred_ctx[row7], 1, h4);
+ } else {
+ b->seg_id = vp8_rac_get_tree(&s->c, ff_vp9_segmentation_tree,
+ s->s.h.segmentation.prob);
+
+ memset(&s->above_segpred_ctx[col], 0, w4);
+ memset(&s->left_segpred_ctx[row7], 0, h4);
+ }
+ if (s->s.h.segmentation.enabled &&
+ (s->s.h.segmentation.update_map || s->s.h.keyframe || s->s.h.intraonly)) {
+ setctx_2d(&s->s.frames[CUR_FRAME].segmentation_map[row * 8 * s->sb_cols + col],
+ bw4, bh4, 8 * s->sb_cols, b->seg_id);
+ }
+
+ b->skip = s->s.h.segmentation.enabled &&
+ s->s.h.segmentation.feat[b->seg_id].skip_enabled;
+ if (!b->skip) {
+ int c = s->left_skip_ctx[row7] + s->above_skip_ctx[col];
+ b->skip = vp56_rac_get_prob(&s->c, s->prob.p.skip[c]);
+ s->counts.skip[c][b->skip]++;
+ }
+
+ if (s->s.h.keyframe || s->s.h.intraonly) {
+ b->intra = 1;
+ } else if (s->s.h.segmentation.enabled && s->s.h.segmentation.feat[b->seg_id].ref_enabled) {
+ b->intra = !s->s.h.segmentation.feat[b->seg_id].ref_val;
+ } else {
+ int c, bit;
+
+ if (have_a && have_l) {
+ c = s->above_intra_ctx[col] + s->left_intra_ctx[row7];
+ c += (c == 2);
+ } else {
+ c = have_a ? 2 * s->above_intra_ctx[col] :
+ have_l ? 2 * s->left_intra_ctx[row7] : 0;
+ }
+ bit = vp56_rac_get_prob(&s->c, s->prob.p.intra[c]);
+ s->counts.intra[c][bit]++;
+ b->intra = !bit;
+ }
+
+ if ((b->intra || !b->skip) && s->s.h.txfmmode == TX_SWITCHABLE) {
+ int c;
+ if (have_a) {
+ if (have_l) {
+ c = (s->above_skip_ctx[col] ? max_tx :
+ s->above_txfm_ctx[col]) +
+ (s->left_skip_ctx[row7] ? max_tx :
+ s->left_txfm_ctx[row7]) > max_tx;
+ } else {
+ c = s->above_skip_ctx[col] ? 1 :
+ (s->above_txfm_ctx[col] * 2 > max_tx);
+ }
+ } else if (have_l) {
+ c = s->left_skip_ctx[row7] ? 1 :
+ (s->left_txfm_ctx[row7] * 2 > max_tx);
+ } else {
+ c = 1;
+ }
+ switch (max_tx) {
+ case TX_32X32:
+ b->tx = vp56_rac_get_prob(&s->c, s->prob.p.tx32p[c][0]);
+ if (b->tx) {
+ b->tx += vp56_rac_get_prob(&s->c, s->prob.p.tx32p[c][1]);
+ if (b->tx == 2)
+ b->tx += vp56_rac_get_prob(&s->c, s->prob.p.tx32p[c][2]);
+ }
+ s->counts.tx32p[c][b->tx]++;
+ break;
+ case TX_16X16:
+ b->tx = vp56_rac_get_prob(&s->c, s->prob.p.tx16p[c][0]);
+ if (b->tx)
+ b->tx += vp56_rac_get_prob(&s->c, s->prob.p.tx16p[c][1]);
+ s->counts.tx16p[c][b->tx]++;
+ break;
+ case TX_8X8:
+ b->tx = vp56_rac_get_prob(&s->c, s->prob.p.tx8p[c]);
+ s->counts.tx8p[c][b->tx]++;
+ break;
+ case TX_4X4:
+ b->tx = TX_4X4;
+ break;
+ }
+ } else {
+ b->tx = FFMIN(max_tx, s->s.h.txfmmode);
+ }
+
+ if (s->s.h.keyframe || s->s.h.intraonly) {
+ uint8_t *a = &s->above_mode_ctx[col * 2];
+ uint8_t *l = &s->left_mode_ctx[(row7) << 1];
+
+ b->comp = 0;
+ if (b->bs > BS_8x8) {
+ // FIXME the memory storage intermediates here aren't really
+ // necessary, they're just there to make the code slightly
+ // simpler for now
+ b->mode[0] = a[0] = vp8_rac_get_tree(&s->c, ff_vp9_intramode_tree,
+ ff_vp9_default_kf_ymode_probs[a[0]][l[0]]);
+ if (b->bs != BS_8x4) {
+ b->mode[1] = vp8_rac_get_tree(&s->c, ff_vp9_intramode_tree,
+ ff_vp9_default_kf_ymode_probs[a[1]][b->mode[0]]);
+ l[0] = a[1] = b->mode[1];
+ } else {
+ l[0] = a[1] = b->mode[1] = b->mode[0];
+ }
+ if (b->bs != BS_4x8) {
+ b->mode[2] = a[0] = vp8_rac_get_tree(&s->c, ff_vp9_intramode_tree,
+ ff_vp9_default_kf_ymode_probs[a[0]][l[1]]);
+ if (b->bs != BS_8x4) {
+ b->mode[3] = vp8_rac_get_tree(&s->c, ff_vp9_intramode_tree,
+ ff_vp9_default_kf_ymode_probs[a[1]][b->mode[2]]);
+ l[1] = a[1] = b->mode[3];
+ } else {
+ l[1] = a[1] = b->mode[3] = b->mode[2];
+ }
+ } else {
+ b->mode[2] = b->mode[0];
+ l[1] = a[1] = b->mode[3] = b->mode[1];
+ }
+ } else {
+ b->mode[0] = vp8_rac_get_tree(&s->c, ff_vp9_intramode_tree,
+ ff_vp9_default_kf_ymode_probs[*a][*l]);
+ b->mode[3] = b->mode[2] = b->mode[1] = b->mode[0];
+ // FIXME this can probably be optimized
+ memset(a, b->mode[0], bwh_tab[0][b->bs][0]);
+ memset(l, b->mode[0], bwh_tab[0][b->bs][1]);
+ }
+ b->uvmode = vp8_rac_get_tree(&s->c, ff_vp9_intramode_tree,
+ ff_vp9_default_kf_uvmode_probs[b->mode[3]]);
+ } else if (b->intra) {
+ b->comp = 0;
+ if (b->bs > BS_8x8) {
+ b->mode[0] = vp8_rac_get_tree(&s->c, ff_vp9_intramode_tree,
+ s->prob.p.y_mode[0]);
+ s->counts.y_mode[0][b->mode[0]]++;
+ if (b->bs != BS_8x4) {
+ b->mode[1] = vp8_rac_get_tree(&s->c, ff_vp9_intramode_tree,
+ s->prob.p.y_mode[0]);
+ s->counts.y_mode[0][b->mode[1]]++;
+ } else {
+ b->mode[1] = b->mode[0];
+ }
+ if (b->bs != BS_4x8) {
+ b->mode[2] = vp8_rac_get_tree(&s->c, ff_vp9_intramode_tree,
+ s->prob.p.y_mode[0]);
+ s->counts.y_mode[0][b->mode[2]]++;
+ if (b->bs != BS_8x4) {
+ b->mode[3] = vp8_rac_get_tree(&s->c, ff_vp9_intramode_tree,
+ s->prob.p.y_mode[0]);
+ s->counts.y_mode[0][b->mode[3]]++;
+ } else {
+ b->mode[3] = b->mode[2];
+ }
+ } else {
+ b->mode[2] = b->mode[0];
+ b->mode[3] = b->mode[1];
+ }
+ } else {
+ static const uint8_t size_group[10] = {
+ 3, 3, 3, 3, 2, 2, 2, 1, 1, 1
+ };
+ int sz = size_group[b->bs];
+
+ b->mode[0] = vp8_rac_get_tree(&s->c, ff_vp9_intramode_tree,
+ s->prob.p.y_mode[sz]);
+ b->mode[1] = b->mode[2] = b->mode[3] = b->mode[0];
+ s->counts.y_mode[sz][b->mode[3]]++;
+ }
+ b->uvmode = vp8_rac_get_tree(&s->c, ff_vp9_intramode_tree,
+ s->prob.p.uv_mode[b->mode[3]]);
+ s->counts.uv_mode[b->mode[3]][b->uvmode]++;
+ } else {
+ static const uint8_t inter_mode_ctx_lut[14][14] = {
+ { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
+ { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
+ { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
+ { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
+ { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
+ { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
+ { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
+ { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
+ { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
+ { 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5 },
+ { 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 2, 2, 1, 3 },
+ { 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 2, 2, 1, 3 },
+ { 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 1, 1, 0, 3 },
+ { 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 3, 3, 3, 4 },
+ };
+
+ if (s->s.h.segmentation.enabled && s->s.h.segmentation.feat[b->seg_id].ref_enabled) {
+ av_assert2(s->s.h.segmentation.feat[b->seg_id].ref_val != 0);
+ b->comp = 0;
+ b->ref[0] = s->s.h.segmentation.feat[b->seg_id].ref_val - 1;
+ } else {
+ // read comp_pred flag
+ if (s->s.h.comppredmode != PRED_SWITCHABLE) {
+ b->comp = s->s.h.comppredmode == PRED_COMPREF;
+ } else {
+ int c;
+
+ // FIXME add intra as ref=0xff (or -1) to make these easier?
+ if (have_a) {
+ if (have_l) {
+ if (s->above_comp_ctx[col] && s->left_comp_ctx[row7]) {
+ c = 4;
+ } else if (s->above_comp_ctx[col]) {
+ c = 2 + (s->left_intra_ctx[row7] ||
+ s->left_ref_ctx[row7] == s->s.h.fixcompref);
+ } else if (s->left_comp_ctx[row7]) {
+ c = 2 + (s->above_intra_ctx[col] ||
+ s->above_ref_ctx[col] == s->s.h.fixcompref);
+ } else {
+ c = (!s->above_intra_ctx[col] &&
+ s->above_ref_ctx[col] == s->s.h.fixcompref) ^
+ (!s->left_intra_ctx[row7] &&
+ s->left_ref_ctx[row & 7] == s->s.h.fixcompref);
+ }
+ } else {
+ c = s->above_comp_ctx[col] ? 3 :
+ (!s->above_intra_ctx[col] && s->above_ref_ctx[col] == s->s.h.fixcompref);
+ }
+ } else if (have_l) {
+ c = s->left_comp_ctx[row7] ? 3 :
+ (!s->left_intra_ctx[row7] && s->left_ref_ctx[row7] == s->s.h.fixcompref);
+ } else {
+ c = 1;
+ }
+ b->comp = vp56_rac_get_prob(&s->c, s->prob.p.comp[c]);
+ s->counts.comp[c][b->comp]++;
+ }
+
+ // read actual references
+ // FIXME probably cache a few variables here to prevent repetitive
+ // memory accesses below
+ if (b->comp) /* two references */ {
+ int fix_idx = s->s.h.signbias[s->s.h.fixcompref], var_idx = !fix_idx, c, bit;
+
+ b->ref[fix_idx] = s->s.h.fixcompref;
+ // FIXME can this codeblob be replaced by some sort of LUT?
+ if (have_a) {
+ if (have_l) {
+ if (s->above_intra_ctx[col]) {
+ if (s->left_intra_ctx[row7]) {
+ c = 2;
+ } else {
+ c = 1 + 2 * (s->left_ref_ctx[row7] != s->s.h.varcompref[1]);
+ }
+ } else if (s->left_intra_ctx[row7]) {
+ c = 1 + 2 * (s->above_ref_ctx[col] != s->s.h.varcompref[1]);
+ } else {
+ int refl = s->left_ref_ctx[row7], refa = s->above_ref_ctx[col];
+
+ if (refl == refa && refa == s->s.h.varcompref[1]) {
+ c = 0;
+ } else if (!s->left_comp_ctx[row7] && !s->above_comp_ctx[col]) {
+ if ((refa == s->s.h.fixcompref && refl == s->s.h.varcompref[0]) ||
+ (refl == s->s.h.fixcompref && refa == s->s.h.varcompref[0])) {
+ c = 4;
+ } else {
+ c = (refa == refl) ? 3 : 1;
+ }
+ } else if (!s->left_comp_ctx[row7]) {
+ if (refa == s->s.h.varcompref[1] && refl != s->s.h.varcompref[1]) {
+ c = 1;
+ } else {
+ c = (refl == s->s.h.varcompref[1] &&
+ refa != s->s.h.varcompref[1]) ? 2 : 4;
+ }
+ } else if (!s->above_comp_ctx[col]) {
+ if (refl == s->s.h.varcompref[1] && refa != s->s.h.varcompref[1]) {
+ c = 1;
+ } else {
+ c = (refa == s->s.h.varcompref[1] &&
+ refl != s->s.h.varcompref[1]) ? 2 : 4;
+ }
+ } else {
+ c = (refl == refa) ? 4 : 2;
+ }
+ }
+ } else {
+ if (s->above_intra_ctx[col]) {
+ c = 2;
+ } else if (s->above_comp_ctx[col]) {
+ c = 4 * (s->above_ref_ctx[col] != s->s.h.varcompref[1]);
+ } else {
+ c = 3 * (s->above_ref_ctx[col] != s->s.h.varcompref[1]);
+ }
+ }
+ } else if (have_l) {
+ if (s->left_intra_ctx[row7]) {
+ c = 2;
+ } else if (s->left_comp_ctx[row7]) {
+ c = 4 * (s->left_ref_ctx[row7] != s->s.h.varcompref[1]);
+ } else {
+ c = 3 * (s->left_ref_ctx[row7] != s->s.h.varcompref[1]);
+ }
+ } else {
+ c = 2;
+ }
+ bit = vp56_rac_get_prob(&s->c, s->prob.p.comp_ref[c]);
+ b->ref[var_idx] = s->s.h.varcompref[bit];
+ s->counts.comp_ref[c][bit]++;
+ } else /* single reference */ {
+ int bit, c;
+
+ if (have_a && !s->above_intra_ctx[col]) {
+ if (have_l && !s->left_intra_ctx[row7]) {
+ if (s->left_comp_ctx[row7]) {
+ if (s->above_comp_ctx[col]) {
+ c = 1 + (!s->s.h.fixcompref || !s->left_ref_ctx[row7] ||
+ !s->above_ref_ctx[col]);
+ } else {
+ c = (3 * !s->above_ref_ctx[col]) +
+ (!s->s.h.fixcompref || !s->left_ref_ctx[row7]);
+ }
+ } else if (s->above_comp_ctx[col]) {
+ c = (3 * !s->left_ref_ctx[row7]) +
+ (!s->s.h.fixcompref || !s->above_ref_ctx[col]);
+ } else {
+ c = 2 * !s->left_ref_ctx[row7] + 2 * !s->above_ref_ctx[col];
+ }
+ } else if (s->above_intra_ctx[col]) {
+ c = 2;
+ } else if (s->above_comp_ctx[col]) {
+ c = 1 + (!s->s.h.fixcompref || !s->above_ref_ctx[col]);
+ } else {
+ c = 4 * (!s->above_ref_ctx[col]);
+ }
+ } else if (have_l && !s->left_intra_ctx[row7]) {
+ if (s->left_intra_ctx[row7]) {
+ c = 2;
+ } else if (s->left_comp_ctx[row7]) {
+ c = 1 + (!s->s.h.fixcompref || !s->left_ref_ctx[row7]);
+ } else {
+ c = 4 * (!s->left_ref_ctx[row7]);
+ }
+ } else {
+ c = 2;
+ }
+ bit = vp56_rac_get_prob(&s->c, s->prob.p.single_ref[c][0]);
+ s->counts.single_ref[c][0][bit]++;
+ if (!bit) {
+ b->ref[0] = 0;
+ } else {
+ // FIXME can this codeblob be replaced by some sort of LUT?
+ if (have_a) {
+ if (have_l) {
+ if (s->left_intra_ctx[row7]) {
+ if (s->above_intra_ctx[col]) {
+ c = 2;
+ } else if (s->above_comp_ctx[col]) {
+ c = 1 + 2 * (s->s.h.fixcompref == 1 ||
+ s->above_ref_ctx[col] == 1);
+ } else if (!s->above_ref_ctx[col]) {
+ c = 3;
+ } else {
+ c = 4 * (s->above_ref_ctx[col] == 1);
+ }
+ } else if (s->above_intra_ctx[col]) {
+ if (s->left_intra_ctx[row7]) {
+ c = 2;
+ } else if (s->left_comp_ctx[row7]) {
+ c = 1 + 2 * (s->s.h.fixcompref == 1 ||
+ s->left_ref_ctx[row7] == 1);
+ } else if (!s->left_ref_ctx[row7]) {
+ c = 3;
+ } else {
+ c = 4 * (s->left_ref_ctx[row7] == 1);
+ }
+ } else if (s->above_comp_ctx[col]) {
+ if (s->left_comp_ctx[row7]) {
+ if (s->left_ref_ctx[row7] == s->above_ref_ctx[col]) {
+ c = 3 * (s->s.h.fixcompref == 1 ||
+ s->left_ref_ctx[row7] == 1);
+ } else {
+ c = 2;
+ }
+ } else if (!s->left_ref_ctx[row7]) {
+ c = 1 + 2 * (s->s.h.fixcompref == 1 ||
+ s->above_ref_ctx[col] == 1);
+ } else {
+ c = 3 * (s->left_ref_ctx[row7] == 1) +
+ (s->s.h.fixcompref == 1 || s->above_ref_ctx[col] == 1);
+ }
+ } else if (s->left_comp_ctx[row7]) {
+ if (!s->above_ref_ctx[col]) {
+ c = 1 + 2 * (s->s.h.fixcompref == 1 ||
+ s->left_ref_ctx[row7] == 1);
+ } else {
+ c = 3 * (s->above_ref_ctx[col] == 1) +
+ (s->s.h.fixcompref == 1 || s->left_ref_ctx[row7] == 1);
+ }
+ } else if (!s->above_ref_ctx[col]) {
+ if (!s->left_ref_ctx[row7]) {
+ c = 3;
+ } else {
+ c = 4 * (s->left_ref_ctx[row7] == 1);
+ }
+ } else if (!s->left_ref_ctx[row7]) {
+ c = 4 * (s->above_ref_ctx[col] == 1);
+ } else {
+ c = 2 * (s->left_ref_ctx[row7] == 1) +
+ 2 * (s->above_ref_ctx[col] == 1);
+ }
+ } else {
+ if (s->above_intra_ctx[col] ||
+ (!s->above_comp_ctx[col] && !s->above_ref_ctx[col])) {
+ c = 2;
+ } else if (s->above_comp_ctx[col]) {
+ c = 3 * (s->s.h.fixcompref == 1 || s->above_ref_ctx[col] == 1);
+ } else {
+ c = 4 * (s->above_ref_ctx[col] == 1);
+ }
+ }
+ } else if (have_l) {
+ if (s->left_intra_ctx[row7] ||
+ (!s->left_comp_ctx[row7] && !s->left_ref_ctx[row7])) {
+ c = 2;
+ } else if (s->left_comp_ctx[row7]) {
+ c = 3 * (s->s.h.fixcompref == 1 || s->left_ref_ctx[row7] == 1);
+ } else {
+ c = 4 * (s->left_ref_ctx[row7] == 1);
+ }
+ } else {
+ c = 2;
+ }
+ bit = vp56_rac_get_prob(&s->c, s->prob.p.single_ref[c][1]);
+ s->counts.single_ref[c][1][bit]++;
+ b->ref[0] = 1 + bit;
+ }
+ }
+ }
+
+ if (b->bs <= BS_8x8) {
+ if (s->s.h.segmentation.enabled && s->s.h.segmentation.feat[b->seg_id].skip_enabled) {
+ b->mode[0] = b->mode[1] = b->mode[2] = b->mode[3] = ZEROMV;
+ } else {
+ static const uint8_t off[10] = {
+ 3, 0, 0, 1, 0, 0, 0, 0, 0, 0
+ };
+
+ // FIXME this needs to use the LUT tables from find_ref_mvs
+ // because not all are -1,0/0,-1
+ int c = inter_mode_ctx_lut[s->above_mode_ctx[col + off[b->bs]]]
+ [s->left_mode_ctx[row7 + off[b->bs]]];
+
+ b->mode[0] = vp8_rac_get_tree(&s->c, ff_vp9_inter_mode_tree,
+ s->prob.p.mv_mode[c]);
+ b->mode[1] = b->mode[2] = b->mode[3] = b->mode[0];
+ s->counts.mv_mode[c][b->mode[0] - 10]++;
+ }
+ }
+
+ if (s->s.h.filtermode == FILTER_SWITCHABLE) {
+ int c;
+
+ if (have_a && s->above_mode_ctx[col] >= NEARESTMV) {
+ if (have_l && s->left_mode_ctx[row7] >= NEARESTMV) {
+ c = s->above_filter_ctx[col] == s->left_filter_ctx[row7] ?
+ s->left_filter_ctx[row7] : 3;
+ } else {
+ c = s->above_filter_ctx[col];
+ }
+ } else if (have_l && s->left_mode_ctx[row7] >= NEARESTMV) {
+ c = s->left_filter_ctx[row7];
+ } else {
+ c = 3;
+ }
+
+ filter_id = vp8_rac_get_tree(&s->c, ff_vp9_filter_tree,
+ s->prob.p.filter[c]);
+ s->counts.filter[c][filter_id]++;
+ b->filter = ff_vp9_filter_lut[filter_id];
+ } else {
+ b->filter = s->s.h.filtermode;
+ }
+
+ if (b->bs > BS_8x8) {
+ int c = inter_mode_ctx_lut[s->above_mode_ctx[col]][s->left_mode_ctx[row7]];
+
+ b->mode[0] = vp8_rac_get_tree(&s->c, ff_vp9_inter_mode_tree,
+ s->prob.p.mv_mode[c]);
+ s->counts.mv_mode[c][b->mode[0] - 10]++;
+ ff_vp9_fill_mv(s, b->mv[0], b->mode[0], 0);
+
+ if (b->bs != BS_8x4) {
+ b->mode[1] = vp8_rac_get_tree(&s->c, ff_vp9_inter_mode_tree,
+ s->prob.p.mv_mode[c]);
+ s->counts.mv_mode[c][b->mode[1] - 10]++;
+ ff_vp9_fill_mv(s, b->mv[1], b->mode[1], 1);
+ } else {
+ b->mode[1] = b->mode[0];
+ AV_COPY32(&b->mv[1][0], &b->mv[0][0]);
+ AV_COPY32(&b->mv[1][1], &b->mv[0][1]);
+ }
+
+ if (b->bs != BS_4x8) {
+ b->mode[2] = vp8_rac_get_tree(&s->c, ff_vp9_inter_mode_tree,
+ s->prob.p.mv_mode[c]);
+ s->counts.mv_mode[c][b->mode[2] - 10]++;
+ ff_vp9_fill_mv(s, b->mv[2], b->mode[2], 2);
+
+ if (b->bs != BS_8x4) {
+ b->mode[3] = vp8_rac_get_tree(&s->c, ff_vp9_inter_mode_tree,
+ s->prob.p.mv_mode[c]);
+ s->counts.mv_mode[c][b->mode[3] - 10]++;
+ ff_vp9_fill_mv(s, b->mv[3], b->mode[3], 3);
+ } else {
+ b->mode[3] = b->mode[2];
+ AV_COPY32(&b->mv[3][0], &b->mv[2][0]);
+ AV_COPY32(&b->mv[3][1], &b->mv[2][1]);
+ }
+ } else {
+ b->mode[2] = b->mode[0];
+ AV_COPY32(&b->mv[2][0], &b->mv[0][0]);
+ AV_COPY32(&b->mv[2][1], &b->mv[0][1]);
+ b->mode[3] = b->mode[1];
+ AV_COPY32(&b->mv[3][0], &b->mv[1][0]);
+ AV_COPY32(&b->mv[3][1], &b->mv[1][1]);
+ }
+ } else {
+ ff_vp9_fill_mv(s, b->mv[0], b->mode[0], -1);
+ AV_COPY32(&b->mv[1][0], &b->mv[0][0]);
+ AV_COPY32(&b->mv[2][0], &b->mv[0][0]);
+ AV_COPY32(&b->mv[3][0], &b->mv[0][0]);
+ AV_COPY32(&b->mv[1][1], &b->mv[0][1]);
+ AV_COPY32(&b->mv[2][1], &b->mv[0][1]);
+ AV_COPY32(&b->mv[3][1], &b->mv[0][1]);
+ }
+
+ vref = b->ref[b->comp ? s->s.h.signbias[s->s.h.varcompref[0]] : 0];
+ }
+
+#if HAVE_FAST_64BIT
+#define SPLAT_CTX(var, val, n) \
+ switch (n) { \
+ case 1: var = val; break; \
+ case 2: AV_WN16A(&var, val * 0x0101); break; \
+ case 4: AV_WN32A(&var, val * 0x01010101); break; \
+ case 8: AV_WN64A(&var, val * 0x0101010101010101ULL); break; \
+ case 16: { \
+ uint64_t v64 = val * 0x0101010101010101ULL; \
+ AV_WN64A( &var, v64); \
+ AV_WN64A(&((uint8_t *) &var)[8], v64); \
+ break; \
+ } \
+ }
+#else
+#define SPLAT_CTX(var, val, n) \
+ switch (n) { \
+ case 1: var = val; break; \
+ case 2: AV_WN16A(&var, val * 0x0101); break; \
+ case 4: AV_WN32A(&var, val * 0x01010101); break; \
+ case 8: { \
+ uint32_t v32 = val * 0x01010101; \
+ AV_WN32A( &var, v32); \
+ AV_WN32A(&((uint8_t *) &var)[4], v32); \
+ break; \
+ } \
+ case 16: { \
+ uint32_t v32 = val * 0x01010101; \
+ AV_WN32A( &var, v32); \
+ AV_WN32A(&((uint8_t *) &var)[4], v32); \
+ AV_WN32A(&((uint8_t *) &var)[8], v32); \
+ AV_WN32A(&((uint8_t *) &var)[12], v32); \
+ break; \
+ } \
+ }
+#endif
+
+ switch (bwh_tab[1][b->bs][0]) {
+#define SET_CTXS(dir, off, n) \
+ do { \
+ SPLAT_CTX(s->dir##_skip_ctx[off], b->skip, n); \
+ SPLAT_CTX(s->dir##_txfm_ctx[off], b->tx, n); \
+ SPLAT_CTX(s->dir##_partition_ctx[off], dir##_ctx[b->bs], n); \
+ if (!s->s.h.keyframe && !s->s.h.intraonly) { \
+ SPLAT_CTX(s->dir##_intra_ctx[off], b->intra, n); \
+ SPLAT_CTX(s->dir##_comp_ctx[off], b->comp, n); \
+ SPLAT_CTX(s->dir##_mode_ctx[off], b->mode[3], n); \
+ if (!b->intra) { \
+ SPLAT_CTX(s->dir##_ref_ctx[off], vref, n); \
+ if (s->s.h.filtermode == FILTER_SWITCHABLE) { \
+ SPLAT_CTX(s->dir##_filter_ctx[off], filter_id, n); \
+ } \
+ } \
+ } \
+ } while (0)
+ case 1: SET_CTXS(above, col, 1); break;
+ case 2: SET_CTXS(above, col, 2); break;
+ case 4: SET_CTXS(above, col, 4); break;
+ case 8: SET_CTXS(above, col, 8); break;
+ }
+ switch (bwh_tab[1][b->bs][1]) {
+ case 1: SET_CTXS(left, row7, 1); break;
+ case 2: SET_CTXS(left, row7, 2); break;
+ case 4: SET_CTXS(left, row7, 4); break;
+ case 8: SET_CTXS(left, row7, 8); break;
+ }
+#undef SPLAT_CTX
+#undef SET_CTXS
+
+ if (!s->s.h.keyframe && !s->s.h.intraonly) {
+ if (b->bs > BS_8x8) {
+ int mv0 = AV_RN32A(&b->mv[3][0]), mv1 = AV_RN32A(&b->mv[3][1]);
+
+ AV_COPY32(&s->left_mv_ctx[row7 * 2 + 0][0], &b->mv[1][0]);
+ AV_COPY32(&s->left_mv_ctx[row7 * 2 + 0][1], &b->mv[1][1]);
+ AV_WN32A(&s->left_mv_ctx[row7 * 2 + 1][0], mv0);
+ AV_WN32A(&s->left_mv_ctx[row7 * 2 + 1][1], mv1);
+ AV_COPY32(&s->above_mv_ctx[col * 2 + 0][0], &b->mv[2][0]);
+ AV_COPY32(&s->above_mv_ctx[col * 2 + 0][1], &b->mv[2][1]);
+ AV_WN32A(&s->above_mv_ctx[col * 2 + 1][0], mv0);
+ AV_WN32A(&s->above_mv_ctx[col * 2 + 1][1], mv1);
+ } else {
+ int n, mv0 = AV_RN32A(&b->mv[3][0]), mv1 = AV_RN32A(&b->mv[3][1]);
+
+ for (n = 0; n < w4 * 2; n++) {
+ AV_WN32A(&s->above_mv_ctx[col * 2 + n][0], mv0);
+ AV_WN32A(&s->above_mv_ctx[col * 2 + n][1], mv1);
+ }
+ for (n = 0; n < h4 * 2; n++) {
+ AV_WN32A(&s->left_mv_ctx[row7 * 2 + n][0], mv0);
+ AV_WN32A(&s->left_mv_ctx[row7 * 2 + n][1], mv1);
+ }
+ }
+ }
+
+ // FIXME kinda ugly
+ for (y = 0; y < h4; y++) {
+ int x, o = (row + y) * s->sb_cols * 8 + col;
+ struct VP9mvrefPair *mv = &s->s.frames[CUR_FRAME].mv[o];
+
+ if (b->intra) {
+ for (x = 0; x < w4; x++) {
+ mv[x].ref[0] =
+ mv[x].ref[1] = -1;
+ }
+ } else if (b->comp) {
+ for (x = 0; x < w4; x++) {
+ mv[x].ref[0] = b->ref[0];
+ mv[x].ref[1] = b->ref[1];
+ AV_COPY32(&mv[x].mv[0], &b->mv[3][0]);
+ AV_COPY32(&mv[x].mv[1], &b->mv[3][1]);
+ }
+ } else {
+ for (x = 0; x < w4; x++) {
+ mv[x].ref[0] = b->ref[0];
+ mv[x].ref[1] = -1;
+ AV_COPY32(&mv[x].mv[0], &b->mv[3][0]);
+ }
+ }
+ }
+}
+
+// FIXME merge cnt/eob arguments?
+static av_always_inline int
+decode_coeffs_b_generic(VP56RangeCoder *c, int16_t *coef, int n_coeffs,
+ int is_tx32x32, int is8bitsperpixel, int bpp, unsigned (*cnt)[6][3],
+ unsigned (*eob)[6][2], uint8_t (*p)[6][11],
+ int nnz, const int16_t *scan, const int16_t (*nb)[2],
+ const int16_t *band_counts, const int16_t *qmul)
+{
+ int i = 0, band = 0, band_left = band_counts[band];
+ uint8_t *tp = p[0][nnz];
+ uint8_t cache[1024];
+
+ do {
+ int val, rc;
+
+ val = vp56_rac_get_prob_branchy(c, tp[0]); // eob
+ eob[band][nnz][val]++;
+ if (!val)
+ break;
+
+ skip_eob:
+ if (!vp56_rac_get_prob_branchy(c, tp[1])) { // zero
+ cnt[band][nnz][0]++;
+ if (!--band_left)
+ band_left = band_counts[++band];
+ cache[scan[i]] = 0;
+ nnz = (1 + cache[nb[i][0]] + cache[nb[i][1]]) >> 1;
+ tp = p[band][nnz];
+ if (++i == n_coeffs)
+ break; //invalid input; blocks should end with EOB
+ goto skip_eob;
+ }
+
+ rc = scan[i];
+ if (!vp56_rac_get_prob_branchy(c, tp[2])) { // one
+ cnt[band][nnz][1]++;
+ val = 1;
+ cache[rc] = 1;
+ } else {
+ // fill in p[3-10] (model fill) - only once per frame for each pos
+ if (!tp[3])
+ memcpy(&tp[3], ff_vp9_model_pareto8[tp[2]], 8);
+
+ cnt[band][nnz][2]++;
+ if (!vp56_rac_get_prob_branchy(c, tp[3])) { // 2, 3, 4
+ if (!vp56_rac_get_prob_branchy(c, tp[4])) {
+ cache[rc] = val = 2;
+ } else {
+ val = 3 + vp56_rac_get_prob(c, tp[5]);
+ cache[rc] = 3;
+ }
+ } else if (!vp56_rac_get_prob_branchy(c, tp[6])) { // cat1/2
+ cache[rc] = 4;
+ if (!vp56_rac_get_prob_branchy(c, tp[7])) {
+ val = 5 + vp56_rac_get_prob(c, 159);
+ } else {
+ val = 7 + (vp56_rac_get_prob(c, 165) << 1);
+ val += vp56_rac_get_prob(c, 145);
+ }
+ } else { // cat 3-6
+ cache[rc] = 5;
+ if (!vp56_rac_get_prob_branchy(c, tp[8])) {
+ if (!vp56_rac_get_prob_branchy(c, tp[9])) {
+ val = 11 + (vp56_rac_get_prob(c, 173) << 2);
+ val += (vp56_rac_get_prob(c, 148) << 1);
+ val += vp56_rac_get_prob(c, 140);
+ } else {
+ val = 19 + (vp56_rac_get_prob(c, 176) << 3);
+ val += (vp56_rac_get_prob(c, 155) << 2);
+ val += (vp56_rac_get_prob(c, 140) << 1);
+ val += vp56_rac_get_prob(c, 135);
+ }
+ } else if (!vp56_rac_get_prob_branchy(c, tp[10])) {
+ val = 35 + (vp56_rac_get_prob(c, 180) << 4);
+ val += (vp56_rac_get_prob(c, 157) << 3);
+ val += (vp56_rac_get_prob(c, 141) << 2);
+ val += (vp56_rac_get_prob(c, 134) << 1);
+ val += vp56_rac_get_prob(c, 130);
+ } else {
+ val = 67;
+ if (!is8bitsperpixel) {
+ if (bpp == 12) {
+ val += vp56_rac_get_prob(c, 255) << 17;
+ val += vp56_rac_get_prob(c, 255) << 16;
+ }
+ val += (vp56_rac_get_prob(c, 255) << 15);
+ val += (vp56_rac_get_prob(c, 255) << 14);
+ }
+ val += (vp56_rac_get_prob(c, 254) << 13);
+ val += (vp56_rac_get_prob(c, 254) << 12);
+ val += (vp56_rac_get_prob(c, 254) << 11);
+ val += (vp56_rac_get_prob(c, 252) << 10);
+ val += (vp56_rac_get_prob(c, 249) << 9);
+ val += (vp56_rac_get_prob(c, 243) << 8);
+ val += (vp56_rac_get_prob(c, 230) << 7);
+ val += (vp56_rac_get_prob(c, 196) << 6);
+ val += (vp56_rac_get_prob(c, 177) << 5);
+ val += (vp56_rac_get_prob(c, 153) << 4);
+ val += (vp56_rac_get_prob(c, 140) << 3);
+ val += (vp56_rac_get_prob(c, 133) << 2);
+ val += (vp56_rac_get_prob(c, 130) << 1);
+ val += vp56_rac_get_prob(c, 129);
+ }
+ }
+ }
+#define STORE_COEF(c, i, v) do { \
+ if (is8bitsperpixel) { \
+ c[i] = v; \
+ } else { \
+ AV_WN32A(&c[i * 2], v); \
+ } \
+} while (0)
+ if (!--band_left)
+ band_left = band_counts[++band];
+ if (is_tx32x32)
+ STORE_COEF(coef, rc, ((vp8_rac_get(c) ? -val : val) * qmul[!!i]) / 2);
+ else
+ STORE_COEF(coef, rc, (vp8_rac_get(c) ? -val : val) * qmul[!!i]);
+ nnz = (1 + cache[nb[i][0]] + cache[nb[i][1]]) >> 1;
+ tp = p[band][nnz];
+ } while (++i < n_coeffs);
+
+ return i;
+}
+
+static int decode_coeffs_b_8bpp(VP9Context *s, int16_t *coef, int n_coeffs,
+ unsigned (*cnt)[6][3], unsigned (*eob)[6][2],
+ uint8_t (*p)[6][11], int nnz, const int16_t *scan,
+ const int16_t (*nb)[2], const int16_t *band_counts,
+ const int16_t *qmul)
+{
+ return decode_coeffs_b_generic(&s->c, coef, n_coeffs, 0, 1, 8, cnt, eob, p,
+ nnz, scan, nb, band_counts, qmul);
+}
+
+static int decode_coeffs_b32_8bpp(VP9Context *s, int16_t *coef, int n_coeffs,
+ unsigned (*cnt)[6][3], unsigned (*eob)[6][2],
+ uint8_t (*p)[6][11], int nnz, const int16_t *scan,
+ const int16_t (*nb)[2], const int16_t *band_counts,
+ const int16_t *qmul)
+{
+ return decode_coeffs_b_generic(&s->c, coef, n_coeffs, 1, 1, 8, cnt, eob, p,
+ nnz, scan, nb, band_counts, qmul);
+}
+
+static int decode_coeffs_b_16bpp(VP9Context *s, int16_t *coef, int n_coeffs,
+ unsigned (*cnt)[6][3], unsigned (*eob)[6][2],
+ uint8_t (*p)[6][11], int nnz, const int16_t *scan,
+ const int16_t (*nb)[2], const int16_t *band_counts,
+ const int16_t *qmul)
+{
+ return decode_coeffs_b_generic(&s->c, coef, n_coeffs, 0, 0, s->s.h.bpp, cnt, eob, p,
+ nnz, scan, nb, band_counts, qmul);
+}
+
+static int decode_coeffs_b32_16bpp(VP9Context *s, int16_t *coef, int n_coeffs,
+ unsigned (*cnt)[6][3], unsigned (*eob)[6][2],
+ uint8_t (*p)[6][11], int nnz, const int16_t *scan,
+ const int16_t (*nb)[2], const int16_t *band_counts,
+ const int16_t *qmul)
+{
+ return decode_coeffs_b_generic(&s->c, coef, n_coeffs, 1, 0, s->s.h.bpp, cnt, eob, p,
+ nnz, scan, nb, band_counts, qmul);
+}
+
+static av_always_inline int decode_coeffs(AVCodecContext *ctx, int is8bitsperpixel)
+{
+ VP9Context *s = ctx->priv_data;
+ VP9Block *b = s->b;
+ int row = s->row, col = s->col;
+ uint8_t (*p)[6][11] = s->prob.coef[b->tx][0 /* y */][!b->intra];
+ unsigned (*c)[6][3] = s->counts.coef[b->tx][0 /* y */][!b->intra];
+ unsigned (*e)[6][2] = s->counts.eob[b->tx][0 /* y */][!b->intra];
+ int w4 = bwh_tab[1][b->bs][0] << 1, h4 = bwh_tab[1][b->bs][1] << 1;
+ int end_x = FFMIN(2 * (s->cols - col), w4);
+ int end_y = FFMIN(2 * (s->rows - row), h4);
+ int n, pl, x, y, res;
+ int16_t (*qmul)[2] = s->s.h.segmentation.feat[b->seg_id].qmul;
+ int tx = 4 * s->s.h.lossless + b->tx;
+ const int16_t * const *yscans = ff_vp9_scans[tx];
+ const int16_t (* const *ynbs)[2] = ff_vp9_scans_nb[tx];
+ const int16_t *uvscan = ff_vp9_scans[b->uvtx][DCT_DCT];
+ const int16_t (*uvnb)[2] = ff_vp9_scans_nb[b->uvtx][DCT_DCT];
+ uint8_t *a = &s->above_y_nnz_ctx[col * 2];
+ uint8_t *l = &s->left_y_nnz_ctx[(row & 7) << 1];
+ static const int16_t band_counts[4][8] = {
+ { 1, 2, 3, 4, 3, 16 - 13 },
+ { 1, 2, 3, 4, 11, 64 - 21 },
+ { 1, 2, 3, 4, 11, 256 - 21 },
+ { 1, 2, 3, 4, 11, 1024 - 21 },
+ };
+ const int16_t *y_band_counts = band_counts[b->tx];
+ const int16_t *uv_band_counts = band_counts[b->uvtx];
+ int bytesperpixel = is8bitsperpixel ? 1 : 2;
+ int total_coeff = 0;
+
+#define MERGE(la, end, step, rd) \
+ for (n = 0; n < end; n += step) \
+ la[n] = !!rd(&la[n])
+#define MERGE_CTX(step, rd) \
+ do { \
+ MERGE(l, end_y, step, rd); \
+ MERGE(a, end_x, step, rd); \
+ } while (0)
+
+#define DECODE_Y_COEF_LOOP(step, mode_index, v) \
+ for (n = 0, y = 0; y < end_y; y += step) { \
+ for (x = 0; x < end_x; x += step, n += step * step) { \
+ enum TxfmType txtp = ff_vp9_intra_txfm_type[b->mode[mode_index]]; \
+ res = (is8bitsperpixel ? decode_coeffs_b##v##_8bpp : decode_coeffs_b##v##_16bpp) \
+ (s, s->block + 16 * n * bytesperpixel, 16 * step * step, \
+ c, e, p, a[x] + l[y], yscans[txtp], \
+ ynbs[txtp], y_band_counts, qmul[0]); \
+ a[x] = l[y] = !!res; \
+ total_coeff |= !!res; \
+ if (step >= 4) { \
+ AV_WN16A(&s->eob[n], res); \
+ } else { \
+ s->eob[n] = res; \
+ } \
+ } \
+ }
+
+#define SPLAT(la, end, step, cond) \
+ if (step == 2) { \
+ for (n = 1; n < end; n += step) \
+ la[n] = la[n - 1]; \
+ } else if (step == 4) { \
+ if (cond) { \
+ for (n = 0; n < end; n += step) \
+ AV_WN32A(&la[n], la[n] * 0x01010101); \
+ } else { \
+ for (n = 0; n < end; n += step) \
+ memset(&la[n + 1], la[n], FFMIN(end - n - 1, 3)); \
+ } \
+ } else /* step == 8 */ { \
+ if (cond) { \
+ if (HAVE_FAST_64BIT) { \
+ for (n = 0; n < end; n += step) \
+ AV_WN64A(&la[n], la[n] * 0x0101010101010101ULL); \
+ } else { \
+ for (n = 0; n < end; n += step) { \
+ uint32_t v32 = la[n] * 0x01010101; \
+ AV_WN32A(&la[n], v32); \
+ AV_WN32A(&la[n + 4], v32); \
+ } \
+ } \
+ } else { \
+ for (n = 0; n < end; n += step) \
+ memset(&la[n + 1], la[n], FFMIN(end - n - 1, 7)); \
+ } \
+ }
+#define SPLAT_CTX(step) \
+ do { \
+ SPLAT(a, end_x, step, end_x == w4); \
+ SPLAT(l, end_y, step, end_y == h4); \
+ } while (0)
+
+ /* y tokens */
+ switch (b->tx) {
+ case TX_4X4:
+ DECODE_Y_COEF_LOOP(1, b->bs > BS_8x8 ? n : 0,);
+ break;
+ case TX_8X8:
+ MERGE_CTX(2, AV_RN16A);
+ DECODE_Y_COEF_LOOP(2, 0,);
+ SPLAT_CTX(2);
+ break;
+ case TX_16X16:
+ MERGE_CTX(4, AV_RN32A);
+ DECODE_Y_COEF_LOOP(4, 0,);
+ SPLAT_CTX(4);
+ break;
+ case TX_32X32:
+ MERGE_CTX(8, AV_RN64A);
+ DECODE_Y_COEF_LOOP(8, 0, 32);
+ SPLAT_CTX(8);
+ break;
+ }
+
+#define DECODE_UV_COEF_LOOP(step, v) \
+ for (n = 0, y = 0; y < end_y; y += step) { \
+ for (x = 0; x < end_x; x += step, n += step * step) { \
+ res = (is8bitsperpixel ? decode_coeffs_b##v##_8bpp : decode_coeffs_b##v##_16bpp) \
+ (s, s->uvblock[pl] + 16 * n * bytesperpixel, \
+ 16 * step * step, c, e, p, a[x] + l[y], \
+ uvscan, uvnb, uv_band_counts, qmul[1]); \
+ a[x] = l[y] = !!res; \
+ total_coeff |= !!res; \
+ if (step >= 4) { \
+ AV_WN16A(&s->uveob[pl][n], res); \
+ } else { \
+ s->uveob[pl][n] = res; \
+ } \
+ } \
+ }
+
+ p = s->prob.coef[b->uvtx][1 /* uv */][!b->intra];
+ c = s->counts.coef[b->uvtx][1 /* uv */][!b->intra];
+ e = s->counts.eob[b->uvtx][1 /* uv */][!b->intra];
+ w4 >>= s->ss_h;
+ end_x >>= s->ss_h;
+ h4 >>= s->ss_v;
+ end_y >>= s->ss_v;
+ for (pl = 0; pl < 2; pl++) {
+ a = &s->above_uv_nnz_ctx[pl][col << !s->ss_h];
+ l = &s->left_uv_nnz_ctx[pl][(row & 7) << !s->ss_v];
+ switch (b->uvtx) {
+ case TX_4X4:
+ DECODE_UV_COEF_LOOP(1,);
+ break;
+ case TX_8X8:
+ MERGE_CTX(2, AV_RN16A);
+ DECODE_UV_COEF_LOOP(2,);
+ SPLAT_CTX(2);
+ break;
+ case TX_16X16:
+ MERGE_CTX(4, AV_RN32A);
+ DECODE_UV_COEF_LOOP(4,);
+ SPLAT_CTX(4);
+ break;
+ case TX_32X32:
+ MERGE_CTX(8, AV_RN64A);
+ DECODE_UV_COEF_LOOP(8, 32);
+ SPLAT_CTX(8);
+ break;
+ }
+ }
+
+ return total_coeff;
+}
+
+static int decode_coeffs_8bpp(AVCodecContext *ctx)
+{
+ return decode_coeffs(ctx, 1);
+}
+
+static int decode_coeffs_16bpp(AVCodecContext *ctx)
+{
+ return decode_coeffs(ctx, 0);
+}
+
+static av_always_inline int check_intra_mode(VP9Context *s, int mode, uint8_t **a,
+ uint8_t *dst_edge, ptrdiff_t stride_edge,
+ uint8_t *dst_inner, ptrdiff_t stride_inner,
+ uint8_t *l, int col, int x, int w,
+ int row, int y, enum TxfmMode tx,
+ int p, int ss_h, int ss_v, int bytesperpixel)
+{
+ int have_top = row > 0 || y > 0;
+ int have_left = col > s->tile_col_start || x > 0;
+ int have_right = x < w - 1;
+ int bpp = s->s.h.bpp;
+ static const uint8_t mode_conv[10][2 /* have_left */][2 /* have_top */] = {
+ [VERT_PRED] = { { DC_127_PRED, VERT_PRED },
+ { DC_127_PRED, VERT_PRED } },
+ [HOR_PRED] = { { DC_129_PRED, DC_129_PRED },
+ { HOR_PRED, HOR_PRED } },
+ [DC_PRED] = { { DC_128_PRED, TOP_DC_PRED },
+ { LEFT_DC_PRED, DC_PRED } },
+ [DIAG_DOWN_LEFT_PRED] = { { DC_127_PRED, DIAG_DOWN_LEFT_PRED },
+ { DC_127_PRED, DIAG_DOWN_LEFT_PRED } },
+ [DIAG_DOWN_RIGHT_PRED] = { { DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_RIGHT_PRED },
+ { DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_RIGHT_PRED } },
+ [VERT_RIGHT_PRED] = { { VERT_RIGHT_PRED, VERT_RIGHT_PRED },
+ { VERT_RIGHT_PRED, VERT_RIGHT_PRED } },
+ [HOR_DOWN_PRED] = { { HOR_DOWN_PRED, HOR_DOWN_PRED },
+ { HOR_DOWN_PRED, HOR_DOWN_PRED } },
+ [VERT_LEFT_PRED] = { { DC_127_PRED, VERT_LEFT_PRED },
+ { DC_127_PRED, VERT_LEFT_PRED } },
+ [HOR_UP_PRED] = { { DC_129_PRED, DC_129_PRED },
+ { HOR_UP_PRED, HOR_UP_PRED } },
+ [TM_VP8_PRED] = { { DC_129_PRED, VERT_PRED },
+ { HOR_PRED, TM_VP8_PRED } },
+ };
+ static const struct {
+ uint8_t needs_left:1;
+ uint8_t needs_top:1;
+ uint8_t needs_topleft:1;
+ uint8_t needs_topright:1;
+ uint8_t invert_left:1;
+ } edges[N_INTRA_PRED_MODES] = {
+ [VERT_PRED] = { .needs_top = 1 },
+ [HOR_PRED] = { .needs_left = 1 },
+ [DC_PRED] = { .needs_top = 1, .needs_left = 1 },
+ [DIAG_DOWN_LEFT_PRED] = { .needs_top = 1, .needs_topright = 1 },
+ [DIAG_DOWN_RIGHT_PRED] = { .needs_left = 1, .needs_top = 1, .needs_topleft = 1 },
+ [VERT_RIGHT_PRED] = { .needs_left = 1, .needs_top = 1, .needs_topleft = 1 },
+ [HOR_DOWN_PRED] = { .needs_left = 1, .needs_top = 1, .needs_topleft = 1 },
+ [VERT_LEFT_PRED] = { .needs_top = 1, .needs_topright = 1 },
+ [HOR_UP_PRED] = { .needs_left = 1, .invert_left = 1 },
+ [TM_VP8_PRED] = { .needs_left = 1, .needs_top = 1, .needs_topleft = 1 },
+ [LEFT_DC_PRED] = { .needs_left = 1 },
+ [TOP_DC_PRED] = { .needs_top = 1 },
+ [DC_128_PRED] = { 0 },
+ [DC_127_PRED] = { 0 },
+ [DC_129_PRED] = { 0 }
+ };
+
+ av_assert2(mode >= 0 && mode < 10);
+ mode = mode_conv[mode][have_left][have_top];
+ if (edges[mode].needs_top) {
+ uint8_t *top, *topleft;
+ int n_px_need = 4 << tx, n_px_have = (((s->cols - col) << !ss_h) - x) * 4;
+ int n_px_need_tr = 0;
+
+ if (tx == TX_4X4 && edges[mode].needs_topright && have_right)
+ n_px_need_tr = 4;
+
+ // if top of sb64-row, use s->intra_pred_data[] instead of
+ // dst[-stride] for intra prediction (it contains pre- instead of
+ // post-loopfilter data)
+ if (have_top) {
+ top = !(row & 7) && !y ?
+ s->intra_pred_data[p] + (col * (8 >> ss_h) + x * 4) * bytesperpixel :
+ y == 0 ? &dst_edge[-stride_edge] : &dst_inner[-stride_inner];
+ if (have_left)
+ topleft = !(row & 7) && !y ?
+ s->intra_pred_data[p] + (col * (8 >> ss_h) + x * 4) * bytesperpixel :
+ y == 0 || x == 0 ? &dst_edge[-stride_edge] :
+ &dst_inner[-stride_inner];
+ }
+
+ if (have_top &&
+ (!edges[mode].needs_topleft || (have_left && top == topleft)) &&
+ (tx != TX_4X4 || !edges[mode].needs_topright || have_right) &&
+ n_px_need + n_px_need_tr <= n_px_have) {
+ *a = top;
+ } else {
+ if (have_top) {
+ if (n_px_need <= n_px_have) {
+ memcpy(*a, top, n_px_need * bytesperpixel);
+ } else {
+#define memset_bpp(c, i1, v, i2, num) do { \
+ if (bytesperpixel == 1) { \
+ memset(&(c)[(i1)], (v)[(i2)], (num)); \
+ } else { \
+ int n, val = AV_RN16A(&(v)[(i2) * 2]); \
+ for (n = 0; n < (num); n++) { \
+ AV_WN16A(&(c)[((i1) + n) * 2], val); \
+ } \
+ } \
+} while (0)
+ memcpy(*a, top, n_px_have * bytesperpixel);
+ memset_bpp(*a, n_px_have, (*a), n_px_have - 1, n_px_need - n_px_have);
+ }
+ } else {
+#define memset_val(c, val, num) do { \
+ if (bytesperpixel == 1) { \
+ memset((c), (val), (num)); \
+ } else { \
+ int n; \
+ for (n = 0; n < (num); n++) { \
+ AV_WN16A(&(c)[n * 2], (val)); \
+ } \
+ } \
+} while (0)
+ memset_val(*a, (128 << (bpp - 8)) - 1, n_px_need);
+ }
+ if (edges[mode].needs_topleft) {
+ if (have_left && have_top) {
+#define assign_bpp(c, i1, v, i2) do { \
+ if (bytesperpixel == 1) { \
+ (c)[(i1)] = (v)[(i2)]; \
+ } else { \
+ AV_COPY16(&(c)[(i1) * 2], &(v)[(i2) * 2]); \
+ } \
+} while (0)
+ assign_bpp(*a, -1, topleft, -1);
+ } else {
+#define assign_val(c, i, v) do { \
+ if (bytesperpixel == 1) { \
+ (c)[(i)] = (v); \
+ } else { \
+ AV_WN16A(&(c)[(i) * 2], (v)); \
+ } \
+} while (0)
+ assign_val((*a), -1, (128 << (bpp - 8)) + (have_top ? +1 : -1));
+ }
+ }
+ if (tx == TX_4X4 && edges[mode].needs_topright) {
+ if (have_top && have_right &&
+ n_px_need + n_px_need_tr <= n_px_have) {
+ memcpy(&(*a)[4 * bytesperpixel], &top[4 * bytesperpixel], 4 * bytesperpixel);
+ } else {
+ memset_bpp(*a, 4, *a, 3, 4);
+ }
+ }
+ }
+ }
+ if (edges[mode].needs_left) {
+ if (have_left) {
+ int n_px_need = 4 << tx, i, n_px_have = (((s->rows - row) << !ss_v) - y) * 4;
+ uint8_t *dst = x == 0 ? dst_edge : dst_inner;
+ ptrdiff_t stride = x == 0 ? stride_edge : stride_inner;
+
+ if (edges[mode].invert_left) {
+ if (n_px_need <= n_px_have) {
+ for (i = 0; i < n_px_need; i++)
+ assign_bpp(l, i, &dst[i * stride], -1);
+ } else {
+ for (i = 0; i < n_px_have; i++)
+ assign_bpp(l, i, &dst[i * stride], -1);
+ memset_bpp(l, n_px_have, l, n_px_have - 1, n_px_need - n_px_have);
+ }
+ } else {
+ if (n_px_need <= n_px_have) {
+ for (i = 0; i < n_px_need; i++)
+ assign_bpp(l, n_px_need - 1 - i, &dst[i * stride], -1);
+ } else {
+ for (i = 0; i < n_px_have; i++)
+ assign_bpp(l, n_px_need - 1 - i, &dst[i * stride], -1);
+ memset_bpp(l, 0, l, n_px_need - n_px_have, n_px_need - n_px_have);
+ }
+ }
+ } else {
+ memset_val(l, (128 << (bpp - 8)) + 1, 4 << tx);
+ }
+ }
+
+ return mode;
+}
+
+static av_always_inline void intra_recon(AVCodecContext *ctx, ptrdiff_t y_off,
+ ptrdiff_t uv_off, int bytesperpixel)
+{
+ VP9Context *s = ctx->priv_data;
+ VP9Block *b = s->b;
+ int row = s->row, col = s->col;
+ int w4 = bwh_tab[1][b->bs][0] << 1, step1d = 1 << b->tx, n;
+ int h4 = bwh_tab[1][b->bs][1] << 1, x, y, step = 1 << (b->tx * 2);
+ int end_x = FFMIN(2 * (s->cols - col), w4);
+ int end_y = FFMIN(2 * (s->rows - row), h4);
+ int tx = 4 * s->s.h.lossless + b->tx, uvtx = b->uvtx + 4 * s->s.h.lossless;
+ int uvstep1d = 1 << b->uvtx, p;
+ uint8_t *dst = s->dst[0], *dst_r = s->s.frames[CUR_FRAME].tf.f->data[0] + y_off;
+ LOCAL_ALIGNED_32(uint8_t, a_buf, [96]);
+ LOCAL_ALIGNED_32(uint8_t, l, [64]);
+
+ for (n = 0, y = 0; y < end_y; y += step1d) {
+ uint8_t *ptr = dst, *ptr_r = dst_r;
+ for (x = 0; x < end_x; x += step1d, ptr += 4 * step1d * bytesperpixel,
+ ptr_r += 4 * step1d * bytesperpixel, n += step) {
+ int mode = b->mode[b->bs > BS_8x8 && b->tx == TX_4X4 ?
+ y * 2 + x : 0];
+ uint8_t *a = &a_buf[32];
+ enum TxfmType txtp = ff_vp9_intra_txfm_type[mode];
+ int eob = b->skip ? 0 : b->tx > TX_8X8 ? AV_RN16A(&s->eob[n]) : s->eob[n];
+
+ mode = check_intra_mode(s, mode, &a, ptr_r,
+ s->s.frames[CUR_FRAME].tf.f->linesize[0],
+ ptr, s->y_stride, l,
+ col, x, w4, row, y, b->tx, 0, 0, 0, bytesperpixel);
+ s->dsp.intra_pred[b->tx][mode](ptr, s->y_stride, l, a);
+ if (eob)
+ s->dsp.itxfm_add[tx][txtp](ptr, s->y_stride,
+ s->block + 16 * n * bytesperpixel, eob);
+ }
+ dst_r += 4 * step1d * s->s.frames[CUR_FRAME].tf.f->linesize[0];
+ dst += 4 * step1d * s->y_stride;
+ }
+
+ // U/V
+ w4 >>= s->ss_h;
+ end_x >>= s->ss_h;
+ end_y >>= s->ss_v;
+ step = 1 << (b->uvtx * 2);
+ for (p = 0; p < 2; p++) {
+ dst = s->dst[1 + p];
+ dst_r = s->s.frames[CUR_FRAME].tf.f->data[1 + p] + uv_off;
+ for (n = 0, y = 0; y < end_y; y += uvstep1d) {
+ uint8_t *ptr = dst, *ptr_r = dst_r;
+ for (x = 0; x < end_x; x += uvstep1d, ptr += 4 * uvstep1d * bytesperpixel,
+ ptr_r += 4 * uvstep1d * bytesperpixel, n += step) {
+ int mode = b->uvmode;
+ uint8_t *a = &a_buf[32];
+ int eob = b->skip ? 0 : b->uvtx > TX_8X8 ? AV_RN16A(&s->uveob[p][n]) : s->uveob[p][n];
+
+ mode = check_intra_mode(s, mode, &a, ptr_r,
+ s->s.frames[CUR_FRAME].tf.f->linesize[1],
+ ptr, s->uv_stride, l, col, x, w4, row, y,
+ b->uvtx, p + 1, s->ss_h, s->ss_v, bytesperpixel);
+ s->dsp.intra_pred[b->uvtx][mode](ptr, s->uv_stride, l, a);
+ if (eob)
+ s->dsp.itxfm_add[uvtx][DCT_DCT](ptr, s->uv_stride,
+ s->uvblock[p] + 16 * n * bytesperpixel, eob);
+ }
+ dst_r += 4 * uvstep1d * s->s.frames[CUR_FRAME].tf.f->linesize[1];
+ dst += 4 * uvstep1d * s->uv_stride;
+ }
+ }
+}
+
+static void intra_recon_8bpp(AVCodecContext *ctx, ptrdiff_t y_off, ptrdiff_t uv_off)
+{
+ intra_recon(ctx, y_off, uv_off, 1);
+}
+
+static void intra_recon_16bpp(AVCodecContext *ctx, ptrdiff_t y_off, ptrdiff_t uv_off)
+{
+ intra_recon(ctx, y_off, uv_off, 2);
+}
+
+static av_always_inline void mc_luma_unscaled(VP9Context *s, vp9_mc_func (*mc)[2],
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const uint8_t *ref, ptrdiff_t ref_stride,
+ ThreadFrame *ref_frame,
+ ptrdiff_t y, ptrdiff_t x, const VP56mv *mv,
+ int bw, int bh, int w, int h, int bytesperpixel)
+{
+ int mx = mv->x, my = mv->y, th;
+
+ y += my >> 3;
+ x += mx >> 3;
+ ref += y * ref_stride + x * bytesperpixel;
+ mx &= 7;
+ my &= 7;
+ // FIXME bilinear filter only needs 0/1 pixels, not 3/4
+ // we use +7 because the last 7 pixels of each sbrow can be changed in
+ // the longest loopfilter of the next sbrow
+ th = (y + bh + 4 * !!my + 7) >> 6;
+ ff_thread_await_progress(ref_frame, FFMAX(th, 0), 0);
+ // The arm/aarch64 _hv filters read one more row than what actually is
+ // needed, so switch to emulated edge one pixel sooner vertically
+ // (!!my * 5) than horizontally (!!mx * 4).
+ if (x < !!mx * 3 || y < !!my * 3 ||
+ x + !!mx * 4 > w - bw || y + !!my * 5 > h - bh) {
+ s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
+ ref - !!my * 3 * ref_stride - !!mx * 3 * bytesperpixel,
+ 160, ref_stride,
+ bw + !!mx * 7, bh + !!my * 7,
+ x - !!mx * 3, y - !!my * 3, w, h);
+ ref = s->edge_emu_buffer + !!my * 3 * 160 + !!mx * 3 * bytesperpixel;
+ ref_stride = 160;
+ }
+ mc[!!mx][!!my](dst, dst_stride, ref, ref_stride, bh, mx << 1, my << 1);
+}
+
+static av_always_inline void mc_chroma_unscaled(VP9Context *s, vp9_mc_func (*mc)[2],
+ uint8_t *dst_u, uint8_t *dst_v,
+ ptrdiff_t dst_stride,
+ const uint8_t *ref_u, ptrdiff_t src_stride_u,
+ const uint8_t *ref_v, ptrdiff_t src_stride_v,
+ ThreadFrame *ref_frame,
+ ptrdiff_t y, ptrdiff_t x, const VP56mv *mv,
+ int bw, int bh, int w, int h, int bytesperpixel)
+{
+ int mx = mv->x * (1 << !s->ss_h), my = mv->y * (1 << !s->ss_v), th;
+
+ y += my >> 4;
+ x += mx >> 4;
+ ref_u += y * src_stride_u + x * bytesperpixel;
+ ref_v += y * src_stride_v + x * bytesperpixel;
+ mx &= 15;
+ my &= 15;
+ // FIXME bilinear filter only needs 0/1 pixels, not 3/4
+ // we use +7 because the last 7 pixels of each sbrow can be changed in
+ // the longest loopfilter of the next sbrow
+ th = (y + bh + 4 * !!my + 7) >> (6 - s->ss_v);
+ ff_thread_await_progress(ref_frame, FFMAX(th, 0), 0);
+ // The arm/aarch64 _hv filters read one more row than what actually is
+ // needed, so switch to emulated edge one pixel sooner vertically
+ // (!!my * 5) than horizontally (!!mx * 4).
+ if (x < !!mx * 3 || y < !!my * 3 ||
+ x + !!mx * 4 > w - bw || y + !!my * 5 > h - bh) {
+ s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
+ ref_u - !!my * 3 * src_stride_u - !!mx * 3 * bytesperpixel,
+ 160, src_stride_u,
+ bw + !!mx * 7, bh + !!my * 7,
+ x - !!mx * 3, y - !!my * 3, w, h);
+ ref_u = s->edge_emu_buffer + !!my * 3 * 160 + !!mx * 3 * bytesperpixel;
+ mc[!!mx][!!my](dst_u, dst_stride, ref_u, 160, bh, mx, my);
+
+ s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
+ ref_v - !!my * 3 * src_stride_v - !!mx * 3 * bytesperpixel,
+ 160, src_stride_v,
+ bw + !!mx * 7, bh + !!my * 7,
+ x - !!mx * 3, y - !!my * 3, w, h);
+ ref_v = s->edge_emu_buffer + !!my * 3 * 160 + !!mx * 3 * bytesperpixel;
+ mc[!!mx][!!my](dst_v, dst_stride, ref_v, 160, bh, mx, my);
+ } else {
+ mc[!!mx][!!my](dst_u, dst_stride, ref_u, src_stride_u, bh, mx, my);
+ mc[!!mx][!!my](dst_v, dst_stride, ref_v, src_stride_v, bh, mx, my);
+ }
+}
+
+#define mc_luma_dir(s, mc, dst, dst_ls, src, src_ls, tref, row, col, mv, \
+ px, py, pw, ph, bw, bh, w, h, i) \
+ mc_luma_unscaled(s, s->dsp.mc, dst, dst_ls, src, src_ls, tref, row, col, \
+ mv, bw, bh, w, h, bytesperpixel)
+#define mc_chroma_dir(s, mc, dstu, dstv, dst_ls, srcu, srcu_ls, srcv, srcv_ls, tref, \
+ row, col, mv, px, py, pw, ph, bw, bh, w, h, i) \
+ mc_chroma_unscaled(s, s->dsp.mc, dstu, dstv, dst_ls, srcu, srcu_ls, srcv, srcv_ls, tref, \
+ row, col, mv, bw, bh, w, h, bytesperpixel)
+#define SCALED 0
+#define FN(x) x##_8bpp
+#define BYTES_PER_PIXEL 1
+#include "vp9_mc_template.c"
+#undef FN
+#undef BYTES_PER_PIXEL
+#define FN(x) x##_16bpp
+#define BYTES_PER_PIXEL 2
+#include "vp9_mc_template.c"
+#undef mc_luma_dir
+#undef mc_chroma_dir
+#undef FN
+#undef BYTES_PER_PIXEL
+#undef SCALED
+
+static av_always_inline void mc_luma_scaled(VP9Context *s, vp9_scaled_mc_func smc,
+ vp9_mc_func (*mc)[2],
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const uint8_t *ref, ptrdiff_t ref_stride,
+ ThreadFrame *ref_frame,
+ ptrdiff_t y, ptrdiff_t x, const VP56mv *in_mv,
+ int px, int py, int pw, int ph,
+ int bw, int bh, int w, int h, int bytesperpixel,
+ const uint16_t *scale, const uint8_t *step)
+{
+ if (s->s.frames[CUR_FRAME].tf.f->width == ref_frame->f->width &&
+ s->s.frames[CUR_FRAME].tf.f->height == ref_frame->f->height) {
+ mc_luma_unscaled(s, mc, dst, dst_stride, ref, ref_stride, ref_frame,
+ y, x, in_mv, bw, bh, w, h, bytesperpixel);
+ } else {
+#define scale_mv(n, dim) (((int64_t)(n) * scale[dim]) >> 14)
+ int mx, my;
+ int refbw_m1, refbh_m1;
+ int th;
+ VP56mv mv;
+
+ mv.x = av_clip(in_mv->x, -(x + pw - px + 4) * 8, (s->cols * 8 - x + px + 3) * 8);
+ mv.y = av_clip(in_mv->y, -(y + ph - py + 4) * 8, (s->rows * 8 - y + py + 3) * 8);
+ // BUG libvpx seems to scale the two components separately. This introduces
+ // rounding errors but we have to reproduce them to be exactly compatible
+ // with the output from libvpx...
+ mx = scale_mv(mv.x * 2, 0) + scale_mv(x * 16, 0);
+ my = scale_mv(mv.y * 2, 1) + scale_mv(y * 16, 1);
+
+ y = my >> 4;
+ x = mx >> 4;
+ ref += y * ref_stride + x * bytesperpixel;
+ mx &= 15;
+ my &= 15;
+ refbw_m1 = ((bw - 1) * step[0] + mx) >> 4;
+ refbh_m1 = ((bh - 1) * step[1] + my) >> 4;
+ // FIXME bilinear filter only needs 0/1 pixels, not 3/4
+ // we use +7 because the last 7 pixels of each sbrow can be changed in
+ // the longest loopfilter of the next sbrow
+ th = (y + refbh_m1 + 4 + 7) >> 6;
+ ff_thread_await_progress(ref_frame, FFMAX(th, 0), 0);
+ // The arm/aarch64 _hv filters read one more row than what actually is
+ // needed, so switch to emulated edge one pixel sooner vertically
+ // (y + 5 >= h - refbh_m1) than horizontally (x + 4 >= w - refbw_m1).
+ if (x < 3 || y < 3 || x + 4 >= w - refbw_m1 || y + 5 >= h - refbh_m1) {
+ s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
+ ref - 3 * ref_stride - 3 * bytesperpixel,
+ 288, ref_stride,
+ refbw_m1 + 8, refbh_m1 + 8,
+ x - 3, y - 3, w, h);
+ ref = s->edge_emu_buffer + 3 * 288 + 3 * bytesperpixel;
+ ref_stride = 288;
+ }
+ smc(dst, dst_stride, ref, ref_stride, bh, mx, my, step[0], step[1]);
+ }
+}
+
+static av_always_inline void mc_chroma_scaled(VP9Context *s, vp9_scaled_mc_func smc,
+ vp9_mc_func (*mc)[2],
+ uint8_t *dst_u, uint8_t *dst_v,
+ ptrdiff_t dst_stride,
+ const uint8_t *ref_u, ptrdiff_t src_stride_u,
+ const uint8_t *ref_v, ptrdiff_t src_stride_v,
+ ThreadFrame *ref_frame,
+ ptrdiff_t y, ptrdiff_t x, const VP56mv *in_mv,
+ int px, int py, int pw, int ph,
+ int bw, int bh, int w, int h, int bytesperpixel,
+ const uint16_t *scale, const uint8_t *step)
+{
+ if (s->s.frames[CUR_FRAME].tf.f->width == ref_frame->f->width &&
+ s->s.frames[CUR_FRAME].tf.f->height == ref_frame->f->height) {
+ mc_chroma_unscaled(s, mc, dst_u, dst_v, dst_stride, ref_u, src_stride_u,
+ ref_v, src_stride_v, ref_frame,
+ y, x, in_mv, bw, bh, w, h, bytesperpixel);
+ } else {
+ int mx, my;
+ int refbw_m1, refbh_m1;
+ int th;
+ VP56mv mv;
+
+ if (s->ss_h) {
+ // BUG https://code.google.com/p/webm/issues/detail?id=820
+ mv.x = av_clip(in_mv->x, -(x + pw - px + 4) * 16, (s->cols * 4 - x + px + 3) * 16);
+ mx = scale_mv(mv.x, 0) + (scale_mv(x * 16, 0) & ~15) + (scale_mv(x * 32, 0) & 15);
+ } else {
+ mv.x = av_clip(in_mv->x, -(x + pw - px + 4) * 8, (s->cols * 8 - x + px + 3) * 8);
+ mx = scale_mv(mv.x * 2, 0) + scale_mv(x * 16, 0);
+ }
+ if (s->ss_v) {
+ // BUG https://code.google.com/p/webm/issues/detail?id=820
+ mv.y = av_clip(in_mv->y, -(y + ph - py + 4) * 16, (s->rows * 4 - y + py + 3) * 16);
+ my = scale_mv(mv.y, 1) + (scale_mv(y * 16, 1) & ~15) + (scale_mv(y * 32, 1) & 15);
+ } else {
+ mv.y = av_clip(in_mv->y, -(y + ph - py + 4) * 8, (s->rows * 8 - y + py + 3) * 8);
+ my = scale_mv(mv.y * 2, 1) + scale_mv(y * 16, 1);
+ }
+#undef scale_mv
+ y = my >> 4;
+ x = mx >> 4;
+ ref_u += y * src_stride_u + x * bytesperpixel;
+ ref_v += y * src_stride_v + x * bytesperpixel;
+ mx &= 15;
+ my &= 15;
+ refbw_m1 = ((bw - 1) * step[0] + mx) >> 4;
+ refbh_m1 = ((bh - 1) * step[1] + my) >> 4;
+ // FIXME bilinear filter only needs 0/1 pixels, not 3/4
+ // we use +7 because the last 7 pixels of each sbrow can be changed in
+ // the longest loopfilter of the next sbrow
+ th = (y + refbh_m1 + 4 + 7) >> (6 - s->ss_v);
+ ff_thread_await_progress(ref_frame, FFMAX(th, 0), 0);
+ // The arm/aarch64 _hv filters read one more row than what actually is
+ // needed, so switch to emulated edge one pixel sooner vertically
+ // (y + 5 >= h - refbh_m1) than horizontally (x + 4 >= w - refbw_m1).
+ if (x < 3 || y < 3 || x + 4 >= w - refbw_m1 || y + 5 >= h - refbh_m1) {
+ s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
+ ref_u - 3 * src_stride_u - 3 * bytesperpixel,
+ 288, src_stride_u,
+ refbw_m1 + 8, refbh_m1 + 8,
+ x - 3, y - 3, w, h);
+ ref_u = s->edge_emu_buffer + 3 * 288 + 3 * bytesperpixel;
+ smc(dst_u, dst_stride, ref_u, 288, bh, mx, my, step[0], step[1]);
+
+ s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
+ ref_v - 3 * src_stride_v - 3 * bytesperpixel,
+ 288, src_stride_v,
+ refbw_m1 + 8, refbh_m1 + 8,
+ x - 3, y - 3, w, h);
+ ref_v = s->edge_emu_buffer + 3 * 288 + 3 * bytesperpixel;
+ smc(dst_v, dst_stride, ref_v, 288, bh, mx, my, step[0], step[1]);
+ } else {
+ smc(dst_u, dst_stride, ref_u, src_stride_u, bh, mx, my, step[0], step[1]);
+ smc(dst_v, dst_stride, ref_v, src_stride_v, bh, mx, my, step[0], step[1]);
+ }
+ }
+}
+
+#define mc_luma_dir(s, mc, dst, dst_ls, src, src_ls, tref, row, col, mv, \
+ px, py, pw, ph, bw, bh, w, h, i) \
+ mc_luma_scaled(s, s->dsp.s##mc, s->dsp.mc, dst, dst_ls, src, src_ls, tref, row, col, \
+ mv, px, py, pw, ph, bw, bh, w, h, bytesperpixel, \
+ s->mvscale[b->ref[i]], s->mvstep[b->ref[i]])
+#define mc_chroma_dir(s, mc, dstu, dstv, dst_ls, srcu, srcu_ls, srcv, srcv_ls, tref, \
+ row, col, mv, px, py, pw, ph, bw, bh, w, h, i) \
+ mc_chroma_scaled(s, s->dsp.s##mc, s->dsp.mc, dstu, dstv, dst_ls, srcu, srcu_ls, srcv, srcv_ls, tref, \
+ row, col, mv, px, py, pw, ph, bw, bh, w, h, bytesperpixel, \
+ s->mvscale[b->ref[i]], s->mvstep[b->ref[i]])
+#define SCALED 1
+#define FN(x) x##_scaled_8bpp
+#define BYTES_PER_PIXEL 1
+#include "vp9_mc_template.c"
+#undef FN
+#undef BYTES_PER_PIXEL
+#define FN(x) x##_scaled_16bpp
+#define BYTES_PER_PIXEL 2
+#include "vp9_mc_template.c"
+#undef mc_luma_dir
+#undef mc_chroma_dir
+#undef FN
+#undef BYTES_PER_PIXEL
+#undef SCALED
+
+static av_always_inline void inter_recon(AVCodecContext *ctx, int bytesperpixel)
+{
+ VP9Context *s = ctx->priv_data;
+ VP9Block *b = s->b;
+ int row = s->row, col = s->col;
+
+ if (s->mvscale[b->ref[0]][0] || (b->comp && s->mvscale[b->ref[1]][0])) {
+ if (bytesperpixel == 1) {
+ inter_pred_scaled_8bpp(ctx);
+ } else {
+ inter_pred_scaled_16bpp(ctx);
+ }
+ } else {
+ if (bytesperpixel == 1) {
+ inter_pred_8bpp(ctx);
+ } else {
+ inter_pred_16bpp(ctx);
+ }
+ }
+ if (!b->skip) {
+ /* mostly copied intra_recon() */
+
+ int w4 = bwh_tab[1][b->bs][0] << 1, step1d = 1 << b->tx, n;
+ int h4 = bwh_tab[1][b->bs][1] << 1, x, y, step = 1 << (b->tx * 2);
+ int end_x = FFMIN(2 * (s->cols - col), w4);
+ int end_y = FFMIN(2 * (s->rows - row), h4);
+ int tx = 4 * s->s.h.lossless + b->tx, uvtx = b->uvtx + 4 * s->s.h.lossless;
+ int uvstep1d = 1 << b->uvtx, p;
+ uint8_t *dst = s->dst[0];
+
+ // y itxfm add
+ for (n = 0, y = 0; y < end_y; y += step1d) {
+ uint8_t *ptr = dst;
+ for (x = 0; x < end_x; x += step1d,
+ ptr += 4 * step1d * bytesperpixel, n += step) {
+ int eob = b->tx > TX_8X8 ? AV_RN16A(&s->eob[n]) : s->eob[n];
+
+ if (eob)
+ s->dsp.itxfm_add[tx][DCT_DCT](ptr, s->y_stride,
+ s->block + 16 * n * bytesperpixel, eob);
+ }
+ dst += 4 * s->y_stride * step1d;
+ }
+
+ // uv itxfm add
+ end_x >>= s->ss_h;
+ end_y >>= s->ss_v;
+ step = 1 << (b->uvtx * 2);
+ for (p = 0; p < 2; p++) {
+ dst = s->dst[p + 1];
+ for (n = 0, y = 0; y < end_y; y += uvstep1d) {
+ uint8_t *ptr = dst;
+ for (x = 0; x < end_x; x += uvstep1d,
+ ptr += 4 * uvstep1d * bytesperpixel, n += step) {
+ int eob = b->uvtx > TX_8X8 ? AV_RN16A(&s->uveob[p][n]) : s->uveob[p][n];
+
+ if (eob)
+ s->dsp.itxfm_add[uvtx][DCT_DCT](ptr, s->uv_stride,
+ s->uvblock[p] + 16 * n * bytesperpixel, eob);
+ }
+ dst += 4 * uvstep1d * s->uv_stride;
+ }
+ }
+ }
+}
+
+static void inter_recon_8bpp(AVCodecContext *ctx)
+{
+ inter_recon(ctx, 1);
+}
+
+static void inter_recon_16bpp(AVCodecContext *ctx)
+{
+ inter_recon(ctx, 2);
+}
+
+static av_always_inline void mask_edges(uint8_t (*mask)[8][4], int ss_h, int ss_v,
+ int row_and_7, int col_and_7,
+ int w, int h, int col_end, int row_end,
+ enum TxfmMode tx, int skip_inter)
+{
+ static const unsigned wide_filter_col_mask[2] = { 0x11, 0x01 };
+ static const unsigned wide_filter_row_mask[2] = { 0x03, 0x07 };
+
+ // FIXME I'm pretty sure all loops can be replaced by a single LUT if
+ // we make VP9Filter.mask uint64_t (i.e. row/col all single variable)
+ // and make the LUT 5-indexed (bl, bp, is_uv, tx and row/col), and then
+ // use row_and_7/col_and_7 as shifts (1*col_and_7+8*row_and_7)
+
+ // the intended behaviour of the vp9 loopfilter is to work on 8-pixel
+ // edges. This means that for UV, we work on two subsampled blocks at
+ // a time, and we only use the topleft block's mode information to set
+ // things like block strength. Thus, for any block size smaller than
+ // 16x16, ignore the odd portion of the block.
+ if (tx == TX_4X4 && (ss_v | ss_h)) {
+ if (h == ss_v) {
+ if (row_and_7 & 1)
+ return;
+ if (!row_end)
+ h += 1;
+ }
+ if (w == ss_h) {
+ if (col_and_7 & 1)
+ return;
+ if (!col_end)
+ w += 1;
+ }
+ }
+
+ if (tx == TX_4X4 && !skip_inter) {
+ int t = 1 << col_and_7, m_col = (t << w) - t, y;
+ // on 32-px edges, use the 8-px wide loopfilter; else, use 4-px wide
+ int m_row_8 = m_col & wide_filter_col_mask[ss_h], m_row_4 = m_col - m_row_8;
+
+ for (y = row_and_7; y < h + row_and_7; y++) {
+ int col_mask_id = 2 - !(y & wide_filter_row_mask[ss_v]);
+
+ mask[0][y][1] |= m_row_8;
+ mask[0][y][2] |= m_row_4;
+ // for odd lines, if the odd col is not being filtered,
+ // skip odd row also:
+ // .---. <-- a
+ // | |
+ // |___| <-- b
+ // ^ ^
+ // c d
+ //
+ // if a/c are even row/col and b/d are odd, and d is skipped,
+ // e.g. right edge of size-66x66.webm, then skip b also (bug)
+ if ((ss_h & ss_v) && (col_end & 1) && (y & 1)) {
+ mask[1][y][col_mask_id] |= (t << (w - 1)) - t;
+ } else {
+ mask[1][y][col_mask_id] |= m_col;
+ }
+ if (!ss_h)
+ mask[0][y][3] |= m_col;
+ if (!ss_v) {
+ if (ss_h && (col_end & 1))
+ mask[1][y][3] |= (t << (w - 1)) - t;
+ else
+ mask[1][y][3] |= m_col;
+ }
+ }
+ } else {
+ int y, t = 1 << col_and_7, m_col = (t << w) - t;
+
+ if (!skip_inter) {
+ int mask_id = (tx == TX_8X8);
+ static const unsigned masks[4] = { 0xff, 0x55, 0x11, 0x01 };
+ int l2 = tx + ss_h - 1, step1d;
+ int m_row = m_col & masks[l2];
+
+ // at odd UV col/row edges tx16/tx32 loopfilter edges, force
+ // 8wd loopfilter to prevent going off the visible edge.
+ if (ss_h && tx > TX_8X8 && (w ^ (w - 1)) == 1) {
+ int m_row_16 = ((t << (w - 1)) - t) & masks[l2];
+ int m_row_8 = m_row - m_row_16;
+
+ for (y = row_and_7; y < h + row_and_7; y++) {
+ mask[0][y][0] |= m_row_16;
+ mask[0][y][1] |= m_row_8;
+ }
+ } else {
+ for (y = row_and_7; y < h + row_and_7; y++)
+ mask[0][y][mask_id] |= m_row;
+ }
+
+ l2 = tx + ss_v - 1;
+ step1d = 1 << l2;
+ if (ss_v && tx > TX_8X8 && (h ^ (h - 1)) == 1) {
+ for (y = row_and_7; y < h + row_and_7 - 1; y += step1d)
+ mask[1][y][0] |= m_col;
+ if (y - row_and_7 == h - 1)
+ mask[1][y][1] |= m_col;
+ } else {
+ for (y = row_and_7; y < h + row_and_7; y += step1d)
+ mask[1][y][mask_id] |= m_col;
+ }
+ } else if (tx != TX_4X4) {
+ int mask_id;
+
+ mask_id = (tx == TX_8X8) || (h == ss_v);
+ mask[1][row_and_7][mask_id] |= m_col;
+ mask_id = (tx == TX_8X8) || (w == ss_h);
+ for (y = row_and_7; y < h + row_and_7; y++)
+ mask[0][y][mask_id] |= t;
+ } else {
+ int t8 = t & wide_filter_col_mask[ss_h], t4 = t - t8;
+
+ for (y = row_and_7; y < h + row_and_7; y++) {
+ mask[0][y][2] |= t4;
+ mask[0][y][1] |= t8;
+ }
+ mask[1][row_and_7][2 - !(row_and_7 & wide_filter_row_mask[ss_v])] |= m_col;
+ }
+ }
+}
+
+void ff_vp9_decode_block(AVCodecContext *ctx, int row, int col,
+ struct VP9Filter *lflvl, ptrdiff_t yoff, ptrdiff_t uvoff,
+ enum BlockLevel bl, enum BlockPartition bp)
+{
+ VP9Context *s = ctx->priv_data;
+ VP9Block *b = s->b;
+ enum BlockSize bs = bl * 3 + bp;
+ int bytesperpixel = s->bytesperpixel;
+ int w4 = bwh_tab[1][bs][0], h4 = bwh_tab[1][bs][1], lvl;
+ int emu[2];
+ AVFrame *f = s->s.frames[CUR_FRAME].tf.f;
+
+ s->row = row;
+ s->row7 = row & 7;
+ s->col = col;
+ s->col7 = col & 7;
+ s->min_mv.x = -(128 + col * 64);
+ s->min_mv.y = -(128 + row * 64);
+ s->max_mv.x = 128 + (s->cols - col - w4) * 64;
+ s->max_mv.y = 128 + (s->rows - row - h4) * 64;
+ if (s->pass < 2) {
+ b->bs = bs;
+ b->bl = bl;
+ b->bp = bp;
+ decode_mode(ctx);
+ b->uvtx = b->tx - ((s->ss_h && w4 * 2 == (1 << b->tx)) ||
+ (s->ss_v && h4 * 2 == (1 << b->tx)));
+
+ if (!b->skip) {
+ int has_coeffs;
+
+ if (bytesperpixel == 1) {
+ has_coeffs = decode_coeffs_8bpp(ctx);
+ } else {
+ has_coeffs = decode_coeffs_16bpp(ctx);
+ }
+ if (!has_coeffs && b->bs <= BS_8x8 && !b->intra) {
+ b->skip = 1;
+ memset(&s->above_skip_ctx[col], 1, w4);
+ memset(&s->left_skip_ctx[s->row7], 1, h4);
+ }
+ } else {
+ int row7 = s->row7;
+
+#define SPLAT_ZERO_CTX(v, n) \
+ switch (n) { \
+ case 1: v = 0; break; \
+ case 2: AV_ZERO16(&v); break; \
+ case 4: AV_ZERO32(&v); break; \
+ case 8: AV_ZERO64(&v); break; \
+ case 16: AV_ZERO128(&v); break; \
+ }
+#define SPLAT_ZERO_YUV(dir, var, off, n, dir2) \
+ do { \
+ SPLAT_ZERO_CTX(s->dir##_y_##var[off * 2], n * 2); \
+ if (s->ss_##dir2) { \
+ SPLAT_ZERO_CTX(s->dir##_uv_##var[0][off], n); \
+ SPLAT_ZERO_CTX(s->dir##_uv_##var[1][off], n); \
+ } else { \
+ SPLAT_ZERO_CTX(s->dir##_uv_##var[0][off * 2], n * 2); \
+ SPLAT_ZERO_CTX(s->dir##_uv_##var[1][off * 2], n * 2); \
+ } \
+ } while (0)
+
+ switch (w4) {
+ case 1: SPLAT_ZERO_YUV(above, nnz_ctx, col, 1, h); break;
+ case 2: SPLAT_ZERO_YUV(above, nnz_ctx, col, 2, h); break;
+ case 4: SPLAT_ZERO_YUV(above, nnz_ctx, col, 4, h); break;
+ case 8: SPLAT_ZERO_YUV(above, nnz_ctx, col, 8, h); break;
+ }
+ switch (h4) {
+ case 1: SPLAT_ZERO_YUV(left, nnz_ctx, row7, 1, v); break;
+ case 2: SPLAT_ZERO_YUV(left, nnz_ctx, row7, 2, v); break;
+ case 4: SPLAT_ZERO_YUV(left, nnz_ctx, row7, 4, v); break;
+ case 8: SPLAT_ZERO_YUV(left, nnz_ctx, row7, 8, v); break;
+ }
+ }
+
+ if (s->pass == 1) {
+ s->b++;
+ s->block += w4 * h4 * 64 * bytesperpixel;
+ s->uvblock[0] += w4 * h4 * 64 * bytesperpixel >> (s->ss_h + s->ss_v);
+ s->uvblock[1] += w4 * h4 * 64 * bytesperpixel >> (s->ss_h + s->ss_v);
+ s->eob += 4 * w4 * h4;
+ s->uveob[0] += 4 * w4 * h4 >> (s->ss_h + s->ss_v);
+ s->uveob[1] += 4 * w4 * h4 >> (s->ss_h + s->ss_v);
+
+ return;
+ }
+ }
+
+ // emulated overhangs if the stride of the target buffer can't hold. This
+ // makes it possible to support emu-edge and so on even if we have large block
+ // overhangs
+ emu[0] = (col + w4) * 8 * bytesperpixel > f->linesize[0] ||
+ (row + h4) > s->rows;
+ emu[1] = ((col + w4) * 8 >> s->ss_h) * bytesperpixel > f->linesize[1] ||
+ (row + h4) > s->rows;
+ if (emu[0]) {
+ s->dst[0] = s->tmp_y;
+ s->y_stride = 128;
+ } else {
+ s->dst[0] = f->data[0] + yoff;
+ s->y_stride = f->linesize[0];
+ }
+ if (emu[1]) {
+ s->dst[1] = s->tmp_uv[0];
+ s->dst[2] = s->tmp_uv[1];
+ s->uv_stride = 128;
+ } else {
+ s->dst[1] = f->data[1] + uvoff;
+ s->dst[2] = f->data[2] + uvoff;
+ s->uv_stride = f->linesize[1];
+ }
+ if (b->intra) {
+ if (s->s.h.bpp > 8) {
+ intra_recon_16bpp(ctx, yoff, uvoff);
+ } else {
+ intra_recon_8bpp(ctx, yoff, uvoff);
+ }
+ } else {
+ if (s->s.h.bpp > 8) {
+ inter_recon_16bpp(ctx);
+ } else {
+ inter_recon_8bpp(ctx);
+ }
+ }
+ if (emu[0]) {
+ int w = FFMIN(s->cols - col, w4) * 8, h = FFMIN(s->rows - row, h4) * 8, n, o = 0;
+
+ for (n = 0; o < w; n++) {
+ int bw = 64 >> n;
+
+ av_assert2(n <= 4);
+ if (w & bw) {
+ s->dsp.mc[n][0][0][0][0](f->data[0] + yoff + o * bytesperpixel, f->linesize[0],
+ s->tmp_y + o * bytesperpixel, 128, h, 0, 0);
+ o += bw;
+ }
+ }
+ }
+ if (emu[1]) {
+ int w = FFMIN(s->cols - col, w4) * 8 >> s->ss_h;
+ int h = FFMIN(s->rows - row, h4) * 8 >> s->ss_v, n, o = 0;
+
+ for (n = s->ss_h; o < w; n++) {
+ int bw = 64 >> n;
+
+ av_assert2(n <= 4);
+ if (w & bw) {
+ s->dsp.mc[n][0][0][0][0](f->data[1] + uvoff + o * bytesperpixel, f->linesize[1],
+ s->tmp_uv[0] + o * bytesperpixel, 128, h, 0, 0);
+ s->dsp.mc[n][0][0][0][0](f->data[2] + uvoff + o * bytesperpixel, f->linesize[2],
+ s->tmp_uv[1] + o * bytesperpixel, 128, h, 0, 0);
+ o += bw;
+ }
+ }
+ }
+
+ // pick filter level and find edges to apply filter to
+ if (s->s.h.filter.level &&
+ (lvl = s->s.h.segmentation.feat[b->seg_id].lflvl[b->intra ? 0 : b->ref[0] + 1]
+ [b->mode[3] != ZEROMV]) > 0) {
+ int x_end = FFMIN(s->cols - col, w4), y_end = FFMIN(s->rows - row, h4);
+ int skip_inter = !b->intra && b->skip, col7 = s->col7, row7 = s->row7;
+
+ setctx_2d(&lflvl->level[row7 * 8 + col7], w4, h4, 8, lvl);
+ mask_edges(lflvl->mask[0], 0, 0, row7, col7, x_end, y_end, 0, 0, b->tx, skip_inter);
+ if (s->ss_h || s->ss_v)
+ mask_edges(lflvl->mask[1], s->ss_h, s->ss_v, row7, col7, x_end, y_end,
+ s->cols & 1 && col + w4 >= s->cols ? s->cols & 7 : 0,
+ s->rows & 1 && row + h4 >= s->rows ? s->rows & 7 : 0,
+ b->uvtx, skip_inter);
+
+ if (!s->filter_lut.lim_lut[lvl]) {
+ int sharp = s->s.h.filter.sharpness;
+ int limit = lvl;
+
+ if (sharp > 0) {
+ limit >>= (sharp + 3) >> 2;
+ limit = FFMIN(limit, 9 - sharp);
+ }
+ limit = FFMAX(limit, 1);
+
+ s->filter_lut.lim_lut[lvl] = limit;
+ s->filter_lut.mblim_lut[lvl] = 2 * (lvl + 2) + limit;
+ }
+ }
+
+ if (s->pass == 2) {
+ s->b++;
+ s->block += w4 * h4 * 64 * bytesperpixel;
+ s->uvblock[0] += w4 * h4 * 64 * bytesperpixel >> (s->ss_v + s->ss_h);
+ s->uvblock[1] += w4 * h4 * 64 * bytesperpixel >> (s->ss_v + s->ss_h);
+ s->eob += 4 * w4 * h4;
+ s->uveob[0] += 4 * w4 * h4 >> (s->ss_v + s->ss_h);
+ s->uveob[1] += 4 * w4 * h4 >> (s->ss_v + s->ss_h);
+ }
+}
--- /dev/null
+/*
+ * Copyright (C) 2013 Ronald S. Bultje <rsbultje gmail com>
+ * Copyright (C) 2013 Clément Bœsch <u pkh me>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "vp9.h"
+#include "vp9data.h"
+
+const int8_t ff_vp9_partition_tree[3][2] = {
+ { -PARTITION_NONE, 1 }, // '0'
+ { -PARTITION_H, 2 }, // '10'
+ { -PARTITION_V, -PARTITION_SPLIT }, // '110', '111'
+};
+
+const uint8_t ff_vp9_default_kf_partition_probs[4][4][3] = {
+ { /* 64x64 -> 32x32 */
+ { 174, 35, 49 } /* a/l both not split */,
+ { 68, 11, 27 } /* a split, l not split */,
+ { 57, 15, 9 } /* l split, a not split */,
+ { 12, 3, 3 } /* a/l both split */
+ }, { /* 32x32 -> 16x16 */
+ { 150, 40, 39 } /* a/l both not split */,
+ { 78, 12, 26 } /* a split, l not split */,
+ { 67, 33, 11 } /* l split, a not split */,
+ { 24, 7, 5 } /* a/l both split */,
+ }, { /* 16x16 -> 8x8 */
+ { 149, 53, 53 } /* a/l both not split */,
+ { 94, 20, 48 } /* a split, l not split */,
+ { 83, 53, 24 } /* l split, a not split */,
+ { 52, 18, 18 } /* a/l both split */,
+ }, { /* 8x8 -> 4x4 */
+ { 158, 97, 94 } /* a/l both not split */,
+ { 93, 24, 99 } /* a split, l not split */,
+ { 85, 119, 44 } /* l split, a not split */,
+ { 62, 59, 67 } /* a/l both split */,
+ },
+};
+
+const int8_t ff_vp9_segmentation_tree[7][2] = {
+ { 1, 2 },
+ { 3, 4 },
+ { 5, 6 },
+ { -0, -1 }, // '00x'
+ { -2, -3 }, // '01x'
+ { -4, -5 }, // '10x'
+ { -6, -7 }, // '11x'
+};
+
+const int8_t ff_vp9_intramode_tree[9][2] = {
+ { -DC_PRED, 1 }, // '0'
+ { -TM_VP8_PRED, 2 }, // '10'
+ { -VERT_PRED, 3 }, // '110'
+ { 4, 6 },
+ { -HOR_PRED, 5 }, // '11100'
+ { -DIAG_DOWN_RIGHT_PRED, -VERT_RIGHT_PRED }, // '11101x'
+ { -DIAG_DOWN_LEFT_PRED, 7 }, // '11110'
+ { -VERT_LEFT_PRED, 8 }, // '111110'
+ { -HOR_DOWN_PRED, -HOR_UP_PRED }, // '111111x'
+};
+
+const uint8_t ff_vp9_default_kf_ymode_probs[10][10][9] = {
+ { /* above = v */
+ { 43, 46, 168, 134, 107, 128, 69, 142, 92 } /* left = v */,
+ { 44, 29, 68, 159, 201, 177, 50, 57, 77 } /* left = h */,
+ { 63, 36, 126, 146, 123, 158, 60, 90, 96 } /* left = dc */,
+ { 58, 38, 76, 114, 97, 172, 78, 133, 92 } /* left = d45 */,
+ { 46, 41, 76, 140, 63, 184, 69, 112, 57 } /* left = d135 */,
+ { 38, 32, 85, 140, 46, 112, 54, 151, 133 } /* left = d117 */,
+ { 39, 27, 61, 131, 110, 175, 44, 75, 136 } /* left = d153 */,
+ { 47, 35, 80, 100, 74, 143, 64, 163, 74 } /* left = d63 */,
+ { 52, 30, 74, 113, 130, 175, 51, 64, 58 } /* left = d27 */,
+ { 36, 61, 116, 114, 128, 162, 80, 125, 82 } /* left = tm */
+ }, { /* above = h */
+ { 55, 44, 68, 166, 179, 192, 57, 57, 108 } /* left = v */,
+ { 42, 26, 11, 199, 241, 228, 23, 15, 85 } /* left = h */,
+ { 82, 26, 26, 171, 208, 204, 44, 32, 105 } /* left = dc */,
+ { 68, 42, 19, 131, 160, 199, 55, 52, 83 } /* left = d45 */,
+ { 58, 50, 25, 139, 115, 232, 39, 52, 118 } /* left = d135 */,
+ { 50, 35, 33, 153, 104, 162, 64, 59, 131 } /* left = d117 */,
+ { 44, 24, 16, 150, 177, 202, 33, 19, 156 } /* left = d153 */,
+ { 53, 49, 21, 110, 116, 168, 59, 80, 76 } /* left = d63 */,
+ { 55, 27, 12, 153, 203, 218, 26, 27, 49 } /* left = d27 */,
+ { 38, 72, 19, 168, 203, 212, 50, 50, 107 } /* left = tm */
+ }, { /* above = dc */
+ { 92, 45, 102, 136, 116, 180, 74, 90, 100 } /* left = v */,
+ { 73, 32, 19, 187, 222, 215, 46, 34, 100 } /* left = h */,
+ { 137, 30, 42, 148, 151, 207, 70, 52, 91 } /* left = dc */,
+ { 91, 30, 32, 116, 121, 186, 93, 86, 94 } /* left = d45 */,
+ { 72, 35, 36, 149, 68, 206, 68, 63, 105 } /* left = d135 */,
+ { 73, 31, 28, 138, 57, 124, 55, 122, 151 } /* left = d117 */,
+ { 67, 23, 21, 140, 126, 197, 40, 37, 171 } /* left = d153 */,
+ { 74, 32, 27, 107, 86, 160, 63, 134, 102 } /* left = d63 */,
+ { 86, 27, 28, 128, 154, 212, 45, 43, 53 } /* left = d27 */,
+ { 59, 67, 44, 140, 161, 202, 78, 67, 119 } /* left = tm */
+ }, { /* above = d45 */
+ { 59, 38, 83, 112, 103, 162, 98, 136, 90 } /* left = v */,
+ { 62, 30, 23, 158, 200, 207, 59, 57, 50 } /* left = h */,
+ { 103, 26, 36, 129, 132, 201, 83, 80, 93 } /* left = dc */,
+ { 67, 30, 29, 84, 86, 191, 102, 91, 59 } /* left = d45 */,
+ { 60, 32, 33, 112, 71, 220, 64, 89, 104 } /* left = d135 */,
+ { 53, 26, 34, 130, 56, 149, 84, 120, 103 } /* left = d117 */,
+ { 53, 21, 23, 133, 109, 210, 56, 77, 172 } /* left = d153 */,
+ { 61, 29, 29, 93, 97, 165, 83, 175, 162 } /* left = d63 */,
+ { 77, 19, 29, 112, 142, 228, 55, 66, 36 } /* left = d27 */,
+ { 47, 47, 43, 114, 137, 181, 100, 99, 95 } /* left = tm */
+ }, { /* above = d135 */
+ { 53, 40, 55, 139, 69, 183, 61, 80, 110 } /* left = v */,
+ { 40, 29, 19, 161, 180, 207, 43, 24, 91 } /* left = h */,
+ { 69, 23, 29, 128, 83, 199, 46, 44, 101 } /* left = dc */,
+ { 60, 34, 19, 105, 61, 198, 53, 64, 89 } /* left = d45 */,
+ { 52, 31, 22, 158, 40, 209, 58, 62, 89 } /* left = d135 */,
+ { 44, 31, 29, 147, 46, 158, 56, 102, 198 } /* left = d117 */,
+ { 35, 19, 12, 135, 87, 209, 41, 45, 167 } /* left = d153 */,
+ { 51, 38, 25, 113, 58, 164, 70, 93, 97 } /* left = d63 */,
+ { 55, 25, 21, 118, 95, 215, 38, 39, 66 } /* left = d27 */,
+ { 47, 54, 34, 146, 108, 203, 72, 103, 151 } /* left = tm */
+ }, { /* above = d117 */
+ { 46, 27, 80, 150, 55, 124, 55, 121, 135 } /* left = v */,
+ { 36, 23, 27, 165, 149, 166, 54, 64, 118 } /* left = h */,
+ { 64, 19, 37, 156, 66, 138, 49, 95, 133 } /* left = dc */,
+ { 53, 21, 36, 131, 63, 163, 60, 109, 81 } /* left = d45 */,
+ { 40, 26, 35, 154, 40, 185, 51, 97, 123 } /* left = d135 */,
+ { 35, 19, 34, 179, 19, 97, 48, 129, 124 } /* left = d117 */,
+ { 36, 20, 26, 136, 62, 164, 33, 77, 154 } /* left = d153 */,
+ { 45, 26, 28, 129, 45, 129, 49, 147, 123 } /* left = d63 */,
+ { 45, 18, 32, 130, 90, 157, 40, 79, 91 } /* left = d27 */,
+ { 38, 44, 51, 136, 74, 162, 57, 97, 121 } /* left = tm */
+ }, { /* above = d153 */
+ { 56, 39, 58, 133, 117, 173, 48, 53, 187 } /* left = v */,
+ { 35, 21, 12, 161, 212, 207, 20, 23, 145 } /* left = h */,
+ { 75, 17, 22, 136, 138, 185, 32, 34, 166 } /* left = dc */,
+ { 56, 29, 19, 117, 109, 181, 55, 68, 112 } /* left = d45 */,
+ { 47, 29, 17, 153, 64, 220, 59, 51, 114 } /* left = d135 */,
+ { 46, 16, 24, 136, 76, 147, 41, 64, 172 } /* left = d117 */,
+ { 34, 17, 11, 108, 152, 187, 13, 15, 209 } /* left = d153 */,
+ { 55, 30, 18, 122, 79, 179, 44, 88, 116 } /* left = d63 */,
+ { 51, 24, 14, 115, 133, 209, 32, 26, 104 } /* left = d27 */,
+ { 37, 49, 25, 129, 168, 164, 41, 54, 148 } /* left = tm */
+ }, { /* above = d63 */
+ { 48, 34, 86, 101, 92, 146, 78, 179, 134 } /* left = v */,
+ { 47, 22, 24, 138, 187, 178, 68, 69, 59 } /* left = h */,
+ { 78, 23, 39, 111, 117, 170, 74, 124, 94 } /* left = dc */,
+ { 56, 25, 33, 105, 112, 187, 95, 177, 129 } /* left = d45 */,
+ { 48, 31, 27, 114, 63, 183, 82, 116, 56 } /* left = d135 */,
+ { 43, 28, 37, 121, 63, 123, 61, 192, 169 } /* left = d117 */,
+ { 42, 17, 24, 109, 97, 177, 56, 76, 122 } /* left = d153 */,
+ { 46, 23, 32, 74, 86, 150, 67, 183, 88 } /* left = d63 */,
+ { 58, 18, 28, 105, 139, 182, 70, 92, 63 } /* left = d27 */,
+ { 36, 38, 48, 92, 122, 165, 88, 137, 91 } /* left = tm */
+ }, { /* above = d27 */
+ { 62, 44, 61, 123, 105, 189, 48, 57, 64 } /* left = v */,
+ { 47, 25, 17, 175, 222, 220, 24, 30, 86 } /* left = h */,
+ { 82, 22, 32, 127, 143, 213, 39, 41, 70 } /* left = dc */,
+ { 68, 36, 17, 106, 102, 206, 59, 74, 74 } /* left = d45 */,
+ { 57, 39, 23, 151, 68, 216, 55, 63, 58 } /* left = d135 */,
+ { 49, 30, 35, 141, 70, 168, 82, 40, 115 } /* left = d117 */,
+ { 51, 25, 15, 136, 129, 202, 38, 35, 139 } /* left = d153 */,
+ { 59, 39, 19, 114, 75, 180, 77, 104, 42 } /* left = d63 */,
+ { 68, 26, 16, 111, 141, 215, 29, 28, 28 } /* left = d27 */,
+ { 40, 61, 26, 126, 152, 206, 61, 59, 93 } /* left = tm */
+ }, { /* above = tm */
+ { 44, 78, 115, 132, 119, 173, 71, 112, 93 } /* left = v */,
+ { 39, 38, 21, 184, 227, 206, 42, 32, 64 } /* left = h */,
+ { 65, 70, 60, 155, 159, 199, 61, 60, 81 } /* left = dc */,
+ { 58, 47, 36, 124, 137, 193, 80, 82, 78 } /* left = d45 */,
+ { 49, 50, 35, 144, 95, 205, 63, 78, 59 } /* left = d135 */,
+ { 41, 53, 52, 148, 71, 142, 65, 128, 51 } /* left = d117 */,
+ { 40, 36, 28, 143, 143, 202, 40, 55, 137 } /* left = d153 */,
+ { 42, 44, 44, 104, 105, 164, 64, 130, 80 } /* left = d63 */,
+ { 52, 34, 29, 129, 183, 227, 42, 35, 43 } /* left = d27 */,
+ { 43, 81, 53, 140, 169, 204, 68, 84, 72 } /* left = tm */
+ }
+};
+
+const uint8_t ff_vp9_default_kf_uvmode_probs[10][9] = {
+ { 118, 15, 123, 148, 131, 101, 44, 93, 131 } /* y = v */,
+ { 113, 12, 23, 188, 226, 142, 26, 32, 125 } /* y = h */,
+ { 144, 11, 54, 157, 195, 130, 46, 58, 108 } /* y = dc */,
+ { 120, 11, 50, 123, 163, 135, 64, 77, 103 } /* y = d45 */,
+ { 113, 9, 36, 155, 111, 157, 32, 44, 161 } /* y = d135 */,
+ { 116, 9, 55, 176, 76, 96, 37, 61, 149 } /* y = d117 */,
+ { 115, 9, 28, 141, 161, 167, 21, 25, 193 } /* y = d153 */,
+ { 116, 12, 64, 120, 140, 125, 49, 115, 121 } /* y = d63 */,
+ { 120, 12, 32, 145, 195, 142, 32, 38, 86 } /* y = d27 */,
+ { 102, 19, 66, 162, 182, 122, 35, 59, 128 } /* y = tm */
+};
+
+const int8_t ff_vp9_inter_mode_tree[3][2] = {
+ { -ZEROMV, 1 }, // '0'
+ { -NEARESTMV, 2 }, // '10'
+ { -NEARMV, -NEWMV }, // '11x'
+};
+
+const int8_t ff_vp9_filter_tree[2][2] = {
+ { -0, 1 }, // '0'
+ { -1, -2 }, // '1x'
+};
+
+const enum FilterMode ff_vp9_filter_lut[3] = {
+ FILTER_8TAP_REGULAR,
+ FILTER_8TAP_SMOOTH,
+ FILTER_8TAP_SHARP,
+};
+
+const int16_t ff_vp9_dc_qlookup[3][256] = {
+ {
+ 4, 8, 8, 9, 10, 11, 12, 12,
+ 13, 14, 15, 16, 17, 18, 19, 19,
+ 20, 21, 22, 23, 24, 25, 26, 26,
+ 27, 28, 29, 30, 31, 32, 32, 33,
+ 34, 35, 36, 37, 38, 38, 39, 40,
+ 41, 42, 43, 43, 44, 45, 46, 47,
+ 48, 48, 49, 50, 51, 52, 53, 53,
+ 54, 55, 56, 57, 57, 58, 59, 60,
+ 61, 62, 62, 63, 64, 65, 66, 66,
+ 67, 68, 69, 70, 70, 71, 72, 73,
+ 74, 74, 75, 76, 77, 78, 78, 79,
+ 80, 81, 81, 82, 83, 84, 85, 85,
+ 87, 88, 90, 92, 93, 95, 96, 98,
+ 99, 101, 102, 104, 105, 107, 108, 110,
+ 111, 113, 114, 116, 117, 118, 120, 121,
+ 123, 125, 127, 129, 131, 134, 136, 138,
+ 140, 142, 144, 146, 148, 150, 152, 154,
+ 156, 158, 161, 164, 166, 169, 172, 174,
+ 177, 180, 182, 185, 187, 190, 192, 195,
+ 199, 202, 205, 208, 211, 214, 217, 220,
+ 223, 226, 230, 233, 237, 240, 243, 247,
+ 250, 253, 257, 261, 265, 269, 272, 276,
+ 280, 284, 288, 292, 296, 300, 304, 309,
+ 313, 317, 322, 326, 330, 335, 340, 344,
+ 349, 354, 359, 364, 369, 374, 379, 384,
+ 389, 395, 400, 406, 411, 417, 423, 429,
+ 435, 441, 447, 454, 461, 467, 475, 482,
+ 489, 497, 505, 513, 522, 530, 539, 549,
+ 559, 569, 579, 590, 602, 614, 626, 640,
+ 654, 668, 684, 700, 717, 736, 755, 775,
+ 796, 819, 843, 869, 896, 925, 955, 988,
+ 1022, 1058, 1098, 1139, 1184, 1232, 1282, 1336,
+ }, {
+ 4, 9, 10, 13, 15, 17, 20, 22,
+ 25, 28, 31, 34, 37, 40, 43, 47,
+ 50, 53, 57, 60, 64, 68, 71, 75,
+ 78, 82, 86, 90, 93, 97, 101, 105,
+ 109, 113, 116, 120, 124, 128, 132, 136,
+ 140, 143, 147, 151, 155, 159, 163, 166,
+ 170, 174, 178, 182, 185, 189, 193, 197,
+ 200, 204, 208, 212, 215, 219, 223, 226,
+ 230, 233, 237, 241, 244, 248, 251, 255,
+ 259, 262, 266, 269, 273, 276, 280, 283,
+ 287, 290, 293, 297, 300, 304, 307, 310,
+ 314, 317, 321, 324, 327, 331, 334, 337,
+ 343, 350, 356, 362, 369, 375, 381, 387,
+ 394, 400, 406, 412, 418, 424, 430, 436,
+ 442, 448, 454, 460, 466, 472, 478, 484,
+ 490, 499, 507, 516, 525, 533, 542, 550,
+ 559, 567, 576, 584, 592, 601, 609, 617,
+ 625, 634, 644, 655, 666, 676, 687, 698,
+ 708, 718, 729, 739, 749, 759, 770, 782,
+ 795, 807, 819, 831, 844, 856, 868, 880,
+ 891, 906, 920, 933, 947, 961, 975, 988,
+ 1001, 1015, 1030, 1045, 1061, 1076, 1090, 1105,
+ 1120, 1137, 1153, 1170, 1186, 1202, 1218, 1236,
+ 1253, 1271, 1288, 1306, 1323, 1342, 1361, 1379,
+ 1398, 1416, 1436, 1456, 1476, 1496, 1516, 1537,
+ 1559, 1580, 1601, 1624, 1647, 1670, 1692, 1717,
+ 1741, 1766, 1791, 1817, 1844, 1871, 1900, 1929,
+ 1958, 1990, 2021, 2054, 2088, 2123, 2159, 2197,
+ 2236, 2276, 2319, 2363, 2410, 2458, 2508, 2561,
+ 2616, 2675, 2737, 2802, 2871, 2944, 3020, 3102,
+ 3188, 3280, 3375, 3478, 3586, 3702, 3823, 3953,
+ 4089, 4236, 4394, 4559, 4737, 4929, 5130, 5347,
+ }, {
+ 4, 12, 18, 25, 33, 41, 50, 60,
+ 70, 80, 91, 103, 115, 127, 140, 153,
+ 166, 180, 194, 208, 222, 237, 251, 266,
+ 281, 296, 312, 327, 343, 358, 374, 390,
+ 405, 421, 437, 453, 469, 484, 500, 516,
+ 532, 548, 564, 580, 596, 611, 627, 643,
+ 659, 674, 690, 706, 721, 737, 752, 768,
+ 783, 798, 814, 829, 844, 859, 874, 889,
+ 904, 919, 934, 949, 964, 978, 993, 1008,
+ 1022, 1037, 1051, 1065, 1080, 1094, 1108, 1122,
+ 1136, 1151, 1165, 1179, 1192, 1206, 1220, 1234,
+ 1248, 1261, 1275, 1288, 1302, 1315, 1329, 1342,
+ 1368, 1393, 1419, 1444, 1469, 1494, 1519, 1544,
+ 1569, 1594, 1618, 1643, 1668, 1692, 1717, 1741,
+ 1765, 1789, 1814, 1838, 1862, 1885, 1909, 1933,
+ 1957, 1992, 2027, 2061, 2096, 2130, 2165, 2199,
+ 2233, 2267, 2300, 2334, 2367, 2400, 2434, 2467,
+ 2499, 2532, 2575, 2618, 2661, 2704, 2746, 2788,
+ 2830, 2872, 2913, 2954, 2995, 3036, 3076, 3127,
+ 3177, 3226, 3275, 3324, 3373, 3421, 3469, 3517,
+ 3565, 3621, 3677, 3733, 3788, 3843, 3897, 3951,
+ 4005, 4058, 4119, 4181, 4241, 4301, 4361, 4420,
+ 4479, 4546, 4612, 4677, 4742, 4807, 4871, 4942,
+ 5013, 5083, 5153, 5222, 5291, 5367, 5442, 5517,
+ 5591, 5665, 5745, 5825, 5905, 5984, 6063, 6149,
+ 6234, 6319, 6404, 6495, 6587, 6678, 6769, 6867,
+ 6966, 7064, 7163, 7269, 7376, 7483, 7599, 7715,
+ 7832, 7958, 8085, 8214, 8352, 8492, 8635, 8788,
+ 8945, 9104, 9275, 9450, 9639, 9832, 10031, 10245,
+ 10465, 10702, 10946, 11210, 11482, 11776, 12081, 12409,
+ 12750, 13118, 13501, 13913, 14343, 14807, 15290, 15812,
+ 16356, 16943, 17575, 18237, 18949, 19718, 20521, 21387,
+ }
+};
+
+const int16_t ff_vp9_ac_qlookup[3][256] = {
+ {
+ 4, 8, 9, 10, 11, 12, 13, 14,
+ 15, 16, 17, 18, 19, 20, 21, 22,
+ 23, 24, 25, 26, 27, 28, 29, 30,
+ 31, 32, 33, 34, 35, 36, 37, 38,
+ 39, 40, 41, 42, 43, 44, 45, 46,
+ 47, 48, 49, 50, 51, 52, 53, 54,
+ 55, 56, 57, 58, 59, 60, 61, 62,
+ 63, 64, 65, 66, 67, 68, 69, 70,
+ 71, 72, 73, 74, 75, 76, 77, 78,
+ 79, 80, 81, 82, 83, 84, 85, 86,
+ 87, 88, 89, 90, 91, 92, 93, 94,
+ 95, 96, 97, 98, 99, 100, 101, 102,
+ 104, 106, 108, 110, 112, 114, 116, 118,
+ 120, 122, 124, 126, 128, 130, 132, 134,
+ 136, 138, 140, 142, 144, 146, 148, 150,
+ 152, 155, 158, 161, 164, 167, 170, 173,
+ 176, 179, 182, 185, 188, 191, 194, 197,
+ 200, 203, 207, 211, 215, 219, 223, 227,
+ 231, 235, 239, 243, 247, 251, 255, 260,
+ 265, 270, 275, 280, 285, 290, 295, 300,
+ 305, 311, 317, 323, 329, 335, 341, 347,
+ 353, 359, 366, 373, 380, 387, 394, 401,
+ 408, 416, 424, 432, 440, 448, 456, 465,
+ 474, 483, 492, 501, 510, 520, 530, 540,
+ 550, 560, 571, 582, 593, 604, 615, 627,
+ 639, 651, 663, 676, 689, 702, 715, 729,
+ 743, 757, 771, 786, 801, 816, 832, 848,
+ 864, 881, 898, 915, 933, 951, 969, 988,
+ 1007, 1026, 1046, 1066, 1087, 1108, 1129, 1151,
+ 1173, 1196, 1219, 1243, 1267, 1292, 1317, 1343,
+ 1369, 1396, 1423, 1451, 1479, 1508, 1537, 1567,
+ 1597, 1628, 1660, 1692, 1725, 1759, 1793, 1828,
+ }, {
+ 4, 9, 11, 13, 16, 18, 21, 24,
+ 27, 30, 33, 37, 40, 44, 48, 51,
+ 55, 59, 63, 67, 71, 75, 79, 83,
+ 88, 92, 96, 100, 105, 109, 114, 118,
+ 122, 127, 131, 136, 140, 145, 149, 154,
+ 158, 163, 168, 172, 177, 181, 186, 190,
+ 195, 199, 204, 208, 213, 217, 222, 226,
+ 231, 235, 240, 244, 249, 253, 258, 262,
+ 267, 271, 275, 280, 284, 289, 293, 297,
+ 302, 306, 311, 315, 319, 324, 328, 332,
+ 337, 341, 345, 349, 354, 358, 362, 367,
+ 371, 375, 379, 384, 388, 392, 396, 401,
+ 409, 417, 425, 433, 441, 449, 458, 466,
+ 474, 482, 490, 498, 506, 514, 523, 531,
+ 539, 547, 555, 563, 571, 579, 588, 596,
+ 604, 616, 628, 640, 652, 664, 676, 688,
+ 700, 713, 725, 737, 749, 761, 773, 785,
+ 797, 809, 825, 841, 857, 873, 889, 905,
+ 922, 938, 954, 970, 986, 1002, 1018, 1038,
+ 1058, 1078, 1098, 1118, 1138, 1158, 1178, 1198,
+ 1218, 1242, 1266, 1290, 1314, 1338, 1362, 1386,
+ 1411, 1435, 1463, 1491, 1519, 1547, 1575, 1603,
+ 1631, 1663, 1695, 1727, 1759, 1791, 1823, 1859,
+ 1895, 1931, 1967, 2003, 2039, 2079, 2119, 2159,
+ 2199, 2239, 2283, 2327, 2371, 2415, 2459, 2507,
+ 2555, 2603, 2651, 2703, 2755, 2807, 2859, 2915,
+ 2971, 3027, 3083, 3143, 3203, 3263, 3327, 3391,
+ 3455, 3523, 3591, 3659, 3731, 3803, 3876, 3952,
+ 4028, 4104, 4184, 4264, 4348, 4432, 4516, 4604,
+ 4692, 4784, 4876, 4972, 5068, 5168, 5268, 5372,
+ 5476, 5584, 5692, 5804, 5916, 6032, 6148, 6268,
+ 6388, 6512, 6640, 6768, 6900, 7036, 7172, 7312,
+ }, {
+ 4, 13, 19, 27, 35, 44, 54, 64,
+ 75, 87, 99, 112, 126, 139, 154, 168,
+ 183, 199, 214, 230, 247, 263, 280, 297,
+ 314, 331, 349, 366, 384, 402, 420, 438,
+ 456, 475, 493, 511, 530, 548, 567, 586,
+ 604, 623, 642, 660, 679, 698, 716, 735,
+ 753, 772, 791, 809, 828, 846, 865, 884,
+ 902, 920, 939, 957, 976, 994, 1012, 1030,
+ 1049, 1067, 1085, 1103, 1121, 1139, 1157, 1175,
+ 1193, 1211, 1229, 1246, 1264, 1282, 1299, 1317,
+ 1335, 1352, 1370, 1387, 1405, 1422, 1440, 1457,
+ 1474, 1491, 1509, 1526, 1543, 1560, 1577, 1595,
+ 1627, 1660, 1693, 1725, 1758, 1791, 1824, 1856,
+ 1889, 1922, 1954, 1987, 2020, 2052, 2085, 2118,
+ 2150, 2183, 2216, 2248, 2281, 2313, 2346, 2378,
+ 2411, 2459, 2508, 2556, 2605, 2653, 2701, 2750,
+ 2798, 2847, 2895, 2943, 2992, 3040, 3088, 3137,
+ 3185, 3234, 3298, 3362, 3426, 3491, 3555, 3619,
+ 3684, 3748, 3812, 3876, 3941, 4005, 4069, 4149,
+ 4230, 4310, 4390, 4470, 4550, 4631, 4711, 4791,
+ 4871, 4967, 5064, 5160, 5256, 5352, 5448, 5544,
+ 5641, 5737, 5849, 5961, 6073, 6185, 6297, 6410,
+ 6522, 6650, 6778, 6906, 7034, 7162, 7290, 7435,
+ 7579, 7723, 7867, 8011, 8155, 8315, 8475, 8635,
+ 8795, 8956, 9132, 9308, 9484, 9660, 9836, 10028,
+ 10220, 10412, 10604, 10812, 11020, 11228, 11437, 11661,
+ 11885, 12109, 12333, 12573, 12813, 13053, 13309, 13565,
+ 13821, 14093, 14365, 14637, 14925, 15213, 15502, 15806,
+ 16110, 16414, 16734, 17054, 17390, 17726, 18062, 18414,
+ 18766, 19134, 19502, 19886, 20270, 20670, 21070, 21486,
+ 21902, 22334, 22766, 23214, 23662, 24126, 24590, 25070,
+ 25551, 26047, 26559, 27071, 27599, 28143, 28687, 29247,
+ }
+};
+
+const enum TxfmType ff_vp9_intra_txfm_type[14] = {
+ [VERT_PRED] = ADST_DCT,
+ [HOR_PRED] = DCT_ADST,
+ [DC_PRED] = DCT_DCT,
+ [DIAG_DOWN_LEFT_PRED] = DCT_DCT,
+ [DIAG_DOWN_RIGHT_PRED] = ADST_ADST,
+ [VERT_RIGHT_PRED] = ADST_DCT,
+ [HOR_DOWN_PRED] = DCT_ADST,
+ [VERT_LEFT_PRED] = ADST_DCT,
+ [HOR_UP_PRED] = DCT_ADST,
+ [TM_VP8_PRED] = ADST_ADST,
+ [NEARESTMV] = DCT_DCT,
+ [NEARMV] = DCT_DCT,
+ [ZEROMV] = DCT_DCT,
+ [NEWMV] = DCT_DCT,
+};
+
+const int16_t ff_vp9_default_scan_4x4[16] = {
+ 0, 1, 4, 5,
+ 2, 8, 3, 6,
+ 12, 9, 7, 10,
+ 13, 11, 14, 15,
+};
+
+const int16_t ff_vp9_col_scan_4x4[16] = {
+ 0, 1, 2, 4,
+ 3, 5, 6, 8,
+ 7, 9, 10, 12,
+ 13, 11, 14, 15,
+};
+
+const int16_t ff_vp9_row_scan_4x4[16] = {
+ 0, 4, 1, 8,
+ 5, 12, 9, 2,
+ 6, 13, 3, 10,
+ 7, 14, 11, 15,
+};
+
+const int16_t ff_vp9_default_scan_8x8[64] = {
+ 0, 1, 8, 2, 9, 16, 10, 3,
+ 17, 24, 18, 11, 4, 25, 32, 19,
+ 12, 26, 5, 33, 20, 27, 40, 13,
+ 34, 6, 41, 28, 21, 35, 42, 48,
+ 14, 7, 36, 29, 43, 56, 49, 22,
+ 15, 37, 50, 44, 57, 30, 23, 51,
+ 45, 58, 38, 31, 52, 59, 39, 46,
+ 53, 60, 47, 54, 61, 55, 62, 63,
+};
+
+const int16_t ff_vp9_col_scan_8x8[64] = {
+ 0, 1, 2, 8, 3, 9, 4, 10,
+ 16, 5, 11, 17, 12, 18, 6, 24,
+ 19, 13, 25, 7, 26, 20, 32, 14,
+ 27, 21, 33, 28, 34, 15, 22, 35,
+ 40, 29, 41, 36, 23, 30, 42, 37,
+ 48, 43, 31, 44, 49, 38, 50, 56,
+ 45, 39, 51, 57, 52, 46, 58, 53,
+ 59, 47, 60, 54, 61, 55, 62, 63,
+};
+
+const int16_t ff_vp9_row_scan_8x8[64] = {
+ 0, 8, 16, 1, 9, 24, 2, 17,
+ 32, 10, 25, 3, 40, 18, 11, 33,
+ 26, 19, 4, 48, 41, 34, 12, 27,
+ 56, 20, 5, 42, 35, 13, 49, 28,
+ 6, 21, 43, 36, 14, 50, 29, 57,
+ 7, 44, 22, 37, 51, 15, 58, 30,
+ 23, 45, 52, 38, 59, 31, 46, 53,
+ 39, 60, 47, 61, 54, 62, 55, 63,
+};
+
+const int16_t ff_vp9_default_scan_16x16[256] = {
+ 0, 1, 16, 2, 17, 32, 3, 18, 33, 48, 4, 34, 19, 49, 20, 5,
+ 35, 64, 50, 36, 65, 21, 6, 51, 80, 66, 37, 22, 52, 7, 81, 67,
+ 38, 82, 53, 23, 96, 68, 8, 83, 97, 54, 39, 69, 112, 24, 98, 84,
+ 70, 55, 9, 40, 85, 99, 113, 128, 25, 114, 100, 71, 86, 56, 10, 41,
+ 115, 101, 129, 116, 72, 87, 26, 130, 144, 102, 57, 11, 42, 117, 131, 145,
+ 88, 103, 27, 73, 132, 118, 146, 58, 160, 12, 43, 133, 147, 104, 89, 119,
+ 161, 74, 148, 134, 28, 162, 59, 13, 176, 120, 149, 90, 135, 105, 163, 44,
+ 75, 177, 164, 29, 150, 121, 136, 178, 165, 14, 106, 60, 91, 151, 45, 179,
+ 192, 137, 166, 122, 76, 180, 152, 30, 61, 15, 107, 167, 181, 193, 92, 208,
+ 46, 138, 123, 153, 194, 77, 168, 182, 31, 195, 209, 183, 108, 139, 62, 154,
+ 47, 196, 93, 169, 210, 197, 224, 124, 184, 211, 78, 109, 170, 155, 63, 198,
+ 212, 185, 225, 240, 140, 94, 199, 125, 79, 213, 226, 171, 186, 156, 214, 200,
+ 110, 227, 141, 95, 241, 215, 228, 201, 126, 242, 187, 172, 157, 229, 111, 216,
+ 243, 142, 202, 230, 127, 217, 244, 173, 188, 231, 158, 203, 143, 245, 218, 232,
+ 189, 246, 159, 174, 233, 247, 219, 204, 175, 190, 248, 234, 205, 220, 249, 191,
+ 235, 221, 250, 206, 222, 251, 236, 207, 237, 223, 252, 238, 253, 239, 254, 255,
+};
+
+const int16_t ff_vp9_col_scan_16x16[256] = {
+ 0, 1, 2, 3, 16, 4, 17, 5, 18, 6, 19, 32, 20, 7, 33, 21,
+ 34, 8, 35, 22, 48, 36, 9, 49, 23, 50, 37, 10, 38, 51, 24, 64,
+ 52, 11, 65, 39, 25, 53, 66, 54, 40, 67, 12, 80, 26, 68, 55, 81,
+ 41, 69, 13, 27, 82, 56, 70, 83, 42, 14, 84, 96, 71, 28, 57, 85,
+ 97, 15, 72, 98, 43, 86, 58, 99, 29, 87, 100, 112, 73, 44, 101, 59,
+ 30, 113, 88, 114, 74, 128, 102, 45, 31, 115, 60, 103, 89, 116, 75, 129,
+ 117, 46, 104, 90, 61, 130, 118, 131, 132, 105, 76, 47, 119, 144, 91, 62,
+ 133, 106, 145, 120, 146, 134, 77, 147, 121, 92, 135, 148, 63, 107, 136, 122,
+ 93, 149, 160, 78, 150, 137, 108, 161, 162, 151, 123, 79, 138, 163, 152, 94,
+ 164, 109, 165, 153, 124, 139, 176, 166, 95, 177, 167, 110, 154, 178, 125, 179,
+ 140, 168, 155, 111, 180, 192, 181, 169, 141, 126, 182, 193, 194, 156, 183, 170,
+ 195, 127, 142, 196, 184, 208, 197, 157, 171, 143, 185, 198, 209, 199, 210, 172,
+ 158, 186, 211, 224, 212, 200, 240, 159, 213, 225, 187, 201, 173, 226, 214, 215,
+ 227, 202, 228, 188, 241, 216, 174, 229, 242, 203, 243, 217, 230, 175, 189, 244,
+ 231, 204, 218, 232, 245, 219, 246, 190, 233, 205, 191, 247, 234, 248, 220, 206,
+ 249, 235, 221, 207, 250, 236, 222, 251, 223, 237, 238, 252, 239, 253, 254, 255,
+};
+
+const int16_t ff_vp9_row_scan_16x16[256] = {
+ 0, 16, 32, 1, 48, 17, 64, 33, 2, 80, 18, 49, 96, 34, 3, 65,
+ 19, 112, 50, 81, 35, 4, 128, 66, 20, 97, 51, 82, 5, 144, 36, 67,
+ 113, 98, 21, 52, 160, 83, 129, 37, 68, 6, 114, 176, 99, 53, 22, 84,
+ 145, 38, 69, 130, 7, 115, 192, 100, 54, 23, 85, 161, 146, 131, 39, 70,
+ 208, 116, 8, 101, 177, 55, 86, 24, 162, 147, 132, 71, 224, 117, 40, 102,
+ 9, 148, 56, 87, 193, 163, 240, 133, 178, 25, 118, 72, 41, 103, 164, 10,
+ 149, 88, 134, 209, 179, 57, 119, 194, 26, 73, 165, 150, 104, 42, 135, 11,
+ 180, 120, 89, 225, 195, 58, 27, 210, 151, 181, 166, 74, 43, 105, 12, 136,
+ 90, 59, 241, 121, 28, 196, 167, 211, 152, 44, 182, 137, 75, 13, 226, 106,
+ 122, 60, 197, 91, 168, 29, 183, 153, 14, 76, 212, 138, 45, 107, 15, 198,
+ 92, 227, 169, 30, 123, 154, 61, 242, 184, 213, 139, 46, 77, 31, 108, 170,
+ 199, 185, 124, 228, 93, 155, 214, 62, 140, 243, 78, 47, 200, 109, 186, 171,
+ 201, 94, 63, 215, 229, 156, 79, 125, 141, 110, 216, 187, 172, 244, 202, 230,
+ 217, 95, 157, 126, 245, 111, 142, 231, 188, 127, 158, 218, 173, 232, 246, 233,
+ 203, 143, 247, 174, 189, 159, 219, 204, 248, 234, 249, 175, 190, 220, 205, 250,
+ 235, 191, 221, 251, 236, 206, 252, 222, 207, 237, 223, 253, 238, 254, 239, 255,
+};
+
+const int16_t ff_vp9_default_scan_32x32[1024] = {
+ 0, 1, 32, 2, 33, 64, 3, 34, 65, 4, 96, 35, 66, 5, 36, 97, 67, 128, 98, 68, 37, 6, 129, 99, 7, 160, 69, 38, 130, 100, 161, 131,
+ 39, 70, 8, 101, 162, 132, 192, 71, 40, 9, 102, 163, 133, 193, 72, 224, 103, 41, 164, 10, 194, 134, 165, 73, 104, 135, 225, 42, 195, 11, 256, 166,
+ 226, 196, 74, 105, 136, 43, 12, 167, 197, 227, 257, 75, 106, 137, 228, 44, 198, 168, 258, 288, 13, 229, 76, 107, 199, 138, 259, 169, 289, 45, 230, 260,
+ 200, 108, 14, 170, 139, 320, 290, 77, 231, 261, 46, 201, 140, 291, 109, 232, 321, 262, 171, 78, 292, 15, 322, 202, 263, 352, 172, 293, 233, 141, 323, 110,
+ 47, 203, 264, 234, 294, 353, 324, 16, 79, 204, 265, 295, 325, 173, 354, 142, 235, 384, 48, 296, 111, 266, 355, 326, 80, 17, 205, 236, 174, 356, 385, 327,
+ 143, 297, 267, 357, 386, 112, 49, 328, 298, 206, 416, 237, 358, 387, 81, 175, 18, 329, 359, 388, 299, 330, 389, 113, 417, 238, 360, 50, 207, 418, 390, 331,
+ 19, 448, 361, 82, 419, 391, 239, 51, 362, 420, 114, 449, 480, 421, 83, 363, 450, 422, 512, 451, 423, 115, 452, 481, 453, 482, 454, 544, 483, 455, 513, 484,
+ 514, 485, 515, 486, 545, 576, 487, 546, 547, 608, 577, 578, 579, 609, 610, 611, 20, 144, 268, 392, 516, 640, 21, 52, 145, 176, 269, 300, 393, 424, 517, 548,
+ 641, 672, 22, 53, 84, 146, 177, 208, 270, 301, 332, 394, 425, 456, 518, 549, 580, 642, 673, 704, 23, 54, 85, 116, 147, 178, 209, 240, 271, 302, 333, 364,
+ 395, 426, 457, 488, 519, 550, 581, 612, 643, 674, 705, 736, 55, 86, 117, 179, 210, 241, 303, 334, 365, 427, 458, 489, 551, 582, 613, 675, 706, 737, 87, 118,
+ 211, 242, 335, 366, 459, 490, 583, 614, 707, 738, 119, 243, 367, 491, 615, 739, 24, 148, 272, 396, 520, 644, 768, 25, 56, 149, 180, 273, 304, 397, 428, 521,
+ 552, 645, 676, 769, 800, 26, 57, 88, 150, 181, 212, 274, 305, 336, 398, 429, 460, 522, 553, 584, 646, 677, 708, 770, 801, 832, 27, 58, 89, 120, 151, 182,
+ 213, 244, 275, 306, 337, 368, 399, 430, 461, 492, 523, 554, 585, 616, 647, 678, 709, 740, 771, 802, 833, 864, 59, 90, 121, 183, 214, 245, 307, 338, 369, 431,
+ 462, 493, 555, 586, 617, 679, 710, 741, 803, 834, 865, 91, 122, 215, 246, 339, 370, 463, 494, 587, 618, 711, 742, 835, 866, 123, 247, 371, 495, 619, 743, 867,
+ 28, 152, 276, 400, 524, 648, 772, 896, 29, 60, 153, 184, 277, 308, 401, 432, 525, 556, 649, 680, 773, 804, 897, 928, 30, 61, 92, 154, 185, 216, 278, 309,
+ 340, 402, 433, 464, 526, 557, 588, 650, 681, 712, 774, 805, 836, 898, 929, 960, 31, 62, 93, 124, 155, 186, 217, 248, 279, 310, 341, 372, 403, 434, 465, 496,
+ 527, 558, 589, 620, 651, 682, 713, 744, 775, 806, 837, 868, 899, 930, 961, 992, 63, 94, 125, 187, 218, 249, 311, 342, 373, 435, 466, 497, 559, 590, 621, 683,
+ 714, 745, 807, 838, 869, 931, 962, 993, 95, 126, 219, 250, 343, 374, 467, 498, 591, 622, 715, 746, 839, 870, 963, 994, 127, 251, 375, 499, 623, 747, 871, 995,
+ 156, 280, 404, 528, 652, 776, 900, 157, 188, 281, 312, 405, 436, 529, 560, 653, 684, 777, 808, 901, 932, 158, 189, 220, 282, 313, 344, 406, 437, 468, 530, 561,
+ 592, 654, 685, 716, 778, 809, 840, 902, 933, 964, 159, 190, 221, 252, 283, 314, 345, 376, 407, 438, 469, 500, 531, 562, 593, 624, 655, 686, 717, 748, 779, 810,
+ 841, 872, 903, 934, 965, 996, 191, 222, 253, 315, 346, 377, 439, 470, 501, 563, 594, 625, 687, 718, 749, 811, 842, 873, 935, 966, 997, 223, 254, 347, 378, 471,
+ 502, 595, 626, 719, 750, 843, 874, 967, 998, 255, 379, 503, 627, 751, 875, 999, 284, 408, 532, 656, 780, 904, 285, 316, 409, 440, 533, 564, 657, 688, 781, 812,
+ 905, 936, 286, 317, 348, 410, 441, 472, 534, 565, 596, 658, 689, 720, 782, 813, 844, 906, 937, 968, 287, 318, 349, 380, 411, 442, 473, 504, 535, 566, 597, 628,
+ 659, 690, 721, 752, 783, 814, 845, 876, 907, 938, 969, 1000, 319, 350, 381, 443, 474, 505, 567, 598, 629, 691, 722, 753, 815, 846, 877, 939, 970, 1001, 351, 382,
+ 475, 506, 599, 630, 723, 754, 847, 878, 971, 1002, 383, 507, 631, 755, 879, 1003, 412, 536, 660, 784, 908, 413, 444, 537, 568, 661, 692, 785, 816, 909, 940, 414,
+ 445, 476, 538, 569, 600, 662, 693, 724, 786, 817, 848, 910, 941, 972, 415, 446, 477, 508, 539, 570, 601, 632, 663, 694, 725, 756, 787, 818, 849, 880, 911, 942,
+ 973, 1004, 447, 478, 509, 571, 602, 633, 695, 726, 757, 819, 850, 881, 943, 974, 1005, 479, 510, 603, 634, 727, 758, 851, 882, 975, 1006, 511, 635, 759, 883, 1007,
+ 540, 664, 788, 912, 541, 572, 665, 696, 789, 820, 913, 944, 542, 573, 604, 666, 697, 728, 790, 821, 852, 914, 945, 976, 543, 574, 605, 636, 667, 698, 729, 760,
+ 791, 822, 853, 884, 915, 946, 977, 1008, 575, 606, 637, 699, 730, 761, 823, 854, 885, 947, 978, 1009, 607, 638, 731, 762, 855, 886, 979, 1010, 639, 763, 887, 1011,
+ 668, 792, 916, 669, 700, 793, 824, 917, 948, 670, 701, 732, 794, 825, 856, 918, 949, 980, 671, 702, 733, 764, 795, 826, 857, 888, 919, 950, 981, 1012, 703, 734,
+ 765, 827, 858, 889, 951, 982, 1013, 735, 766, 859, 890, 983, 1014, 767, 891, 1015, 796, 920, 797, 828, 921, 952, 798, 829, 860, 922, 953, 984, 799, 830, 861, 892,
+ 923, 954, 985, 1016, 831, 862, 893, 955, 986, 1017, 863, 894, 987, 1018, 895, 1019, 924, 925, 956, 926, 957, 988, 927, 958, 989, 1020, 959, 990, 1021, 991, 1022, 1023,
+};
+
+const int16_t * const ff_vp9_scans[5][4] = {
+ {
+ ff_vp9_default_scan_4x4, ff_vp9_col_scan_4x4,
+ ff_vp9_row_scan_4x4, ff_vp9_default_scan_4x4
+ }, {
+ ff_vp9_default_scan_8x8, ff_vp9_col_scan_8x8,
+ ff_vp9_row_scan_8x8, ff_vp9_default_scan_8x8
+ }, {
+ ff_vp9_default_scan_16x16, ff_vp9_col_scan_16x16,
+ ff_vp9_row_scan_16x16, ff_vp9_default_scan_16x16
+ }, {
+ ff_vp9_default_scan_32x32, ff_vp9_default_scan_32x32,
+ ff_vp9_default_scan_32x32, ff_vp9_default_scan_32x32
+ }, { // lossless
+ ff_vp9_default_scan_4x4, ff_vp9_default_scan_4x4,
+ ff_vp9_default_scan_4x4, ff_vp9_default_scan_4x4
+ }
+};
+
+const int16_t ff_vp9_default_scan_4x4_nb[16][2] = {
+ { 0, 0 }, { 0, 0 }, { 4, 1 }, { 1, 1 },
+ { 4, 4 }, { 2, 2 }, { 5, 2 }, { 8, 8 },
+ { 8, 5 }, { 6, 3 }, { 9, 6 }, { 12, 9 },
+ { 10, 7 }, { 13, 10 }, { 14, 11 }, { 0, 0 },
+};
+
+const int16_t ff_vp9_col_scan_4x4_nb[16][2] = {
+ { 0, 0 }, { 1, 1 }, { 0, 0 }, { 2, 2 },
+ { 4, 4 }, { 5, 5 }, { 4, 4 }, { 6, 6 },
+ { 8, 8 }, { 9, 9 }, { 8, 8 }, { 12, 12 },
+ { 10, 10 }, { 13, 13 }, { 14, 14 }, { 0, 0 },
+};
+
+const int16_t ff_vp9_row_scan_4x4_nb[16][2] = {
+ { 0, 0 }, { 0, 0 }, { 4, 4 }, { 1, 1 },
+ { 8, 8 }, { 5, 5 }, { 1, 1 }, { 2, 2 },
+ { 9, 9 }, { 2, 2 }, { 6, 6 }, { 3, 3 },
+ { 10, 10 }, { 7, 7 }, { 11, 11 }, { 0, 0 },
+};
+
+const int16_t ff_vp9_default_scan_8x8_nb[64][2] = {
+ { 0, 0 }, { 0, 0 }, { 1, 1 }, { 8, 1 },
+ { 8, 8 }, { 9, 2 }, { 2, 2 }, { 16, 9 },
+ { 16, 16 }, { 17, 10 }, { 10, 3 }, { 3, 3 },
+ { 24, 17 }, { 24, 24 }, { 18, 11 }, { 11, 4 },
+ { 25, 18 }, { 4, 4 }, { 32, 25 }, { 19, 12 },
+ { 26, 19 }, { 32, 32 }, { 12, 5 }, { 33, 26 },
+ { 5, 5 }, { 40, 33 }, { 27, 20 }, { 20, 13 },
+ { 34, 27 }, { 41, 34 }, { 40, 40 }, { 13, 6 },
+ { 6, 6 }, { 35, 28 }, { 28, 21 }, { 42, 35 },
+ { 48, 48 }, { 48, 41 }, { 21, 14 }, { 14, 7 },
+ { 36, 29 }, { 49, 42 }, { 43, 36 }, { 56, 49 },
+ { 29, 22 }, { 22, 15 }, { 50, 43 }, { 44, 37 },
+ { 57, 50 }, { 37, 30 }, { 30, 23 }, { 51, 44 },
+ { 58, 51 }, { 38, 31 }, { 45, 38 }, { 52, 45 },
+ { 59, 52 }, { 46, 39 }, { 53, 46 }, { 60, 53 },
+ { 54, 47 }, { 61, 54 }, { 62, 55 }, { 0, 0 },
+};
+
+const int16_t ff_vp9_col_scan_8x8_nb[64][2] = {
+ { 0, 0 }, { 1, 1 }, { 0, 0 }, { 2, 2 },
+ { 8, 8 }, { 3, 3 }, { 9, 9 }, { 8, 8 },
+ { 4, 4 }, { 10, 10 }, { 16, 16 }, { 11, 11 },
+ { 17, 17 }, { 5, 5 }, { 16, 16 }, { 18, 18 },
+ { 12, 12 }, { 24, 24 }, { 6, 6 }, { 25, 25 },
projects / ffmpeg.git / commitdiff