diff mbox series

[FFmpeg-devel,v6,07/14] vvcdec: add inter prediction

Message ID TYSPR06MB64333C21333A8CE90470A5B9AA85A@TYSPR06MB6433.apcprd06.prod.outlook.com
State Superseded
Headers show
Series [FFmpeg-devel,v6,01/14] vvcdec: add vvc decoder stub | expand

Checks

Context Check Description
yinshiyou/make_loongarch64 success Make finished
yinshiyou/make_fate_loongarch64 success Make fate finished
andriy/make_x86 success Make finished
andriy/make_fate_x86 success Make fate finished

Commit Message

Nuo Mi Dec. 5, 2023, 2:45 p.m. UTC
---
 libavcodec/vvc/Makefile             |    1 +
 libavcodec/vvc/vvc_inter.c          |  939 ++++++++++++++++++++++++
 libavcodec/vvc/vvc_inter.h          |   42 ++
 libavcodec/vvc/vvc_inter_template.c | 1023 +++++++++++++++++++++++++++
 libavcodec/vvc/vvcdec.h             |    5 +
 libavcodec/vvc/vvcdsp.h             |  170 +++++
 6 files changed, 2180 insertions(+)
 create mode 100644 libavcodec/vvc/vvc_inter.c
 create mode 100644 libavcodec/vvc/vvc_inter.h
 create mode 100644 libavcodec/vvc/vvc_inter_template.c
 create mode 100644 libavcodec/vvc/vvcdsp.h
diff mbox series

Patch

diff --git a/libavcodec/vvc/Makefile b/libavcodec/vvc/Makefile
index 912e9f516c..61b1ab39de 100644
--- a/libavcodec/vvc/Makefile
+++ b/libavcodec/vvc/Makefile
@@ -5,6 +5,7 @@  OBJS-$(CONFIG_VVC_DECODER)          +=  vvc/vvcdec.o            \
                                         vvc/vvc_cabac.o         \
                                         vvc/vvc_ctu.o           \
                                         vvc/vvc_data.o          \
+                                        vvc/vvc_inter.o         \
                                         vvc/vvc_mvs.o           \
                                         vvc/vvc_ps.o            \
                                         vvc/vvc_refs.o          \
diff --git a/libavcodec/vvc/vvc_inter.c b/libavcodec/vvc/vvc_inter.c
new file mode 100644
index 0000000000..c494089c6b
--- /dev/null
+++ b/libavcodec/vvc/vvc_inter.c
@@ -0,0 +1,939 @@ 
+/*
+ * VVC inter prediction
+ *
+ * Copyright (C) 2022 Nuo Mi
+ *
+ * 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/frame.h"
+
+#include "vvc_data.h"
+#include "vvc_inter.h"
+#include "vvc_mvs.h"
+#include "vvc_refs.h"
+
+// +1 is enough, + 32 for asm alignment
+#define PROF_TEMP_OFFSET (MAX_PB_SIZE + 32)
+static const int bcw_w_lut[] = {4, 5, 3, 10, -2};
+
+static int emulated_edge(const VVCFrameContext *fc, uint8_t *dst, const uint8_t **src, ptrdiff_t *src_stride,
+    const int x_off, const int y_off, const int block_w, const int block_h, const int is_luma)
+{
+    const int extra_before = is_luma ? LUMA_EXTRA_BEFORE : CHROMA_EXTRA_BEFORE;
+    const int extra_after  = is_luma ? LUMA_EXTRA_AFTER : CHROMA_EXTRA_AFTER;
+    const int extra        = is_luma ? LUMA_EXTRA : CHROMA_EXTRA;
+    const int pic_width    = is_luma ? fc->ps.pps->width  : (fc->ps.pps->width >> fc->ps.sps->hshift[1]);
+    const int pic_height   = is_luma ? fc->ps.pps->height : (fc->ps.pps->height >> fc->ps.sps->vshift[1]);
+
+    if (x_off < extra_before || y_off < extra_before ||
+        x_off >= pic_width - block_w - extra_after ||
+        y_off >= pic_height - block_h - extra_after) {
+        const ptrdiff_t edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << fc->ps.sps->pixel_shift;
+        int offset     = extra_before * *src_stride      + (extra_before << fc->ps.sps->pixel_shift);
+        int buf_offset = extra_before * edge_emu_stride + (extra_before << fc->ps.sps->pixel_shift);
+
+        fc->vdsp.emulated_edge_mc(dst, *src - offset, edge_emu_stride, *src_stride,
+            block_w + extra, block_h + extra, x_off - extra_before, y_off - extra_before,
+            pic_width, pic_height);
+
+        *src = dst + buf_offset;
+        *src_stride = edge_emu_stride;
+        return 1;
+    }
+    return 0;
+}
+
+static void emulated_edge_dmvr(const VVCFrameContext *fc, uint8_t *dst, const uint8_t **src, ptrdiff_t *src_stride,
+    const int x_sb, const int y_sb, const int x_off, const int y_off, const int block_w, const int block_h, const int is_luma)
+{
+    const int extra_before = is_luma ? LUMA_EXTRA_BEFORE : CHROMA_EXTRA_BEFORE;
+    const int extra_after  = is_luma ? LUMA_EXTRA_AFTER : CHROMA_EXTRA_AFTER;
+    const int extra        = is_luma ? LUMA_EXTRA : CHROMA_EXTRA;
+    const int pic_width    = is_luma ? fc->ps.pps->width  : (fc->ps.pps->width >> fc->ps.sps->hshift[1]);
+    const int pic_height   = is_luma ? fc->ps.pps->height : (fc->ps.pps->height >> fc->ps.sps->vshift[1]);
+
+    if (x_off < extra_before || y_off < extra_before ||
+        x_off >= pic_width - block_w - extra_after ||
+        y_off >= pic_height - block_h - extra_after||
+        (x_off != x_sb || y_off !=  y_sb)) {
+        const int ps                    = fc->ps.sps->pixel_shift;
+        const ptrdiff_t edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << ps;
+        const int offset                = extra_before * *src_stride + (extra_before << ps);
+        const int buf_offset            = extra_before * edge_emu_stride + (extra_before << ps);
+
+        const int start_x               = FFMIN(FFMAX(x_sb - extra_before, 0), pic_width  - 1);
+        const int start_y               = FFMIN(FFMAX(y_sb - extra_before, 0), pic_height - 1);
+        const int width                 = FFMAX(FFMIN(pic_width, x_sb + block_w + extra_after) - start_x, 1);
+        const int height                = FFMAX(FFMIN(pic_height, y_sb + block_h + extra_after) - start_y, 1);
+
+        fc->vdsp.emulated_edge_mc(dst, *src - offset, edge_emu_stride, *src_stride, block_w + extra, block_h + extra,
+            x_off - start_x - extra_before, y_off - start_y - extra_before, width, height);
+
+        *src = dst + buf_offset;
+        *src_stride = edge_emu_stride;
+   }
+}
+
+static void emulated_edge_bilinear(const VVCFrameContext *fc, uint8_t *dst, const uint8_t **src, ptrdiff_t *src_stride,
+    const int x_off, const int y_off, const int block_w, const int block_h)
+{
+    int pic_width   = fc->ps.pps->width;
+    int pic_height  = fc->ps.pps->height;
+
+    if (x_off < BILINEAR_EXTRA_BEFORE || y_off < BILINEAR_EXTRA_BEFORE ||
+        x_off >= pic_width - block_w - BILINEAR_EXTRA_AFTER ||
+        y_off >= pic_height - block_h - BILINEAR_EXTRA_AFTER) {
+        const ptrdiff_t edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << fc->ps.sps->pixel_shift;
+        const int offset                = BILINEAR_EXTRA_BEFORE * *src_stride + (BILINEAR_EXTRA_BEFORE << fc->ps.sps->pixel_shift);
+        const int buf_offset            = BILINEAR_EXTRA_BEFORE * edge_emu_stride + (BILINEAR_EXTRA_BEFORE << fc->ps.sps->pixel_shift);
+
+        fc->vdsp.emulated_edge_mc(dst, *src - offset, edge_emu_stride, *src_stride, block_w + BILINEAR_EXTRA, block_h + BILINEAR_EXTRA,
+            x_off - BILINEAR_EXTRA_BEFORE, y_off - BILINEAR_EXTRA_BEFORE,  pic_width, pic_height);
+
+        *src = dst + buf_offset;
+        *src_stride = edge_emu_stride;
+    }
+}
+
+
+#define EMULATED_EDGE_LUMA(dst, src, src_stride, x_off, y_off)                      \
+    emulated_edge(fc, dst, src, src_stride, x_off, y_off, block_w, block_h, 1)
+
+#define EMULATED_EDGE_CHROMA(dst, src, src_stride, x_off, y_off)                    \
+    emulated_edge(fc, dst, src, src_stride, x_off, y_off, block_w, block_h, 0)
+
+#define EMULATED_EDGE_DMVR_LUMA(dst, src, src_stride, x_sb, y_sb, x_off, y_off)     \
+    emulated_edge_dmvr(fc, dst, src, src_stride, x_sb, y_sb, x_off, y_off, block_w, block_h, 1)
+
+#define EMULATED_EDGE_DMVR_CHROMA(dst, src, src_stride, x_sb, y_sb, x_off, y_off)   \
+    emulated_edge_dmvr(fc, dst, src, src_stride, x_sb, y_sb, x_off, y_off, block_w, block_h, 0)
+
+#define EMULATED_EDGE_BILINEAR(dst, src, src_stride, x_off, y_off)                  \
+    emulated_edge_bilinear(fc, dst, src, src_stride, x_off, y_off, pred_w, pred_h)
+
+// part of 8.5.6.6 Weighted sample prediction process
+static int derive_weight_uni(int *denom, int *wx, int *ox,
+    const VVCLocalContext *lc, const MvField *mvf, const int c_idx)
+{
+    const VVCFrameContext *fc   = lc->fc;
+    const VVCPPS *pps           = fc->ps.pps;
+    const VVCSH *sh             = &lc->sc->sh;
+    const int weight_flag       = (IS_P(sh->r) && pps->r->pps_weighted_pred_flag) ||
+                                  (IS_B(sh->r) && pps->r->pps_weighted_bipred_flag);
+    if (weight_flag) {
+        const int lx                = mvf->pred_flag - PF_L0;
+        const PredWeightTable *w    = pps->r->pps_wp_info_in_ph_flag ? &fc->ps.ph.pwt : &sh->pwt;
+
+        *denom = w->log2_denom[c_idx > 0];
+        *wx = w->weight[lx][c_idx][mvf->ref_idx[lx]];
+        *ox = w->offset[lx][c_idx][mvf->ref_idx[lx]];
+    }
+    return weight_flag;
+}
+
+// part of 8.5.6.6 Weighted sample prediction process
+static int derive_weight(int *denom, int *w0, int *w1, int *o0, int *o1,
+    const VVCLocalContext *lc, const MvField *mvf, const int c_idx, const int dmvr_flag)
+{
+    const VVCFrameContext *fc   = lc->fc;
+    const VVCPPS *pps           = fc->ps.pps;
+    const VVCSH *sh             = &lc->sc->sh;
+    const int bcw_idx           = mvf->bcw_idx;
+    const int weight_flag       = (IS_P(sh->r) && pps->r->pps_weighted_pred_flag) ||
+                                  (IS_B(sh->r) && pps->r->pps_weighted_bipred_flag && !dmvr_flag);
+    if ((!weight_flag && !bcw_idx) || (bcw_idx && lc->cu->ciip_flag))
+        return 0;
+
+    if (bcw_idx) {
+        *denom = 2;
+        *w1 = bcw_w_lut[bcw_idx];
+        *w0 = 8 - *w1;
+        *o0 = *o1 = 0;
+    } else {
+        const VVCPPS *pps = fc->ps.pps;
+        const PredWeightTable *w = pps->r->pps_wp_info_in_ph_flag ? &fc->ps.ph.pwt : &sh->pwt;
+
+        *denom = w->log2_denom[c_idx > 0];
+        *w0 = w->weight[L0][c_idx][mvf->ref_idx[L0]];
+        *w1 = w->weight[L1][c_idx][mvf->ref_idx[L1]];
+        *o0 = w->offset[L0][c_idx][mvf->ref_idx[L0]];
+        *o1 = w->offset[L1][c_idx][mvf->ref_idx[L1]];
+    }
+    return 1;
+}
+
+static void luma_mc(VVCLocalContext *lc, int16_t *dst, const AVFrame *ref, const Mv *mv,
+    int x_off, int y_off, const int block_w, const int block_h)
+{
+    const VVCFrameContext *fc   = lc->fc;
+    const uint8_t *src          = ref->data[0];
+    ptrdiff_t src_stride        = ref->linesize[0];
+    const int idx               = av_log2(block_w) - 1;
+    const int mx                = mv->x & 0xf;
+    const int my                = mv->y & 0xf;
+    const int8_t *hf            = ff_vvc_inter_luma_filters[0][mx];
+    const int8_t *vf            = ff_vvc_inter_luma_filters[0][my];
+
+    x_off += mv->x >> 4;
+    y_off += mv->y >> 4;
+    src   += y_off * src_stride + (x_off << fc->ps.sps->pixel_shift);
+
+    EMULATED_EDGE_LUMA(lc->edge_emu_buffer, &src, &src_stride, x_off, y_off);
+
+    fc->vvcdsp.inter.put[LUMA][idx][!!my][!!mx](dst, src, src_stride, block_h, hf, vf, block_w);
+}
+
+static void chroma_mc(VVCLocalContext *lc, int16_t *dst, const AVFrame *ref, const Mv *mv,
+    int x_off, int y_off, const int block_w, const int block_h, const int c_idx)
+{
+    const VVCFrameContext *fc   = lc->fc;
+    const uint8_t *src          = ref->data[c_idx];
+    ptrdiff_t src_stride        = ref->linesize[c_idx];
+    int hs                      = fc->ps.sps->hshift[c_idx];
+    int vs                      = fc->ps.sps->vshift[c_idx];
+    const int idx               = av_log2(block_w) - 1;
+    const intptr_t mx           = av_mod_uintp2(mv->x, 4 + hs) << (1 - hs);
+    const intptr_t my           = av_mod_uintp2(mv->y, 4 + vs) << (1 - vs);
+    const int8_t *hf            = ff_vvc_inter_chroma_filters[0][mx];
+    const int8_t *vf            = ff_vvc_inter_chroma_filters[0][my];
+
+    x_off += mv->x >> (4 + hs);
+    y_off += mv->y >> (4 + vs);
+    src  += y_off * src_stride + (x_off << fc->ps.sps->pixel_shift);
+
+    EMULATED_EDGE_CHROMA(lc->edge_emu_buffer, &src, &src_stride, x_off, y_off);
+    fc->vvcdsp.inter.put[CHROMA][idx][!!my][!!mx](dst, src, src_stride, block_h, hf, vf, block_w);
+}
+
+static void luma_mc_uni(VVCLocalContext *lc, uint8_t *dst, const ptrdiff_t dst_stride,
+    const AVFrame *ref, const MvField *mvf, int x_off, int y_off, const int block_w, const int block_h,
+    const int hf_idx, const int vf_idx)
+{
+    const VVCFrameContext *fc   = lc->fc;
+    const int lx                = mvf->pred_flag - PF_L0;
+    const Mv *mv                = mvf->mv + lx;
+    const uint8_t *src          = ref->data[0];
+    ptrdiff_t src_stride        = ref->linesize[0];
+    const int idx               = av_log2(block_w) - 1;
+    const int mx                = mv->x & 0xf;
+    const int my                = mv->y & 0xf;
+    const int8_t *hf            = ff_vvc_inter_luma_filters[hf_idx][mx];
+    const int8_t *vf            = ff_vvc_inter_luma_filters[vf_idx][my];
+    int denom, wx, ox;
+
+    x_off += mv->x >> 4;
+    y_off += mv->y >> 4;
+    src   += y_off * src_stride + (x_off << fc->ps.sps->pixel_shift);
+
+    EMULATED_EDGE_LUMA(lc->edge_emu_buffer, &src, &src_stride, x_off, y_off);
+
+    if (derive_weight_uni(&denom, &wx, &ox, lc, mvf, LUMA)) {
+        fc->vvcdsp.inter.put_uni_w[LUMA][idx][!!my][!!mx](dst, dst_stride, src, src_stride,
+            block_h, denom, wx, ox, hf, vf, block_w);
+    } else {
+        fc->vvcdsp.inter.put_uni[LUMA][idx][!!my][!!mx](dst, dst_stride, src, src_stride,
+            block_h, hf, vf, block_w);
+    }
+}
+
+static void luma_mc_bi(VVCLocalContext *lc, uint8_t *dst, const ptrdiff_t dst_stride,
+    const AVFrame *ref0, const Mv *mv0, const int x_off, const int y_off, const int block_w, const int block_h,
+    const AVFrame *ref1, const Mv *mv1, const MvField *mvf, const int hf_idx, const int vf_idx,
+    const MvField *orig_mv, const int sb_bdof_flag)
+{
+    const VVCFrameContext *fc   = lc->fc;
+    const PredictionUnit *pu    = &lc->cu->pu;
+    const int idx               = av_log2(block_w) - 1;
+    const AVFrame *ref[]        = { ref0, ref1 };
+    int16_t *tmp[]              = { lc->tmp + sb_bdof_flag * PROF_TEMP_OFFSET, lc->tmp1 + sb_bdof_flag * PROF_TEMP_OFFSET };
+    int denom, w0, w1, o0, o1;
+    const int weight_flag       = derive_weight(&denom, &w0, &w1, &o0, &o1, lc, mvf, LUMA, pu->dmvr_flag);
+
+    for (int i = L0; i <= L1; i++) {
+        const Mv *mv            = mvf->mv + i;
+        const int mx            = mv->x & 0xf;
+        const int my            = mv->y & 0xf;
+        const int ox            = x_off + (mv->x >> 4);
+        const int oy            = y_off + (mv->y >> 4);
+        ptrdiff_t src_stride    = ref[i]->linesize[0];
+        const uint8_t *src      = ref[i]->data[0] + oy * src_stride + (ox << fc->ps.sps->pixel_shift);
+        const int8_t *hf        = ff_vvc_inter_luma_filters[hf_idx][mx];
+        const int8_t *vf        = ff_vvc_inter_luma_filters[vf_idx][my];
+
+        if (pu->dmvr_flag) {
+            const int x_sb = x_off + (orig_mv->mv[i].x >> 4);
+            const int y_sb = y_off + (orig_mv->mv[i].y >> 4);
+
+            EMULATED_EDGE_DMVR_LUMA(lc->edge_emu_buffer, &src, &src_stride, x_sb, y_sb, ox, oy);
+        } else {
+            EMULATED_EDGE_LUMA(lc->edge_emu_buffer, &src, &src_stride, ox, oy);
+        }
+        fc->vvcdsp.inter.put[LUMA][idx][!!my][!!mx](tmp[i], src, src_stride, block_h, hf, vf, block_w);
+        if (sb_bdof_flag)
+            fc->vvcdsp.inter.bdof_fetch_samples(tmp[i], src, src_stride, mx, my, block_w, block_h);
+    }
+
+    if (sb_bdof_flag)
+        fc->vvcdsp.inter.apply_bdof(dst, dst_stride, tmp[L0], tmp[L1], block_w, block_h);
+    else if (weight_flag)
+        fc->vvcdsp.inter.w_avg(dst, dst_stride, tmp[L0], tmp[L1], block_w, block_h, denom, w0, w1, o0, o1);
+    else
+        fc->vvcdsp.inter.avg(dst, dst_stride, tmp[L0], tmp[L1], block_w, block_h);
+}
+
+static void chroma_mc_uni(VVCLocalContext *lc, uint8_t *dst, const ptrdiff_t dst_stride,
+    const uint8_t *src, ptrdiff_t src_stride, int x_off, int y_off,
+    const int block_w, const int block_h, const MvField *mvf, const int c_idx,
+    const int hf_idx, const int vf_idx)
+{
+    const VVCFrameContext *fc   = lc->fc;
+    const int lx                = mvf->pred_flag - PF_L0;
+    const int hs                = fc->ps.sps->hshift[1];
+    const int vs                = fc->ps.sps->vshift[1];
+    const int idx               = av_log2(block_w) - 1;
+    const Mv *mv                = &mvf->mv[lx];
+    const intptr_t mx           = av_mod_uintp2(mv->x, 4 + hs) << (1 - hs);
+    const intptr_t my           = av_mod_uintp2(mv->y, 4 + vs) << (1 - vs);
+    const int8_t *hf            = ff_vvc_inter_chroma_filters[hf_idx][mx];
+    const int8_t *vf            = ff_vvc_inter_chroma_filters[vf_idx][my];
+    int denom, wx, ox;
+
+    x_off += mv->x >> (4 + hs);
+    y_off += mv->y >> (4 + vs);
+    src  += y_off * src_stride + (x_off << fc->ps.sps->pixel_shift);
+
+
+    EMULATED_EDGE_CHROMA(lc->edge_emu_buffer, &src, &src_stride, x_off, y_off);
+    if (derive_weight_uni(&denom, &wx, &ox, lc, mvf, c_idx)) {
+        fc->vvcdsp.inter.put_uni_w[CHROMA][idx][!!my][!!mx](dst, dst_stride, src, src_stride,
+            block_h, denom, wx, ox, hf, vf, block_w);
+    } else {
+        fc->vvcdsp.inter.put_uni[CHROMA][idx][!!my][!!mx](dst, dst_stride, src, src_stride,
+            block_h, hf, vf, block_w);
+    }
+}
+
+static void chroma_mc_bi(VVCLocalContext *lc, uint8_t *dst, const ptrdiff_t dst_stride,
+    const AVFrame *ref0, const AVFrame *ref1, const int x_off, const int y_off,
+    const int block_w, const int block_h,  const MvField *mvf, const int c_idx,
+    const int hf_idx, const int vf_idx, const MvField *orig_mv, const int dmvr_flag, const int ciip_flag)
+{
+    const VVCFrameContext *fc   = lc->fc;
+    const int hs                = fc->ps.sps->hshift[1];
+    const int vs                = fc->ps.sps->vshift[1];
+    const int idx               = av_log2(block_w) - 1;
+    const AVFrame *ref[]        = { ref0, ref1 };
+    int16_t *tmp[]              = { lc->tmp, lc->tmp1 };
+    int denom, w0, w1, o0, o1;
+    const int weight_flag       = derive_weight(&denom, &w0, &w1, &o0, &o1, lc, mvf, c_idx, dmvr_flag);
+
+    for (int i = L0; i <= L1; i++) {
+        const Mv *mv            = mvf->mv + i;
+        const int mx            = av_mod_uintp2(mv->x, 4 + hs) << (1 - hs);
+        const int my            = av_mod_uintp2(mv->y, 4 + vs) << (1 - vs);
+        const int ox            = x_off + (mv->x >> (4 + hs));
+        const int oy            = y_off + (mv->y >> (4 + vs));
+        ptrdiff_t src_stride    = ref[i]->linesize[c_idx];
+        const uint8_t *src      = ref[i]->data[c_idx] + oy * src_stride + (ox << fc->ps.sps->pixel_shift);
+        const int8_t *hf        = ff_vvc_inter_chroma_filters[hf_idx][mx];
+        const int8_t *vf        = ff_vvc_inter_chroma_filters[vf_idx][my];
+        if (dmvr_flag) {
+            const int x_sb = x_off + (orig_mv->mv[i].x >> (4 + hs));
+            const int y_sb = y_off + (orig_mv->mv[i].y >> (4 + vs));
+            EMULATED_EDGE_DMVR_CHROMA(lc->edge_emu_buffer,  &src, &src_stride, x_sb, y_sb, ox, oy);
+        } else {
+            EMULATED_EDGE_CHROMA(lc->edge_emu_buffer, &src, &src_stride, ox, oy);
+        }
+        fc->vvcdsp.inter.put[CHROMA][idx][!!my][!!mx](tmp[i],  src, src_stride, block_h, hf, vf, block_w);
+    }
+    if (weight_flag)
+        fc->vvcdsp.inter.w_avg(dst, dst_stride, tmp[L0], tmp[L1], block_w, block_h, denom, w0, w1, o0, o1);
+    else
+        fc->vvcdsp.inter.avg(dst, dst_stride, tmp[L0], tmp[L1], block_w, block_h);
+}
+
+static void luma_prof_uni(VVCLocalContext *lc, uint8_t *dst, const ptrdiff_t dst_stride,
+    const AVFrame *ref, const MvField *mvf, int x_off, int y_off, const int block_w, const int block_h,
+    const int cb_prof_flag, const int16_t *diff_mv_x, const int16_t *diff_mv_y)
+{
+    const VVCFrameContext *fc   = lc->fc;
+    const uint8_t *src          = ref->data[0];
+    ptrdiff_t src_stride        = ref->linesize[0];
+    uint16_t *prof_tmp          = lc->tmp + PROF_TEMP_OFFSET;
+    const int idx               = av_log2(block_w) - 1;
+    const int lx                = mvf->pred_flag - PF_L0;
+    const Mv *mv                = mvf->mv + lx;
+    const int mx                = mv->x & 0xf;
+    const int my                = mv->y & 0xf;
+    const int8_t *hf            = ff_vvc_inter_luma_filters[2][mx];
+    const int8_t *vf            = ff_vvc_inter_luma_filters[2][my];
+    int denom, wx, ox;
+    const int weight_flag       = derive_weight_uni(&denom, &wx, &ox, lc, mvf, LUMA);
+
+    x_off += mv->x >> 4;
+    y_off += mv->y >> 4;
+    src   += y_off * src_stride + (x_off << fc->ps.sps->pixel_shift);
+
+    EMULATED_EDGE_LUMA(lc->edge_emu_buffer, &src, &src_stride, x_off, y_off);
+    if (cb_prof_flag) {
+        fc->vvcdsp.inter.put[LUMA][idx][!!my][!!mx](prof_tmp, src, src_stride, AFFINE_MIN_BLOCK_SIZE, hf, vf, AFFINE_MIN_BLOCK_SIZE);
+        fc->vvcdsp.inter.fetch_samples(prof_tmp, src, src_stride, mx, my);
+        if (!weight_flag)
+            fc->vvcdsp.inter.apply_prof_uni(dst, dst_stride, prof_tmp, diff_mv_x, diff_mv_y);
+        else
+            fc->vvcdsp.inter.apply_prof_uni_w(dst, dst_stride, prof_tmp, diff_mv_x, diff_mv_y, denom, wx, ox);
+    } else {
+        if (!weight_flag)
+            fc->vvcdsp.inter.put_uni[LUMA][idx][!!my][!!mx](dst, dst_stride, src, src_stride, block_h, hf, vf, block_w);
+        else
+            fc->vvcdsp.inter.put_uni_w[LUMA][idx][!!my][!!mx](dst, dst_stride, src, src_stride, block_h, denom, wx, ox, hf, vf, block_w);
+    }
+}
+
+static void luma_prof_bi(VVCLocalContext *lc, uint8_t *dst, const ptrdiff_t dst_stride,
+    const AVFrame *ref0, const AVFrame *ref1, const MvField *mvf, const int x_off, const int y_off,
+    const int block_w, const int block_h)
+{
+    const VVCFrameContext *fc   = lc->fc;
+    const PredictionUnit *pu    = &lc->cu->pu;
+    const AVFrame *ref[]        = { ref0, ref1 };
+    int16_t *tmp[]              = { lc->tmp, lc->tmp1 };
+    uint16_t *prof_tmp          = lc->tmp2 + PROF_TEMP_OFFSET;
+    const int idx               = av_log2(block_w) - 1;
+    int denom, w0, w1, o0, o1;
+    const int weight_flag       = derive_weight(&denom, &w0, &w1, &o0, &o1, lc, mvf, LUMA, 0);
+
+    for (int i = L0; i <= L1; i++) {
+        const Mv *mv            = mvf->mv + i;
+        const int mx            = mv->x & 0xf;
+        const int my            = mv->y & 0xf;
+        const int ox            = x_off + (mv->x >> 4);
+        const int oy            = y_off + (mv->y >> 4);
+        ptrdiff_t src_stride    = ref[i]->linesize[0];
+        const uint8_t *src      = ref[i]->data[0] + oy * src_stride + (ox << fc->ps.sps->pixel_shift);
+        const int8_t *hf        = ff_vvc_inter_luma_filters[2][mx];
+        const int8_t *vf        = ff_vvc_inter_luma_filters[2][my];
+
+        EMULATED_EDGE_LUMA(lc->edge_emu_buffer, &src, &src_stride, ox, oy);
+        if (!pu->cb_prof_flag[i]) {
+            fc->vvcdsp.inter.put[LUMA][idx][!!my][!!mx](tmp[i], src, src_stride, block_h, hf, vf, block_w);
+        } else {
+            fc->vvcdsp.inter.put[LUMA][idx][!!my][!!mx](prof_tmp, src, src_stride, AFFINE_MIN_BLOCK_SIZE, hf, vf, AFFINE_MIN_BLOCK_SIZE);
+            fc->vvcdsp.inter.fetch_samples(prof_tmp, src, src_stride, mx, my);
+            fc->vvcdsp.inter.apply_prof(tmp[i], prof_tmp, pu->diff_mv_x[i], pu->diff_mv_y[i]);
+        }
+    }
+
+    if (weight_flag)
+        fc->vvcdsp.inter.w_avg(dst, dst_stride, tmp[L0], tmp[L1], block_w, block_h,  denom, w0, w1, o0, o1);
+    else
+        fc->vvcdsp.inter.avg(dst, dst_stride, tmp[L0], tmp[L1], block_w, block_h);
+}
+
+static int pred_get_refs(const VVCLocalContext *lc, VVCFrame *ref[2],  const MvField *mv)
+{
+    const RefPicList *rpl = lc->sc->rpl;
+
+    for (int mask = PF_L0; mask <= PF_L1; mask++) {
+        if (mv->pred_flag & mask) {
+            const int lx = mask - PF_L0;
+            ref[lx] = rpl[lx].ref[mv->ref_idx[lx]];
+            if (!ref[lx])
+                return AVERROR_INVALIDDATA;
+        }
+    }
+    return 0;
+}
+
+#define POS(c_idx, x, y)                                                                            \
+        &fc->frame->data[c_idx][((y) >> fc->ps.sps->vshift[c_idx]) * fc->frame->linesize[c_idx] +   \
+            (((x) >> fc->ps.sps->hshift[c_idx]) << fc->ps.sps->pixel_shift)]
+
+static void pred_gpm_blk(VVCLocalContext *lc)
+{
+    const VVCFrameContext *fc  = lc->fc;
+    const CodingUnit *cu       = lc->cu;
+    const PredictionUnit *pu   = &cu->pu;
+
+    const uint8_t angle_idx   = ff_vvc_gpm_angle_idx[pu->gpm_partition_idx];
+    const uint8_t weights_idx = ff_vvc_gpm_angle_to_weights_idx[angle_idx];
+    const int w = av_log2(cu->cb_width) - 3;
+    const int h = av_log2(cu->cb_height) - 3;
+    const uint8_t off_x = ff_vvc_gpm_weights_offset_x[pu->gpm_partition_idx][h][w];
+    const uint8_t off_y = ff_vvc_gpm_weights_offset_y[pu->gpm_partition_idx][h][w];
+    const uint8_t mirror_type = ff_vvc_gpm_angle_to_mirror[angle_idx];
+    const uint8_t *weights;
+
+    const int c_end = fc->ps.sps->r->sps_chroma_format_idc ? 3 : 1;
+
+    int16_t *tmp[2] = {lc->tmp, lc->tmp1};
+
+    for (int c_idx = 0; c_idx < c_end; c_idx++) {
+        const int hs     = fc->ps.sps->hshift[c_idx];
+        const int vs     = fc->ps.sps->vshift[c_idx];
+        const int x      = lc->cu->x0  >> hs;
+        const int y      = lc->cu->y0  >> vs;
+        const int width  = cu->cb_width >> hs;
+        const int height = cu->cb_height >> vs;
+        uint8_t *dst = POS(c_idx, lc->cu->x0, lc->cu->y0);
+        ptrdiff_t dst_stride = fc->frame->linesize[c_idx];
+
+        int step_x = 1 << hs;
+        int step_y = VVC_GPM_WEIGHT_SIZE << vs;
+        if (!mirror_type) {
+            weights = &ff_vvc_gpm_weights[weights_idx][off_y * VVC_GPM_WEIGHT_SIZE + off_x];
+        } else if (mirror_type == 1) {
+            step_x = -step_x;
+            weights = &ff_vvc_gpm_weights[weights_idx][off_y * VVC_GPM_WEIGHT_SIZE + VVC_GPM_WEIGHT_SIZE - 1- off_x];
+        } else {
+            step_y = -step_y;
+            weights = &ff_vvc_gpm_weights[weights_idx][(VVC_GPM_WEIGHT_SIZE - 1 - off_y) * VVC_GPM_WEIGHT_SIZE + off_x];
+        }
+
+        for (int i = 0; i < 2; i++) {
+            const MvField *mv = pu->gpm_mv + i;
+            const int lx = mv->pred_flag - PF_L0;
+            VVCFrame *ref = lc->sc->rpl[lx].ref[mv->ref_idx[lx]];
+            if (!ref)
+                return;
+            if (c_idx)
+                chroma_mc(lc, tmp[i], ref->frame, mv->mv + lx, x, y, width, height, c_idx);
+            else
+                luma_mc(lc, tmp[i], ref->frame, mv->mv + lx, x, y, width, height);
+        }
+        fc->vvcdsp.inter.put_gpm(dst, dst_stride, width, height, tmp[0], tmp[1], weights, step_x, step_y);
+    }
+    return;
+}
+
+static int ciip_derive_intra_weight(const VVCLocalContext *lc, const int x0, const int y0,
+    const int width, const int height)
+{
+    const VVCFrameContext *fc   = lc->fc;
+    const VVCSPS *sps           = fc->ps.sps;
+    const int x0b               = av_mod_uintp2(x0, sps->ctb_log2_size_y);
+    const int y0b               = av_mod_uintp2(y0, sps->ctb_log2_size_y);
+    const int available_l       = lc->ctb_left_flag || x0b;
+    const int available_u       = lc->ctb_up_flag || y0b;
+    const int min_pu_width      = fc->ps.pps->min_pu_width;
+
+    int w = 1;
+
+    if (available_u &&fc->tab.mvf[((y0 - 1) >> MIN_PU_LOG2) * min_pu_width + ((x0 - 1 + width)>> MIN_PU_LOG2)].pred_flag == PF_INTRA)
+        w++;
+
+    if (available_l && fc->tab.mvf[((y0 - 1 + height)>> MIN_PU_LOG2) * min_pu_width + ((x0 - 1) >> MIN_PU_LOG2)].pred_flag == PF_INTRA)
+        w++;
+
+    return w;
+}
+
+static void pred_regular_luma(VVCLocalContext *lc, const int hf_idx, const int vf_idx, const MvField *mv,
+    const int x0, const int y0, const int sbw, const int sbh, const MvField *orig_mv, const int sb_bdof_flag)
+{
+    const SliceContext *sc          = lc->sc;
+    const VVCFrameContext *fc       = lc->fc;
+    const int ciip_flag             = lc->cu->ciip_flag;
+    uint8_t *dst                    = POS(0, x0, y0);
+    const ptrdiff_t dst_stride      = fc->frame->linesize[0];
+    uint8_t *inter                  = ciip_flag ? (uint8_t *)lc->ciip_tmp1 : dst;
+    const ptrdiff_t inter_stride    = ciip_flag ? (MAX_PB_SIZE * sizeof(uint16_t)) : dst_stride;
+    VVCFrame *ref[2];
+
+    if (pred_get_refs(lc, ref, mv) < 0)
+        return;
+
+    if (mv->pred_flag != PF_BI) {
+        const int lx = mv->pred_flag - PF_L0;
+        luma_mc_uni(lc, inter, inter_stride, ref[lx]->frame,
+            mv, x0, y0, sbw, sbh, hf_idx, vf_idx);
+    } else {
+        luma_mc_bi(lc, inter, inter_stride, ref[0]->frame,
+            &mv->mv[0], x0, y0, sbw, sbh, ref[1]->frame, &mv->mv[1], mv,
+            hf_idx, vf_idx, orig_mv, sb_bdof_flag);
+    }
+
+    if (ciip_flag) {
+        const int intra_weight = ciip_derive_intra_weight(lc, x0, y0, sbw, sbh);
+        fc->vvcdsp.intra.intra_pred(lc, x0, y0, sbw, sbh, 0);
+        if (sc->sh.r->sh_lmcs_used_flag)
+            fc->vvcdsp.lmcs.filter(inter, inter_stride, sbw, sbh, fc->ps.lmcs.fwd_lut);
+        fc->vvcdsp.inter.put_ciip(dst, dst_stride, sbw, sbh, inter, inter_stride, intra_weight);
+
+    }
+}
+
+static void pred_regular_chroma(VVCLocalContext *lc, const MvField *mv,
+    const int x0, const int y0, const int sbw, const int sbh, const MvField *orig_mv, const int dmvr_flag)
+{
+    const VVCFrameContext *fc   = lc->fc;
+    const int hs                = fc->ps.sps->hshift[1];
+    const int vs                = fc->ps.sps->vshift[1];
+    const int x0_c              = x0 >> hs;
+    const int y0_c              = y0 >> vs;
+    const int w_c               = sbw >> hs;
+    const int h_c               = sbh >> vs;
+    const int do_ciip           = lc->cu->ciip_flag && (w_c > 2);
+
+    uint8_t* dst1               = POS(1, x0, y0);
+    uint8_t* dst2               = POS(2, x0, y0);
+    const ptrdiff_t dst1_stride = fc->frame->linesize[1];
+    const ptrdiff_t dst2_stride = fc->frame->linesize[2];
+
+    uint8_t *inter1 = do_ciip ? (uint8_t *)lc->ciip_tmp1 : dst1;
+    const ptrdiff_t inter1_stride = do_ciip ? (MAX_PB_SIZE * sizeof(uint16_t)) : dst1_stride;
+
+    uint8_t *inter2 = do_ciip ? (uint8_t *)lc->ciip_tmp2 : dst2;
+    const ptrdiff_t inter2_stride = do_ciip ? (MAX_PB_SIZE * sizeof(uint16_t)) : dst2_stride;
+
+    //fix me
+    const int hf_idx = 0;
+    const int vf_idx = 0;
+    VVCFrame *ref[2];
+
+    if (pred_get_refs(lc, ref, mv) < 0)
+        return;
+
+    if (mv->pred_flag != PF_BI) {
+        const int lx = mv->pred_flag - PF_L0;
+        if (!ref[lx])
+            return;
+
+        chroma_mc_uni(lc, inter1, inter1_stride, ref[lx]->frame->data[1], ref[lx]->frame->linesize[1],
+            x0_c, y0_c, w_c, h_c, mv, CB, hf_idx, vf_idx);
+        chroma_mc_uni(lc, inter2, inter2_stride, ref[lx]->frame->data[2], ref[lx]->frame->linesize[2],
+            x0_c, y0_c, w_c, h_c, mv, CR, hf_idx, vf_idx);
+    } else {
+        if (!ref[0] || !ref[1])
+            return;
+
+        chroma_mc_bi(lc, inter1, inter1_stride, ref[0]->frame, ref[1]->frame,
+            x0_c, y0_c, w_c, h_c, mv, CB, hf_idx, vf_idx, orig_mv, dmvr_flag, lc->cu->ciip_flag);
+
+        chroma_mc_bi(lc, inter2, inter2_stride, ref[0]->frame, ref[1]->frame,
+            x0_c, y0_c, w_c, h_c, mv, CR, hf_idx, vf_idx, orig_mv, dmvr_flag, lc->cu->ciip_flag);
+
+    }
+    if (do_ciip) {
+        const int intra_weight = ciip_derive_intra_weight(lc, x0, y0, sbw, sbh);
+        fc->vvcdsp.intra.intra_pred(lc, x0, y0, sbw, sbh, 1);
+        fc->vvcdsp.intra.intra_pred(lc, x0, y0, sbw, sbh, 2);
+        fc->vvcdsp.inter.put_ciip(dst1, dst1_stride, w_c, h_c, inter1, inter1_stride, intra_weight);
+        fc->vvcdsp.inter.put_ciip(dst2, dst2_stride, w_c, h_c, inter2, inter2_stride, intra_weight);
+
+    }
+}
+
+// 8.5.3.5 Parametric motion vector refinement process
+static int parametric_mv_refine(const int *sad, const int stride)
+{
+    const int sad_minus  = sad[-stride];
+    const int sad_center = sad[0];
+    const int sad_plus   = sad[stride];
+    int dmvc;
+    int denom = (( sad_minus + sad_plus) - (sad_center << 1 ) ) << 3;
+    if (!denom)
+        dmvc = 0;
+    else {
+        if (sad_minus == sad_center)
+            dmvc = -8;
+        else if (sad_plus == sad_center)
+            dmvc = 8;
+        else {
+            int num = ( sad_minus - sad_plus ) << 4;
+            int sign_num = 0;
+            int quotient = 0;
+            int counter = 3;
+            if (num < 0 ) {
+                num = - num;
+                sign_num = 1;
+            }
+            while (counter > 0) {
+                counter = counter - 1;
+                quotient = quotient << 1;
+                if ( num >= denom ) {
+                    num = num - denom;
+                    quotient = quotient + 1;
+                }
+                denom = (denom >> 1);
+            }
+            if (sign_num == 1 )
+                dmvc = -quotient;
+            else
+                dmvc = quotient;
+        }
+    }
+    return dmvc;
+}
+
+#define SAD_ARRAY_SIZE 5
+//8.5.3 Decoder-side motion vector refinement process
+static void dmvr_mv_refine(VVCLocalContext *lc, MvField *mvf, MvField *orig_mv, int *sb_bdof_flag,
+    const AVFrame *ref0, const AVFrame *ref1, const int x_off, const int y_off, const int block_w, const int block_h)
+{
+    const VVCFrameContext *fc   = lc->fc;
+    const int sr_range          = 2;
+    const AVFrame *ref[]        = { ref0, ref1 };
+    int16_t *tmp[]              = { lc->tmp, lc->tmp1 };
+    int sad[SAD_ARRAY_SIZE][SAD_ARRAY_SIZE];
+    int min_dx, min_dy, min_sad, dx, dy;
+
+    *orig_mv = *mvf;
+    min_dx = min_dy = dx = dy = 2;
+
+    for (int i = L0; i <= L1; i++) {
+        const int pred_w        = block_w + 2 * sr_range;
+        const int pred_h        = block_h + 2 * sr_range;
+        const Mv *mv            = mvf->mv + i;
+        const int mx            = mv->x & 0xf;
+        const int my            = mv->y & 0xf;
+        const int ox            = x_off + (mv->x >> 4) - sr_range;
+        const int oy            = y_off + (mv->y >> 4) - sr_range;
+        ptrdiff_t src_stride    = ref[i]->linesize[LUMA];
+        const uint8_t *src      = ref[i]->data[LUMA] + oy * src_stride + (ox << fc->ps.sps->pixel_shift);
+        EMULATED_EDGE_BILINEAR(lc->edge_emu_buffer, &src, &src_stride, ox, oy);
+        fc->vvcdsp.inter.dmvr[!!my][!!mx](tmp[i], src, src_stride, pred_h, mx, my, pred_w);
+    }
+
+    min_sad = fc->vvcdsp.inter.sad(tmp[L0], tmp[L1], dx, dy, block_w, block_h);
+    min_sad -= min_sad >> 2;
+    sad[dy][dx] = min_sad;
+
+    if (min_sad >= block_w * block_h) {
+        int dmv[2];
+        // 8.5.3.4 Array entry selection process
+        for (dy = 0; dy < SAD_ARRAY_SIZE; dy++) {
+            for (dx = 0; dx < SAD_ARRAY_SIZE; dx++) {
+                if (dx != sr_range || dy != sr_range) {
+                    sad[dy][dx] = fc->vvcdsp.inter.sad(lc->tmp, lc->tmp1, dx, dy, block_w, block_h);
+                    if (sad[dy][dx] < min_sad) {
+                        min_sad = sad[dy][dx];
+                        min_dx = dx;
+                        min_dy = dy;
+                    }
+                }
+            }
+        }
+        dmv[0] = (min_dx - sr_range) << 4;
+        dmv[1] = (min_dy - sr_range) << 4;
+        if (min_dx != 0 && min_dx != 4 && min_dy != 0 && min_dy != 4) {
+            dmv[0] += parametric_mv_refine(&sad[min_dy][min_dx], 1);
+            dmv[1] += parametric_mv_refine(&sad[min_dy][min_dx], SAD_ARRAY_SIZE);
+        }
+
+        for (int i = L0; i <= L1; i++) {
+            Mv *mv = mvf->mv + i;
+            mv->x += (1 - 2 * i) * dmv[0];
+            mv->y += (1 - 2 * i) * dmv[1];
+            ff_vvc_clip_mv(mv);
+        }
+    }
+    if (min_sad < 2 * block_w * block_h) {
+        *sb_bdof_flag = 0;
+    }
+}
+
+static void set_dmvr_info(VVCFrameContext *fc, const int x0, const int y0,
+    const int width, const int height, const MvField *mvf)
+
+{
+    const VVCPPS *pps = fc->ps.pps;
+
+    for (int y = y0; y < y0 + height; y += MIN_PU_SIZE) {
+        for (int x = x0; x < x0 + width; x += MIN_PU_SIZE) {
+            const int idx = pps->min_pu_width * (y >> MIN_PU_LOG2) + (x >> MIN_PU_LOG2);
+            fc->ref->tab_dmvr_mvf[idx] = *mvf;
+        }
+    }
+}
+
+static void fill_dmvr_info(const VVCFrameContext *fc, const int x0, const int y0,
+    const int width, const int height)
+{
+    const VVCPPS *pps = fc->ps.pps;
+    const int w = width >> MIN_PU_LOG2;
+
+    for (int y = y0 >> MIN_PU_LOG2; y < (y0 + height) >> MIN_PU_LOG2; y++) {
+        const int idx = pps->min_pu_width * y + (x0 >> MIN_PU_LOG2);
+        const MvField *mvf = fc->tab.mvf + idx;
+        MvField *dmvr_mvf  = fc->ref->tab_dmvr_mvf + idx;
+        memcpy(dmvr_mvf, mvf, sizeof(MvField) * w);
+    }
+}
+
+static void derive_sb_mv(VVCLocalContext *lc, MvField *mv, MvField *orig_mv, int *sb_bdof_flag,
+    const int x0, const int y0, const int sbw, const int sbh)
+{
+    VVCFrameContext *fc      = lc->fc;
+    const PredictionUnit *pu = &lc->cu->pu;
+
+    *orig_mv = *mv = *ff_vvc_get_mvf(fc, x0, y0);
+    if (pu->bdof_flag)
+        *sb_bdof_flag = 1;
+    if (pu->dmvr_flag) {
+        VVCFrame* ref[2];
+        if (pred_get_refs(lc, ref, mv) < 0)
+            return;
+        dmvr_mv_refine(lc, mv, orig_mv, sb_bdof_flag, ref[0]->frame, ref[1]->frame, x0, y0, sbw, sbh);
+        set_dmvr_info(fc, x0, y0, sbw, sbh, mv);
+    }
+}
+
+static void pred_regular_blk(VVCLocalContext *lc, const int skip_ciip)
+{
+    const VVCFrameContext *fc   = lc->fc;
+    const CodingUnit *cu        = lc->cu;
+    PredictionUnit *pu          = &lc->cu->pu;
+    const MotionInfo *mi        = &pu->mi;
+    MvField mv, orig_mv;
+    int sbw, sbh, sb_bdof_flag = 0;
+
+    if (cu->ciip_flag && skip_ciip)
+        return;
+
+    sbw = cu->cb_width / mi->num_sb_x;
+    sbh = cu->cb_height / mi->num_sb_y;
+
+    for (int sby = 0; sby < mi->num_sb_y; sby++) {
+        for (int sbx = 0; sbx < mi->num_sb_x; sbx++) {
+            const int x0 = cu->x0 + sbx * sbw;
+            const int y0 = cu->y0 + sby * sbh;
+
+            if (cu->ciip_flag)
+                ff_vvc_set_neighbour_available(lc, x0, y0, sbw, sbh);
+
+            derive_sb_mv(lc, &mv, &orig_mv, &sb_bdof_flag, x0, y0, sbw, sbh);
+            pred_regular_luma(lc, mi->hpel_if_idx, mi->hpel_if_idx, &mv, x0, y0, sbw, sbh, &orig_mv, sb_bdof_flag);
+            if (fc->ps.sps->r->sps_chroma_format_idc)
+                pred_regular_chroma(lc, &mv, x0, y0, sbw, sbh, &orig_mv, pu->dmvr_flag);
+        }
+    }
+}
+
+static void derive_affine_mvc(MvField *mvc, const VVCFrameContext *fc, const MvField *mv,
+    const int x0, const int y0, const int sbw, const int sbh)
+{
+    const int hs = fc->ps.sps->hshift[1];
+    const int vs = fc->ps.sps->vshift[1];
+    const MvField* mv2 = ff_vvc_get_mvf(fc, x0 + hs * sbw, y0 + vs * sbh);
+    *mvc = *mv;
+    mvc->mv[0].x += mv2->mv[0].x;
+    mvc->mv[0].y += mv2->mv[0].y;
+    mvc->mv[1].x += mv2->mv[1].x;
+    mvc->mv[1].y += mv2->mv[1].y;
+    ff_vvc_round_mv(mvc->mv + 0, 0, 1);
+    ff_vvc_round_mv(mvc->mv + 1, 0, 1);
+}
+
+static void pred_affine_blk(VVCLocalContext *lc)
+{
+    const VVCFrameContext *fc  = lc->fc;
+    const CodingUnit *cu       = lc->cu;
+    const PredictionUnit *pu   = &cu->pu;
+    const MotionInfo *mi       = &pu->mi;
+    const int x0   = cu->x0;
+    const int y0   = cu->y0;
+    const int sbw  = cu->cb_width / mi->num_sb_x;
+    const int sbh  = cu->cb_height / mi->num_sb_y;
+    const int hs = fc->ps.sps->hshift[1];
+    const int vs = fc->ps.sps->vshift[1];
+
+    for (int sby = 0; sby < mi->num_sb_y; sby++) {
+        for (int sbx = 0; sbx < mi->num_sb_x; sbx++) {
+            const int x = x0 + sbx * sbw;
+            const int y = y0 + sby * sbh;
+
+            uint8_t *dst0 = POS(0, x, y);
+            const MvField *mv = ff_vvc_get_mvf(fc, x, y);
+            VVCFrame *ref[2];
+
+            if (pred_get_refs(lc, ref, mv) < 0)
+                return;
+
+            if (mi->pred_flag != PF_BI) {
+                const int lx = mi->pred_flag - PF_L0;
+                luma_prof_uni(lc, dst0, fc->frame->linesize[0], ref[lx]->frame,
+                    mv, x, y, sbw, sbh, pu->cb_prof_flag[lx],
+                    pu->diff_mv_x[lx], pu->diff_mv_y[lx]);
+            } else {
+                luma_prof_bi(lc, dst0, fc->frame->linesize[0], ref[0]->frame, ref[1]->frame,
+                    mv, x, y, sbw, sbh);
+            }
+            if (fc->ps.sps->r->sps_chroma_format_idc) {
+                if (!av_mod_uintp2(sby, vs) && !av_mod_uintp2(sbx, hs)) {
+                    MvField mvc;
+                    derive_affine_mvc(&mvc, fc, mv, x, y, sbw, sbh);
+                    pred_regular_chroma(lc, &mvc, x, y, sbw<<hs, sbh<<vs, NULL, 0);
+
+                }
+            }
+
+        }
+    }
+}
+
+static void predict_inter(VVCLocalContext *lc)
+{
+    const VVCFrameContext *fc   = lc->fc;
+    const CodingUnit *cu        = lc->cu;
+    const PredictionUnit *pu    = &cu->pu;
+
+    if (pu->merge_gpm_flag)
+        pred_gpm_blk(lc);
+    else if (pu->inter_affine_flag)
+        pred_affine_blk(lc);
+    else
+        pred_regular_blk(lc, 1);    //intra block is not ready yet, skip ciip
+
+    if (!pu->dmvr_flag)
+        fill_dmvr_info(fc, cu->x0, cu->y0, cu->cb_width, cu->cb_height);
+    if (lc->sc->sh.r->sh_lmcs_used_flag && !cu->ciip_flag) {
+        uint8_t* dst0 = POS(0, cu->x0, cu->y0);
+        fc->vvcdsp.lmcs.filter(dst0, fc->frame->linesize[LUMA], cu->cb_width, cu->cb_height, fc->ps.lmcs.fwd_lut);
+    }
+}
+
+static int has_inter_luma(const CodingUnit *cu)
+{
+    return cu->pred_mode != MODE_INTRA && cu->pred_mode != MODE_PLT && cu->tree_type != DUAL_TREE_CHROMA;
+}
+
+int ff_vvc_predict_inter(VVCLocalContext *lc, const int rs)
+{
+    const VVCFrameContext *fc   = lc->fc;
+    const CTU *ctu              = fc->tab.ctus + rs;
+    CodingUnit *cu              = ctu->cus;
+
+    while (cu) {
+        lc->cu = cu;
+        if (has_inter_luma(cu))
+            predict_inter(lc);
+        cu = cu->next;
+    }
+
+    return 0;
+}
+
+void ff_vvc_predict_ciip(VVCLocalContext *lc)
+{
+    av_assert0(lc->cu->ciip_flag);
+
+    //todo: refact out ciip from pred_regular_blk
+    pred_regular_blk(lc, 0);
+}
+
+#undef POS
diff --git a/libavcodec/vvc/vvc_inter.h b/libavcodec/vvc/vvc_inter.h
new file mode 100644
index 0000000000..61fc44b3a0
--- /dev/null
+++ b/libavcodec/vvc/vvc_inter.h
@@ -0,0 +1,42 @@ 
+/*
+ * VVC inter prediction
+ *
+ * Copyright (C) 2023 Nuo Mi
+ *
+ * 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
+ */
+
+#ifndef AVCODEC_VVC_VVC_INTER_H
+#define AVCODEC_VVC_VVC_INTER_H
+
+#include "vvc_ctu.h"
+
+/**
+ * Loop entire CTU to predict all inter coding blocks
+ * @param lc local context for CTU
+ * @param rs raster order for the CTU
+ * @return AVERROR
+ */
+int ff_vvc_predict_inter(VVCLocalContext *lc, int rs);
+
+/**
+ * CIIP(Combined Inter-Intra Prediction) for a coding block
+ * @param lc local context for CTU
+ */
+void ff_vvc_predict_ciip(VVCLocalContext *lc);
+
+#endif // AVCODEC_VVC_VVC_INTER_H
diff --git a/libavcodec/vvc/vvc_inter_template.c b/libavcodec/vvc/vvc_inter_template.c
new file mode 100644
index 0000000000..b67b66a2dc
--- /dev/null
+++ b/libavcodec/vvc/vvc_inter_template.c
@@ -0,0 +1,1023 @@ 
+/*
+ * VVC inter prediction DSP
+ *
+ * Copyright (C) 2022 Nuo Mi
+ *
+ * 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
+ */
+
+////////////////////////////////////////////////////////////////////////////////
+//
+////////////////////////////////////////////////////////////////////////////////
+static void FUNC(put_pixels)(int16_t *dst,
+    const uint8_t *_src, const ptrdiff_t _src_stride,
+    const int height, const int8_t *hf, const int8_t *vf, const int width)
+{
+    const pixel *src            = (const pixel *)_src;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = src[x] << (14 - BIT_DEPTH);
+        src += src_stride;
+        dst += MAX_PB_SIZE;
+    }
+}
+
+static void FUNC(put_uni_pixels)(uint8_t *_dst, const ptrdiff_t _dst_stride,
+    const uint8_t *_src, const ptrdiff_t _src_stride, const int height,
+     const int8_t *hf, const int8_t *vf, const int width)
+{
+    const pixel *src            = (const pixel *)_src;
+    pixel *dst                  = (pixel *)_dst;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const ptrdiff_t dst_stride  = _dst_stride / sizeof(pixel);
+
+    for (int y = 0; y < height; y++) {
+        memcpy(dst, src, width * sizeof(pixel));
+        src += src_stride;
+        dst += dst_stride;
+    }
+}
+
+static void FUNC(put_uni_w_pixels)(uint8_t *_dst, const ptrdiff_t _dst_stride,
+    const uint8_t *_src, const ptrdiff_t _src_stride, const int height,
+    const int denom, const int wx, const int _ox,  const int8_t *hf, const int8_t *vf,
+    const int width)
+{
+    const pixel *src            = (const pixel *)_src;
+    pixel *dst                  = (pixel *)_dst;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const ptrdiff_t dst_stride  = _dst_stride / sizeof(pixel);
+    const int shift             = denom + 14 - BIT_DEPTH;
+#if BIT_DEPTH < 14
+    const int offset            = 1 << (shift - 1);
+#else
+    const int offset            = 0;
+#endif
+    const int ox                = _ox * (1 << (BIT_DEPTH - 8));
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++) {
+            const int v = (src[x] << (14 - BIT_DEPTH));
+            dst[x] = av_clip_pixel(((v * wx + offset) >> shift) + ox);
+        }
+        src += src_stride;
+        dst += dst_stride;
+    }
+}
+
+////////////////////////////////////////////////////////////////////////////////
+//
+////////////////////////////////////////////////////////////////////////////////
+#define LUMA_FILTER(src, stride)                                               \
+    (filter[0] * src[x - 3 * stride] +                                         \
+     filter[1] * src[x - 2 * stride] +                                         \
+     filter[2] * src[x -     stride] +                                         \
+     filter[3] * src[x             ] +                                         \
+     filter[4] * src[x +     stride] +                                         \
+     filter[5] * src[x + 2 * stride] +                                         \
+     filter[6] * src[x + 3 * stride] +                                         \
+     filter[7] * src[x + 4 * stride])
+
+static void FUNC(put_luma_h)(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
+    const int height, const int8_t *hf, const int8_t *vf, const int width)
+{
+    const pixel *src           = (const pixel*)_src;
+    const ptrdiff_t src_stride = _src_stride / sizeof(pixel);
+    const int8_t *filter       = hf;
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = LUMA_FILTER(src, 1) >> (BIT_DEPTH - 8);
+        src += src_stride;
+        dst += MAX_PB_SIZE;
+    }
+}
+
+static void FUNC(put_luma_v)(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
+    const int height, const int8_t *hf, const int8_t *vf, const int width)
+{
+    const pixel *src           = (pixel*)_src;
+    const ptrdiff_t src_stride = _src_stride / sizeof(pixel);
+    const int8_t *filter       = vf;
+
+    for (int y = 0; y < height; y++)  {
+        for (int x = 0; x < width; x++)
+            dst[x] = LUMA_FILTER(src, src_stride) >> (BIT_DEPTH - 8);
+        src += src_stride;
+        dst += MAX_PB_SIZE;
+    }
+}
+
+static void FUNC(put_luma_hv)(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
+    const int height, const int8_t *hf, const int8_t *vf, const int width)
+{
+    int16_t tmp_array[(MAX_PB_SIZE + LUMA_EXTRA) * MAX_PB_SIZE];
+    int16_t *tmp                = tmp_array;
+    const pixel *src            = (const pixel*)_src;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const int8_t *filter        = hf;
+
+    src   -= LUMA_EXTRA_BEFORE * src_stride;
+    for (int y = 0; y < height + LUMA_EXTRA; y++) {
+        for (int x = 0; x < width; x++)
+            tmp[x] = LUMA_FILTER(src, 1) >> (BIT_DEPTH - 8);
+        src += src_stride;
+        tmp += MAX_PB_SIZE;
+    }
+
+    tmp    = tmp_array + LUMA_EXTRA_BEFORE * MAX_PB_SIZE;
+    filter = vf;
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = LUMA_FILTER(tmp, MAX_PB_SIZE) >> 6;
+        tmp += MAX_PB_SIZE;
+        dst += MAX_PB_SIZE;
+    }
+}
+
+static void FUNC(put_uni_luma_h)(uint8_t *_dst,  const ptrdiff_t _dst_stride,
+    const uint8_t *_src, const ptrdiff_t _src_stride,
+    const int height, const int8_t *hf, const int8_t *vf, const int width)
+{
+    const pixel *src           = (const pixel*)_src;
+    pixel *dst                 = (pixel *)_dst;
+    const ptrdiff_t src_stride = _src_stride / sizeof(pixel);
+    const ptrdiff_t dst_stride = _dst_stride / sizeof(pixel);
+    const int8_t *filter       = hf;
+    const int shift            = 14 - BIT_DEPTH;
+#if BIT_DEPTH < 14
+    const int offset           = 1 << (shift - 1);
+#else
+    const int offset           = 0;
+#endif
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++) {
+            const int val = LUMA_FILTER(src, 1) >> (BIT_DEPTH - 8);
+            dst[x]        = av_clip_pixel((val + offset) >> shift);
+        }
+        src   += src_stride;
+        dst   += dst_stride;
+    }
+}
+
+static void FUNC(put_uni_luma_v)(uint8_t *_dst,  const ptrdiff_t _dst_stride,
+    const uint8_t *_src, const ptrdiff_t _src_stride,
+    const int height, const int8_t *hf, const int8_t *vf, const int width)
+{
+
+    const pixel *src            = (const pixel*)_src;
+    pixel *dst                  = (pixel *)_dst;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const ptrdiff_t dst_stride  = _dst_stride / sizeof(pixel);
+    const int8_t *filter        = vf;
+    const int shift             = 14 - BIT_DEPTH;
+#if BIT_DEPTH < 14
+    const int offset            = 1 << (shift - 1);
+#else
+    const int offset            = 0;
+#endif
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++) {
+            const int val = LUMA_FILTER(src, src_stride) >> (BIT_DEPTH - 8);
+            dst[x]        = av_clip_pixel((val + offset) >> shift);
+        }
+        src   += src_stride;
+        dst   += dst_stride;
+    }
+}
+
+static void FUNC(put_uni_luma_hv)(uint8_t *_dst, const ptrdiff_t _dst_stride,
+    const uint8_t *_src, const ptrdiff_t _src_stride,
+    const int height, const int8_t *hf, const int8_t *vf, const int width)
+{
+    int16_t tmp_array[(MAX_PB_SIZE + LUMA_EXTRA) * MAX_PB_SIZE];
+    int16_t *tmp                = tmp_array;
+    const pixel *src            = (const pixel*)_src;
+    pixel *dst                  = (pixel *)_dst;
+    const ptrdiff_t dst_stride  = _dst_stride / sizeof(pixel);
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const int8_t *filter        = hf;
+    const int shift             =  14 - BIT_DEPTH;
+#if BIT_DEPTH < 14
+    const int offset            = 1 << (shift - 1);
+#else
+    const int offset            = 0;
+#endif
+
+    src   -= LUMA_EXTRA_BEFORE * src_stride;
+    for (int y = 0; y < height + LUMA_EXTRA; y++) {
+        for (int x = 0; x < width; x++)
+            tmp[x] = LUMA_FILTER(src, 1) >> (BIT_DEPTH - 8);
+        src += src_stride;
+        tmp += MAX_PB_SIZE;
+    }
+
+    tmp    = tmp_array + LUMA_EXTRA_BEFORE * MAX_PB_SIZE;
+    filter = vf;
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++) {
+            const int val = LUMA_FILTER(tmp, MAX_PB_SIZE) >> 6;
+            dst[x]  = av_clip_pixel((val  + offset) >> shift);
+        }
+        tmp += MAX_PB_SIZE;
+        dst += dst_stride;
+    }
+
+}
+
+static void FUNC(put_uni_luma_w_h)(uint8_t *_dst,  const ptrdiff_t _dst_stride,
+    const uint8_t *_src, const ptrdiff_t _src_stride, int height,
+    const int denom, const int wx, const int _ox, const int8_t *hf, const int8_t *vf,
+    const int width)
+{
+    const pixel *src            = (const pixel*)_src;
+    pixel *dst                  = (pixel *)_dst;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const ptrdiff_t dst_stride  = _dst_stride / sizeof(pixel);
+    const int8_t *filter        = hf;
+    const int ox                = _ox * (1 << (BIT_DEPTH - 8));
+    const int shift             = denom + 14 - BIT_DEPTH;
+#if BIT_DEPTH < 14
+    const int offset            = 1 << (shift - 1);
+#else
+    const int offset            = 0;
+#endif
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = av_clip_pixel((((LUMA_FILTER(src, 1) >> (BIT_DEPTH - 8)) * wx + offset) >> shift) + ox);
+        src += src_stride;
+        dst += dst_stride;
+    }
+}
+
+static void FUNC(put_uni_luma_w_v)(uint8_t *_dst,  const ptrdiff_t _dst_stride,
+    const uint8_t *_src, const ptrdiff_t _src_stride, const int height,
+    const int denom, const int wx, const int _ox, const int8_t *hf, const int8_t *vf,
+    const int width)
+{
+    const pixel *src            = (const pixel*)_src;
+    pixel *dst                  = (pixel *)_dst;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const ptrdiff_t dst_stride  = _dst_stride / sizeof(pixel);
+    const int8_t *filter        = vf;
+    const int ox                = _ox * (1 << (BIT_DEPTH - 8));
+    const int shift             = denom + 14 - BIT_DEPTH;
+#if BIT_DEPTH < 14
+    const int offset            = 1 << (shift - 1);
+#else
+    const int offset            = 0;
+#endif
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = av_clip_pixel((((LUMA_FILTER(src, src_stride) >> (BIT_DEPTH - 8)) * wx + offset) >> shift) + ox);
+        src += src_stride;
+        dst += dst_stride;
+    }
+}
+
+static void FUNC(put_uni_luma_w_hv)(uint8_t *_dst,  const ptrdiff_t _dst_stride,
+    const uint8_t *_src, const ptrdiff_t _src_stride, const int height, const int denom,
+    const int wx, const int _ox, const int8_t *hf, const int8_t *vf, const int width)
+{
+    int16_t tmp_array[(MAX_PB_SIZE + LUMA_EXTRA) * MAX_PB_SIZE];
+    int16_t *tmp                = tmp_array;
+    const pixel *src            = (const pixel*)_src;
+    pixel *dst                  = (pixel *)_dst;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const ptrdiff_t dst_stride  = _dst_stride / sizeof(pixel);
+    const int8_t *filter        = hf;
+    const int ox                = _ox * (1 << (BIT_DEPTH - 8));
+    const int shift             = denom + 14 - BIT_DEPTH;
+#if BIT_DEPTH < 14
+    const int offset            = 1 << (shift - 1);
+#else
+    const int offset            = 0;
+#endif
+
+    src   -= LUMA_EXTRA_BEFORE * src_stride;
+    for (int y = 0; y < height + LUMA_EXTRA; y++) {
+        for (int x = 0; x < width; x++)
+            tmp[x] = LUMA_FILTER(src, 1) >> (BIT_DEPTH - 8);
+        src += src_stride;
+        tmp += MAX_PB_SIZE;
+    }
+
+    tmp    = tmp_array + LUMA_EXTRA_BEFORE * MAX_PB_SIZE;
+    filter = vf;
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = av_clip_pixel((((LUMA_FILTER(tmp, MAX_PB_SIZE) >> 6) * wx + offset) >> shift) + ox);
+        tmp += MAX_PB_SIZE;
+        dst += dst_stride;
+    }
+}
+
+////////////////////////////////////////////////////////////////////////////////
+//
+////////////////////////////////////////////////////////////////////////////////
+#define CHROMA_FILTER(src, stride)                                               \
+    (filter[0] * src[x - stride] +                                             \
+     filter[1] * src[x]          +                                             \
+     filter[2] * src[x + stride] +                                             \
+     filter[3] * src[x + 2 * stride])
+
+static void FUNC(put_chroma_h)(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
+    const int height, const int8_t *hf, const int8_t *vf, const int width)
+{
+    const pixel *src            = (const pixel *)_src;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const int8_t *filter        = hf;
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = CHROMA_FILTER(src, 1) >> (BIT_DEPTH - 8);
+        src += src_stride;
+        dst += MAX_PB_SIZE;
+    }
+}
+
+static void FUNC(put_chroma_v)(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
+    const int height, const int8_t *hf, const int8_t *vf, const int width)
+{
+    const pixel *src            = (const pixel *)_src;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const int8_t *filter        = vf;
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = CHROMA_FILTER(src, src_stride) >> (BIT_DEPTH - 8);
+        src += src_stride;
+        dst += MAX_PB_SIZE;
+    }
+}
+
+static void FUNC(put_chroma_hv)(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
+    const int height, const int8_t *hf, const int8_t *vf, const int width)
+{
+    int16_t tmp_array[(MAX_PB_SIZE + CHROMA_EXTRA) * MAX_PB_SIZE];
+    int16_t *tmp                = tmp_array;
+    const pixel *src            = (const pixel *)_src;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const int8_t *filter        = hf;
+
+    src -= CHROMA_EXTRA_BEFORE * src_stride;
+
+    for (int y = 0; y < height + CHROMA_EXTRA; y++) {
+        for (int x = 0; x < width; x++)
+            tmp[x] = CHROMA_FILTER(src, 1) >> (BIT_DEPTH - 8);
+        src += src_stride;
+        tmp += MAX_PB_SIZE;
+    }
+
+    tmp    = tmp_array + CHROMA_EXTRA_BEFORE * MAX_PB_SIZE;
+    filter = vf;
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = CHROMA_FILTER(tmp, MAX_PB_SIZE) >> 6;
+        tmp += MAX_PB_SIZE;
+        dst += MAX_PB_SIZE;
+    }
+}
+
+static void FUNC(put_uni_chroma_h)(uint8_t *_dst, const ptrdiff_t _dst_stride,
+    const uint8_t *_src, const ptrdiff_t _src_stride,
+    const int height, const int8_t *hf, const int8_t *vf, const int width)
+{
+    const pixel *src            = (const pixel *)_src;
+    pixel *dst                  = (pixel *)_dst;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const ptrdiff_t dst_stride  = _dst_stride / sizeof(pixel);
+    const int8_t *filter        = hf;
+    const int shift             = 14 - BIT_DEPTH;
+#if BIT_DEPTH < 14
+    const int offset            = 1 << (shift - 1);
+#else
+    const int offset            = 0;
+#endif
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = av_clip_pixel(((CHROMA_FILTER(src, 1) >> (BIT_DEPTH - 8)) + offset) >> shift);
+        src += src_stride;
+        dst += dst_stride;
+    }
+}
+
+static void FUNC(put_uni_chroma_v)(uint8_t *_dst, const ptrdiff_t _dst_stride,
+    const uint8_t *_src, const ptrdiff_t _src_stride,
+    const int height, const int8_t *hf, const int8_t *vf, const int width)
+{
+    const pixel *src            = (const pixel *)_src;
+    pixel *dst                  = (pixel *)_dst;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const ptrdiff_t dst_stride  = _dst_stride / sizeof(pixel);
+    const int8_t *filter        = vf;
+    const int shift             = 14 - BIT_DEPTH;
+#if BIT_DEPTH < 14
+    const int offset            = 1 << (shift - 1);
+#else
+    const int offset            = 0;
+#endif
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = av_clip_pixel(((CHROMA_FILTER(src, src_stride) >> (BIT_DEPTH - 8)) + offset) >> shift);
+        src += src_stride;
+        dst += dst_stride;
+    }
+}
+
+static void FUNC(put_uni_chroma_hv)(uint8_t *_dst, const ptrdiff_t _dst_stride,
+    const uint8_t *_src, const ptrdiff_t _src_stride,
+    const int height, const int8_t *hf, const int8_t *vf, const int width)
+{
+    int16_t tmp_array[(MAX_PB_SIZE + CHROMA_EXTRA) * MAX_PB_SIZE];
+    int16_t *tmp                = tmp_array;
+    const pixel *src            = (const pixel *)_src;
+    pixel *dst                  = (pixel *)_dst;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const ptrdiff_t dst_stride  = _dst_stride / sizeof(pixel);
+    const int8_t *filter        = hf;
+    const int shift             = 14 - BIT_DEPTH;
+#if BIT_DEPTH < 14
+    const int offset            = 1 << (shift - 1);
+#else
+    const int offset            = 0;
+#endif
+
+    src -= CHROMA_EXTRA_BEFORE * src_stride;
+
+    for (int y = 0; y < height + CHROMA_EXTRA; y++) {
+        for (int x = 0; x < width; x++)
+            tmp[x] = CHROMA_FILTER(src, 1) >> (BIT_DEPTH - 8);
+        src += src_stride;
+        tmp += MAX_PB_SIZE;
+    }
+
+    tmp    = tmp_array + CHROMA_EXTRA_BEFORE * MAX_PB_SIZE;
+    filter = vf;
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = av_clip_pixel(((CHROMA_FILTER(tmp, MAX_PB_SIZE) >> 6) + offset) >> shift);
+        tmp += MAX_PB_SIZE;
+        dst += dst_stride;
+    }
+}
+
+static void FUNC(put_uni_chroma_w_h)(uint8_t *_dst, ptrdiff_t _dst_stride,
+    const uint8_t *_src, ptrdiff_t _src_stride, int height, int denom, int wx, int ox,
+    const int8_t *hf, const int8_t *vf, int width)
+{
+    const pixel *src            = (const pixel *)_src;
+    pixel *dst                  = (pixel *)_dst;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const ptrdiff_t dst_stride  = _dst_stride / sizeof(pixel);
+    const int8_t *filter        = hf;
+    const int shift             = denom + 14 - BIT_DEPTH;
+#if BIT_DEPTH < 14
+    const int offset            = 1 << (shift - 1);
+#else
+    const int offset            = 0;
+#endif
+
+    ox     = ox * (1 << (BIT_DEPTH - 8));
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++) {
+            dst[x] = av_clip_pixel((((CHROMA_FILTER(src, 1) >> (BIT_DEPTH - 8)) * wx + offset) >> shift) + ox);
+        }
+        dst += dst_stride;
+        src += src_stride;
+    }
+}
+
+static void FUNC(put_uni_chroma_w_v)(uint8_t *_dst, const ptrdiff_t _dst_stride,
+    const uint8_t *_src, const ptrdiff_t _src_stride, const int height,
+    const int denom, const int wx, const int _ox, const int8_t *hf, const int8_t *vf,
+    const int width)
+{
+    const pixel *src            = (const pixel *)_src;
+    pixel *dst                  = (pixel *)_dst;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const ptrdiff_t dst_stride  = _dst_stride / sizeof(pixel);
+    const int8_t *filter        = vf;
+    const int shift             = denom + 14 - BIT_DEPTH;
+    const int ox                = _ox * (1 << (BIT_DEPTH - 8));
+#if BIT_DEPTH < 14
+    int offset                  = 1 << (shift - 1);
+#else
+    int offset                  = 0;
+#endif
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++) {
+            dst[x] = av_clip_pixel((((CHROMA_FILTER(src, src_stride) >> (BIT_DEPTH - 8)) * wx + offset) >> shift) + ox);
+        }
+        dst += dst_stride;
+        src += src_stride;
+    }
+}
+
+static void FUNC(put_uni_chroma_w_hv)(uint8_t *_dst, ptrdiff_t _dst_stride,
+     const uint8_t *_src, ptrdiff_t _src_stride,  int height, int denom, int wx, int ox,
+     const int8_t *hf, const int8_t *vf, int width)
+{
+    int16_t tmp_array[(MAX_PB_SIZE + CHROMA_EXTRA) * MAX_PB_SIZE];
+    int16_t *tmp                = tmp_array;
+    const pixel *src            = (const pixel *)_src;
+    pixel *dst                  = (pixel *)_dst;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const ptrdiff_t dst_stride  = _dst_stride / sizeof(pixel);
+    const int8_t *filter        = hf;
+    const int shift             = denom + 14 - BIT_DEPTH;
+#if BIT_DEPTH < 14
+    const int offset            = 1 << (shift - 1);
+#else
+    const int offset            = 0;
+#endif
+
+    src -= CHROMA_EXTRA_BEFORE * src_stride;
+
+    for (int y = 0; y < height + CHROMA_EXTRA; y++) {
+        for (int x = 0; x < width; x++)
+            tmp[x] = CHROMA_FILTER(src, 1) >> (BIT_DEPTH - 8);
+        src += src_stride;
+        tmp += MAX_PB_SIZE;
+    }
+
+    tmp    = tmp_array + CHROMA_EXTRA_BEFORE * MAX_PB_SIZE;
+    filter = vf;
+
+    ox     = ox * (1 << (BIT_DEPTH - 8));
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = av_clip_pixel((((CHROMA_FILTER(tmp, MAX_PB_SIZE) >> 6) * wx + offset) >> shift) + ox);
+        tmp += MAX_PB_SIZE;
+        dst += dst_stride;
+    }
+}
+
+static void FUNC(avg)(uint8_t *_dst, const ptrdiff_t _dst_stride,
+    const int16_t *src0, const int16_t *src1, const int width, const int height)
+{
+    pixel *dst                  = (pixel*)_dst;
+    const ptrdiff_t dst_stride  = _dst_stride / sizeof(pixel);
+    const int shift             = FFMAX(3, 15 - BIT_DEPTH);
+    const int offset            = 1 << (shift - 1);
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = av_clip_pixel((src0[x] + src1[x] + offset) >> shift);
+        src0 += MAX_PB_SIZE;
+        src1 += MAX_PB_SIZE;
+        dst  += dst_stride;
+    }
+}
+
+static void FUNC(w_avg)(uint8_t *_dst, const ptrdiff_t _dst_stride,
+    const int16_t *src0, const int16_t *src1, const int width, const int height,
+    const int denom, const int w0, const int w1, const int o0, const int o1)
+{
+    pixel *dst                  = (pixel*)_dst;
+    const ptrdiff_t dst_stride  = _dst_stride / sizeof(pixel);
+    const int shift             = denom + FFMAX(3, 15 - BIT_DEPTH);
+    const int offset            = (((o0 + o1) << (BIT_DEPTH - 8)) + 1) << (shift - 1);
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = av_clip_pixel((src0[x] * w0 + src1[x] * w1 + offset) >> shift);
+        src0 += MAX_PB_SIZE;
+        src1 += MAX_PB_SIZE;
+        dst  += dst_stride;
+    }
+}
+
+static void FUNC(put_ciip)(uint8_t *_dst, const ptrdiff_t _dst_stride,
+    const int width, const int height,
+    const uint8_t *_inter, const ptrdiff_t _inter_stride, const int intra_weight)
+{
+    pixel *dst                = (pixel *)_dst;
+    pixel *inter              = (pixel *)_inter;
+    const size_t dst_stride   = _dst_stride / sizeof(pixel);
+    const size_t inter_stride = _inter_stride / sizeof(pixel);
+    const int inter_weight    = 4 - intra_weight;
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = (dst[x] * intra_weight + inter[x] * inter_weight + 2) >> 2;
+        dst   += dst_stride;
+        inter += inter_stride;
+    }
+}
+
+static void FUNC(put_gpm)(uint8_t *_dst, ptrdiff_t dst_stride,
+    const int width, const int height,
+    const int16_t *src0, const int16_t *src1,
+    const uint8_t *weights, const int step_x, const int step_y)
+{
+    const int shift  = FFMAX(5, 17 - BIT_DEPTH);
+    const int offset = 1 << (shift - 1);
+    pixel *dst       = (pixel *)_dst;
+
+    dst_stride /= sizeof(pixel);
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++) {
+            const uint8_t w = weights[x * step_x];
+            dst[x] = av_clip_pixel((src0[x] * w + src1[x] * (8 - w) + offset) >> shift);
+        }
+        dst     += dst_stride;
+        src0    += MAX_PB_SIZE;
+        src1    += MAX_PB_SIZE;
+        weights += step_y;
+    }
+}
+
+//8.5.6.3.3 Luma integer sample fetching process, add one extra pad line
+static void FUNC(bdof_fetch_samples)(int16_t *_dst, const uint8_t *_src, const ptrdiff_t _src_stride,
+    const int x_frac, const int y_frac, const int width, const int height)
+{
+    const int x_off             = (x_frac >> 3) - 1;
+    const int y_off             = (y_frac >> 3) - 1;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const pixel *src            = (pixel*)_src + (x_off) + y_off * src_stride;
+    int16_t *dst                = _dst - 1 - MAX_PB_SIZE;
+    const int shift             = 14 - BIT_DEPTH;
+    const int bdof_width        = width + 2 * BDOF_BORDER_EXT;
+
+    // top
+    for (int i = 0; i < bdof_width; i++)
+        dst[i] = src[i] << shift;
+
+    dst += MAX_PB_SIZE;
+    src += src_stride;
+
+    for (int i = 0; i < height; i++) {
+        dst[0] = src[0] << shift;
+        dst[1 + width] = src[1 + width] << shift;
+        dst += MAX_PB_SIZE;
+        src += src_stride;
+    }
+    for (int i = 0; i < bdof_width; i++)
+        dst[i] = src[i] << shift;
+}
+
+//8.5.6.3.3 Luma integer sample fetching process
+static void FUNC(fetch_samples)(int16_t *_dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int x_frac, const int y_frac)
+{
+    FUNC(bdof_fetch_samples)(_dst, _src, _src_stride, x_frac, y_frac, AFFINE_MIN_BLOCK_SIZE, AFFINE_MIN_BLOCK_SIZE);
+}
+
+static void FUNC(prof_grad_filter)(int16_t *_gradient_h, int16_t *_gradient_v, const ptrdiff_t gradient_stride,
+    const int16_t *_src, const ptrdiff_t src_stride, const int width, const int height, const int pad)
+{
+    const int shift     = 6;
+    const int16_t *src  = _src;
+    int16_t *gradient_h = _gradient_h + pad * (1 + gradient_stride);
+    int16_t *gradient_v = _gradient_v + pad * (1 + gradient_stride);
+
+    for (int y = 0; y < height; y++) {
+        const int16_t *p = src;
+        for (int x = 0; x < width; x++) {
+            gradient_h[x] = (p[1] >> shift) - (p[-1] >> shift);
+            gradient_v[x] = (p[src_stride] >> shift) - (p[-src_stride] >> shift);
+            p++;
+        }
+        gradient_h += gradient_stride;
+        gradient_v += gradient_stride;
+        src += src_stride;
+    }
+    if (pad) {
+        pad_int16(_gradient_h + 1 + gradient_stride, gradient_stride, width, height);
+        pad_int16(_gradient_v + 1 + gradient_stride, gradient_stride, width, height);
+    }
+}
+
+static void FUNC(apply_prof)(int16_t *dst, const int16_t *src, const int16_t *diff_mv_x, const int16_t *diff_mv_y)
+{
+    const int limit     = (1 << FFMAX(13, BIT_DEPTH + 1));          ///< dILimit
+
+    int16_t gradient_h[AFFINE_MIN_BLOCK_SIZE * AFFINE_MIN_BLOCK_SIZE];
+    int16_t gradient_v[AFFINE_MIN_BLOCK_SIZE * AFFINE_MIN_BLOCK_SIZE];
+    FUNC(prof_grad_filter)(gradient_h, gradient_v, AFFINE_MIN_BLOCK_SIZE, src, MAX_PB_SIZE, AFFINE_MIN_BLOCK_SIZE, AFFINE_MIN_BLOCK_SIZE, 0);
+
+    for (int y = 0; y < AFFINE_MIN_BLOCK_SIZE; y++) {
+        for (int x = 0; x < AFFINE_MIN_BLOCK_SIZE; x++) {
+            const int o = y * AFFINE_MIN_BLOCK_SIZE + x;
+            const int di = gradient_h[o] * diff_mv_x[o] + gradient_v[o] * diff_mv_y[o];
+            const int val = src[x] + av_clip(di, -limit, limit - 1);
+            dst[x] = val;
+
+        }
+        src += MAX_PB_SIZE;
+        dst += MAX_PB_SIZE;
+    }
+}
+
+static void FUNC(apply_prof_uni)(uint8_t *_dst, const ptrdiff_t _dst_stride, const int16_t *src, const int16_t *diff_mv_x, const int16_t *diff_mv_y)
+{
+    const int limit             = (1 << FFMAX(13, BIT_DEPTH + 1));          ///< dILimit
+    pixel *dst                  = (pixel*)_dst;
+    const ptrdiff_t dst_stride  = _dst_stride / sizeof(pixel);
+    const int shift             = 14 - BIT_DEPTH;
+#if BIT_DEPTH < 14
+    const int offset            = 1 << (shift - 1);
+#else
+    const int offset            = 0;
+#endif
+    int16_t gradient_h[AFFINE_MIN_BLOCK_SIZE * AFFINE_MIN_BLOCK_SIZE];
+    int16_t gradient_v[AFFINE_MIN_BLOCK_SIZE * AFFINE_MIN_BLOCK_SIZE];
+
+    FUNC(prof_grad_filter)(gradient_h, gradient_v, AFFINE_MIN_BLOCK_SIZE, src, MAX_PB_SIZE, AFFINE_MIN_BLOCK_SIZE, AFFINE_MIN_BLOCK_SIZE, 0);
+
+    for (int y = 0; y < AFFINE_MIN_BLOCK_SIZE; y++) {
+        for (int x = 0; x < AFFINE_MIN_BLOCK_SIZE; x++) {
+            const int o = y * AFFINE_MIN_BLOCK_SIZE + x;
+            const int di = gradient_h[o] * diff_mv_x[o] + gradient_v[o] * diff_mv_y[o];
+            const int val = src[x] + av_clip(di, -limit, limit - 1);
+            dst[x] = av_clip_pixel((val + offset) >> shift);
+
+        }
+        src += MAX_PB_SIZE;
+        dst += dst_stride;
+    }
+}
+
+static void FUNC(apply_prof_uni_w)(uint8_t *_dst, const ptrdiff_t _dst_stride,
+    const int16_t *src, const int16_t *diff_mv_x, const int16_t *diff_mv_y,
+    const int denom, const int wx, const int _ox)
+{
+    const int limit             = (1 << FFMAX(13, BIT_DEPTH + 1));          ///< dILimit
+    pixel *dst                  = (pixel*)_dst;
+    const ptrdiff_t dst_stride  = _dst_stride / sizeof(pixel);
+    const int shift             = denom + FFMAX(2, 14 - BIT_DEPTH);
+    const int offset            = 1 << (shift - 1);
+    const int ox                = _ox * (1 << (BIT_DEPTH - 8));
+    int16_t gradient_h[AFFINE_MIN_BLOCK_SIZE * AFFINE_MIN_BLOCK_SIZE];
+    int16_t gradient_v[AFFINE_MIN_BLOCK_SIZE * AFFINE_MIN_BLOCK_SIZE];
+
+    FUNC(prof_grad_filter)(gradient_h, gradient_v, AFFINE_MIN_BLOCK_SIZE, src, MAX_PB_SIZE, AFFINE_MIN_BLOCK_SIZE, AFFINE_MIN_BLOCK_SIZE, 0);
+
+    for (int y = 0; y < AFFINE_MIN_BLOCK_SIZE; y++) {
+        for (int x = 0; x < AFFINE_MIN_BLOCK_SIZE; x++) {
+            const int o = y * AFFINE_MIN_BLOCK_SIZE + x;
+            const int di = gradient_h[o] * diff_mv_x[o] + gradient_v[o] * diff_mv_y[o];
+            const int val = src[x] + av_clip(di, -limit, limit - 1);
+            dst[x] = av_clip_pixel(((val * wx + offset) >>  shift)  + ox);
+        }
+        src += MAX_PB_SIZE;
+        dst += dst_stride;
+    }
+}
+
+static void FUNC(derive_bdof_vx_vy)(const int16_t *_src0, const int16_t *_src1,
+    const int16_t **gradient_h, const int16_t **gradient_v, ptrdiff_t gradient_stride,
+    int* vx, int* vy)
+{
+    const int shift2 = 4;
+    const int shift3 = 1;
+    const int thres = 1 << 4;
+    int sgx2 = 0, sgy2 = 0, sgxgy = 0, sgxdi = 0, sgydi = 0;
+    const int16_t *src0 = _src0 - 1 - MAX_PB_SIZE;
+    const int16_t *src1 = _src1 - 1 - MAX_PB_SIZE;
+
+    for (int y = 0; y < BDOF_GRADIENT_SIZE; y++) {
+        for (int x = 0; x < BDOF_GRADIENT_SIZE; x++) {
+            const int diff = (src0[x] >> shift2) - (src1[x] >> shift2);
+            const int idx = gradient_stride * y  + x;
+            const int temph = (gradient_h[0][idx] + gradient_h[1][idx]) >> shift3;
+            const int tempv = (gradient_v[0][idx] + gradient_v[1][idx]) >> shift3;
+            sgx2 += FFABS(temph);
+            sgy2 += FFABS(tempv);
+            sgxgy += VVC_SIGN(tempv) * temph;
+            sgxdi += -VVC_SIGN(temph) * diff;
+            sgydi += -VVC_SIGN(tempv) * diff;
+        }
+        src0 += MAX_PB_SIZE;
+        src1 += MAX_PB_SIZE;
+    }
+    *vx = sgx2 > 0 ? av_clip((sgxdi << 2) >> av_log2(sgx2) , -thres + 1, thres - 1) : 0;
+    *vy = sgy2 > 0 ? av_clip(((sgydi << 2) - ((*vx * sgxgy) >> 1)) >> av_log2(sgy2), -thres + 1, thres - 1) : 0;
+}
+
+static void FUNC(apply_bdof_min_block)(pixel* dst, const ptrdiff_t dst_stride, const int16_t *src0, const int16_t *src1,
+    const int16_t **gradient_h, const int16_t **gradient_v, const int vx, const int vy)
+{
+    const int shift4 = 15 - BIT_DEPTH;
+    const int offset4 = 1 << (shift4 - 1);
+
+    const int16_t* gh[] = { gradient_h[0] + 1 + BDOF_PADDED_SIZE, gradient_h[1] + 1 + BDOF_PADDED_SIZE };
+    const int16_t* gv[] = { gradient_v[0] + 1 + BDOF_PADDED_SIZE, gradient_v[1] + 1 + BDOF_PADDED_SIZE };
+
+    for (int y = 0; y < BDOF_BLOCK_SIZE; y++) {
+        for (int x = 0; x < BDOF_BLOCK_SIZE; x++) {
+            const int idx = y * BDOF_PADDED_SIZE + x;
+            const int bdof_offset = vx * (gh[0][idx] - gh[1][idx]) + vy * (gv[0][idx] - gv[1][idx]);
+            dst[x] = av_clip_pixel((src0[x] + offset4 + src1[x] + bdof_offset) >> shift4);
+        }
+        dst  += dst_stride;
+        src0 += MAX_PB_SIZE;
+        src1 += MAX_PB_SIZE;
+    }
+}
+
+static void FUNC(apply_bdof)(uint8_t *_dst, const ptrdiff_t _dst_stride, int16_t *_src0, int16_t *_src1,
+    const int block_w, const int block_h)
+{
+    int16_t gradient_h[2][BDOF_PADDED_SIZE * BDOF_PADDED_SIZE];
+    int16_t gradient_v[2][BDOF_PADDED_SIZE * BDOF_PADDED_SIZE];
+    int vx, vy;
+    const ptrdiff_t dst_stride  = _dst_stride / sizeof(pixel);
+    pixel* dst                  = (pixel*)_dst;
+
+    FUNC(prof_grad_filter)(gradient_h[0], gradient_v[0], BDOF_PADDED_SIZE,
+        _src0, MAX_PB_SIZE, block_w, block_h, 1);
+    pad_int16(_src0, MAX_PB_SIZE, block_w, block_h);
+    FUNC(prof_grad_filter)(gradient_h[1], gradient_v[1], BDOF_PADDED_SIZE,
+        _src1, MAX_PB_SIZE, block_w, block_h, 1);
+    pad_int16(_src1, MAX_PB_SIZE, block_w, block_h);
+
+    for (int y = 0; y < block_h; y += BDOF_BLOCK_SIZE) {
+        for (int x = 0; x < block_w; x += BDOF_BLOCK_SIZE) {
+            const int16_t* src0 = _src0 + y * MAX_PB_SIZE + x;
+            const int16_t* src1 = _src1 + y * MAX_PB_SIZE + x;
+            pixel *d            = dst + x;
+            const int idx       = BDOF_PADDED_SIZE * y  + x;
+            const int16_t* gh[] = { gradient_h[0] + idx, gradient_h[1] + idx };
+            const int16_t* gv[] = { gradient_v[0] + idx, gradient_v[1] + idx };
+            FUNC(derive_bdof_vx_vy)(src0, src1, gh, gv, BDOF_PADDED_SIZE, &vx, &vy);
+            FUNC(apply_bdof_min_block)(d, dst_stride, src0, src1, gh, gv, vx, vy);
+        }
+        dst += BDOF_BLOCK_SIZE * dst_stride;
+    }
+}
+
+#define DMVR_FILTER(src, stride)                                                \
+    (filter[0] * src[x] +                                                       \
+     filter[1] * src[x + stride])
+
+//8.5.3.2.2 Luma sample bilinear interpolation process
+static void FUNC(dmvr)(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
+    const int height, const intptr_t mx, const intptr_t my, const int width)
+{
+    const pixel *src            = (const pixel *)_src;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+#if BIT_DEPTH > 10
+    const int shift4            = BIT_DEPTH - 10;
+    const int offset4           = 1 << (shift4 - 1);
+    #define DMVR_SHIFT(s)       (((s) + offset4) >> shift4)
+#else
+    #define DMVR_SHIFT(s)       ((s) << (10 - BIT_DEPTH))
+#endif
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = DMVR_SHIFT(src[x]);
+        src += src_stride;
+        dst += MAX_PB_SIZE;
+    }
+#undef DMVR_SHIFT
+}
+
+//8.5.3.2.2 Luma sample bilinear interpolation process
+static void FUNC(dmvr_h)(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
+    const int height, const intptr_t mx, const intptr_t my, const int width)
+{
+    const pixel *src            = (const pixel*)_src;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const int8_t *filter        = ff_vvc_inter_luma_dmvr_filters[mx];
+    const int shift1            = BIT_DEPTH - 6;
+    const int offset1           = 1 << (shift1 - 1);
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = (DMVR_FILTER(src, 1) + offset1) >> shift1;
+        src += src_stride;
+        dst += MAX_PB_SIZE;
+    }
+}
+
+//8.5.3.2.2 Luma sample bilinear interpolation process
+static void FUNC(dmvr_v)(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
+    const int height, const intptr_t mx, const intptr_t my, const int width)
+{
+    const pixel *src            = (pixel*)_src;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const int8_t *filter        = ff_vvc_inter_luma_dmvr_filters[my];
+    const int shift1            = BIT_DEPTH - 6;
+    const int offset1           = 1 << (shift1 - 1);
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = (DMVR_FILTER(src, src_stride) + offset1) >> shift1;
+        src += src_stride;
+        dst += MAX_PB_SIZE;
+    }
+
+}
+
+//8.5.3.2.2 Luma sample bilinear interpolation process
+static void FUNC(dmvr_hv)(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
+    const int height, const intptr_t mx, const intptr_t my, const int width)
+{
+    int16_t tmp_array[(MAX_PB_SIZE + BILINEAR_EXTRA) * MAX_PB_SIZE];
+    int16_t *tmp                = tmp_array;
+    const pixel *src            = (const pixel*)_src;
+    const ptrdiff_t src_stride  = _src_stride / sizeof(pixel);
+    const int8_t *filter        = ff_vvc_inter_luma_dmvr_filters[mx];
+    const int shift1            = BIT_DEPTH - 6;
+    const int offset1           = 1 << (shift1 - 1);
+    const int shift2            = 4;
+    const int offset2           = 1 << (shift2 - 1);
+
+    src   -= BILINEAR_EXTRA_BEFORE * src_stride;
+    for (int y = 0; y < height + BILINEAR_EXTRA; y++) {
+        for (int x = 0; x < width; x++)
+            tmp[x] = (DMVR_FILTER(src, 1) + offset1) >> shift1;
+        src += src_stride;
+        tmp += MAX_PB_SIZE;
+    }
+
+    tmp    = tmp_array + BILINEAR_EXTRA_BEFORE * MAX_PB_SIZE;
+    filter = ff_vvc_inter_luma_dmvr_filters[my];
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++)
+            dst[x] = (DMVR_FILTER(tmp, MAX_PB_SIZE) + offset2) >> shift2;
+        tmp += MAX_PB_SIZE;
+        dst += MAX_PB_SIZE;
+    }
+}
+
+#define PEL_FUNC(dst, C, idx1, idx2, a)                                         \
+    do {                                                                        \
+        for (int w = 0; w < 7; w++)                                             \
+            inter->dst[C][w][idx1][idx2] = FUNC(a);                             \
+    } while (0)                                                                 \
+
+#define DIR_FUNCS(d, C, c)                                                      \
+        PEL_FUNC(put_##d, C, 0, 0, put_##d##_pixels);                           \
+        PEL_FUNC(put_##d, C, 0, 1, put_##d##_##c##_h);                          \
+        PEL_FUNC(put_##d, C, 1, 0, put_##d##_##c##_v);                          \
+        PEL_FUNC(put_##d, C, 1, 1, put_##d##_##c##_hv);                         \
+        PEL_FUNC(put_##d##_w, C, 0, 0, put_##d##_w_pixels);                     \
+        PEL_FUNC(put_##d##_w, C, 0, 1, put_##d##_##c##_w_h);                    \
+        PEL_FUNC(put_##d##_w, C, 1, 0, put_##d##_##c##_w_v);                    \
+        PEL_FUNC(put_##d##_w, C, 1, 1, put_##d##_##c##_w_hv);
+
+#define FUNCS(C, c)                                                             \
+        PEL_FUNC(put, C, 0, 0, put_pixels);                                     \
+        PEL_FUNC(put, C, 0, 1, put_##c##_h);                                    \
+        PEL_FUNC(put, C, 1, 0, put_##c##_v);                                    \
+        PEL_FUNC(put, C, 1, 1, put_##c##_hv);                                   \
+        DIR_FUNCS(uni, C, c);                                                   \
+
+static void FUNC(ff_vvc_inter_dsp_init)(VVCInterDSPContext *const inter)
+{
+    FUNCS(LUMA, luma);
+    FUNCS(CHROMA, chroma);
+
+    inter->avg                  = FUNC(avg);
+    inter->w_avg                = FUNC(w_avg);
+
+    inter->dmvr[0][0]           = FUNC(dmvr);
+    inter->dmvr[0][1]           = FUNC(dmvr_h);
+    inter->dmvr[1][0]           = FUNC(dmvr_v);
+    inter->dmvr[1][1]           = FUNC(dmvr_hv);
+
+    inter->put_ciip             = FUNC(put_ciip);
+    inter->put_gpm              = FUNC(put_gpm);
+
+    inter->fetch_samples        = FUNC(fetch_samples);
+    inter->bdof_fetch_samples   = FUNC(bdof_fetch_samples);
+    inter->apply_prof           = FUNC(apply_prof);
+    inter->apply_prof_uni       = FUNC(apply_prof_uni);
+    inter->apply_prof_uni_w     = FUNC(apply_prof_uni_w);
+    inter->apply_bdof           = FUNC(apply_bdof);
+    inter->prof_grad_filter     = FUNC(prof_grad_filter);
+    inter->sad                  = vvc_sad;
+}
+
+#undef FUNCS
+#undef PEL_FUNC
+#undef DMVR_FUNCS
diff --git a/libavcodec/vvc/vvcdec.h b/libavcodec/vvc/vvcdec.h
index cd8db24a69..0453f49ee5 100644
--- a/libavcodec/vvc/vvcdec.h
+++ b/libavcodec/vvc/vvcdec.h
@@ -24,9 +24,11 @@ 
 #ifndef AVCODEC_VVC_VVCDEC_H
 #define AVCODEC_VVC_VVCDEC_H
 
+#include "libavcodec/videodsp.h"
 #include "libavcodec/vvc.h"
 
 #include "vvc_ps.h"
+#include "vvcdsp.h"
 
 #define LUMA                    0
 #define CHROMA                  1
@@ -103,6 +105,9 @@  typedef struct VVCFrameContext {
 
     VVCFrame *ref;
 
+    VVCDSPContext vvcdsp;
+    VideoDSPContext vdsp;
+
     struct VVCFrameThread *ft;
 
     uint64_t decode_order;
diff --git a/libavcodec/vvc/vvcdsp.h b/libavcodec/vvc/vvcdsp.h
new file mode 100644
index 0000000000..95e228dcc4
--- /dev/null
+++ b/libavcodec/vvc/vvcdsp.h
@@ -0,0 +1,170 @@ 
+/*
+ * VVC DSP
+ *
+ * Copyright (C) 2021 Nuo Mi
+ *
+ *
+ * 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
+ */
+
+#ifndef AVCODEC_VVC_VVCDSP_H
+#define AVCODEC_VVC_VVCDSP_H
+
+#include <stdint.h>
+#include <stddef.h>
+
+enum TxType {
+    DCT2,
+    DST7,
+    DCT8,
+    N_TX_TYPE,
+};
+
+enum TxSize {
+    TX_SIZE_2,
+    TX_SIZE_4,
+    TX_SIZE_8,
+    TX_SIZE_16,
+    TX_SIZE_32,
+    TX_SIZE_64,
+    N_TX_SIZE,
+};
+
+typedef struct VVCInterDSPContext {
+    void (*put[2 /* luma, chroma */][7 /* log2(width) - 1 */][2 /* int, frac */][2 /* int, frac */])(
+        int16_t *dst, const uint8_t *src, ptrdiff_t src_stride, int height,
+        const int8_t *hf, const int8_t *vf, int width);
+
+    void (*put_uni[2 /* luma, chroma */][7 /* log2(width) - 1 */][2 /* int, frac */][2 /* int, frac */])(
+        uint8_t *dst, ptrdiff_t dst_stride, const uint8_t *src, ptrdiff_t src_stride, int height,
+        const int8_t *hf, const int8_t *vf, int width);
+
+    void (*put_uni_w[2 /* luma, chroma */][7 /* log2(width) - 1 */][2 /* int, frac */][2 /* int, frac */])(
+        uint8_t *dst, ptrdiff_t dst_stride, const uint8_t *src, ptrdiff_t src_stride, int height,
+        int denom, int wx, int ox, const int8_t *hf, const int8_t *vf, int width);
+
+    void (*avg)(uint8_t *dst, ptrdiff_t dst_stride,
+        const int16_t *src0, const int16_t *src1, int width, int height);
+
+    void (*w_avg)(uint8_t *_dst, const ptrdiff_t _dst_stride,
+        const int16_t *src0, const int16_t *src1, int width, int height,
+        int denom, int w0, int w1, int o0, int o1);
+
+    void (*put_ciip)(uint8_t *dst, ptrdiff_t dst_stride, int width, int height,
+        const uint8_t *inter, ptrdiff_t inter_stride, int inter_weight);
+
+    void (*put_gpm)(uint8_t *dst, ptrdiff_t dst_stride, int width, int height,
+        const int16_t *src0, const int16_t *src1,
+        const uint8_t *weights, int step_x, int step_y);
+
+    void (*fetch_samples)(int16_t *dst, const uint8_t *src, ptrdiff_t src_stride, int x_frac, int y_frac);
+    void (*bdof_fetch_samples)(int16_t *dst, const uint8_t *src, ptrdiff_t src_stride, int x_frac, int y_frac,
+        int width, int height);
+
+    void (*prof_grad_filter)(int16_t *gradient_h, int16_t *gradient_v, const ptrdiff_t gradient_stride,
+        const int16_t *src, const ptrdiff_t src_stride, int width, int height, const int pad);
+    void (*apply_prof)(int16_t *dst, const int16_t *src, const int16_t *diff_mv_x, const int16_t *diff_mv_y);
+
+    void (*apply_prof_uni)(uint8_t *dst, ptrdiff_t dst_stride, const int16_t *src,
+        const int16_t *diff_mv_x, const int16_t *diff_mv_y);
+    void (*apply_prof_uni_w)(uint8_t *dst, const ptrdiff_t dst_stride, const int16_t *src,
+        const int16_t *diff_mv_x, const int16_t *diff_mv_y, int denom, int wx, int ox);
+
+    void (*apply_bdof)(uint8_t *dst, ptrdiff_t dst_stride, int16_t *src0, int16_t *src1, int block_w, int block_h);
+
+    int (*sad)(const int16_t *src0, const int16_t *src1, int dx, int dy, int block_w, int block_h);
+    void (*dmvr[2][2])(int16_t *dst, const uint8_t *src, ptrdiff_t src_stride, int height,
+        intptr_t mx, intptr_t my, int width);
+} VVCInterDSPContext;
+
+struct VVCLocalContext;
+
+typedef struct VVCIntraDSPContext {
+    void (*intra_cclm_pred)(const struct VVCLocalContext *lc, int x0, int y0, int w, int h);
+    void (*lmcs_scale_chroma)(struct VVCLocalContext *lc, int *dst, const int *coeff, int w, int h, int x0_cu, int y0_cu);
+    void (*intra_pred)(const struct VVCLocalContext *lc, int x0, int y0, int w, int h, int c_idx);
+    void (*pred_planar)(uint8_t *src, const uint8_t *top, const uint8_t *left, int w, int h, ptrdiff_t stride);
+    void (*pred_mip)(uint8_t *src, const uint8_t *top, const uint8_t *left, int w, int h, ptrdiff_t stride,
+        int mode_id, int is_transpose);
+    void (*pred_dc)(uint8_t *src, const uint8_t *top, const uint8_t *left, int w, int h, ptrdiff_t stride);
+    void (*pred_v)(uint8_t *src, const uint8_t *_top, int w, int h, ptrdiff_t stride);
+    void (*pred_h)(uint8_t *src, const uint8_t *_left, int w, int h, ptrdiff_t stride);
+    void (*pred_angular_v)(uint8_t *src, const uint8_t *_top, const uint8_t *_left,
+        int w, int h, ptrdiff_t stride, int c_idx, int mode, int ref_idx, int filter_flag, int need_pdpc);
+    void (*pred_angular_h)(uint8_t *src, const uint8_t *_top, const uint8_t *_left, int w, int h, ptrdiff_t stride,
+        int c_idx, int mode, int ref_idx, int filter_flag, int need_pdpc);
+} VVCIntraDSPContext;
+
+typedef struct VVCItxDSPContext {
+    void (*add_residual)(uint8_t *dst, const int *res, int width, int height, ptrdiff_t stride);
+    void (*add_residual_joint)(uint8_t *dst, const int *res, int width, int height, ptrdiff_t stride, int c_sign, int shift);
+    void (*pred_residual_joint)(int *buf, int width, int height, int c_sign, int shift);
+
+    void (*itx[N_TX_TYPE][N_TX_SIZE])(int *out, ptrdiff_t out_step, const int *in, ptrdiff_t in_step);
+    void (*transform_bdpcm)(int *coeffs, int width, int height, int vertical, int log2_transform_range);
+} VVCItxDSPContext;
+
+typedef struct VVCLMCSDSPContext {
+    void (*filter)(uint8_t *dst, ptrdiff_t dst_stride, int width, int height, const uint8_t *lut);
+} VVCLMCSDSPContext;
+
+typedef struct VVCLFDSPContext {
+    int (*ladf_level[2 /* h, v */])(const uint8_t *pix, ptrdiff_t stride);
+
+    void (*filter_luma[2 /* h, v */])(uint8_t *pix, ptrdiff_t stride, const int32_t *beta, const int32_t *tc,
+        const uint8_t *no_p, const uint8_t *no_q, const uint8_t *max_len_p, const uint8_t *max_len_q, int hor_ctu_edge);
+    void (*filter_chroma[2 /* h, v */])(uint8_t *pix, ptrdiff_t stride, const int32_t *beta, const int32_t *tc,
+        const uint8_t *no_p, const uint8_t *no_q, const uint8_t *max_len_p, const uint8_t *max_len_q, int shift);
+} VVCLFDSPContext;
+
+struct SAOParams;
+typedef struct VVCSAODSPContext {
+    void (*band_filter[9])(uint8_t *dst, const uint8_t *src, ptrdiff_t dst_stride, ptrdiff_t src_stride,
+        const int16_t *sao_offset_val, int sao_left_class, int width, int height);
+    /* implicit src_stride parameter has value of 2 * MAX_PB_SIZE + AV_INPUT_BUFFER_PADDING_SIZE */
+    void (*edge_filter[9])(uint8_t *dst /* align 16 */, const uint8_t *src /* align 32 */, ptrdiff_t dst_stride,
+        const int16_t *sao_offset_val, int sao_eo_class, int width, int height);
+    void (*edge_restore[2])(uint8_t *dst, const uint8_t *src, ptrdiff_t dst_stride, ptrdiff_t src_stride,
+        const struct SAOParams *sao, const int *borders, int width, int height, int c_idx,
+        const uint8_t *vert_edge, const uint8_t *horiz_edge, const uint8_t *diag_edge);
+} VVCSAODSPContext;
+
+typedef struct VVCALFDSPContext {
+    void (*filter[2 /* luma, chroma */])(uint8_t *dst, ptrdiff_t dst_stride, const uint8_t *src,  ptrdiff_t src_stride,
+        int width, int height, const int16_t *filter, const int16_t *clip, int vb_pos);
+    void (*filter_cc)(uint8_t *dst, ptrdiff_t dst_stride, const uint8_t *luma, ptrdiff_t luma_stride,
+        int width, int height, int hs, int vs, const int16_t *filter, int vb_pos);
+
+    void (*classify)(int *class_idx, int *transpose_idx, const uint8_t *src, ptrdiff_t src_stride, int width, int height,
+        int vb_pos, int *gradient_tmp);
+    void (*recon_coeff_and_clip)(int16_t *coeff, int16_t *clip, const int *class_idx, const int *transpose_idx, int size,
+        const int16_t *coeff_set, const uint8_t *clip_idx_set, const uint8_t *class_to_filt);
+} VVCALFDSPContext;
+
+typedef struct VVCDSPContext {
+    VVCInterDSPContext inter;
+    VVCIntraDSPContext intra;
+    VVCItxDSPContext itx;
+    VVCLMCSDSPContext lmcs;
+    VVCLFDSPContext lf;
+    VVCSAODSPContext sao;
+    VVCALFDSPContext alf;
+} VVCDSPContext;
+
+void ff_vvc_dsp_init(VVCDSPContext *hpc, int bit_depth);
+
+#endif /* AVCODEC_VVC_VVCDSP_H */