hashirama/uvg266
1
0
Fork 0
mirror of https://github.com/ultravideo/uvg266.git synced 2024-11-27 19:24:06 +00:00
Code Issues Actions Projects Releases Wiki Activity

[fme] Enable 1/16th luma pixel fme

Browse source
This commit is contained in:
Marko Viitanen 2021-11-22 16:04:57 +02:00
parent f0994be9a2
commit 64705a27ee
5 changed files with 103 additions and 72 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

48
src/filter.c
View file

@ -46,24 +46,60 @@ const uint8_t kvz_g_beta_table_8x8[64] =
58, 60, 62, 64, 66, 68, 70, 72, 74, 76, 78, 80, 82, 84, 86, 88
};
const int8_t kvz_g_luma_filter[4][8] =
const int8_t kvz_g_luma_filter[16][8] =
{
{ 0, 0, 0, 64, 0, 0, 0, 0 },
{ -1, 4, -10, 58, 17, -5, 1, 0 },
{ -1, 4, -11, 40, 40, -11, 4, -1 },
{ 0, 1, -5, 17, 58, -10, 4, -1 }
{ 0, 1, -3, 63, 4, -2, 1, 0 },
{ -1, 2, -5, 62, 8, -3, 1, 0 },
{ -1, 3, -8, 60, 13, -4, 1, 0 },
{ -1, 4, -10, 58, 17, -5, 1, 0 }, //1/4
{ -1, 4, -11, 52, 26, -8, 3, -1 },
{ -1, 3, -9, 47, 31, -10, 4, -1 },
{ -1, 4, -11, 45, 34, -10, 4, -1 },
{ -1, 4, -11, 40, 40, -11, 4, -1 }, //1/2
{ -1, 4, -10, 34, 45, -11, 4, -1 },
{ -1, 4, -10, 31, 47, -9, 3, -1 },
{ -1, 3, -8, 26, 52, -11, 4, -1 },
{ 0, 1, -5, 17, 58, -10, 4, -1 }, //3/4
{ 0, 1, -4, 13, 60, -8, 3, -1 },
{ 0, 1, -3, 8, 62, -5, 2, -1 },
{ 0, 1, -2, 4, 63, -3, 1, 0 }
};
const int8_t kvz_g_chroma_filter[8][4] =
const int8_t kvz_g_chroma_filter[32][4] =
{
{ 0, 64, 0, 0 },
{ -1, 63, 2, 0 },
{ -2, 62, 4, 0 },
{ -2, 60, 7, -1 },
{ -2, 58, 10, -2 },
{ -3, 57, 12, -2 },
{ -4, 56, 14, -2 },
{ -4, 55, 15, -2 },
{ -4, 54, 16, -2 },
{ -5, 53, 18, -2 },
{ -6, 52, 20, -2 },
{ -6, 49, 24, -3 },
{ -6, 46, 28, -4 },
{ -5, 44, 29, -4 },
{ -4, 42, 30, -4 },
{ -4, 39, 33, -4 },
{ -4, 36, 36, -4 },
{ -4, 33, 39, -4 },
{ -4, 30, 42, -4 },
{ -4, 29, 44, -5 },
{ -4, 28, 46, -6 },
{ -3, 24, 49, -6 },
{ -2, 20, 52, -6 },
{ -2, 18, 53, -5 },
{ -2, 16, 54, -4 },
{ -2, 10, 58, -2 }
{ -2, 15, 55, -4 },
{ -2, 14, 56, -4 },
{ -2, 12, 57, -3 },
{ -2, 10, 58, -2 },
{ -1, 7, 60, -2 },
{ 0, 4, 62, -2 },
{ 0, 2, 63, -1 },
};
//////////////////////////////////////////////////////////////////////////

52
src/inter.c
View file

@ -62,8 +62,8 @@ static void inter_recon_frac_luma(const encoder_state_t *const state,
const mv_t mv_param[2],
lcu_t *lcu)
{
int mv_frac_x = (mv_param[0] & 3);
int mv_frac_y = (mv_param[1] & 3);
int mv_frac_x = (mv_param[0] & 15);
int mv_frac_y = (mv_param[1] & 15);
// Space for extrapolated pixels and the part from the picture.
// Some extra for AVX2.
@ -77,8 +77,8 @@ static void inter_recon_frac_luma(const encoder_state_t *const state,
.src_w = ref->width,
.src_h = ref->height,
.src_s = ref->stride,
.blk_x = state->tile->offset_x + xpos + (mv_param[0] >> 2),
.blk_y = state->tile->offset_y + ypos + (mv_param[1] >> 2),
.blk_x = state->tile->offset_x + xpos + (mv_param[0] >> INTERNAL_MV_PREC),
.blk_y = state->tile->offset_y + ypos + (mv_param[1] >> INTERNAL_MV_PREC),
.blk_w = block_width,
.blk_h = block_height,
.pad_l = KVZ_LUMA_FILTER_OFFSET,
@ -117,8 +117,8 @@ static void inter_recon_frac_luma_hi(const encoder_state_t *const state,
const mv_t mv_param[2],
hi_prec_buf_t *hi_prec_out)
{
int mv_frac_x = (mv_param[0] & 3);
int mv_frac_y = (mv_param[1] & 3);
int mv_frac_x = (mv_param[0] & 15);
int mv_frac_y = (mv_param[1] & 15);
// Space for extrapolated pixels and the part from the picture.
// Some extra for AVX2.
@ -132,8 +132,8 @@ static void inter_recon_frac_luma_hi(const encoder_state_t *const state,
.src_w = ref->width,
.src_h = ref->height,
.src_s = ref->stride,
.blk_x = state->tile->offset_x + xpos + (mv_param[0] >> 2),
.blk_y = state->tile->offset_y + ypos + (mv_param[1] >> 2),
.blk_x = state->tile->offset_x + xpos + (mv_param[0] >> INTERNAL_MV_PREC),
.blk_y = state->tile->offset_y + ypos + (mv_param[1] >> INTERNAL_MV_PREC),
.blk_w = block_width,
.blk_h = block_height,
.pad_l = KVZ_LUMA_FILTER_OFFSET,
@ -172,8 +172,8 @@ static void inter_recon_frac_chroma(const encoder_state_t *const state,
const mv_t mv_param[2],
lcu_t *lcu)
{
int mv_frac_x = (mv_param[0] & 7);
int mv_frac_y = (mv_param[1] & 7);
int mv_frac_x = (mv_param[0] & 31);
int mv_frac_y = (mv_param[1] & 31);
// Space for extrapolated pixels and the part from the picture.
// Some extra for AVX2.
@ -190,8 +190,8 @@ static void inter_recon_frac_chroma(const encoder_state_t *const state,
.src_w = ref->width / 2,
.src_h = ref->height / 2,
.src_s = ref->stride / 2,
.blk_x = (state->tile->offset_x + xpos) / 2 + (mv_param[0] >> 3),
.blk_y = (state->tile->offset_y + ypos) / 2 + (mv_param[1] >> 3),
.blk_x = (state->tile->offset_x + xpos) / 2 + (mv_param[0] >> (INTERNAL_MV_PREC + 1)),
.blk_y = (state->tile->offset_y + ypos) / 2 + (mv_param[1] >> (INTERNAL_MV_PREC + 1)),
.blk_w = block_width / 2,
.blk_h = block_height / 2,
.pad_l = KVZ_CHROMA_FILTER_OFFSET,
@ -244,8 +244,8 @@ static void inter_recon_frac_chroma_hi(const encoder_state_t *const state,
const mv_t mv_param[2],
hi_prec_buf_t *hi_prec_out)
{
int mv_frac_x = (mv_param[0] & 7);
int mv_frac_y = (mv_param[1] & 7);
int mv_frac_x = (mv_param[0] & 31);
int mv_frac_y = (mv_param[1] & 31);
// Space for extrapolated pixels and the part from the picture.
// Some extra for AVX2.
@ -262,8 +262,8 @@ static void inter_recon_frac_chroma_hi(const encoder_state_t *const state,
.src_w = ref->width / 2,
.src_h = ref->height / 2,
.src_s = ref->stride / 2,
.blk_x = (state->tile->offset_x + xpos) / 2 + (mv_param[0] >> 3),
.blk_y = (state->tile->offset_y + ypos) / 2 + (mv_param[1] >> 3),
.blk_x = (state->tile->offset_x + xpos) / 2 + (mv_param[0] >> (INTERNAL_MV_PREC + 1) ),
.blk_y = (state->tile->offset_y + ypos) / 2 + (mv_param[1] >> (INTERNAL_MV_PREC + 1) ),
.blk_w = block_width / 2,
.blk_h = block_height / 2,
.pad_l = KVZ_CHROMA_FILTER_OFFSET,
@ -372,9 +372,7 @@ static void inter_recon_unipred(const encoder_state_t * const state,
bool predict_luma,
bool predict_chroma)
{
mv_t mv_param_qpel[2] = { mv_param[0], mv_param[1] };
mv_t mv_param_fpel[2] = { mv_param[0], mv_param[1] };
kvz_change_precision(INTERNAL_MV_PREC, 2, &mv_param_qpel[0], &mv_param_qpel[1]);
kvz_change_precision(INTERNAL_MV_PREC, 0, &mv_param_fpel[0], &mv_param_fpel[1]);
@ -394,7 +392,7 @@ static void inter_recon_unipred(const encoder_state_t * const state,
// With 420, odd coordinates need interpolation.
const int8_t fractional_chroma = (mv_in_pu.x & 1) || (mv_in_pu.y & 1);
const int8_t fractional_luma = ((mv_param_qpel[0] & 3) || (mv_param_qpel[1] & 3));
const int8_t fractional_luma = ((mv_param[0] & 15) || (mv_param[1] & 15));
// Generate prediction for luma.
if (predict_luma) {
@ -404,13 +402,13 @@ static void inter_recon_unipred(const encoder_state_t * const state,
inter_recon_frac_luma_hi(state, ref,
pu_in_tile.x, pu_in_tile.y,
width, height,
mv_param_qpel, hi_prec_out);
mv_param, hi_prec_out);
}
else {
inter_recon_frac_luma(state, ref,
pu_in_tile.x, pu_in_tile.y,
width, height,
mv_param_qpel, lcu);
mv_param, lcu);
}
} else {
// With an integer MV, copy pixels directly from the reference.
@ -443,12 +441,12 @@ static void inter_recon_unipred(const encoder_state_t * const state,
inter_recon_frac_chroma_hi(state, ref,
pu_in_tile.x, pu_in_tile.y,
width, height,
mv_param_qpel, hi_prec_out);
mv_param, hi_prec_out);
} else {
inter_recon_frac_chroma(state, ref,
pu_in_tile.x, pu_in_tile.y,
width, height,
mv_param_qpel, lcu);
mv_param, lcu);
}
} else {
// With an integer MV, copy pixels directly from the reference.
@ -511,11 +509,11 @@ void kvz_inter_recon_bipred(const encoder_state_t * const state,
kvz_pixel temp_lcu_u[LCU_WIDTH_C*LCU_WIDTH_C];
kvz_pixel temp_lcu_v[LCU_WIDTH_C*LCU_WIDTH_C];
const int hi_prec_luma_rec0 = (mv_param[0][0]>>2) & 3 || (mv_param[0][1]>>2) & 3;
const int hi_prec_luma_rec1 = (mv_param[1][0]>>2) & 3 || (mv_param[1][1]>>2) & 3;
const int hi_prec_luma_rec0 = (mv_param[0][0] & 15) || (mv_param[0][1] & 15);
const int hi_prec_luma_rec1 = (mv_param[1][0] & 15) || (mv_param[1][1] & 15);
const int hi_prec_chroma_rec0 = (mv_param[0][0]>>2) & 7 || (mv_param[0][1]>>2) & 7;
const int hi_prec_chroma_rec1 = (mv_param[1][0]>>2) & 7 || (mv_param[1][1]>>2) & 7;
const int hi_prec_chroma_rec0 = (mv_param[0][0] & 31) || (mv_param[0][1] & 31);
const int hi_prec_chroma_rec1 = (mv_param[1][0] & 31) || (mv_param[1][1] & 31);
hi_prec_buf_t* high_precision_rec0 = 0;
hi_prec_buf_t* high_precision_rec1 = 0;

9
src/search_inter.c
View file

@ -272,10 +272,10 @@ static bool mv_in_merge(const inter_search_info_t *info, vector2d_t mv)
for (int i = 0; i < info->num_merge_cand; ++i) {
if (info->merge_cand[i].dir == 3) continue;
const vector2d_t merge_mv = {
(info->merge_cand[i].mv[info->merge_cand[i].dir - 1][0] + 2) >> INTERNAL_MV_PREC,
(info->merge_cand[i].mv[info->merge_cand[i].dir - 1][1] + 2) >> INTERNAL_MV_PREC
info->merge_cand[i].mv[info->merge_cand[i].dir - 1][0],
info->merge_cand[i].mv[info->merge_cand[i].dir - 1][1]
};
if (merge_mv.x == mv.x && merge_mv.y == mv.y) {
if (merge_mv.x == mv.x * (1 << (INTERNAL_MV_PREC)) && merge_mv.y == mv.y * (1 << (INTERNAL_MV_PREC))) {
return true;
}
}
@ -1684,9 +1684,6 @@ static void search_pu_inter(encoder_state_t * const state,
for (int merge_idx = 0; merge_idx < info.num_merge_cand; ++merge_idx) {
inter_merge_cand_t *cur_cand = &info.merge_cand[merge_idx];
if ((cur_cand->dir & 1 && (cur_cand->mv[0][0] & 3 || cur_cand->mv[0][1] & 3)) ||
(cur_cand->dir & 2 && (cur_cand->mv[1][0] & 3 || cur_cand->mv[1][1] & 3))) continue;
cur_cu->inter.mv_dir = cur_cand->dir;
cur_cu->inter.mv_ref[0] = cur_cand->ref[0];
cur_cu->inter.mv_ref[1] = cur_cand->ref[1];

30
src/strategies/avx2/ipol-avx2.c
View file

@ -38,8 +38,8 @@
#include "strategies/generic/ipol-generic.h"
extern int8_t kvz_g_luma_filter[4][8];
extern int8_t kvz_g_chroma_filter[8][4];
extern int8_t kvz_g_luma_filter[16][8];
extern int8_t kvz_g_chroma_filter[32][4];
static int32_t kvz_eight_tap_filter_hor_avx2(int8_t *filter, kvz_pixel *data)
{
@ -635,7 +635,7 @@ static void kvz_filter_hpel_blocks_hor_ver_luma_avx2(const encoder_control_t * e
int32_t wp_offset1 = 1 << (wp_shift1 - 1);
int8_t *fir0 = kvz_g_luma_filter[0];
int8_t *fir2 = kvz_g_luma_filter[2];
int8_t *fir2 = kvz_g_luma_filter[8];
int16_t dst_stride = LCU_WIDTH;
int16_t hor_stride = LCU_WIDTH;
@ -745,7 +745,7 @@ static void kvz_filter_hpel_blocks_diag_luma_avx2(const encoder_control_t * enco
int32_t wp_shift1 = 14 - KVZ_BIT_DEPTH;
int32_t wp_offset1 = 1 << (wp_shift1 - 1);
int8_t *fir2 = kvz_g_luma_filter[2];
int8_t *fir2 = kvz_g_luma_filter[8];
int16_t dst_stride = LCU_WIDTH;
int16_t hor_stride = LCU_WIDTH;
@ -826,9 +826,9 @@ static void kvz_filter_qpel_blocks_hor_ver_luma_avx2(const encoder_control_t * e
int32_t wp_offset1 = 1 << (wp_shift1 - 1);
int8_t *fir0 = kvz_g_luma_filter[0];
int8_t *fir2 = kvz_g_luma_filter[2];
int8_t *fir1 = kvz_g_luma_filter[1];
int8_t *fir3 = kvz_g_luma_filter[3];
int8_t *fir2 = kvz_g_luma_filter[8];
int8_t *fir1 = kvz_g_luma_filter[4];
int8_t *fir3 = kvz_g_luma_filter[12];
// Horiziontal positions. Positions 0 and 2 have already been calculated in filtered.
int16_t *hor_pos0 = hor_intermediate[0];
@ -1137,8 +1137,8 @@ static void kvz_sample_quarterpel_luma_avx2(const encoder_control_t * const enco
const mv_t mv[2])
{
// TODO: horizontal and vertical only filtering
int8_t *hor_fir = kvz_g_luma_filter[mv[0] & 3];
int8_t *ver_fir = kvz_g_luma_filter[mv[1] & 3];
int8_t *hor_fir = kvz_g_luma_filter[mv[0] & 15];
int8_t *ver_fir = kvz_g_luma_filter[mv[1] & 15];
// Buffer for intermediate values with one extra row
// because the loop writes two rows each iteration.
@ -1162,8 +1162,8 @@ static void kvz_sample_quarterpel_luma_hi_avx2(const encoder_control_t * const e
const mv_t mv[2])
{
// TODO: horizontal and vertical only filtering
int8_t *hor_fir = kvz_g_luma_filter[mv[0] & 3];
int8_t *ver_fir = kvz_g_luma_filter[mv[1] & 3];
int8_t *hor_fir = kvz_g_luma_filter[mv[0] & 15];
int8_t *ver_fir = kvz_g_luma_filter[mv[1] & 15];
// Buffer for intermediate values with one extra row
// because the loop writes two rows each iteration.
@ -1191,8 +1191,8 @@ static void kvz_sample_octpel_chroma_avx2(const encoder_control_t *const encoder
kvz_sample_octpel_chroma_generic(encoder, src, src_stride, width, height, dst, dst_stride, hor_flag, ver_flag, mv);
return;
}
int8_t *hor_fir = kvz_g_chroma_filter[mv[0] & 7];
int8_t *ver_fir = kvz_g_chroma_filter[mv[1] & 7];
int8_t *hor_fir = kvz_g_chroma_filter[mv[0] & 31];
int8_t *ver_fir = kvz_g_chroma_filter[mv[1] & 31];
// Buffer for intermediate values with 3 extra rows
// because the loop writes four rows each iteration.
@ -1219,8 +1219,8 @@ static void kvz_sample_octpel_chroma_hi_avx2(const encoder_control_t *const enco
kvz_sample_octpel_chroma_hi_generic(encoder, src, src_stride, width, height, dst, dst_stride, hor_flag, ver_flag, mv);
return;
}
int8_t *hor_fir = kvz_g_chroma_filter[mv[0] & 7];
int8_t *ver_fir = kvz_g_chroma_filter[mv[1] & 7];
int8_t *hor_fir = kvz_g_chroma_filter[mv[0] & 31];
int8_t *ver_fir = kvz_g_chroma_filter[mv[1] & 31];
// Buffer for intermediate values with 3 extra rows
// because the loop writes four rows each iteration.

34
src/strategies/generic/ipol-generic.c
View file

@ -28,8 +28,8 @@
#include "strategies/strategies-ipol.h"
#include "strategyselector.h"
extern int8_t kvz_g_luma_filter[4][8];
extern int8_t kvz_g_chroma_filter[8][4];
extern int8_t kvz_g_luma_filter[16][8];
extern int8_t kvz_g_chroma_filter[32][4];
int32_t kvz_eight_tap_filter_hor_generic(int8_t *filter, kvz_pixel *data)
{
@ -133,8 +133,8 @@ void kvz_sample_quarterpel_luma_generic(const encoder_control_t * const encoder,
int32_t wp_offset1 = 1 << (wp_shift1 - 1);
// Select filters according to the fractional part of the x and y mv components
int8_t *hor_filter = kvz_g_luma_filter[mv[0] & 3];
int8_t *ver_filter = kvz_g_luma_filter[mv[1] & 3];
int8_t *hor_filter = kvz_g_luma_filter[mv[0] & 15];
int8_t *ver_filter = kvz_g_luma_filter[mv[1] & 15];
int16_t hor_filtered[KVZ_EXT_BLOCK_W_LUMA][LCU_WIDTH];
int16_t hor_stride = LCU_WIDTH;
@ -166,8 +166,8 @@ void kvz_sample_quarterpel_luma_hi_generic(const encoder_control_t * const encod
int32_t shift2 = 6;
// Select filters according to the fractional part of the x and y mv components
int8_t *hor_filter = kvz_g_luma_filter[mv[0] & 3];
int8_t *ver_filter = kvz_g_luma_filter[mv[1] & 3];
int8_t *hor_filter = kvz_g_luma_filter[mv[0] & 15];
int8_t *ver_filter = kvz_g_luma_filter[mv[1] & 15];
int16_t hor_filtered[KVZ_EXT_BLOCK_W_LUMA][LCU_WIDTH];
int16_t hor_stride = LCU_WIDTH;
@ -210,7 +210,7 @@ void kvz_filter_hpel_blocks_hor_ver_luma_generic(const encoder_control_t * encod
int32_t wp_offset1 = 1 << (wp_shift1 - 1);
int8_t *fir0 = kvz_g_luma_filter[0];
int8_t *fir2 = kvz_g_luma_filter[2];
int8_t *fir2 = kvz_g_luma_filter[8];
int16_t dst_stride = LCU_WIDTH;
int16_t hor_stride = LCU_WIDTH;
@ -324,7 +324,7 @@ void kvz_filter_hpel_blocks_diag_luma_generic(const encoder_control_t * encoder,
int32_t wp_shift1 = 14 - KVZ_BIT_DEPTH;
int32_t wp_offset1 = 1 << (wp_shift1 - 1);
int8_t *fir2 = kvz_g_luma_filter[2];
int8_t *fir2 = kvz_g_luma_filter[8];
int16_t dst_stride = LCU_WIDTH;
int16_t hor_stride = LCU_WIDTH;
@ -407,9 +407,9 @@ void kvz_filter_qpel_blocks_hor_ver_luma_generic(const encoder_control_t * encod
int32_t wp_offset1 = 1 << (wp_shift1 - 1);
int8_t *fir0 = kvz_g_luma_filter[0];
int8_t *fir2 = kvz_g_luma_filter[2];
int8_t *fir1 = kvz_g_luma_filter[1];
int8_t *fir3 = kvz_g_luma_filter[3];
int8_t *fir2 = kvz_g_luma_filter[8];
int8_t *fir1 = kvz_g_luma_filter[4];
int8_t *fir3 = kvz_g_luma_filter[12];
// Horiziontal positions. Positions 0 and 2 have already been calculated in filtered.
int16_t *hor_pos0 = hor_intermediate[0];
@ -565,8 +565,8 @@ void kvz_filter_qpel_blocks_diag_luma_generic(const encoder_control_t * encoder,
int32_t wp_shift1 = 14 - KVZ_BIT_DEPTH;
int32_t wp_offset1 = 1 << (wp_shift1 - 1);
int8_t *fir1 = kvz_g_luma_filter[1];
int8_t *fir3 = kvz_g_luma_filter[3];
int8_t *fir1 = kvz_g_luma_filter[4];
int8_t *fir3 = kvz_g_luma_filter[12];
// Horiziontal positions.
int16_t *hor_pos_l = hor_intermediate[3];
@ -671,8 +671,8 @@ void kvz_sample_octpel_chroma_generic(const encoder_control_t * const encoder, k
int32_t wp_offset1 = 1 << (wp_shift1 - 1);
// Select filters according to the fractional part of the x and y mv components
int8_t *hor_filter = kvz_g_chroma_filter[mv[0] & 7];
int8_t *ver_filter = kvz_g_chroma_filter[mv[1] & 7];
int8_t *hor_filter = kvz_g_chroma_filter[mv[0] & 31];
int8_t *ver_filter = kvz_g_chroma_filter[mv[1] & 31];
int16_t hor_filtered[KVZ_EXT_BLOCK_W_CHROMA][LCU_WIDTH_C];
int16_t hor_stride = LCU_WIDTH_C;
@ -704,8 +704,8 @@ void kvz_sample_octpel_chroma_hi_generic(const encoder_control_t * const encoder
int32_t shift2 = 6;
// Select filters according to the fractional part of the x and y mv components
int8_t *hor_filter = kvz_g_chroma_filter[mv[0] & 7];
int8_t *ver_filter = kvz_g_chroma_filter[mv[1] & 7];
int8_t *hor_filter = kvz_g_chroma_filter[mv[0] & 31];
int8_t *ver_filter = kvz_g_chroma_filter[mv[1] & 31];
int16_t hor_filtered[KVZ_EXT_BLOCK_W_CHROMA][LCU_WIDTH_C];
int16_t hor_stride = LCU_WIDTH_C;