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

Implement inter recon for non-square blocks.

Browse source 
Adds parameter height to functions kvz_inter_recon_lcu and
kvz_inter_recon_lcu_bipred and makes them work on non-square sizes.
Fractional reconstruction functions do not handle non-square blocks yet.
This commit is contained in:
Arttu Ylä-Outinen 2015-09-16 10:37:45 +03:00
parent f874c8614e
commit dc4525c0e3
4 changed files with 73 additions and 16 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

40
src/inter.c
View file

@ -172,12 +172,21 @@ void kvz_inter_recon_14bit_frac_chroma(const encoder_state_t * const state, cons
* \param xpos block x position
* \param ypos block y position
* \param width block width
* \param height block height
* \param mv[2] motion vector
* \param lcu destination lcu
* \param hi_prec destination of high precision output (null if not needed)
* \returns Void
*/
void kvz_inter_recon_lcu(const encoder_state_t * const state, const kvz_picture * const ref, int32_t xpos, int32_t ypos,int32_t width, const int16_t mv_param[2], lcu_t *lcu, hi_prec_buf_t *hi_prec_out)
void kvz_inter_recon_lcu(const encoder_state_t * const state,
const kvz_picture * const ref,
int32_t xpos,
int32_t ypos,
int32_t width,
int32_t height,
const int16_t mv_param[2],
lcu_t *lcu,
hi_prec_buf_t *hi_prec_out)
{
int x,y,coord_x,coord_y;
int16_t mv[2] = { mv_param[0], mv_param[1] };
@ -191,7 +200,7 @@ void kvz_inter_recon_lcu(const encoder_state_t * const state, const kvz_picture
// positive overflow flag
int8_t overflow_pos_x = (state->tile->lcu_offset_x * LCU_WIDTH + xpos + (mv[0]>>2) + width > ref->width )?1:0;
int8_t overflow_pos_y = (state->tile->lcu_offset_y * LCU_WIDTH + ypos + (mv[1]>>2) + width > ref->height)?1:0;
int8_t overflow_pos_y = (state->tile->lcu_offset_y * LCU_WIDTH + ypos + (mv[1]>>2) + height > ref->height)?1:0;
int8_t chroma_halfpel = ((mv[0]>>2)&1) || ((mv[1]>>2)&1); //!< (luma integer mv) lsb is set -> chroma is half-pel
// Luma quarter-pel
@ -224,7 +233,7 @@ void kvz_inter_recon_lcu(const encoder_state_t * const state, const kvz_picture
// With overflow present, more checking
if (overflow_neg_x || overflow_neg_y || overflow_pos_x || overflow_pos_y) {
// Copy Luma with boundary checking
for (y = ypos; y < ypos + width; y++) {
for (y = ypos; y < ypos + height; y++) {
for (x = xpos; x < xpos + width; x++) {
int x_in_lcu = (x & ((LCU_WIDTH)-1));
int y_in_lcu = (y & ((LCU_WIDTH)-1));
@ -258,7 +267,7 @@ void kvz_inter_recon_lcu(const encoder_state_t * const state, const kvz_picture
if(!chroma_halfpel) {
// Copy Chroma with boundary checking
for (y = ypos>>1; y < (ypos + width)>>1; y++) {
for (y = ypos>>1; y < (ypos + height)>>1; y++) {
for (x = xpos>>1; x < (xpos + width)>>1; x++) {
int x_in_lcu = (x & ((LCU_WIDTH>>1)-1));
int y_in_lcu = (y & ((LCU_WIDTH>>1)-1));
@ -294,7 +303,7 @@ void kvz_inter_recon_lcu(const encoder_state_t * const state, const kvz_picture
}
} else { //If no overflow, we can copy without checking boundaries
// Copy Luma
for (y = ypos; y < ypos + width; y++) {
for (y = ypos; y < ypos + height; y++) {
int y_in_lcu = (y & ((LCU_WIDTH)-1));
coord_y = ((y + state->tile->lcu_offset_y * LCU_WIDTH) + mv[1]) * ref->width; // pre-calculate
for (x = xpos; x < xpos + width; x++) {
@ -307,7 +316,7 @@ void kvz_inter_recon_lcu(const encoder_state_t * const state, const kvz_picture
if(!chroma_halfpel) {
// Copy Chroma
// TODO: chroma fractional pixel interpolation
for (y = ypos>>1; y < (ypos + width)>>1; y++) {
for (y = ypos>>1; y < (ypos + height)>>1; y++) {
int y_in_lcu = (y & ((LCU_WIDTH>>1)-1));
coord_y = ((y + state->tile->lcu_offset_y * (LCU_WIDTH>>1)) + (mv[1]>>1)) * ref_width_c; // pre-calculate
for (x = xpos>>1; x < (xpos + width)>>1; x++) {
@ -328,12 +337,21 @@ void kvz_inter_recon_lcu(const encoder_state_t * const state, const kvz_picture
* \param xpos block x position
* \param ypos block y position
* \param width block width
* \param height block height
* \param mv[2][2] motion vectors
* \param lcu destination lcu
* \returns Void
*/
void kvz_inter_recon_lcu_bipred(const encoder_state_t * const state, const kvz_picture * ref1, const kvz_picture * ref2, int32_t xpos, int32_t ypos, int32_t width, int16_t mv_param[2][2], lcu_t* lcu) {
void kvz_inter_recon_lcu_bipred(const encoder_state_t * const state,
const kvz_picture * ref1,
const kvz_picture * ref2,
int32_t xpos,
int32_t ypos,
int32_t width,
int32_t height,
int16_t mv_param[2][2],
lcu_t* lcu) {
kvz_pixel temp_lcu_y[LCU_WIDTH*LCU_WIDTH];
kvz_pixel temp_lcu_u[LCU_WIDTH_C*LCU_WIDTH_C];
kvz_pixel temp_lcu_v[LCU_WIDTH_C*LCU_WIDTH_C];
@ -352,7 +370,7 @@ void kvz_inter_recon_lcu_bipred(const encoder_state_t * const state, const kvz_p
if (hi_prec_chroma_rec0) high_precision_rec0 = kvz_hi_prec_buf_t_alloc(LCU_WIDTH*LCU_WIDTH);
if (hi_prec_chroma_rec1) high_precision_rec1 = kvz_hi_prec_buf_t_alloc(LCU_WIDTH*LCU_WIDTH);
//Reconstruct both predictors
kvz_inter_recon_lcu(state, ref1, xpos, ypos, width, mv_param[0], lcu, high_precision_rec0);
kvz_inter_recon_lcu(state, ref1, xpos, ypos, width, height, mv_param[0], lcu, high_precision_rec0);
if (!hi_prec_luma_rec0){
memcpy(temp_lcu_y, lcu->rec.y, sizeof(kvz_pixel) * 64 * 64);
}
@ -360,10 +378,10 @@ void kvz_inter_recon_lcu_bipred(const encoder_state_t * const state, const kvz_p
memcpy(temp_lcu_u, lcu->rec.u, sizeof(kvz_pixel) * 32 * 32);
memcpy(temp_lcu_v, lcu->rec.v, sizeof(kvz_pixel) * 32 * 32);
}
kvz_inter_recon_lcu(state, ref2, xpos, ypos, width, mv_param[1], lcu, high_precision_rec1);
kvz_inter_recon_lcu(state, ref2, xpos, ypos, width, height, mv_param[1], lcu, high_precision_rec1);
// After reconstruction, merge the predictors by taking an average of each pixel
for (temp_y = 0; temp_y < width; ++temp_y) {
for (temp_y = 0; temp_y < height; ++temp_y) {
int y_in_lcu = ((ypos + temp_y) & ((LCU_WIDTH)-1));
for (temp_x = 0; temp_x < width; ++temp_x) {
int x_in_lcu = ((xpos + temp_x) & ((LCU_WIDTH)-1));
@ -373,7 +391,7 @@ void kvz_inter_recon_lcu_bipred(const encoder_state_t * const state, const kvz_p
}
}
for (temp_y = 0; temp_y < width >> 1; ++temp_y) {
for (temp_y = 0; temp_y < height >> 1; ++temp_y) {
int y_in_lcu = (((ypos >> 1) + temp_y) & (LCU_WIDTH_C - 1));
for (temp_x = 0; temp_x < width >> 1; ++temp_x) {
int x_in_lcu = (((xpos >> 1) + temp_x) & (LCU_WIDTH_C - 1));

22
src/inter.h
View file

@ -40,8 +40,26 @@ typedef struct {
//void kvz_inter_set_block(image* im,uint32_t x_cu, uint32_t y_cu, uint8_t depth, cu_info *cur_cu);
void kvz_inter_recon_lcu(const encoder_state_t * const state, const kvz_picture * ref, int32_t xpos, int32_t ypos, int32_t width, const int16_t mv_param[2], lcu_t* lcu, hi_prec_buf_t *hi_prec_out);
void kvz_inter_recon_lcu_bipred(const encoder_state_t * const state, const kvz_picture * ref1, const kvz_picture * ref2, int32_t xpos, int32_t ypos, int32_t width, int16_t mv_param[2][2], lcu_t* lcu);
void kvz_inter_recon_lcu(const encoder_state_t * const state,
const kvz_picture * ref,
int32_t xpos,
int32_t ypos,
int32_t width,
int32_t height,
const int16_t mv_param[2],
lcu_t* lcu,
hi_prec_buf_t *hi_prec_out);
void kvz_inter_recon_lcu_bipred(const encoder_state_t * const state,
const kvz_picture * ref1,
const kvz_picture * ref2,
int32_t xpos,
int32_t ypos,
int32_t width,
int32_t height,
int16_t mv_param[2][2],
lcu_t* lcu);
void kvz_inter_get_spatial_merge_candidates(int32_t x,
int32_t y,

18
src/search.c
View file

@ -573,9 +573,23 @@ static double search_cu(encoder_state_t * const state, int x, int y, int depth,
kvz_lcu_set_trdepth(&work_tree[depth], x, y, depth, tr_depth);
if (cur_cu->inter.mv_dir == 3) {
kvz_inter_recon_lcu_bipred(state, state->global->ref->images[cur_cu->inter.mv_ref[0]], state->global->ref->images[cur_cu->inter.mv_ref[1]], x, y, LCU_WIDTH >> depth, cur_cu->inter.mv, &work_tree[depth]);
kvz_inter_recon_lcu_bipred(state,
state->global->ref->images[cur_cu->inter.mv_ref[0]],
state->global->ref->images[cur_cu->inter.mv_ref[1]],
x, y,
LCU_WIDTH >> depth,
LCU_WIDTH >> depth,
cur_cu->inter.mv,
&work_tree[depth]);
} else {
kvz_inter_recon_lcu(state, state->global->ref->images[cur_cu->inter.mv_ref[cur_cu->inter.mv_dir - 1]], x, y, LCU_WIDTH >> depth, cur_cu->inter.mv[cur_cu->inter.mv_dir - 1], &work_tree[depth], 0);
kvz_inter_recon_lcu(state,
state->global->ref->images[cur_cu->inter.mv_ref[cur_cu->inter.mv_dir - 1]],
x, y,
LCU_WIDTH >> depth,
LCU_WIDTH >> depth,
cur_cu->inter.mv[cur_cu->inter.mv_dir - 1],
&work_tree[depth],
0);
}
kvz_quantize_lcu_luma_residual(state, x, y, depth, NULL, &work_tree[depth]);

9
src/search_inter.c
View file

@ -1199,7 +1199,14 @@ int kvz_search_cu_inter(const encoder_state_t * const state, int x, int y, int d
if (ceil_mv_l0 < max_mv || ceil_mv_l1 < max_mv) continue;
}
kvz_inter_recon_lcu_bipred(state, state->global->ref->images[merge_cand[i].ref[0]], state->global->ref->images[merge_cand[j].ref[1]], x, y, LCU_WIDTH >> depth, mv, templcu);
kvz_inter_recon_lcu_bipred(state,
state->global->ref->images[merge_cand[i].ref[0]],
state->global->ref->images[merge_cand[j].ref[1]],
x, y,
LCU_WIDTH >> depth,
LCU_WIDTH >> depth,
mv,
templcu);
for (int ypos = 0; ypos < LCU_WIDTH >> depth; ++ypos) {
int dst_y = ypos*(LCU_WIDTH >> depth);