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

Simplify the prediction mode search recursion.

Browse source 
No need to be so clever for something this simple.

I moved the cost initialization outside the recursing function because it
relied on the clever recursion to work. It should eventually be moved to its
own function that also initializes all the other fields just to be safe. I
didn't do that yet because I want to do it per-LCU and these functions are
still working on per-frame basis.
This commit is contained in:
Ari Koivula 2014-02-04 18:07:48 +02:00
parent 337a565232
commit ea312082a4
1 changed files with 65 additions and 43 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

108
src/search.c
View file

@ -365,59 +365,65 @@ void search_intra(encoder_control *encoder, uint16_t x_ctb, uint16_t y_ctb, uint
} }
/** /**
* \brief * \brief Search best modes at each depth for the whole picture.
*
* This function fills the cur_pic->cu_array of the current picture
* with the best mode and it's cost for each CU at each depth for the whole
* frame.
*/ */
void search_tree(encoder_control *encoder, void search_tree(encoder_control *encoder,
uint16_t x_ctb, uint16_t y_ctb, uint8_t depth) uint16_t x_ctb, uint16_t y_ctb, uint8_t depth)
{ {
uint8_t border_x = ((encoder->in.width) < (x_ctb * (LCU_WIDTH >> MAX_DEPTH) + (LCU_WIDTH >> depth))) ? 1 : 0; int cu_width = LCU_WIDTH >> depth;
uint8_t border_y = ((encoder->in.height) < (y_ctb * (LCU_WIDTH >> MAX_DEPTH) + (LCU_WIDTH >> depth))) ? 1 : 0; int x = x_ctb * (LCU_WIDTH >> MAX_DEPTH);
uint8_t border_split_x = ((encoder->in.width) < ((x_ctb + 1) * (LCU_WIDTH >> MAX_DEPTH) + (LCU_WIDTH >> (depth + 1)))) ? 0 : 1; int y = y_ctb * (LCU_WIDTH >> MAX_DEPTH);
uint8_t border_split_y = ((encoder->in.height) < ((y_ctb + 1) * (LCU_WIDTH >> MAX_DEPTH) + (LCU_WIDTH >> (depth + 1)))) ? 0 : 1;
uint8_t border = border_x | border_y; // are we in any border CU
picture *cur_pic = encoder->in.cur_pic; // Stop recursion if the CU is completely outside the frame.
cu_info *cur_cu = &cur_pic->cu_array[depth][x_ctb + y_ctb * (encoder->in.width_in_lcu << MAX_DEPTH)]; if (x >= encoder->in.width || y >= encoder->in.height) {
return;
}
cur_cu->intra[0].cost = 0xffffffff; // If the CU is partially outside the frame, split.
cur_cu->inter.cost = 0xffffffff; if (x + cu_width > encoder->in.width ||
y + cu_width > encoder->in.height)
{
int half_cu = 1 << (MAX_DEPTH - (depth + 1));
// Force split on border search_tree(encoder, x_ctb, y_ctb, depth + 1);
if (depth != MAX_DEPTH) { search_tree(encoder, x_ctb + half_cu, y_ctb, depth + 1);
if (border) { search_tree(encoder, x_ctb, y_ctb + half_cu, depth + 1);
uint8_t change = 1 << (MAX_DEPTH - 1 - depth); search_tree(encoder, x_ctb + half_cu, y_ctb + half_cu, depth + 1);
search_tree(encoder, x_ctb, y_ctb, depth + 1);
if (!border_x || border_split_x) { return;
search_tree(encoder, x_ctb + change, y_ctb, depth + 1); }
}
if (!border_y || border_split_y) { // CU is completely inside the frame, so search for best prediction mode at
search_tree(encoder, x_ctb, y_ctb + change, depth + 1); // this depth.
} {
if (!border || (border_split_x && border_split_y)) { picture *cur_pic = encoder->in.cur_pic;
search_tree(encoder, x_ctb + change, y_ctb + change, depth + 1);
} if (cur_pic->slicetype != SLICE_I &&
return; depth >= MIN_INTER_SEARCH_DEPTH &&
depth <= MAX_INTER_SEARCH_DEPTH)
{
search_inter(encoder, x_ctb, y_ctb, depth);
}
if ((encoder->in.cur_pic->slicetype == SLICE_I || USE_INTRA_IN_P) &&
depth >= MIN_INTRA_SEARCH_DEPTH &&
depth <= MAX_INTRA_SEARCH_DEPTH)
{
search_intra(encoder, x_ctb, y_ctb, depth);
} }
} }
if (cur_pic->slicetype != SLICE_I // Recurse to max search depth.
&& depth >= MIN_INTER_SEARCH_DEPTH && depth <= MAX_INTER_SEARCH_DEPTH) {
search_inter(encoder, x_ctb, y_ctb, depth);
}
if (depth >= MIN_INTRA_SEARCH_DEPTH && depth <= MAX_INTRA_SEARCH_DEPTH
&& (encoder->in.cur_pic->slicetype == SLICE_I || USE_INTRA_IN_P)) {
search_intra(encoder, x_ctb, y_ctb, depth);
}
// Split and search to max_depth
if (depth < MAX_INTRA_SEARCH_DEPTH && depth < MAX_INTER_SEARCH_DEPTH) { if (depth < MAX_INTRA_SEARCH_DEPTH && depth < MAX_INTER_SEARCH_DEPTH) {
// Split blocks and remember to change x and y block positions int half_cu = 1 << (MAX_DEPTH - (depth + 1));
uint8_t change = 1 << (MAX_DEPTH - 1 - depth); search_tree(encoder, x_ctb, y_ctb, depth + 1);
search_tree(encoder, x_ctb, y_ctb, depth + 1); search_tree(encoder, x_ctb + half_cu, y_ctb, depth + 1);
search_tree(encoder, x_ctb + change, y_ctb, depth + 1); search_tree(encoder, x_ctb, y_ctb + half_cu, depth + 1);
search_tree(encoder, x_ctb, y_ctb + change, depth + 1); search_tree(encoder, x_ctb + half_cu, y_ctb + half_cu, depth + 1);
search_tree(encoder, x_ctb + change, y_ctb + change, depth + 1);
} }
} }
@ -462,7 +468,6 @@ uint32_t search_best_mode(encoder_control *encoder,
} }
/** /**
* \brief * \brief
*/ */
@ -471,6 +476,22 @@ void search_slice_data(encoder_control *encoder)
int16_t x_lcu, y_lcu; int16_t x_lcu, y_lcu;
FILE *fp = 0, *fp2 = 0; FILE *fp = 0, *fp2 = 0;
// Initialize the costs in the cu-array used for searching.
{
int d, x_cu, y_cu;
for (y_cu = 0; y_cu < encoder->in.height / CU_MIN_SIZE_PIXELS; ++y_cu) {
for (x_cu = 0; x_cu < encoder->in.width / CU_MIN_SIZE_PIXELS; ++x_cu) {
for (d = 0; d <= MAX_DEPTH; ++d) {
picture *cur_pic = encoder->in.cur_pic;
cu_info *cur_cu = &cur_pic->cu_array[d][x_cu + y_cu * (encoder->in.width_in_lcu << MAX_DEPTH)];
cur_cu->intra[0].cost = -1;
cur_cu->inter.cost = -1;
}
}
}
}
if (RENDER_CU) { if (RENDER_CU) {
fp = open_cu_file("cu_search.html"); fp = open_cu_file("cu_search.html");
fp2 = open_cu_file("cu_best.html"); fp2 = open_cu_file("cu_best.html");
@ -480,6 +501,7 @@ void search_slice_data(encoder_control *encoder)
for (y_lcu = 0; y_lcu < encoder->in.height_in_lcu; y_lcu++) { for (y_lcu = 0; y_lcu < encoder->in.height_in_lcu; y_lcu++) {
for (x_lcu = 0; x_lcu < encoder->in.width_in_lcu; x_lcu++) { for (x_lcu = 0; x_lcu < encoder->in.width_in_lcu; x_lcu++) {
uint8_t depth = 0; uint8_t depth = 0;
// Recursive function for looping through all the sub-blocks // Recursive function for looping through all the sub-blocks
search_tree(encoder, x_lcu << MAX_DEPTH, y_lcu << MAX_DEPTH, depth); search_tree(encoder, x_lcu << MAX_DEPTH, y_lcu << MAX_DEPTH, depth);
if (RENDER_CU) { if (RENDER_CU) {