Drop temporary arrays in bipred search

Changes bipred search to use the original source and reconstruction
arrays directly instead of copying them.
This commit is contained in:
Arttu Ylä-Outinen 2018-02-14 10:26:46 +02:00
parent 0ea516ba30
commit cb06cfeadb

View file

@ -1405,19 +1405,16 @@ static void search_pu_inter_bipred(inter_search_info_t *info,
if (i >= info->num_merge_cand || j >= info->num_merge_cand) break; if (i >= info->num_merge_cand || j >= info->num_merge_cand) break;
// Find one L0 and L1 candidate according to the priority list // Find one L0 and L1 candidate according to the priority list
if ((merge_cand[i].dir & 0x1) && (merge_cand[j].dir & 0x2)) { if (!(merge_cand[i].dir & 0x1) || !(merge_cand[j].dir & 0x2)) continue;
if (ref_LX[0][merge_cand[i].ref[0]] !=
ref_LX[1][merge_cand[j].ref[1]] ||
merge_cand[i].mv[0][0] != merge_cand[j].mv[1][0] || if (ref_LX[0][merge_cand[i].ref[0]] == ref_LX[1][merge_cand[j].ref[1]] &&
merge_cand[i].mv[0][1] != merge_cand[j].mv[1][1]) merge_cand[i].mv[0][0] == merge_cand[j].mv[1][0] &&
merge_cand[i].mv[0][1] == merge_cand[j].mv[1][1])
{ {
uint32_t bitcost[2]; continue;
uint32_t cost = 0; }
int16_t mv[2][2];
kvz_pixel tmp_block[64 * 64];
kvz_pixel tmp_pic[64 * 64];
int16_t mv[2][2];
mv[0][0] = merge_cand[i].mv[0][0]; mv[0][0] = merge_cand[i].mv[0][0];
mv[0][1] = merge_cand[i].mv[0][1]; mv[0][1] = merge_cand[i].mv[0][1];
mv[1][0] = merge_cand[j].mv[1][0]; mv[1][0] = merge_cand[j].mv[1][0];
@ -1430,7 +1427,6 @@ static void search_pu_inter_bipred(inter_search_info_t *info,
continue; continue;
} }
lcu_t templcu;
kvz_inter_recon_bipred(info->state, kvz_inter_recon_bipred(info->state,
ref->images[ref_LX[0][merge_cand[i].ref[0]]], ref->images[ref_LX[0][merge_cand[i].ref[0]]],
ref->images[ref_LX[1][merge_cand[j].ref[1]]], ref->images[ref_LX[1][merge_cand[j].ref[1]]],
@ -1438,18 +1434,14 @@ static void search_pu_inter_bipred(inter_search_info_t *info,
width, width,
height, height,
mv, mv,
&templcu); lcu);
for (int ypos = 0; ypos < height; ++ypos) { const kvz_pixel *rec = &lcu->rec.y[SUB_SCU(y) * LCU_WIDTH + SUB_SCU(x)];
int dst_y = ypos * width; const kvz_pixel *src = &frame->source->y[x + y * frame->source->width];
for (int xpos = 0; xpos < width; ++xpos) { uint32_t cost =
tmp_block[dst_y + xpos] = templcu.rec.y[ kvz_satd_any_size(cu_width, cu_width, rec, LCU_WIDTH, src, frame->source->width);
SUB_SCU(y + ypos) * LCU_WIDTH + SUB_SCU(x + xpos)];
tmp_pic[dst_y + xpos] = frame->source->y[x + xpos + (y + ypos)*frame->source->width];
}
}
cost = kvz_satd_any_size(cu_width, cu_width, tmp_pic, cu_width, tmp_block, cu_width); uint32_t bitcost[2] = { 0, 0 };
cost += info->mvd_cost_func(info->state, cost += info->mvd_cost_func(info->state,
merge_cand[i].mv[0][0], merge_cand[i].mv[0][0],
@ -1473,7 +1465,6 @@ static void search_pu_inter_bipred(inter_search_info_t *info,
const int extra_bits = mv_ref_coded[0] + mv_ref_coded[1] + 2 /* mv dir cost */; const int extra_bits = mv_ref_coded[0] + mv_ref_coded[1] + 2 /* mv dir cost */;
cost += info->state->lambda_sqrt * extra_bits + 0.5; cost += info->state->lambda_sqrt * extra_bits + 0.5;
if (cost < *inter_cost) { if (cost < *inter_cost) {
cur_cu->inter.mv_dir = 3; cur_cu->inter.mv_dir = 3;
@ -1517,8 +1508,6 @@ static void search_pu_inter_bipred(inter_search_info_t *info,
*inter_bitcost = bitcost[0] + bitcost[1] + extra_bits; *inter_bitcost = bitcost[0] + bitcost[1] + extra_bits;
} }
} }
}
}
} }