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

[SIMD] Loop transformations for SIMD processing

Browse source
This commit is contained in:
Ari Lemmetti 2021-09-06 22:30:36 +03:00
parent c195d906d3
commit 22da8cfe65
1 changed files with 18 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

24
src/strategies/avx2/intra-avx2.c
View file

@ -976,13 +976,25 @@ static void kvz_pdpc_planar_dc_avx2(
// TODO: replace latter log2_width with log2_height
const int scale = ((log2_width - 2 + log2_width - 2 + 2) >> 2);
int16_t wLT[LCU_WIDTH];
// Same weights regardless of axis, compute once
for (int i = 0; i < width; ++i) {
wLT[i] = 32 >> MIN(31, ((i << 1) >> scale));
}
// Process in 4x4 blocks
// TODO: replace width with height
for (int y = 0; y < width; y++) {
int wT = 32 >> MIN(31, ((y << 1) >> scale));
for (int x = 0; x < width; x++) {
int wL = 32 >> MIN(31, ((x << 1) >> scale));
dst[x + y * width] = dst[x + y * width] + ((wL * (used_ref->left[y + 1] - dst[x + y * width])
+ wT * (used_ref->top[x + 1] - dst[x + y * width]) + 32) >> 6);
for (int y = 0; y < width; y += 4) {
for (int x = 0; x < width; x += 4) {
for (int yy = 0; yy < 4; ++yy) {
for (int xx = 0; xx < 4; ++xx) {
dst[(x + xx) + (y + yy) * width] += ((
wLT[(x + xx)] * (used_ref->left[(y + yy) + 1] - dst[(x + xx) + (y + yy) * width]) +
wLT[(y + yy)] * (used_ref->top [(x + xx) + 1] - dst[(x + xx) + (y + yy) * width]) +
32) >> 6);
}
}
}
}
}