One may interleave even more vertical and horizontal passes, but this
is less straightforward (more code changes than just moves).
--
Christophe
From b48828b956d3a183d358ef9ad8a12f354c87da7d Mon Sep 17 00:00:00 2001
From: Christophe Gisquet <christophe.gisq...@gmail.com>
Date: Sat, 26 Jul 2014 21:40:53 +0200
Subject: [PATCH 13/13] hevc_filter: run vertical and horizontal together
This should help cache locality. On win64:
Before: 1397x cycles, 16216 bytes
After: 1369x cycles, 16040 bytes
---
libavcodec/hevc_filter.c | 90 ++++++++++++++++++++++++------------------------
1 file changed, 45 insertions(+), 45 deletions(-)
diff --git a/libavcodec/hevc_filter.c b/libavcodec/hevc_filter.c
index 7a25b9f..d98f04c 100644
--- a/libavcodec/hevc_filter.c
+++ b/libavcodec/hevc_filter.c
@@ -375,8 +375,11 @@ static void deblocking_filter_CTB(HEVCContext *s, int x0, int y0)
tc_offset = cur_tc_offset;
beta_offset = cur_beta_offset;
- // vertical filtering luma
+ x_end2 = x_end;
+ if (x_end2 != s->sps->width)
+ x_end2 -= 8;
for (y = y0; y < y_end; y += 8) {
+ // vertical filtering luma
for (x = x0 ? x0 : 8; x < x_end; x += 8) {
const int bs0 = s->vertical_bs[(x + y * s->bs_width) >> 2];
const int bs1 = s->vertical_bs[(x + (y + 4) * s->bs_width) >> 2];
@@ -402,47 +405,12 @@ static void deblocking_filter_CTB(HEVCContext *s, int x0, int y0)
beta, tc, no_p, no_q);
}
}
- }
-
- // vertical filtering chroma
- for (chroma = 1; chroma <= 2; chroma++) {
- int h = 1 << s->sps->hshift[chroma];
- int v = 1 << s->sps->vshift[chroma];
- for (y = y0; y < y_end; y += (8 * v)) {
- for (x = x0 ? x0 : 8 * h; x < x_end; x += (8 * h)) {
- const int bs0 = s->vertical_bs[(x + y * s->bs_width) >> 2];
- const int bs1 = s->vertical_bs[(x + (y + (4 * v)) * s->bs_width) >> 2];
- if ((bs0 == 2) || (bs1 == 2)) {
- const int qp0 = (get_qPy(s, x - 1, y) + get_qPy(s, x, y) + 1) >> 1;
- const int qp1 = (get_qPy(s, x - 1, y + (4 * v)) + get_qPy(s, x, y + (4 * v)) + 1) >> 1;
+ if(!y)
+ continue;
- c_tc[0] = (bs0 == 2) ? chroma_tc(s, qp0, chroma, tc_offset) : 0;
- c_tc[1] = (bs1 == 2) ? chroma_tc(s, qp1, chroma, tc_offset) : 0;
- src = &s->frame->data[chroma][(y >> s->sps->vshift[chroma]) * s->frame->linesize[chroma] + ((x >> s->sps->hshift[chroma]) << s->sps->pixel_shift)];
- if (pcmf) {
- no_p[0] = get_pcm(s, x - 1, y);
- no_p[1] = get_pcm(s, x - 1, y + (4 * v));
- no_q[0] = get_pcm(s, x, y);
- no_q[1] = get_pcm(s, x, y + (4 * v));
- s->hevcdsp.hevc_v_loop_filter_chroma_c(src,
- s->frame->linesize[chroma],
- c_tc, no_p, no_q);
- } else
- s->hevcdsp.hevc_v_loop_filter_chroma(src,
- s->frame->linesize[chroma],
- c_tc, no_p, no_q);
- }
- }
- }
- }
-
- // horizontal filtering luma
- x_end2 = x_end;
- if (x_end != s->sps->width)
- x_end -= 8;
- for (y = y0 ? y0 : 8; y < y_end; y += 8) {
- for (x = x0 ? x0 - 8 : 0; x < x_end; x += 8) {
+ // horizontal filtering luma
+ for (x = x0 ? x0 - 8 : 0; x < x_end2; x += 8) {
const int bs0 = s->horizontal_bs[( x + y * s->bs_width) >> 2];
const int bs1 = s->horizontal_bs[((x + 4) + y * s->bs_width) >> 2];
if (bs0 || bs1) {
@@ -471,15 +439,47 @@ static void deblocking_filter_CTB(HEVCContext *s, int x0, int y0)
}
}
- // horizontal filtering chroma
for (chroma = 1; chroma <= 2; chroma++) {
int h = 1 << s->sps->hshift[chroma];
int v = 1 << s->sps->vshift[chroma];
- if (x_end2 != s->sps->width)
- x_end = x_end2 - 8 * h;
- for (y = y0 ? y0 : 8 * v; y < y_end; y += (8 * v)) {
+
+ // vertical filtering chroma
+ for (y = y0; y < y_end; y += (8 * v)) {
+ for (x = x0 ? x0 : 8 * h; x < x_end; x += (8 * h)) {
+ const int bs0 = s->vertical_bs[(x + y * s->bs_width) >> 2];
+ const int bs1 = s->vertical_bs[(x + (y + (4 * v)) * s->bs_width) >> 2];
+
+ if ((bs0 == 2) || (bs1 == 2)) {
+ const int qp0 = (get_qPy(s, x - 1, y) + get_qPy(s, x, y) + 1) >> 1;
+ const int qp1 = (get_qPy(s, x - 1, y + (4 * v)) + get_qPy(s, x, y + (4 * v)) + 1) >> 1;
+
+ c_tc[0] = (bs0 == 2) ? chroma_tc(s, qp0, chroma, tc_offset) : 0;
+ c_tc[1] = (bs1 == 2) ? chroma_tc(s, qp1, chroma, tc_offset) : 0;
+ src = &s->frame->data[chroma][(y >> s->sps->vshift[chroma]) * s->frame->linesize[chroma] + ((x >> s->sps->hshift[chroma]) << s->sps->pixel_shift)];
+ if (pcmf) {
+ no_p[0] = get_pcm(s, x - 1, y);
+ no_p[1] = get_pcm(s, x - 1, y + (4 * v));
+ no_q[0] = get_pcm(s, x, y);
+ no_q[1] = get_pcm(s, x, y + (4 * v));
+ s->hevcdsp.hevc_v_loop_filter_chroma_c(src,
+ s->frame->linesize[chroma],
+ c_tc, no_p, no_q);
+ } else
+ s->hevcdsp.hevc_v_loop_filter_chroma(src,
+ s->frame->linesize[chroma],
+ c_tc, no_p, no_q);
+ }
+ }
+
+ if(!y)
+ continue;
+
+ // horizontal filtering chroma
tc_offset = x0 ? left_tc_offset : cur_tc_offset;
- for (x = x0 ? x0 - 8 * h : 0; x < x_end; x += (8 * h)) {
+ x_end2 = x_end;
+ if (x_end != s->sps->width)
+ x_end2 = x_end - 8 * h;
+ for (x = x0 ? x0 - 8 * h : 0; x < x_end2; x += (8 * h)) {
const int bs0 = s->horizontal_bs[( x + y * s->bs_width) >> 2];
const int bs1 = s->horizontal_bs[((x + 4 * h) + y * s->bs_width) >> 2];
if ((bs0 == 2) || (bs1 == 2)) {
--
1.9.2.msysgit.0
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
