Skip to content

GitLab

A
aom-rav1e
Commit b2be25f4 authored by Steinar Midtskogen's avatar Steinar Midtskogen
Browse files
Options

Added high bit-depth support in CLPF. 

Change-Id: Ic5eadb323227a820ad876c32d4dc296e05db6ece
parent fad338b3
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 1036 additions and 477 deletions
aom_dsp/aom_dsp_rtcd_defs.pl
View file @ b2be25f4
......@@ -625,6 +625,14 @@ add_proto qw/void aom_lpf_horizontal_4_dual/, "uint8_t *s, int pitch, const uint
specialize qw/aom_lpf_horizontal_4_dual sse2 neon dspr2 msa/;
if (aom_config("CONFIG_CLPF") eq "yes") {
if (aom_config("CONFIG_AOM_HIGHBITDEPTH") eq "yes") {
add_proto qw/void aom_clpf_block_hbd/, "const uint16_t *src, uint16_t *dst, int sstride, int dstride, int x0, int y0, int sizex, int sizey, int width, int height, unsigned int strength";
specialize qw/aom_clpf_block_hbd sse2 ssse3 sse4_1 neon/;
add_proto qw/void aom_clpf_detect_hbd/, "const uint16_t *rec, const uint16_t *org, int rstride, int ostride, int x0, int y0, int width, int height, int *sum0, int *sum1, unsigned int strength, int shift";
specialize qw/aom_clpf_detect_hbd sse2 ssse3 sse4_1 neon/;
add_proto qw/void aom_clpf_detect_multi_hbd/, "const uint16_t *rec, const uint16_t *org, int rstride, int ostride, int x0, int y0, int width, int height, int *sum, int shift";
specialize qw/aom_clpf_detect_multi_hbd sse2 ssse3 sse4_1 neon/;
}
add_proto qw/void aom_clpf_block/, "const uint8_t *src, uint8_t *dst, int sstride, int dstride, int x0, int y0, int sizex, int sizey, int width, int height, unsigned int strength";
specialize qw/aom_clpf_block sse2 ssse3 sse4_1 neon/;
add_proto qw/void aom_clpf_detect/, "const uint8_t *rec, const uint8_t *org, int rstride, int ostride, int x0, int y0, int width, int height, int *sum0, int *sum1, unsigned int strength";
......
av1/common/clpf.c
View file @ b2be25f4
......@@ -8,6 +8,7 @@
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <assert.h>
#include "av1/common/clpf.h"
#include "./aom_dsp_rtcd.h"
#include "aom_dsp/aom_dsp_common.h"
......@@ -47,6 +48,29 @@ void aom_clpf_block_c(const uint8_t *src, uint8_t *dst, int sstride,
}
}
#if CONFIG_AOM_HIGHBITDEPTH
// Identical to aom_clpf_block_c() apart from "src" and "dst".
void aom_clpf_block_hbd_c(const uint16_t *src, uint16_t *dst, int sstride,
int dstride, int x0, int y0, int sizex, int sizey,
int width, int height, unsigned int strength) {
int x, y;
for (y = y0; y < y0 + sizey; y++) {
for (x = x0; x < x0 + sizex; x++) {
int X = src[y * sstride + x];
int A = src[AOMMAX(0, y - 1) * sstride + x];
int B = src[y * sstride + AOMMAX(0, x - 2)];
int C = src[y * sstride + AOMMAX(0, x - 1)];
int D = src[y * sstride + AOMMIN(width - 1, x + 1)];
int E = src[y * sstride + AOMMIN(width - 1, x + 2)];
int F = src[AOMMIN(height - 1, y + 1) * sstride + x];
int delta;
delta = av1_clpf_sample(X, A, B, C, D, E, F, strength);
dst[y * dstride + x] = X + delta;
}
}
}
#endif
// Return number of filtered blocks
int av1_clpf_frame(const YV12_BUFFER_CONFIG *orig_dst,
const YV12_BUFFER_CONFIG *rec, const YV12_BUFFER_CONFIG *org,
......@@ -75,15 +99,27 @@ int av1_clpf_frame(const YV12_BUFFER_CONFIG *orig_dst,
const int cache_blocks = cache_size / (bs * bs);
YV12_BUFFER_CONFIG dst = *orig_dst;
assert(bs == 8); // Optimised code assumes this.
#if CONFIG_AOM_HIGHBITDEPTH
strength <<= (cm->bit_depth - 8);
#endif
// Make buffer space for in-place filtering
if (rec->y_buffer == dst.y_buffer) {
#if CONFIG_AOM_HIGHBITDEPTH
CHECK_MEM_ERROR(cm, cache,
aom_malloc(cache_size << !!cm->use_highbitdepth));
dst.y_buffer = cm->use_highbitdepth ? CONVERT_TO_BYTEPTR(cache) : cache;
#else
CHECK_MEM_ERROR(cm, cache, aom_malloc(cache_size));
dst.y_buffer = cache;
#endif
CHECK_MEM_ERROR(cm, cache_ptr,
aom_malloc(cache_blocks * sizeof(*cache_ptr)));
CHECK_MEM_ERROR(cm, cache_dst,
aom_malloc(cache_blocks * sizeof(*cache_dst)));
memset(cache_ptr, 0, cache_blocks * sizeof(*cache_dst));
dst.y_buffer = cache;
dstride = bs;
}
......@@ -125,34 +161,108 @@ int av1_clpf_frame(const YV12_BUFFER_CONFIG *orig_dst,
// Temporary buffering needed if filtering in-place
if (cache) {
if (cache_ptr[cache_idx]) {
// Copy filtered block back into the frame
// Copy filtered block back into the frame
#if CONFIG_AOM_HIGHBITDEPTH
if (cm->use_highbitdepth) {
uint16_t *const d =
CONVERT_TO_SHORTPTR(cache_dst[cache_idx]);
for (c = 0; c < bs; c++) {
*(uint64_t *)(d + c * sstride) =
*(uint64_t *)(cache_ptr[cache_idx] + c * bs * 2);
*(uint64_t *)(d + c * sstride + 4) =
*(uint64_t *)(cache_ptr[cache_idx] + c * bs * 2 + 8);
}
} else {
for (c = 0; c < bs; c++)
*(uint64_t *)(cache_dst[cache_idx] + c * sstride) =
*(uint64_t *)(cache_ptr[cache_idx] + c * bs);
}
#else
for (c = 0; c < bs; c++)
*(uint64_t *)(cache_dst[cache_idx] + c * sstride) =
*(uint64_t *)(cache_ptr[cache_idx] + c * bs);
#endif
}
#if CONFIG_AOM_HIGHBITDEPTH
if (cm->use_highbitdepth) {
cache_ptr[cache_idx] = cache + cache_idx * bs * bs * 2;
dst.y_buffer = CONVERT_TO_BYTEPTR(cache_ptr[cache_idx]) -
ypos * bs - xpos;
} else {
cache_ptr[cache_idx] = cache + cache_idx * bs * bs;
dst.y_buffer = cache_ptr[cache_idx] - ypos * bs - xpos;
}
#else
cache_ptr[cache_idx] = cache + cache_idx * bs * bs;
dst.y_buffer = cache_ptr[cache_idx] - ypos * bs - xpos;
#endif
cache_dst[cache_idx] = rec->y_buffer + ypos * sstride + xpos;
if (++cache_idx >= cache_blocks) cache_idx = 0;
}
// Apply the filter
// Apply the filter
#if CONFIG_AOM_HIGHBITDEPTH
if (cm->use_highbitdepth) {
aom_clpf_block_hbd(CONVERT_TO_SHORTPTR(rec->y_buffer),
CONVERT_TO_SHORTPTR(dst.y_buffer), sstride,
dstride, xpos, ypos, bs, bs, width, height,
strength);
} else {
aom_clpf_block(rec->y_buffer, dst.y_buffer, sstride, dstride,
xpos, ypos, bs, bs, width, height, strength);
}
#else
aom_clpf_block(rec->y_buffer, dst.y_buffer, sstride, dstride,
xpos, ypos, bs, bs, width, height, strength);
#endif
} else { // Skip block, copy instead
if (!cache)
if (!cache) {
#if CONFIG_AOM_HIGHBITDEPTH
if (cm->use_highbitdepth) {
uint16_t *const d = CONVERT_TO_SHORTPTR(dst.y_buffer);
const uint16_t *const s = CONVERT_TO_SHORTPTR(rec->y_buffer);
for (c = 0; c < bs; c++) {
*(uint64_t *)(d + (ypos + c) * dstride + xpos) =
*(uint64_t *)(s + (ypos + c) * sstride + xpos);
*(uint64_t *)(d + (ypos + c) * dstride + xpos + 4) =
*(uint64_t *)(s + (ypos + c) * sstride + xpos + 4);
}
} else {
for (c = 0; c < bs; c++)
*(uint64_t *)(dst.y_buffer + (ypos + c) * dstride + xpos) =
*(uint64_t *)(rec->y_buffer + (ypos + c) * sstride +
xpos);
}
#else
for (c = 0; c < bs; c++)
*(uint64_t *)(dst.y_buffer + (ypos + c) * dstride + xpos) = *(
uint64_t *)(rec->y_buffer + (ypos + c) * sstride + xpos);
#endif
}
}
}
}
} else { // Entire filter block is skip, copy
if (!cache)
if (!cache) {
#if CONFIG_AOM_HIGHBITDEPTH
if (cm->use_highbitdepth) {
for (m = 0; m < h; m++)
memcpy(CONVERT_TO_SHORTPTR(dst.y_buffer) + (yoff + m) * dstride +
xoff,
CONVERT_TO_SHORTPTR(rec->y_buffer) + (yoff + m) * sstride +
xoff,
w * 2);
} else {
for (m = 0; m < h; m++)
memcpy(dst.y_buffer + (yoff + m) * dstride + xoff,
rec->y_buffer + (yoff + m) * sstride + xoff, w);
}
#else
for (m = 0; m < h; m++)
memcpy(dst.y_buffer + (yoff + m) * dstride + xoff,
rec->y_buffer + (yoff + m) * sstride + xoff, w);
#endif
}
}
block_index += !allskip; // Count number of blocks filtered
}
......@@ -161,10 +271,27 @@ int av1_clpf_frame(const YV12_BUFFER_CONFIG *orig_dst,
if (cache) {
// Copy remaining blocks into the frame
for (cache_idx = 0; cache_idx < cache_blocks && cache_ptr[cache_idx];
cache_idx++)
cache_idx++) {
#if CONFIG_AOM_HIGHBITDEPTH
if (cm->use_highbitdepth) {
uint16_t *const d = CONVERT_TO_SHORTPTR(cache_dst[cache_idx]);
for (c = 0; c < bs; c++) {
*(uint64_t *)(d + c * sstride) =
*(uint64_t *)(cache_ptr[cache_idx] + c * bs * 2);
*(uint64_t *)(d + c * sstride + 4) =
*(uint64_t *)(cache_ptr[cache_idx] + c * bs * 2 + 8);
}
} else {
for (c = 0; c < bs; c++)
*(uint64_t *)(cache_dst[cache_idx] + c * sstride) =
*(uint64_t *)(cache_ptr[cache_idx] + c * bs);
}
#else
for (c = 0; c < bs; c++)
*(uint64_t *)(cache_dst[cache_idx] + c * sstride) =
*(uint64_t *)(cache_ptr[cache_idx] + c * bs);
#endif
}
aom_free(cache);
aom_free(cache_ptr);
......
av1/common/clpf_simd.h
View file @ b2be25f4
This diff is collapsed.
av1/encoder/clpf_rdo.c
View file @ b2be25f4
......@@ -66,6 +66,62 @@ void aom_clpf_detect_multi_c(const uint8_t *rec, const uint8_t *org,
}
}
#if CONFIG_AOM_HIGHBITDEPTH
// Identical to aom_clpf_detect_c() apart from "rec" and "org".
void aom_clpf_detect_hbd_c(const uint16_t *rec, const uint16_t *org,
int rstride, int ostride, int x0, int y0, int width,
int height, int *sum0, int *sum1,
unsigned int strength, int shift) {
int x, y;
for (y = y0; y < y0 + 8; y++) {
for (x = x0; x < x0 + 8; x++) {
int O = org[y * ostride + x] >> shift;
int X = rec[y * rstride + x] >> shift;
int A = rec[AOMMAX(0, y - 1) * rstride + x] >> shift;
int B = rec[y * rstride + AOMMAX(0, x - 2)] >> shift;
int C = rec[y * rstride + AOMMAX(0, x - 1)] >> shift;
int D = rec[y * rstride + AOMMIN(width - 1, x + 1)] >> shift;
int E = rec[y * rstride + AOMMIN(width - 1, x + 2)] >> shift;
int F = rec[AOMMIN(height - 1, y + 1) * rstride + x] >> shift;
int delta = av1_clpf_sample(X, A, B, C, D, E, F, strength >> shift);
int Y = X + delta;
*sum0 += (O - X) * (O - X);
*sum1 += (O - Y) * (O - Y);
}
}
}
// aom_clpf_detect_multi_c() apart from "rec" and "org".
void aom_clpf_detect_multi_hbd_c(const uint16_t *rec, const uint16_t *org,
int rstride, int ostride, int x0, int y0,
int width, int height, int *sum, int shift) {
int x, y;
for (y = y0; y < y0 + 8; y++) {
for (x = x0; x < x0 + 8; x++) {
int O = org[y * ostride + x] >> shift;
int X = rec[y * rstride + x] >> shift;
int A = rec[AOMMAX(0, y - 1) * rstride + x] >> shift;
int B = rec[y * rstride + AOMMAX(0, x - 2)] >> shift;
int C = rec[y * rstride + AOMMAX(0, x - 1)] >> shift;
int D = rec[y * rstride + AOMMIN(width - 1, x + 1)] >> shift;
int E = rec[y * rstride + AOMMIN(width - 1, x + 2)] >> shift;
int F = rec[AOMMIN(height - 1, y + 1) * rstride + x] >> shift;
int delta1 = av1_clpf_sample(X, A, B, C, D, E, F, 1);
int delta2 = av1_clpf_sample(X, A, B, C, D, E, F, 2);
int delta3 = av1_clpf_sample(X, A, B, C, D, E, F, 4);
int F1 = X + delta1;
int F2 = X + delta2;
int F3 = X + delta3;
sum[0] += (O - X) * (O - X);
sum[1] += (O - F1) * (O - F1);
sum[2] += (O - F2) * (O - F2);
sum[3] += (O - F3) * (O - F3);
}
}
}
#endif
int av1_clpf_decision(int k, int l, const YV12_BUFFER_CONFIG *rec,
const YV12_BUFFER_CONFIG *org, const AV1_COMMON *cm,
int block_size, int w, int h, unsigned int strength,
......@@ -77,10 +133,25 @@ int av1_clpf_decision(int k, int l, const YV12_BUFFER_CONFIG *rec,
int ypos = (k << fb_size_log2) + m * block_size;
const int bs = MAX_MIB_SIZE;
if (!cm->mi_grid_visible[ypos / bs * cm->mi_stride + xpos / bs]
->mbmi.skip)
->mbmi.skip) {
#if CONFIG_AOM_HIGHBITDEPTH
if (cm->use_highbitdepth) {
aom_clpf_detect_hbd(CONVERT_TO_SHORTPTR(rec->y_buffer),
CONVERT_TO_SHORTPTR(org->y_buffer), rec->y_stride,
org->y_stride, xpos, ypos, rec->y_crop_width,
rec->y_crop_height, &sum0, &sum1, strength,
cm->bit_depth - 8);
} else {
aom_clpf_detect(rec->y_buffer, org->y_buffer, rec->y_stride,
org->y_stride, xpos, ypos, rec->y_crop_width,
rec->y_crop_height, &sum0, &sum1, strength);
}
#else
aom_clpf_detect(rec->y_buffer, org->y_buffer, rec->y_stride,
org->y_stride, xpos, ypos, rec->y_crop_width,
rec->y_crop_height, &sum0, &sum1, strength);
#endif
}
}
}
*res = sum1 < sum0;
......@@ -145,9 +216,23 @@ static int clpf_rdo(int y, int x, const YV12_BUFFER_CONFIG *rec,
if (!cm->mi_grid_visible[ypos / MAX_MIB_SIZE * cm->mi_stride +
xpos / MAX_MIB_SIZE]
->mbmi.skip) {
#if CONFIG_AOM_HIGHBITDEPTH
if (cm->use_highbitdepth) {
aom_clpf_detect_multi_hbd(CONVERT_TO_SHORTPTR(rec->y_buffer),
CONVERT_TO_SHORTPTR(org->y_buffer),
rec->y_stride, org->y_stride, xpos, ypos,
rec->y_crop_width, rec->y_crop_height, sum,
cm->bit_depth - 8);
} else {
aom_clpf_detect_multi(rec->y_buffer, org->y_buffer, rec->y_stride,
org->y_stride, xpos, ypos, rec->y_crop_width,
rec->y_crop_height, sum);
}
#else
aom_clpf_detect_multi(rec->y_buffer, org->y_buffer, rec->y_stride,
org->y_stride, xpos, ypos, rec->y_crop_width,
rec->y_crop_height, sum);
#endif
filtered = 1;
}
}
......
av1/encoder/clpf_rdo_simd.h
View file @ b2be25f4
This diff is collapsed.
test/clpf_test.cc
View file @ b2be25f4
This diff is collapsed.
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment