generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Added high bit-depth support in CLPF.

Browse Source 
Change-Id: Ic5eadb323227a820ad876c32d4dc296e05db6ece
This commit is contained in:
Steinar Midtskogen
2016-09-09 15:23:35 +02:00
committed by Yaowu Xu
parent 9351b2f792
commit 3dbd55a6c4
6 changed files with 1038 additions and 479 deletions
Download Patch File Download Diff File Expand all files Collapse all files

141
av1/common/clpf.c
View File

@@ -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);