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

Merge "Rewrite loop_filter_info_n struct"

Browse Source
This commit is contained in:
Yunqing Wang 2013-10-29 09:49:36 -07:00 committed by Gerrit Code Review
commit c634ec6a56
2 changed files with 27 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

44
vp9/common/vp9_loopfilter.c
View File

@ -16,12 +16,6 @@
#include "vp9/common/vp9_seg_common.h" #include "vp9/common/vp9_seg_common.h"
struct loop_filter_info {
const uint8_t *mblim;
const uint8_t *lim;
const uint8_t *hev_thr;
};
// This structure holds bit masks for all 8x8 blocks in a 64x64 region. // This structure holds bit masks for all 8x8 blocks in a 64x64 region.
// Each 1 bit represents a position in which we want to apply the loop filter. // Each 1 bit represents a position in which we want to apply the loop filter.
// Left_ entries refer to whether we apply a filter on the border to the // Left_ entries refer to whether we apply a filter on the border to the
@ -259,8 +253,8 @@ static void update_sharpness(loop_filter_info_n *lfi, int sharpness_lvl) {
if (block_inside_limit < 1) if (block_inside_limit < 1)
block_inside_limit = 1; block_inside_limit = 1;
vpx_memset(lfi->lim[lvl], block_inside_limit, SIMD_WIDTH); vpx_memset(lfi->lfthr[lvl].lim, block_inside_limit, SIMD_WIDTH);
vpx_memset(lfi->mblim[lvl], (2 * (lvl + 2) + block_inside_limit), vpx_memset(lfi->lfthr[lvl].mblim, (2 * (lvl + 2) + block_inside_limit),
SIMD_WIDTH); SIMD_WIDTH);
} }
} }
@ -268,7 +262,7 @@ static void update_sharpness(loop_filter_info_n *lfi, int sharpness_lvl) {
void vp9_loop_filter_init(VP9_COMMON *cm) { void vp9_loop_filter_init(VP9_COMMON *cm) {
loop_filter_info_n *lfi = &cm->lf_info; loop_filter_info_n *lfi = &cm->lf_info;
struct loopfilter *lf = &cm->lf; struct loopfilter *lf = &cm->lf;
int i; int lvl;
// init limits for given sharpness // init limits for given sharpness
update_sharpness(lfi, lf->sharpness_level); update_sharpness(lfi, lf->sharpness_level);
@ -278,8 +272,8 @@ void vp9_loop_filter_init(VP9_COMMON *cm) {
lf_init_lut(lfi); lf_init_lut(lfi);
// init hev threshold const vectors // init hev threshold const vectors
for (i = 0; i < 4; i++) for (lvl = 0; lvl <= MAX_LOOP_FILTER; lvl++)
vpx_memset(lfi->hev_thr[i], i, SIMD_WIDTH); vpx_memset(lfi->lfthr[lvl].hev_thr, (lvl >> 4), SIMD_WIDTH);
} }
void vp9_loop_filter_frame_init(VP9_COMMON *cm, int default_filt_lvl) { void vp9_loop_filter_frame_init(VP9_COMMON *cm, int default_filt_lvl) {
@ -330,16 +324,14 @@ void vp9_loop_filter_frame_init(VP9_COMMON *cm, int default_filt_lvl) {
static int build_lfi(const loop_filter_info_n *lfi_n, static int build_lfi(const loop_filter_info_n *lfi_n,
const MB_MODE_INFO *mbmi, const MB_MODE_INFO *mbmi,
struct loop_filter_info *lfi) { const loop_filter_thresh **lfi) {
const int seg = mbmi->segment_id; const int seg = mbmi->segment_id;
const int ref = mbmi->ref_frame[0]; const int ref = mbmi->ref_frame[0];
const int mode = lfi_n->mode_lf_lut[mbmi->mode]; const int mode = lfi_n->mode_lf_lut[mbmi->mode];
const int filter_level = lfi_n->lvl[seg][ref][mode]; const int filter_level = lfi_n->lvl[seg][ref][mode];
if (filter_level > 0) { if (filter_level > 0) {
lfi->mblim = lfi_n->mblim[filter_level]; *lfi = &lfi_n->lfthr[filter_level];
lfi->lim = lfi_n->lim[filter_level];
lfi->hev_thr = lfi_n->hev_thr[filter_level >> 4];
return 1; return 1;
} else { } else {
return 0; return 0;
@ -351,11 +343,13 @@ static void filter_selectively_vert(uint8_t *s, int pitch,
unsigned int mask_8x8, unsigned int mask_8x8,
unsigned int mask_4x4, unsigned int mask_4x4,
unsigned int mask_4x4_int, unsigned int mask_4x4_int,
const struct loop_filter_info *lfi) { const loop_filter_thresh **p_lfi) {
unsigned int mask; unsigned int mask;
for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int;
mask; mask >>= 1) { mask; mask >>= 1) {
const loop_filter_thresh *lfi = *p_lfi;
if (mask & 1) { if (mask & 1) {
if (mask_16x16 & 1) { if (mask_16x16 & 1) {
vp9_mb_lpf_vertical_edge_w(s, pitch, lfi->mblim, lfi->lim, vp9_mb_lpf_vertical_edge_w(s, pitch, lfi->mblim, lfi->lim,
@ -379,7 +373,7 @@ static void filter_selectively_vert(uint8_t *s, int pitch,
vp9_loop_filter_vertical_edge(s + 4, pitch, lfi->mblim, lfi->lim, vp9_loop_filter_vertical_edge(s + 4, pitch, lfi->mblim, lfi->lim,
lfi->hev_thr, 1); lfi->hev_thr, 1);
s += 8; s += 8;
lfi++; p_lfi++;
mask_16x16 >>= 1; mask_16x16 >>= 1;
mask_8x8 >>= 1; mask_8x8 >>= 1;
mask_4x4 >>= 1; mask_4x4 >>= 1;
@ -393,12 +387,14 @@ static void filter_selectively_horiz(uint8_t *s, int pitch,
unsigned int mask_4x4, unsigned int mask_4x4,
unsigned int mask_4x4_int, unsigned int mask_4x4_int,
int only_4x4_1, int only_4x4_1,
const struct loop_filter_info *lfi) { const loop_filter_thresh **p_lfi) {
unsigned int mask; unsigned int mask;
int count; int count;
for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int; for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int;
mask; mask >>= count) { mask; mask >>= count) {
const loop_filter_thresh *lfi = *p_lfi;
count = 1; count = 1;
if (mask & 1) { if (mask & 1) {
if (!only_4x4_1) { if (!only_4x4_1) {
@ -432,7 +428,7 @@ static void filter_selectively_horiz(uint8_t *s, int pitch,
lfi->lim, lfi->hev_thr, 1); lfi->lim, lfi->hev_thr, 1);
} }
s += 8 * count; s += 8 * count;
lfi += count; p_lfi += count;
mask_16x16 >>= count; mask_16x16 >>= count;
mask_8x8 >>= count; mask_8x8 >>= count;
mask_4x4 >>= count; mask_4x4 >>= count;
@ -805,7 +801,7 @@ static void filter_block_plane_non420(VP9_COMMON *cm,
unsigned int mask_8x8[MI_BLOCK_SIZE] = {0}; unsigned int mask_8x8[MI_BLOCK_SIZE] = {0};
unsigned int mask_4x4[MI_BLOCK_SIZE] = {0}; unsigned int mask_4x4[MI_BLOCK_SIZE] = {0};
unsigned int mask_4x4_int[MI_BLOCK_SIZE] = {0}; unsigned int mask_4x4_int[MI_BLOCK_SIZE] = {0};
struct loop_filter_info lfi[MI_BLOCK_SIZE][MI_BLOCK_SIZE]; const loop_filter_thresh *lfi[MI_BLOCK_SIZE][MI_BLOCK_SIZE];
int r, c; int r, c;
for (r = 0; r < MI_BLOCK_SIZE && mi_row + r < cm->mi_rows; r += row_step) { for (r = 0; r < MI_BLOCK_SIZE && mi_row + r < cm->mi_rows; r += row_step) {
@ -834,7 +830,7 @@ static void filter_block_plane_non420(VP9_COMMON *cm,
const int skip_border_4x4_r = ss_y && mi_row + r == cm->mi_rows - 1; const int skip_border_4x4_r = ss_y && mi_row + r == cm->mi_rows - 1;
// Filter level can vary per MI // Filter level can vary per MI
if (!build_lfi(&cm->lf_info, &mi[0].mbmi, lfi[r] + (c >> ss_x))) if (!build_lfi(&cm->lf_info, &mi[0].mbmi, &lfi[r][c >> ss_x]))
continue; continue;
// Build masks based on the transform size of each block // Build masks based on the transform size of each block
@ -925,7 +921,7 @@ static void filter_block_plane(VP9_COMMON *const cm,
struct buf_2d *const dst = &plane->dst; struct buf_2d *const dst = &plane->dst;
uint8_t* const dst0 = dst->buf; uint8_t* const dst0 = dst->buf;
unsigned int mask_4x4_int[MI_BLOCK_SIZE] = {0}; unsigned int mask_4x4_int[MI_BLOCK_SIZE] = {0};
struct loop_filter_info lfi[MI_BLOCK_SIZE][MI_BLOCK_SIZE]; const loop_filter_thresh *lfi[MI_BLOCK_SIZE][MI_BLOCK_SIZE];
int r, c; int r, c;
int row_shift = 3 - ss_x; int row_shift = 3 - ss_x;
int row_mask = 0xff >> (ss_x << 2); int row_mask = 0xff >> (ss_x << 2);
@ -938,8 +934,8 @@ static void filter_block_plane(VP9_COMMON *const cm,
// Determine the vertical edges that need filtering // Determine the vertical edges that need filtering
for (c = 0; c < MI_BLOCK_SIZE && mi_col + c < cm->mi_cols; c += col_step) { for (c = 0; c < MI_BLOCK_SIZE && mi_col + c < cm->mi_cols; c += col_step) {
const MODE_INFO *mi = mi_8x8[c]; const MODE_INFO *mi = mi_8x8[c];
if (!build_lfi(&cm->lf_info, &mi[0].mbmi, lfi[r] + (c >> ss_x)))
continue; build_lfi(&cm->lf_info, &mi[0].mbmi, &lfi[r][c >> ss_x]);
} }
if (!plane->plane_type) { if (!plane->plane_type) {
mask_4x4_int[r] = MASK_ROW(lfm->int_4x4_y); mask_4x4_int[r] = MASK_ROW(lfm->int_4x4_y);

13
vp9/common/vp9_loopfilter.h
View File

@ -46,12 +46,13 @@ struct loopfilter {
// Need to align this structure so when it is declared and // Need to align this structure so when it is declared and
// passed it can be loaded into vector registers. // passed it can be loaded into vector registers.
typedef struct { typedef struct {
DECLARE_ALIGNED(SIMD_WIDTH, uint8_t, DECLARE_ALIGNED(SIMD_WIDTH, uint8_t, mblim[SIMD_WIDTH]);
mblim[MAX_LOOP_FILTER + 1][SIMD_WIDTH]); DECLARE_ALIGNED(SIMD_WIDTH, uint8_t, lim[SIMD_WIDTH]);
DECLARE_ALIGNED(SIMD_WIDTH, uint8_t, DECLARE_ALIGNED(SIMD_WIDTH, uint8_t, hev_thr[SIMD_WIDTH]);
lim[MAX_LOOP_FILTER + 1][SIMD_WIDTH]); } loop_filter_thresh;
DECLARE_ALIGNED(SIMD_WIDTH, uint8_t,
hev_thr[4][SIMD_WIDTH]); typedef struct {
loop_filter_thresh lfthr[MAX_LOOP_FILTER + 1];
uint8_t lvl[MAX_SEGMENTS][MAX_REF_FRAMES][MAX_MODE_LF_DELTAS]; uint8_t lvl[MAX_SEGMENTS][MAX_REF_FRAMES][MAX_MODE_LF_DELTAS];
uint8_t mode_lf_lut[MB_MODE_COUNT]; uint8_t mode_lf_lut[MB_MODE_COUNT];
} loop_filter_info_n; } loop_filter_info_n;