CDEF cleanup
Name changes and code moves to bring code more in line with the
design doc and an upcoming single-pass patch. No functional changes.
Change-Id: I2bccd58c644e534b139f420b623390aa971fbdb0
diff --git a/av1/encoder/bitstream.c b/av1/encoder/bitstream.c
index 9e20819..a7e1dfc 100644
--- a/av1/encoder/bitstream.c
+++ b/av1/encoder/bitstream.c
@@ -3474,8 +3474,8 @@
#if CONFIG_CDEF
static void encode_cdef(const AV1_COMMON *cm, struct aom_write_bit_buffer *wb) {
int i;
- aom_wb_write_literal(wb, cm->cdef_dering_damping - 5, 1);
- aom_wb_write_literal(wb, cm->cdef_clpf_damping - 3, 2);
+ aom_wb_write_literal(wb, cm->cdef_pri_damping - 5, 1);
+ aom_wb_write_literal(wb, cm->cdef_sec_damping - 3, 2);
aom_wb_write_literal(wb, cm->cdef_bits, 2);
for (i = 0; i < cm->nb_cdef_strengths; i++) {
aom_wb_write_literal(wb, cm->cdef_strengths[i], CDEF_STRENGTH_BITS);
diff --git a/av1/encoder/pickcdef.c b/av1/encoder/pickcdef.c
index e4ec388..443e9e5 100644
--- a/av1/encoder/pickcdef.c
+++ b/av1/encoder/pickcdef.c
@@ -19,11 +19,11 @@
#include "av1/common/reconinter.h"
#include "av1/encoder/encoder.h"
-#define REDUCED_STRENGTHS 8
-#define REDUCED_TOTAL_STRENGTHS (REDUCED_STRENGTHS * CLPF_STRENGTHS)
-#define TOTAL_STRENGTHS (DERING_STRENGTHS * CLPF_STRENGTHS)
+#define REDUCED_PRI_STRENGTHS 8
+#define REDUCED_TOTAL_STRENGTHS (REDUCED_PRI_STRENGTHS * CDEF_SEC_STRENGTHS)
+#define TOTAL_STRENGTHS (CDEF_PRI_STRENGTHS * CDEF_SEC_STRENGTHS)
-static int priconv[REDUCED_STRENGTHS] = { 0, 1, 2, 3, 4, 7, 12, 25 };
+static int priconv[REDUCED_PRI_STRENGTHS] = { 0, 1, 2, 3, 4, 7, 12, 25 };
/* Search for the best strength to add as an option, knowing we
already selected nb_strengths options. */
@@ -232,13 +232,13 @@
}
/* Compute MSE only on the blocks we filtered. */
-uint64_t compute_dering_dist(uint16_t *dst, int dstride, uint16_t *src,
- dering_list *dlist, int dering_count,
- BLOCK_SIZE bsize, int coeff_shift, int pli) {
+uint64_t compute_cdef_dist(uint16_t *dst, int dstride, uint16_t *src,
+ cdef_list *dlist, int cdef_count, BLOCK_SIZE bsize,
+ int coeff_shift, int pli) {
uint64_t sum = 0;
int bi, bx, by;
if (bsize == BLOCK_8X8) {
- for (bi = 0; bi < dering_count; bi++) {
+ for (bi = 0; bi < cdef_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
if (pli == 0) {
@@ -250,7 +250,7 @@
}
}
} else if (bsize == BLOCK_4X8) {
- for (bi = 0; bi < dering_count; bi++) {
+ for (bi = 0; bi < cdef_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
sum += mse_4x4_16bit(&dst[(by << 3) * dstride + (bx << 2)], dstride,
@@ -259,7 +259,7 @@
&src[(bi << (3 + 2)) + 4 * 4], 4);
}
} else if (bsize == BLOCK_8X4) {
- for (bi = 0; bi < dering_count; bi++) {
+ for (bi = 0; bi < cdef_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
sum += mse_4x4_16bit(&dst[(by << 2) * dstride + (bx << 3)], dstride,
@@ -269,7 +269,7 @@
}
} else {
assert(bsize == BLOCK_4X4);
- for (bi = 0; bi < dering_count; bi++) {
+ for (bi = 0; bi < cdef_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
sum += mse_4x4_16bit(&dst[(by << 2) * dstride + (bx << 2)], dstride,
@@ -282,12 +282,12 @@
void av1_cdef_search(YV12_BUFFER_CONFIG *frame, const YV12_BUFFER_CONFIG *ref,
AV1_COMMON *cm, MACROBLOCKD *xd, int fast) {
int r, c;
- int sbr, sbc;
+ int fbr, fbc;
uint16_t *src[3];
uint16_t *ref_coeff[3];
- dering_list dlist[MI_SIZE_64X64 * MI_SIZE_64X64];
- int dir[OD_DERING_NBLOCKS][OD_DERING_NBLOCKS] = { { 0 } };
- int var[OD_DERING_NBLOCKS][OD_DERING_NBLOCKS] = { { 0 } };
+ cdef_list dlist[MI_SIZE_64X64 * MI_SIZE_64X64];
+ int dir[CDEF_NBLOCKS][CDEF_NBLOCKS] = { { 0 } };
+ int var[CDEF_NBLOCKS][CDEF_NBLOCKS] = { { 0 } };
int stride[3];
int bsize[3];
int mi_wide_l2[3];
@@ -295,18 +295,18 @@
int xdec[3];
int ydec[3];
int pli;
- int dering_count;
+ int cdef_count;
int coeff_shift = AOMMAX(cm->bit_depth - 8, 0);
uint64_t best_tot_mse = (uint64_t)1 << 63;
uint64_t tot_mse;
int sb_count;
- int nvsb = (cm->mi_rows + MI_SIZE_64X64 - 1) / MI_SIZE_64X64;
- int nhsb = (cm->mi_cols + MI_SIZE_64X64 - 1) / MI_SIZE_64X64;
- int *sb_index = aom_malloc(nvsb * nhsb * sizeof(*sb_index));
- int *selected_strength = aom_malloc(nvsb * nhsb * sizeof(*sb_index));
+ int nvfb = (cm->mi_rows + MI_SIZE_64X64 - 1) / MI_SIZE_64X64;
+ int nhfb = (cm->mi_cols + MI_SIZE_64X64 - 1) / MI_SIZE_64X64;
+ int *sb_index = aom_malloc(nvfb * nhfb * sizeof(*sb_index));
+ int *selected_strength = aom_malloc(nvfb * nhfb * sizeof(*sb_index));
uint64_t(*mse[2])[TOTAL_STRENGTHS];
- int clpf_damping = 3 + (cm->base_qindex >> 6);
- int dering_damping = 6;
+ int pri_damping = 6;
+ int sec_damping = 3 + (cm->base_qindex >> 6);
int i;
int nb_strengths;
int nb_strength_bits;
@@ -314,19 +314,18 @@
double lambda;
int nplanes = 3;
const int total_strengths = fast ? REDUCED_TOTAL_STRENGTHS : TOTAL_STRENGTHS;
- DECLARE_ALIGNED(32, uint16_t, inbuf[OD_DERING_INBUF_SIZE]);
+ DECLARE_ALIGNED(32, uint16_t, inbuf[CDEF_INBUF_SIZE]);
uint16_t *in;
- DECLARE_ALIGNED(32, uint16_t, tmp_dst[MAX_SB_SQUARE]);
- int chroma_dering =
- xd->plane[1].subsampling_x == xd->plane[1].subsampling_y &&
- xd->plane[2].subsampling_x == xd->plane[2].subsampling_y;
+ DECLARE_ALIGNED(32, uint16_t, tmp_dst[CDEF_BLOCKSIZE * CDEF_BLOCKSIZE]);
+ int chroma_cdef = xd->plane[1].subsampling_x == xd->plane[1].subsampling_y &&
+ xd->plane[2].subsampling_x == xd->plane[2].subsampling_y;
quantizer =
av1_ac_quant(cm->base_qindex, 0, cm->bit_depth) >> (cm->bit_depth - 8);
lambda = .12 * quantizer * quantizer / 256.;
av1_setup_dst_planes(xd->plane, cm->sb_size, frame, 0, 0);
- mse[0] = aom_malloc(sizeof(**mse) * nvsb * nhsb);
- mse[1] = aom_malloc(sizeof(**mse) * nvsb * nhsb);
+ mse[0] = aom_malloc(sizeof(**mse) * nvfb * nhfb);
+ mse[1] = aom_malloc(sizeof(**mse) * nvfb * nhfb);
for (pli = 0; pli < nplanes; pli++) {
uint8_t *ref_buffer;
int ref_stride;
@@ -380,65 +379,64 @@
}
}
}
- in = inbuf + OD_FILT_VBORDER * OD_FILT_BSTRIDE + OD_FILT_HBORDER;
+ in = inbuf + CDEF_VBORDER * CDEF_BSTRIDE + CDEF_HBORDER;
sb_count = 0;
- for (sbr = 0; sbr < nvsb; ++sbr) {
- for (sbc = 0; sbc < nhsb; ++sbc) {
+ for (fbr = 0; fbr < nvfb; ++fbr) {
+ for (fbc = 0; fbc < nhfb; ++fbc) {
int nvb, nhb;
int gi;
int dirinit = 0;
- nhb = AOMMIN(MI_SIZE_64X64, cm->mi_cols - MI_SIZE_64X64 * sbc);
- nvb = AOMMIN(MI_SIZE_64X64, cm->mi_rows - MI_SIZE_64X64 * sbr);
- cm->mi_grid_visible[MI_SIZE_64X64 * sbr * cm->mi_stride +
- MI_SIZE_64X64 * sbc]
+ nhb = AOMMIN(MI_SIZE_64X64, cm->mi_cols - MI_SIZE_64X64 * fbc);
+ nvb = AOMMIN(MI_SIZE_64X64, cm->mi_rows - MI_SIZE_64X64 * fbr);
+ cm->mi_grid_visible[MI_SIZE_64X64 * fbr * cm->mi_stride +
+ MI_SIZE_64X64 * fbc]
->mbmi.cdef_strength = -1;
- if (sb_all_skip(cm, sbr * MI_SIZE_64X64, sbc * MI_SIZE_64X64)) continue;
- dering_count = sb_compute_dering_list(cm, sbr * MI_SIZE_64X64,
- sbc * MI_SIZE_64X64, dlist, 1);
+ if (sb_all_skip(cm, fbr * MI_SIZE_64X64, fbc * MI_SIZE_64X64)) continue;
+ cdef_count = sb_compute_cdef_list(cm, fbr * MI_SIZE_64X64,
+ fbc * MI_SIZE_64X64, dlist, 1);
for (pli = 0; pli < nplanes; pli++) {
- for (i = 0; i < OD_DERING_INBUF_SIZE; i++)
- inbuf[i] = OD_DERING_VERY_LARGE;
+ for (i = 0; i < CDEF_INBUF_SIZE; i++) inbuf[i] = CDEF_VERY_LARGE;
for (gi = 0; gi < total_strengths; gi++) {
int threshold;
uint64_t curr_mse;
- int clpf_strength;
- threshold = gi / CLPF_STRENGTHS;
+ int sec_strength;
+ threshold = gi / CDEF_SEC_STRENGTHS;
if (fast) threshold = priconv[threshold];
- if (pli > 0 && !chroma_dering) threshold = 0;
+ if (pli > 0 && !chroma_cdef) threshold = 0;
/* We avoid filtering the pixels for which some of the pixels to
average
are outside the frame. We could change the filter instead, but it
would add special cases for any future vectorization. */
- int yoff = OD_FILT_VBORDER * (sbr != 0);
- int xoff = OD_FILT_HBORDER * (sbc != 0);
+ int yoff = CDEF_VBORDER * (fbr != 0);
+ int xoff = CDEF_HBORDER * (fbc != 0);
int ysize = (nvb << mi_high_l2[pli]) +
- OD_FILT_VBORDER * (sbr != nvsb - 1) + yoff;
+ CDEF_VBORDER * (fbr != nvfb - 1) + yoff;
int xsize = (nhb << mi_wide_l2[pli]) +
- OD_FILT_HBORDER * (sbc != nhsb - 1) + xoff;
- clpf_strength = gi % CLPF_STRENGTHS;
- if (clpf_strength == 0)
- copy_sb16_16(&in[(-yoff * OD_FILT_BSTRIDE - xoff)], OD_FILT_BSTRIDE,
+ CDEF_HBORDER * (fbc != nhfb - 1) + xoff;
+ sec_strength = gi % CDEF_SEC_STRENGTHS;
+ if (sec_strength == 0)
+ copy_sb16_16(&in[(-yoff * CDEF_BSTRIDE - xoff)], CDEF_BSTRIDE,
src[pli],
- (sbr * MI_SIZE_64X64 << mi_high_l2[pli]) - yoff,
- (sbc * MI_SIZE_64X64 << mi_wide_l2[pli]) - xoff,
+ (fbr * MI_SIZE_64X64 << mi_high_l2[pli]) - yoff,
+ (fbc * MI_SIZE_64X64 << mi_wide_l2[pli]) - xoff,
stride[pli], ysize, xsize);
- od_dering(clpf_strength ? NULL : (uint8_t *)in, OD_FILT_BSTRIDE,
- tmp_dst, in, xdec[pli], ydec[pli], dir, &dirinit, var, pli,
- dlist, dering_count, threshold,
- clpf_strength + (clpf_strength == 3), clpf_damping,
- dering_damping, coeff_shift, clpf_strength != 0, 1);
- curr_mse = compute_dering_dist(
+ cdef_filter_fb(sec_strength ? NULL : (uint8_t *)in, CDEF_BSTRIDE,
+ tmp_dst, in, xdec[pli], ydec[pli], dir, &dirinit, var,
+ pli, dlist, cdef_count, threshold,
+ sec_strength + (sec_strength == 3), sec_damping,
+ pri_damping, coeff_shift, sec_strength != 0, 1);
+ curr_mse = compute_cdef_dist(
ref_coeff[pli] +
- (sbr * MI_SIZE_64X64 << mi_high_l2[pli]) * stride[pli] +
- (sbc * MI_SIZE_64X64 << mi_wide_l2[pli]),
- stride[pli], tmp_dst, dlist, dering_count, bsize[pli],
- coeff_shift, pli);
+ (fbr * MI_SIZE_64X64 << mi_high_l2[pli]) * stride[pli] +
+ (fbc * MI_SIZE_64X64 << mi_wide_l2[pli]),
+ stride[pli], tmp_dst, dlist, cdef_count, bsize[pli], coeff_shift,
+ pli);
if (pli < 2)
mse[pli][sb_count][gi] = curr_mse;
else
mse[1][sb_count][gi] += curr_mse;
sb_index[sb_count] =
- MI_SIZE_64X64 * sbr * cm->mi_stride + MI_SIZE_64X64 * sbc;
+ MI_SIZE_64X64 * fbr * cm->mi_stride + MI_SIZE_64X64 * fbc;
}
}
sb_count++;
@@ -494,15 +492,17 @@
if (fast) {
for (int j = 0; j < nb_strengths; j++) {
cm->cdef_strengths[j] =
- priconv[cm->cdef_strengths[j] / CLPF_STRENGTHS] * CLPF_STRENGTHS +
- (cm->cdef_strengths[j] % CLPF_STRENGTHS);
+ priconv[cm->cdef_strengths[j] / CDEF_SEC_STRENGTHS] *
+ CDEF_SEC_STRENGTHS +
+ (cm->cdef_strengths[j] % CDEF_SEC_STRENGTHS);
cm->cdef_uv_strengths[j] =
- priconv[cm->cdef_uv_strengths[j] / CLPF_STRENGTHS] * CLPF_STRENGTHS +
- (cm->cdef_uv_strengths[j] % CLPF_STRENGTHS);
+ priconv[cm->cdef_uv_strengths[j] / CDEF_SEC_STRENGTHS] *
+ CDEF_SEC_STRENGTHS +
+ (cm->cdef_uv_strengths[j] % CDEF_SEC_STRENGTHS);
}
}
- cm->cdef_dering_damping = dering_damping;
- cm->cdef_clpf_damping = clpf_damping;
+ cm->cdef_pri_damping = pri_damping;
+ cm->cdef_sec_damping = sec_damping;
aom_free(mse[0]);
aom_free(mse[1]);
for (pli = 0; pli < nplanes; pli++) {
