ref: 49dee1687934fd218aedd4db261337cdb7940691
parent: 2291563ba84b312ebefe26134dc27236d5ef81fb
parent: dc0f457c942f8f52e4b1cd329bf2f9febd2e6dcc
author: Dmitry Kovalev <dkovalev@google.com>
date: Wed Jun 26 06:23:27 EDT 2013
Merge "Using get_plane_block_{width, height} instead of custom code."
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -494,23 +494,26 @@
static int64_t block_error_sby(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize,
int shift) {- const int bwl = b_width_log2(bsize), bhl = b_height_log2(bsize);
+ struct macroblockd_plane *p = &x->e_mbd.plane[0];
+ const int bw = plane_block_width(bsize, p);
+ const int bh = plane_block_height(bsize, p);
return vp9_block_error(x->plane[0].coeff, x->e_mbd.plane[0].dqcoeff,
- 16 << (bwl + bhl)) >> shift;
+ bw * bh) >> shift;
}
static int64_t block_error_sbuv(MACROBLOCK *x, BLOCK_SIZE_TYPE bsize,
int shift) {- const int bwl = b_width_log2(bsize), bhl = b_height_log2(bsize);
int64_t sum = 0;
int plane;
for (plane = 1; plane < MAX_MB_PLANE; plane++) {- const int subsampling = x->e_mbd.plane[plane].subsampling_x +
- x->e_mbd.plane[plane].subsampling_y;
+ struct macroblockd_plane *p = &x->e_mbd.plane[plane];
+ const int bw = plane_block_width(bsize, p);
+ const int bh = plane_block_height(bsize, p);
sum += vp9_block_error(x->plane[plane].coeff, x->e_mbd.plane[plane].dqcoeff,
- 16 << (bwl + bhl - subsampling));
+ bw * bh);
}
+
return sum >> shift;
}
@@ -1092,25 +1095,22 @@
int k;
MACROBLOCKD *xd = &x->e_mbd;
BLOCK_SIZE_TYPE bsize = xd->mode_info_context->mbmi.sb_type;
- int bwl = b_width_log2(bsize), bw = 1 << bwl;
- int bhl = b_height_log2(bsize), bh = 1 << bhl;
+ const int bw = plane_block_width(bsize, &xd->plane[0]);
+ const int bh = plane_block_height(bsize, &xd->plane[0]);
int idx, idy;
const int src_stride = x->plane[0].src.stride;
- uint8_t* const src =
- raster_block_offset_uint8(xd, BLOCK_SIZE_SB8X8, 0, i,
- x->plane[0].src.buf, src_stride);
- int16_t* src_diff =
- raster_block_offset_int16(xd, BLOCK_SIZE_SB8X8, 0, i,
- x->plane[0].src_diff);
+ uint8_t* const src = raster_block_offset_uint8(xd, BLOCK_SIZE_SB8X8, 0, i,
+ x->plane[0].src.buf,
+ src_stride);
+ int16_t* src_diff = raster_block_offset_int16(xd, BLOCK_SIZE_SB8X8, 0, i,
+ x->plane[0].src_diff);
int16_t* coeff = BLOCK_OFFSET(x->plane[0].coeff, 16, i);
- uint8_t* const pre =
- raster_block_offset_uint8(xd, BLOCK_SIZE_SB8X8, 0, i,
- xd->plane[0].pre[0].buf,
- xd->plane[0].pre[0].stride);
- uint8_t* const dst =
- raster_block_offset_uint8(xd, BLOCK_SIZE_SB8X8, 0, i,
- xd->plane[0].dst.buf,
- xd->plane[0].dst.stride);
+ uint8_t* const pre = raster_block_offset_uint8(xd, BLOCK_SIZE_SB8X8, 0, i,
+ xd->plane[0].pre[0].buf,
+ xd->plane[0].pre[0].stride);
+ uint8_t* const dst = raster_block_offset_uint8(xd, BLOCK_SIZE_SB8X8, 0, i,
+ xd->plane[0].dst.buf,
+ xd->plane[0].dst.stride);
int64_t thisdistortion = 0;
int thisrate = 0;
@@ -1123,7 +1123,7 @@
xd->plane[0].dst.stride,
&xd->mode_info_context->bmi[i].as_mv[0],
&xd->scale_factor[0],
- 4 * bw, 4 * bh, 0 /* no avg */, &xd->subpix,
+ bw, bh, 0 /* no avg */, &xd->subpix,
MV_PRECISION_Q3);
// TODO(debargha): Make this work properly with the
@@ -1137,17 +1137,17 @@
vp9_build_inter_predictor(second_pre, xd->plane[0].pre[1].stride,
dst, xd->plane[0].dst.stride,
&xd->mode_info_context->bmi[i].as_mv[1],
- &xd->scale_factor[1], 4 * bw, 4 * bh, 1,
+ &xd->scale_factor[1], bw, bh, 1,
&xd->subpix, MV_PRECISION_Q3);
}
- vp9_subtract_block(4 * bh, 4 * bw, src_diff, 8,
+ vp9_subtract_block(bh, bw, src_diff, 8,
src, src_stride,
dst, xd->plane[0].dst.stride);
k = i;
- for (idy = 0; idy < bh; ++idy) {- for (idx = 0; idx < bw; ++idx) {+ for (idy = 0; idy < bh / 4; ++idy) {+ for (idx = 0; idx < bw / 4; ++idx) {k += (idy * 2 + idx);
src_diff = raster_block_offset_int16(xd, BLOCK_SIZE_SB8X8, 0, k,
x->plane[0].src_diff);
@@ -2233,13 +2233,8 @@
int_mv *frame_mv,
int mi_row, int mi_col,
int_mv single_newmv[MAX_REF_FRAMES]) {- const int bw = 1 << mi_width_log2(bsize), bh = 1 << mi_height_log2(bsize);
-
VP9_COMMON *cm = &cpi->common;
MACROBLOCKD *xd = &x->e_mbd;
- const enum BlockSize block_size = get_plane_block_size(bsize, &xd->plane[0]);
- const enum BlockSize uv_block_size = get_plane_block_size(bsize,
- &xd->plane[1]);
MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
const int is_comp_pred = (mbmi->ref_frame[1] > 0);
const int num_refs = is_comp_pred ? 2 : 1;
@@ -2370,13 +2365,14 @@
int p;
for (p = 0; p < MAX_MB_PLANE; p++) {- const int y = (MI_SIZE * bh) >> xd->plane[p].subsampling_y;
- const int x = (MI_SIZE * bw) >> xd->plane[p].subsampling_x;
+ struct macroblockd_plane *pd = &xd->plane[p];
+ const int bw = plane_block_width(bsize, pd);
+ const int bh = plane_block_height(bsize, pd);
int i;
- for (i = 0; i < y; i++)
- vpx_memcpy(&tmp_buf[p][64 * i],
- xd->plane[p].dst.buf + i * xd->plane[p].dst.stride, x);
+ for (i = 0; i < bh; i++)
+ vpx_memcpy(&tmp_buf[p][64 * i], pd->dst.buf + i * pd->dst.stride,
+ bw);
}
pred_exists = 1;
}
@@ -2394,13 +2390,13 @@
int p;
for (p = 0; p < MAX_MB_PLANE; p++) {- const int y = (MI_SIZE * bh) >> xd->plane[p].subsampling_y;
- const int x = (MI_SIZE * bw) >> xd->plane[p].subsampling_x;
+ struct macroblockd_plane *pd = &xd->plane[p];
+ const int bw = plane_block_width(bsize, pd);
+ const int bh = plane_block_height(bsize, pd);
int i;
- for (i = 0; i < y; i++)
- vpx_memcpy(xd->plane[p].dst.buf + i * xd->plane[p].dst.stride,
- &tmp_buf[p][64 * i], x);
+ for (i = 0; i < bh; i++)
+ vpx_memcpy(pd->dst.buf + i * pd->dst.stride, &tmp_buf[p][64 * i], bw);
}
} else {// Handles the special case when a filter that is not in the
@@ -2414,36 +2410,37 @@
if (cpi->active_map_enabled && x->active_ptr[0] == 0)
x->skip = 1;
else if (x->encode_breakout) {+ const enum BlockSize y_size = get_plane_block_size(bsize, &xd->plane[0]);
+ const enum BlockSize uv_size = get_plane_block_size(bsize, &xd->plane[1]);
+
unsigned int var, sse;
- int threshold = (xd->plane[0].dequant[1]
- * xd->plane[0].dequant[1] >> 4);
+ int threshold = (xd->plane[0].dequant[1] * xd->plane[0].dequant[1] >> 4);
+
if (threshold < x->encode_breakout)
threshold = x->encode_breakout;
- var = cpi->fn_ptr[block_size].vf(x->plane[0].src.buf,
- x->plane[0].src.stride,
- xd->plane[0].dst.buf,
- xd->plane[0].dst.stride,
- &sse);
+ var = cpi->fn_ptr[y_size].vf(x->plane[0].src.buf, x->plane[0].src.stride,
+ xd->plane[0].dst.buf, xd->plane[0].dst.stride,
+ &sse);
if ((int)sse < threshold) {unsigned int q2dc = xd->plane[0].dequant[0];
- /* If there is no codeable 2nd order dc
- or a very small uniform pixel change change */
+ // If there is no codeable 2nd order dc
+ // or a very small uniform pixel change change
if ((sse - var < q2dc * q2dc >> 4) ||
(sse / 2 > var && sse - var < 64)) {// Check u and v to make sure skip is ok
int sse2;
unsigned int sse2u, sse2v;
- var = cpi->fn_ptr[uv_block_size].vf(x->plane[1].src.buf,
- x->plane[1].src.stride,
- xd->plane[1].dst.buf,
- xd->plane[1].dst.stride, &sse2u);
- var = cpi->fn_ptr[uv_block_size].vf(x->plane[2].src.buf,
- x->plane[1].src.stride,
- xd->plane[2].dst.buf,
- xd->plane[1].dst.stride, &sse2v);
+ var = cpi->fn_ptr[uv_size].vf(x->plane[1].src.buf,
+ x->plane[1].src.stride,
+ xd->plane[1].dst.buf,
+ xd->plane[1].dst.stride, &sse2u);
+ var = cpi->fn_ptr[uv_size].vf(x->plane[2].src.buf,
+ x->plane[2].src.stride,
+ xd->plane[2].dst.buf,
+ xd->plane[2].dst.stride, &sse2v);
sse2 = sse2u + sse2v;
if (sse2 * 2 < threshold) {@@ -2451,7 +2448,7 @@
*distortion = sse + sse2;
*rate2 = 500;
- /* for best_yrd calculation */
+ // for best_yrd calculation
*rate_uv = 0;
*distortion_uv = sse2;