Use reconstructed boundary pixel in dcq mode
Change-Id: I50dcce53952d10f1e0f7107e6a86155ab2ec685a
diff --git a/av1/encoder/allintra_vis.c b/av1/encoder/allintra_vis.c
index 01cdf17..2bb2899 100644
--- a/av1/encoder/allintra_vis.c
+++ b/av1/encoder/allintra_vis.c
@@ -9,8 +9,10 @@
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
+#include "av1/common/enums.h"
#include "av1/common/idct.h"
+#include "av1/common/reconinter.h"
#include "av1/encoder/allintra_vis.h"
#include "av1/encoder/hybrid_fwd_txfm.h"
#include "av1/encoder/rdopt_utils.h"
@@ -181,18 +183,20 @@
memset(&mbmi, 0, sizeof(mbmi));
MB_MODE_INFO *mbmi_ptr = &mbmi;
xd->mi = &mbmi_ptr;
+ xd->cur_buf = cpi->source;
+
+ const SequenceHeader *const seq_params = cm->seq_params;
+ if (aom_realloc_frame_buffer(
+ &cm->cur_frame->buf, cm->width, cm->height, seq_params->subsampling_x,
+ seq_params->subsampling_y, seq_params->use_highbitdepth,
+ cpi->oxcf.border_in_pixels, cm->features.byte_alignment, NULL, NULL,
+ NULL, cpi->oxcf.tool_cfg.enable_global_motion))
+ aom_internal_error(cm->error, AOM_CODEC_MEM_ERROR,
+ "Failed to allocate frame buffer");
cm->quant_params.base_qindex = cpi->oxcf.rc_cfg.cq_level;
av1_frame_init_quantizer(cpi);
- union {
-#if CONFIG_AV1_HIGHBITDEPTH
- DECLARE_ALIGNED(32, uint16_t, zero_pred16[32 * 32]);
-#endif
- DECLARE_ALIGNED(32, uint8_t, zero_pred8[32 * 32]);
- } pred_buffer_mem;
- uint8_t *pred_buf;
-
DECLARE_ALIGNED(32, int16_t, src_diff[32 * 32]);
DECLARE_ALIGNED(32, tran_low_t, coeff[32 * 32]);
DECLARE_ALIGNED(32, tran_low_t, qcoeff[32 * 32]);
@@ -203,27 +207,11 @@
const int block_size = tx_size_wide[tx_size];
const int coeff_count = block_size * block_size;
-#if CONFIG_AV1_HIGHBITDEPTH
- xd->cur_buf = cpi->source;
- if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
- pred_buf = CONVERT_TO_BYTEPTR(pred_buffer_mem.zero_pred16);
- memset(pred_buffer_mem.zero_pred16, 0,
- sizeof(*pred_buffer_mem.zero_pred16) * coeff_count);
- } else {
- pred_buf = pred_buffer_mem.zero_pred8;
- memset(pred_buffer_mem.zero_pred8, 0,
- sizeof(*pred_buffer_mem.zero_pred8) * coeff_count);
- }
-#else
- pred_buf = pred_buffer_mem.zero_pred8;
- memset(pred_buffer_mem.zero_pred8, 0,
- sizeof(*pred_buffer_mem.zero_pred8) * coeff_count);
-#endif
-
const BitDepthInfo bd_info = get_bit_depth_info(xd);
cpi->norm_wiener_variance = 0;
int mb_step = mi_size_wide[BLOCK_16X16];
+ BLOCK_SIZE bsize = BLOCK_16X16;
for (mb_row = 0; mb_row < cpi->frame_info.mb_rows; ++mb_row) {
for (mb_col = 0; mb_col < cpi->frame_info.mb_cols; ++mb_col) {
@@ -236,20 +224,36 @@
xd->up_available = mi_row > 0;
xd->left_available = mi_col > 0;
- int dst_mb_offset = mi_row * MI_SIZE * buf_stride + mi_col * MI_SIZE;
- uint8_t *dst_buffer = xd->cur_buf->y_buffer + dst_mb_offset;
+ const int mi_width = mi_size_wide[bsize];
+ const int mi_height = mi_size_high[bsize];
+ set_mode_info_offsets(&cpi->common.mi_params, &cpi->mbmi_ext_info, x, xd,
+ mi_row, mi_col);
+ set_mi_row_col(xd, &xd->tile, mi_row, mi_height, mi_col, mi_width,
+ cm->mi_params.mi_rows, cm->mi_params.mi_cols);
+ set_plane_n4(xd, mi_size_wide[bsize], mi_size_high[bsize],
+ av1_num_planes(cm));
+ xd->mi[0]->bsize = bsize;
+ xd->mi[0]->motion_mode = SIMPLE_TRANSLATION;
+
+ av1_setup_dst_planes(xd->plane, bsize, &cm->cur_frame->buf, mi_row,
+ mi_col, 0, av1_num_planes(cm));
+
+ int dst_buffer_stride = xd->plane[0].dst.stride;
+ uint8_t *dst_buffer = xd->plane[0].dst.buf;
+ uint8_t *mb_buffer =
+ buffer + mi_row * MI_SIZE * buf_stride + mi_col * MI_SIZE;
for (PREDICTION_MODE mode = INTRA_MODE_START; mode < INTRA_MODE_END;
++mode) {
- av1_predict_intra_block(xd, cm->seq_params->sb_size,
- cm->seq_params->enable_intra_edge_filter,
- block_size, block_size, tx_size, mode, 0, 0,
- FILTER_INTRA_MODES, dst_buffer, buf_stride,
- pred_buf, block_size, 0, 0, 0);
+ av1_predict_intra_block(
+ xd, cm->seq_params->sb_size,
+ cm->seq_params->enable_intra_edge_filter, block_size, block_size,
+ tx_size, mode, 0, 0, FILTER_INTRA_MODES, dst_buffer,
+ dst_buffer_stride, dst_buffer, dst_buffer_stride, 0, 0, 0);
av1_subtract_block(bd_info, block_size, block_size, src_diff,
- block_size, dst_buffer, buf_stride, pred_buf,
- block_size);
+ block_size, mb_buffer, buf_stride, dst_buffer,
+ dst_buffer_stride);
av1_quick_txfm(0, tx_size, bd_info, src_diff, block_size, coeff);
int intra_cost = aom_satd(coeff, coeff_count);
if (intra_cost < best_intra_cost) {
@@ -260,15 +264,14 @@
int idx;
int16_t median_val = 0;
- uint8_t *mb_buffer =
- buffer + mb_row * block_size * buf_stride + mb_col * block_size;
int64_t wiener_variance = 0;
- av1_predict_intra_block(
- xd, cm->seq_params->sb_size, cm->seq_params->enable_intra_edge_filter,
- block_size, block_size, tx_size, best_mode, 0, 0, FILTER_INTRA_MODES,
- dst_buffer, buf_stride, pred_buf, block_size, 0, 0, 0);
+ av1_predict_intra_block(xd, cm->seq_params->sb_size,
+ cm->seq_params->enable_intra_edge_filter,
+ block_size, block_size, tx_size, best_mode, 0, 0,
+ FILTER_INTRA_MODES, dst_buffer, dst_buffer_stride,
+ dst_buffer, dst_buffer_stride, 0, 0, 0);
av1_subtract_block(bd_info, block_size, block_size, src_diff, block_size,
- mb_buffer, buf_stride, pred_buf, block_size);
+ mb_buffer, buf_stride, dst_buffer, dst_buffer_stride);
av1_quick_txfm(0, tx_size, bd_info, src_diff, block_size, coeff);
const struct macroblock_plane *const p = &x->plane[0];
@@ -289,9 +292,8 @@
av1_quantize_fp_facade(coeff, pix_num, p, qcoeff, dqcoeff, &eob,
scan_order, &quant_param);
#endif // CONFIG_AV1_HIGHBITDEPTH
- av1_inverse_transform_block(xd, dqcoeff, 0, DCT_DCT, tx_size, pred_buf,
- block_size, eob, 0);
-
+ av1_inverse_transform_block(xd, dqcoeff, 0, DCT_DCT, tx_size, dst_buffer,
+ dst_buffer_stride, eob, 0);
WeberStats *weber_stats =
&cpi->mb_weber_stats[mb_row * cpi->frame_info.mb_cols + mb_col];
@@ -310,17 +312,17 @@
int src_pix, rec_pix;
#if CONFIG_AV1_HIGHBITDEPTH
if (is_cur_buf_hbd(xd)) {
- uint16_t *dst = CONVERT_TO_SHORTPTR(dst_buffer);
- uint16_t *rec = CONVERT_TO_SHORTPTR(pred_buf);
- src_pix = dst[pix_row * buf_stride + pix_col];
- rec_pix = rec[pix_row * block_size + pix_col];
+ uint16_t *src = CONVERT_TO_SHORTPTR(mb_buffer);
+ uint16_t *rec = CONVERT_TO_SHORTPTR(dst_buffer);
+ src_pix = src[pix_row * buf_stride + pix_col];
+ rec_pix = rec[pix_row * dst_buffer_stride + pix_col];
} else {
- src_pix = dst_buffer[pix_row * buf_stride + pix_col];
- rec_pix = pred_buf[pix_row * block_size + pix_col];
+ src_pix = mb_buffer[pix_row * buf_stride + pix_col];
+ rec_pix = dst_buffer[pix_row * dst_buffer_stride + pix_col];
}
#else
- src_pix = dst_buffer[pix_row * buf_stride + pix_col];
- rec_pix = pred_buf[pix_row * block_size + pix_col];
+ src_pix = mb_buffer[pix_row * buf_stride + pix_col];
+ rec_pix = dst_buffer[pix_row * dst_buffer_stride + pix_col];
#endif
src_mean += src_pix;
rec_mean += rec_pix;
@@ -395,6 +397,8 @@
if (sb_count > 0)
cpi->norm_wiener_variance = (int64_t)(exp(sb_wiener_log / sb_count));
cpi->norm_wiener_variance = AOMMAX(1, cpi->norm_wiener_variance);
+
+ aom_free_frame_buffer(&cm->cur_frame->buf);
}
int av1_get_sbq_perceptual_ai(AV1_COMP *const cpi, BLOCK_SIZE bsize, int mi_row,
