Add pointers to temp buffers in 'xd'.
Based on comments in
https://aomedia-review.googlesource.com/c/aom/+/70022
This avoids passing the temp buffer pointers and/or 'x' all the way
through the call chain.
BUG=aomedia:2135
Change-Id: I32f0a0b006d9fb24ee1b85ff1f33012bdb0eeb42
diff --git a/av1/decoder/decodeframe.c b/av1/decoder/decodeframe.c
index 30f5277..03cedef 100644
--- a/av1/decoder/decodeframe.c
+++ b/av1/decoder/decodeframe.c
@@ -191,11 +191,10 @@
(void)tx_size;
}
-static void predict_inter_block_void(AV1_COMMON *const cm, const ThreadData *td,
+static void predict_inter_block_void(AV1_COMMON *const cm,
MACROBLOCKD *const xd, int mi_row,
int mi_col, BLOCK_SIZE bsize) {
(void)cm;
- (void)td;
(void)xd;
(void)mi_row;
(void)mi_col;
@@ -625,7 +624,6 @@
}
static INLINE void dec_build_inter_predictors(const AV1_COMMON *cm,
- CONV_BUF_TYPE *tmp_conv_dst,
MACROBLOCKD *xd, int plane,
const MB_MODE_INFO *mi,
int build_for_obmc, int bw,
@@ -692,7 +690,7 @@
int tmp_dst_stride = 8;
assert(bw < 8 || bh < 8);
ConvolveParams conv_params = get_conv_params_no_round(
- 0, plane, tmp_conv_dst, tmp_dst_stride, is_compound, xd->bd);
+ 0, plane, xd->tmp_conv_dst, tmp_dst_stride, is_compound, xd->bd);
conv_params.use_jnt_comp_avg = 0;
struct buf_2d *const dst_buf = &pd->dst;
uint8_t *dst = dst_buf->buf + dst_buf->stride * y + x;
@@ -796,7 +794,7 @@
}
ConvolveParams conv_params = get_conv_params_no_round(
- 0, plane, tmp_conv_dst, MAX_SB_SIZE, is_compound, xd->bd);
+ 0, plane, xd->tmp_conv_dst, MAX_SB_SIZE, is_compound, xd->bd);
av1_jnt_comp_weight_assign(cm, mi, 0, &conv_params.fwd_offset,
&conv_params.bck_offset,
&conv_params.use_jnt_comp_avg, is_compound);
@@ -829,9 +827,11 @@
}
}
-static void dec_build_inter_predictors_for_planes(
- const AV1_COMMON *cm, const ThreadData *td, MACROBLOCKD *xd,
- BLOCK_SIZE bsize, int mi_row, int mi_col, int plane_from, int plane_to) {
+static void dec_build_inter_predictors_for_planes(const AV1_COMMON *cm,
+ MACROBLOCKD *xd,
+ BLOCK_SIZE bsize, int mi_row,
+ int mi_col, int plane_from,
+ int plane_to) {
int plane;
const int mi_x = mi_col * MI_SIZE;
const int mi_y = mi_row * MI_SIZE;
@@ -844,18 +844,15 @@
pd->subsampling_y))
continue;
- dec_build_inter_predictors(cm, td->tmp_conv_dst, xd, plane, xd->mi[0], 0,
- bw, bh, mi_x, mi_y);
+ dec_build_inter_predictors(cm, xd, plane, xd->mi[0], 0, bw, bh, mi_x, mi_y);
}
}
static void dec_build_inter_predictors_sby(const AV1_COMMON *cm,
- const ThreadData *td,
MACROBLOCKD *xd, int mi_row,
int mi_col, BUFFER_SET *ctx,
BLOCK_SIZE bsize) {
- dec_build_inter_predictors_for_planes(cm, td, xd, bsize, mi_row, mi_col, 0,
- 0);
+ dec_build_inter_predictors_for_planes(cm, xd, bsize, mi_row, mi_col, 0, 0);
if (is_interintra_pred(xd->mi[0])) {
BUFFER_SET default_ctx = { { xd->plane[0].dst.buf, NULL, NULL },
@@ -867,11 +864,10 @@
}
static void dec_build_inter_predictors_sbuv(const AV1_COMMON *cm,
- const ThreadData *td,
MACROBLOCKD *xd, int mi_row,
int mi_col, BUFFER_SET *ctx,
BLOCK_SIZE bsize) {
- dec_build_inter_predictors_for_planes(cm, td, xd, bsize, mi_row, mi_col, 1,
+ dec_build_inter_predictors_for_planes(cm, xd, bsize, mi_row, mi_col, 1,
MAX_MB_PLANE - 1);
if (is_interintra_pred(xd->mi[0])) {
@@ -886,14 +882,13 @@
}
}
-static void dec_build_inter_predictors_sb(const AV1_COMMON *cm,
- const ThreadData *td, MACROBLOCKD *xd,
+static void dec_build_inter_predictors_sb(const AV1_COMMON *cm, MACROBLOCKD *xd,
int mi_row, int mi_col,
BUFFER_SET *ctx, BLOCK_SIZE bsize) {
const int num_planes = av1_num_planes(cm);
- dec_build_inter_predictors_sby(cm, td, xd, mi_row, mi_col, ctx, bsize);
+ dec_build_inter_predictors_sby(cm, xd, mi_row, mi_col, ctx, bsize);
if (num_planes > 1)
- dec_build_inter_predictors_sbuv(cm, td, xd, mi_row, mi_col, ctx, bsize);
+ dec_build_inter_predictors_sbuv(cm, xd, mi_row, mi_col, ctx, bsize);
}
static INLINE void dec_build_prediction_by_above_pred(
@@ -918,14 +913,14 @@
block_size_high[BLOCK_64X64] >> (pd->subsampling_y + 1));
if (av1_skip_u4x4_pred_in_obmc(bsize, pd, 0)) continue;
- dec_build_inter_predictors(ctxt->cm, ctxt->tmp_conv_dst, xd, j,
- &backup_mbmi, 1, bw, bh, mi_x, mi_y);
+ dec_build_inter_predictors(ctxt->cm, xd, j, &backup_mbmi, 1, bw, bh, mi_x,
+ mi_y);
}
}
static void dec_build_prediction_by_above_preds(
- const AV1_COMMON *cm, const ThreadData *td, MACROBLOCKD *xd, int mi_row,
- int mi_col, uint8_t *tmp_buf[MAX_MB_PLANE], int tmp_width[MAX_MB_PLANE],
+ const AV1_COMMON *cm, MACROBLOCKD *xd, int mi_row, int mi_col,
+ uint8_t *tmp_buf[MAX_MB_PLANE], int tmp_width[MAX_MB_PLANE],
int tmp_height[MAX_MB_PLANE], int tmp_stride[MAX_MB_PLANE]) {
if (!xd->up_available) return;
@@ -936,11 +931,10 @@
int pred_height = AOMMIN(this_height / 2, 32);
xd->mb_to_bottom_edge += (this_height - pred_height) * 8;
- struct build_prediction_ctxt ctxt = {
- cm, td->tmp_conv_dst, mi_row,
- mi_col, tmp_buf, tmp_width,
- tmp_height, tmp_stride, xd->mb_to_right_edge
- };
+ struct build_prediction_ctxt ctxt = { cm, mi_row,
+ mi_col, tmp_buf,
+ tmp_width, tmp_height,
+ tmp_stride, xd->mb_to_right_edge };
BLOCK_SIZE bsize = xd->mi[0]->sb_type;
foreach_overlappable_nb_above(cm, xd, mi_col,
max_neighbor_obmc[mi_size_wide_log2[bsize]],
@@ -972,14 +966,14 @@
int bh = (left_mi_height << MI_SIZE_LOG2) >> pd->subsampling_y;
if (av1_skip_u4x4_pred_in_obmc(bsize, pd, 1)) continue;
- dec_build_inter_predictors(ctxt->cm, ctxt->tmp_conv_dst, xd, j,
- &backup_mbmi, 1, bw, bh, mi_x, mi_y);
+ dec_build_inter_predictors(ctxt->cm, xd, j, &backup_mbmi, 1, bw, bh, mi_x,
+ mi_y);
}
}
static void dec_build_prediction_by_left_preds(
- const AV1_COMMON *cm, const ThreadData *td, MACROBLOCKD *xd, int mi_row,
- int mi_col, uint8_t *tmp_buf[MAX_MB_PLANE], int tmp_width[MAX_MB_PLANE],
+ const AV1_COMMON *cm, MACROBLOCKD *xd, int mi_row, int mi_col,
+ uint8_t *tmp_buf[MAX_MB_PLANE], int tmp_width[MAX_MB_PLANE],
int tmp_height[MAX_MB_PLANE], int tmp_stride[MAX_MB_PLANE]) {
if (!xd->left_available) return;
@@ -990,11 +984,10 @@
int pred_width = AOMMIN(this_width / 2, 32);
xd->mb_to_right_edge += (this_width - pred_width) * 8;
- struct build_prediction_ctxt ctxt = {
- cm, td->tmp_conv_dst, mi_row,
- mi_col, tmp_buf, tmp_width,
- tmp_height, tmp_stride, xd->mb_to_bottom_edge
- };
+ struct build_prediction_ctxt ctxt = { cm, mi_row,
+ mi_col, tmp_buf,
+ tmp_width, tmp_height,
+ tmp_stride, xd->mb_to_bottom_edge };
BLOCK_SIZE bsize = xd->mi[0]->sb_type;
foreach_overlappable_nb_left(cm, xd, mi_row,
max_neighbor_obmc[mi_size_high_log2[bsize]],
@@ -1006,7 +999,6 @@
}
static void dec_build_obmc_inter_predictors_sb(const AV1_COMMON *cm,
- const ThreadData *td,
MACROBLOCKD *xd, int mi_row,
int mi_col) {
const int num_planes = av1_num_planes(cm);
@@ -1020,27 +1012,27 @@
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
int len = sizeof(uint16_t);
- dst_buf1[0] = CONVERT_TO_BYTEPTR(td->tmp_obmc_bufs[0]);
+ dst_buf1[0] = CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[0]);
dst_buf1[1] =
- CONVERT_TO_BYTEPTR(td->tmp_obmc_bufs[0] + MAX_SB_SQUARE * len);
+ CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[0] + MAX_SB_SQUARE * len);
dst_buf1[2] =
- CONVERT_TO_BYTEPTR(td->tmp_obmc_bufs[0] + MAX_SB_SQUARE * 2 * len);
- dst_buf2[0] = CONVERT_TO_BYTEPTR(td->tmp_obmc_bufs[1]);
+ CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[0] + MAX_SB_SQUARE * 2 * len);
+ dst_buf2[0] = CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[1]);
dst_buf2[1] =
- CONVERT_TO_BYTEPTR(td->tmp_obmc_bufs[1] + MAX_SB_SQUARE * len);
+ CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[1] + MAX_SB_SQUARE * len);
dst_buf2[2] =
- CONVERT_TO_BYTEPTR(td->tmp_obmc_bufs[1] + MAX_SB_SQUARE * 2 * len);
+ CONVERT_TO_BYTEPTR(xd->tmp_obmc_bufs[1] + MAX_SB_SQUARE * 2 * len);
} else {
- dst_buf1[0] = td->tmp_obmc_bufs[0];
- dst_buf1[1] = td->tmp_obmc_bufs[0] + MAX_SB_SQUARE;
- dst_buf1[2] = td->tmp_obmc_bufs[0] + MAX_SB_SQUARE * 2;
- dst_buf2[0] = td->tmp_obmc_bufs[1];
- dst_buf2[1] = td->tmp_obmc_bufs[1] + MAX_SB_SQUARE;
- dst_buf2[2] = td->tmp_obmc_bufs[1] + MAX_SB_SQUARE * 2;
+ dst_buf1[0] = xd->tmp_obmc_bufs[0];
+ dst_buf1[1] = xd->tmp_obmc_bufs[0] + MAX_SB_SQUARE;
+ dst_buf1[2] = xd->tmp_obmc_bufs[0] + MAX_SB_SQUARE * 2;
+ dst_buf2[0] = xd->tmp_obmc_bufs[1];
+ dst_buf2[1] = xd->tmp_obmc_bufs[1] + MAX_SB_SQUARE;
+ dst_buf2[2] = xd->tmp_obmc_bufs[1] + MAX_SB_SQUARE * 2;
}
- dec_build_prediction_by_above_preds(cm, td, xd, mi_row, mi_col, dst_buf1,
+ dec_build_prediction_by_above_preds(cm, xd, mi_row, mi_col, dst_buf1,
dst_width1, dst_height1, dst_stride1);
- dec_build_prediction_by_left_preds(cm, td, xd, mi_row, mi_col, dst_buf2,
+ dec_build_prediction_by_left_preds(cm, xd, mi_row, mi_col, dst_buf2,
dst_width2, dst_height2, dst_stride2);
av1_setup_dst_planes(xd->plane, xd->mi[0]->sb_type, get_frame_new_buffer(cm),
mi_row, mi_col, 0, num_planes);
@@ -1055,9 +1047,8 @@
}
}
-static void predict_inter_block(AV1_COMMON *const cm, const ThreadData *td,
- MACROBLOCKD *const xd, int mi_row, int mi_col,
- BLOCK_SIZE bsize) {
+static void predict_inter_block(AV1_COMMON *const cm, MACROBLOCKD *const xd,
+ int mi_row, int mi_col, BLOCK_SIZE bsize) {
MB_MODE_INFO *mbmi = xd->mi[0];
const int num_planes = av1_num_planes(cm);
for (int ref = 0; ref < 1 + has_second_ref(mbmi); ++ref) {
@@ -1075,9 +1066,9 @@
}
}
- dec_build_inter_predictors_sb(cm, td, xd, mi_row, mi_col, NULL, bsize);
+ dec_build_inter_predictors_sb(cm, xd, mi_row, mi_col, NULL, bsize);
if (mbmi->motion_mode == OBMC_CAUSAL) {
- dec_build_obmc_inter_predictors_sb(cm, td, xd, mi_row, mi_col);
+ dec_build_obmc_inter_predictors_sb(cm, xd, mi_row, mi_col);
}
#if CONFIG_MISMATCH_DEBUG
for (int plane = 0; plane < num_planes; ++plane) {
@@ -1164,7 +1155,7 @@
}
}
} else {
- td->predict_inter_block_visit(cm, td, xd, mi_row, mi_col, bsize);
+ td->predict_inter_block_visit(cm, xd, mi_row, mi_col, bsize);
// Reconstruction
if (!mbmi->skip) {
int eobtotal = 0;
@@ -2923,6 +2914,11 @@
td->xd.corrupted = 0;
td->xd.mc_buf[0] = td->mc_buf[0];
td->xd.mc_buf[1] = td->mc_buf[1];
+ td->xd.tmp_conv_dst = td->tmp_conv_dst;
+ for (int j = 0; j < 2; ++j) {
+ td->xd.tmp_obmc_bufs[j] = td->tmp_obmc_bufs[j];
+ }
+
for (tile_row = tile_rows_start; tile_row < tile_rows_end; ++tile_row) {
const int row = inv_row_order ? tile_rows - 1 - tile_row : tile_row;
@@ -3417,6 +3413,10 @@
thread_data->td->xd.corrupted = 0;
thread_data->td->xd.mc_buf[0] = thread_data->td->mc_buf[0];
thread_data->td->xd.mc_buf[1] = thread_data->td->mc_buf[1];
+ thread_data->td->xd.tmp_conv_dst = thread_data->td->tmp_conv_dst;
+ for (int j = 0; j < 2; ++j) {
+ thread_data->td->xd.tmp_obmc_bufs[j] = thread_data->td->tmp_obmc_bufs[j];
+ }
winterface->sync(worker);
worker->hook = worker_hook;
