improving palette throughput
* code the palette color index using 45 wavefront
* interleave the coeff and palette color index in
transform block level
* the above change does not change code efficiency
Details:
The 45 wavefront scan allows to compute the ctx of
the diagonal samples' indices at the same time.
Interleaving palette indices and palette residual
on a transform block basis means that the entropy
decoding and further processing of the palette
residual is not delayed by the entropy decoding
of all the color indices of the palette encoded
block.
Change-Id: Ie9f576002a9a68394b99c23b01e9730df06df070
diff --git a/av1/decoder/detokenize.c b/av1/decoder/detokenize.c
index 4c1408d..bbdc7ad 100644
--- a/av1/decoder/detokenize.c
+++ b/av1/decoder/detokenize.c
@@ -442,6 +442,70 @@
#endif // !CONFIG_PVQ
#if CONFIG_PALETTE
+#if CONFIG_PALETTE_THROUGHPUT
+void av1_decode_palette_tokens(MACROBLOCKD *const xd, int plane,
+ TX_SIZE tx_size, int row, int col,
+ aom_reader *r) {
+ const MODE_INFO *const mi = xd->mi[0];
+ const MB_MODE_INFO *const mbmi = &mi->mbmi;
+ uint8_t color_order[PALETTE_MAX_SIZE];
+ const int n = mbmi->palette_mode_info.palette_size[plane];
+ int i, j;
+ uint8_t *const color_map = xd->plane[plane].color_index_map;
+ const aom_prob(
+ *const prob)[PALETTE_COLOR_INDEX_CONTEXTS][PALETTE_COLORS - 1] =
+ plane ? av1_default_palette_uv_color_index_prob
+ : av1_default_palette_y_color_index_prob;
+ int plane_block_width, plane_block_height, rows, cols;
+
+ const int bsize = txsize_to_bsize[tx_size];
+ const int tx_block_width = 1 << tx_size_wide_log2[0];
+ const int tx_block_height = 1 << tx_size_high_log2[0];
+ av1_get_block_dimensions(mbmi->sb_type, plane, xd, &plane_block_width,
+ &plane_block_height, &rows, &cols);
+ const int block_width =
+ AOMMIN(cols - col * tx_block_width, block_size_wide[bsize]);
+ const int block_height =
+ AOMMIN(rows - row * tx_block_height, block_size_high[bsize]);
+ assert(plane == 0 || plane == 1);
+
+ // run wavefront on the palette map index decoding per transform block
+ for (i = ((row == 0 && col == 0) ? 1 : 0); i < block_width + block_height - 1;
+ ++i) {
+ for (j = AOMMIN(i, block_width - 1); j >= AOMMAX(0, i - block_height + 1);
+ --j) {
+ const int color_ctx = av1_get_palette_color_index_context(
+ color_map, plane_block_width, row * tx_block_height + (i - j),
+ col * tx_block_width + j, n, color_order, NULL);
+ const int color_idx =
+ aom_read_tree(r, av1_palette_color_index_tree[n - 2],
+ prob[n - 2][color_ctx], ACCT_STR);
+ assert(color_idx >= 0 && color_idx < n);
+ color_map[(row * tx_block_height + i - j) * plane_block_width +
+ col * tx_block_width + j] = color_order[color_idx];
+ }
+ }
+ // Copy last column to extra columns.
+ if (block_width < block_size_wide[bsize]) {
+ for (i = 0; i < block_height; ++i) {
+ memset(color_map + (row * tx_block_height + i) * plane_block_width +
+ col * tx_block_width + block_width,
+ color_map[(row * tx_block_height + i) * plane_block_width +
+ col * tx_block_width + block_width - 1],
+ (block_size_wide[bsize] - block_width));
+ }
+ }
+ // Copy last row to extra rows.
+ if (block_height < block_size_high[bsize]) {
+ for (i = block_height; i < block_size_high[bsize]; ++i) {
+ memcpy(color_map + (row * tx_block_height + i) * plane_block_width,
+ color_map +
+ (row * tx_block_height + block_height - 1) * plane_block_width,
+ block_size_wide[bsize]);
+ }
+ }
+}
+#else
void av1_decode_palette_tokens(MACROBLOCKD *const xd, int plane,
aom_reader *r) {
const MODE_INFO *const mi = xd->mi[0];
@@ -479,6 +543,7 @@
color_map + (rows - 1) * plane_block_width, plane_block_width);
}
}
+#endif // CONFIG_PALETTE_THROUGHPUT
#endif // CONFIG_PALETTE
#if !CONFIG_PVQ
