Don't allocate dqcoeff in MACROBLOCKD. The encoder gets its dqcoeff from the context tree. In the decoder move it to directly after MACROBLOCKD. Change-Id: I46c9b76f26956a360d17de0b26ecb994dae34ecb
diff --git a/vp9/common/vp9_blockd.h b/vp9/common/vp9_blockd.h index fd8fe3f..fbdbb44 100644 --- a/vp9/common/vp9_blockd.h +++ b/vp9/common/vp9_blockd.h
@@ -187,12 +187,6 @@ int bd; #endif - /* dqcoeff are shared by all the planes. So planes must be decoded serially */ -#if CONFIG_VP9_ENCODER - DECLARE_ALIGNED(16, tran_low_t, dqcoeff[64 * 64]); -#else - DECLARE_ALIGNED(16, tran_low_t, dqcoeff[32 * 32]); -#endif int lossless; int corrupted;
diff --git a/vp9/common/vp9_onyxc_int.h b/vp9/common/vp9_onyxc_int.h index 8f1240a..ca0dfc8 100644 --- a/vp9/common/vp9_onyxc_int.h +++ b/vp9/common/vp9_onyxc_int.h
@@ -348,11 +348,12 @@ (const vp9_prob (*)[PARTITION_TYPES - 1])cm->fc->partition_prob; } -static INLINE void vp9_init_macroblockd(VP9_COMMON *cm, MACROBLOCKD *xd) { +static INLINE void vp9_init_macroblockd(VP9_COMMON *cm, MACROBLOCKD *xd, + tran_low_t *dqcoeff) { int i; for (i = 0; i < MAX_MB_PLANE; ++i) { - xd->plane[i].dqcoeff = xd->dqcoeff; + xd->plane[i].dqcoeff = dqcoeff; xd->above_context[i] = cm->above_context + i * sizeof(*cm->above_context) * 2 * mi_cols_aligned_to_sb(cm->mi_cols);
diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c index 6c64540..4cdfe87 100644 --- a/vp9/decoder/vp9_decodeframe.c +++ b/vp9/decoder/vp9_decodeframe.c
@@ -1324,11 +1324,12 @@ tile_data->xd.corrupted = 0; tile_data->xd.counts = cm->frame_parallel_decoding_mode ? NULL : &cm->counts; + vp9_zero(tile_data->dqcoeff); vp9_tile_init(&tile_data->xd.tile, tile_data->cm, tile_row, tile_col); setup_token_decoder(buf->data, data_end, buf->size, &cm->error, &tile_data->bit_reader, pbi->decrypt_cb, pbi->decrypt_state); - vp9_init_macroblockd(cm, &tile_data->xd); + vp9_init_macroblockd(cm, &tile_data->xd, tile_data->dqcoeff); } } @@ -1543,12 +1544,13 @@ tile_data->xd.corrupted = 0; tile_data->xd.counts = cm->frame_parallel_decoding_mode ? 0 : &tile_data->counts; + vp9_zero(tile_data->dqcoeff); vp9_tile_init(tile, cm, 0, buf->col); vp9_tile_init(&tile_data->xd.tile, cm, 0, buf->col); setup_token_decoder(buf->data, data_end, buf->size, &cm->error, &tile_data->bit_reader, pbi->decrypt_cb, pbi->decrypt_state); - vp9_init_macroblockd(cm, &tile_data->xd); + vp9_init_macroblockd(cm, &tile_data->xd, tile_data->dqcoeff); worker->had_error = 0; if (i == num_workers - 1 || n == tile_cols - 1) {
diff --git a/vp9/decoder/vp9_decoder.h b/vp9/decoder/vp9_decoder.h index 76a8746..af47f85 100644 --- a/vp9/decoder/vp9_decoder.h +++ b/vp9/decoder/vp9_decoder.h
@@ -32,6 +32,8 @@ VP9_COMMON *cm; vp9_reader bit_reader; DECLARE_ALIGNED(16, MACROBLOCKD, xd); + /* dqcoeff are shared by all the planes. So planes must be decoded serially */ + DECLARE_ALIGNED(16, tran_low_t, dqcoeff[32 * 32]); } TileData; typedef struct TileWorkerData { @@ -39,6 +41,8 @@ vp9_reader bit_reader; FRAME_COUNTS counts; DECLARE_ALIGNED(16, MACROBLOCKD, xd); + /* dqcoeff are shared by all the planes. So planes must be decoded serially */ + DECLARE_ALIGNED(16, tran_low_t, dqcoeff[32 * 32]); struct vpx_internal_error_info error_info; } TileWorkerData;
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c index f27f57a..6c595b7 100644 --- a/vp9/encoder/vp9_encoder.c +++ b/vp9/encoder/vp9_encoder.c
@@ -732,7 +732,7 @@ vp9_set_mb_mi(cm, cm->width, cm->height); vp9_init_context_buffers(cm); - vp9_init_macroblockd(cm, xd); + vp9_init_macroblockd(cm, xd, NULL); cpi->td.mb.mbmi_ext_base = cpi->mbmi_ext_base; memset(cpi->mbmi_ext_base, 0, cm->mi_rows * cm->mi_cols * sizeof(*cpi->mbmi_ext_base));