VP8ThreadData Struct Reference

Public Member Functions
	DECLARE_ALIGNED (16, int16_t, block)[6][4][16]
	DECLARE_ALIGNED (16, int16_t, block_dc)[16]
	DECLARE_ALIGNED (16, uint8_t, non_zero_count_cache)[6][4]
	This is the index plus one of the last non-zero coeff for each of the blocks in the current macroblock. More...
	DECLARE_ALIGNED (8, uint8_t, left_nnz)[9]
	For coeff decode, we need to know whether the above block had non-zero coefficients. More...
	DECLARE_ALIGNED (16, uint8_t, edge_emu_buffer)[21 *EDGE_EMU_LINESIZE]

Public Attributes
int	thread_nr
atomic_int	thread_mb_pos
atomic_int	wait_mb_pos
VP8FilterStrength *	filter_strength
VP8mvbounds	mv_bounds

Member Function Documentation

DECLARE_ALIGNED() [1/2]

VP8ThreadData::DECLARE_ALIGNED	(	16	,
		uint8_t	,
		non_zero_count_cache
	)

This is the index plus one of the last non-zero coeff for each of the blocks in the current macroblock.

So, 0 -> no coeffs 1 -> dc-only (special transform) 2+-> full transform

DECLARE_ALIGNED() [2/2]

VP8ThreadData::DECLARE_ALIGNED	(	8	,
		uint8_t	,
		left_nnz
	)

For coeff decode, we need to know whether the above block had non-zero coefficients.

This means for each macroblock, we need data for 4 luma blocks, 2 u blocks, 2 v blocks, and the luma dc block, for a total of 9 per macroblock. We keep the last row in top_nnz.

---

The documentation for this struct was generated from the following file:

• libavcodec/vp8.h