Marvell International Limited

Zi Gu

Jia Bao

In video decoding using the H.264/AVC standard, the computation of an inverse 4&#xd7;4 integer transform of a coefficient matrix derived by variable length decoding may be carried out using data reduction techniques to reduce the computation load. If the index value of the highest-indexed nonzero coefficient in the matrix is three or higher, the transform is computed conventionally, using two 1D transform operations separated by a transpose operation, and followed by rounding and shifting. If the index value of the highest-indexed nonzero coefficient in the matrix is zero (including the case where there is no nonzero coefficient), the inverse integer transform operation includes only rounding and shifting of that coefficient. If the index value of the highest-indexed nonzero coefficient in the matrix is one or two, then the inverse integer transform operation can be performed using a single integrated 2D transform followed by rounding and shifting.

Fast inverse integer transform for video decoding

A method of filtering a pixels along a block edge during decoding of compressed VC-1 video includes loading edge segments with pixels along the sides of a block edge into registers, calculating a selection mask for a third pixel pair, and filtering the pixels in the edge segments simultaneously in the registers, if the selection mask meets a predetermined criteria. In another embodiment, a method of filtering pixels along a block edge during decoding of compressed VC-1 video includes loading edge segments with pixels along the side of a block edge into registers, swapping a first pair of the pixels with a second pair of the pixels, where each of the pairs of pixels are third pixel pairs of their respective segments, filtering the pairs of pixels simultaneously in the registers, and filtering the remaining pixels of the respective edge segments, if the results of filtering the third pixels pairs meet a predetermined criteria. The registers may be single instruction multiple data (SIMD) registers used with a Wireless MMX processor.

Growing community of inventors

Shanghai, China

Zi Gu