Samsung S3C6400X User Manual page 653

FIMV-MFC V1.0
Inter-Prediction
The Inter-Predictor uses reconstructed motion vector that represents the displacement between the block
currently being decoded and the corresponding location in the reference frame, to calculate interpolated pixel data
for motion compensation.
The Inter-Predictor consists of main controller, interpolator, DMA and local memory. The role of each sub-block is
listed below:
It is controlled by the BIT processor that writes control information (motion vector,
block mode, reference picture index, picture size and run command etc.) into the
Main
control register in main controller. It includes data register for motion vector and
controller
reference picture base address in spreg64x32.
It interpolates reference picture data with 2-pel/4-pel resolution. It uses temporal
Interpolato
memory (spreg64x32e16) for calculating H.264 quarter-pels.
r
Reads reference picture data from SDRAM(decoding) or ME local
memory(encoding) and write it into local buffer memory(dpreg60x96e8,
DMA
dpreg36x96e8)
Inter-predictor supports all H.264 block mode (16x16, 16x8, 8x16, 8x8, 8x4, 4x8, 4x4), half/quarter pixel
resolution, 16 reference frame in H.264 BP and padding pixel when the motion vector points to an area that is
outside the image.
The figure 21.16 is the block diagram of Inter-predictor.
When the BIT processor set the registers (motion vectors, reference frame index, macroblock mode etc) of main
controller, main controller controls DMA and interpolator.
The DMA read reference pixel data from SDRAM in decoding and from ME local memory in encoding to reduce
SDRAM bandwidth. If the motion vectors used to locate the predicted blocks in the reference frame may include
pixel locations that are outside the boundary of the reference frame in decoding mode. In these cases, the out-of-
bounds pixel values are the replicated values of the edge pixel by DMA.
Preliminary product information describe products that are in development,
for which full characterization data and associated errata are not yet available.
Specifications and information herein are subject to change without notice.
MULTI-FORMAT
VIDEO
CODEC
21-27