Renesas SH7781 用户手册
20. Graphics Data Translation Accelerator (GDTA)
Rev.1.00 Jan. 10, 2008 Page 1019 of 1658
REJ09B0261-0100
20.4.2
Explanation of MC Operation
By writing 1 to the MC_EN bit in GACER, registers in the MC register unit can be accessed.
After setting, as initial values, the input frame width/height, input YUV padding size, output frame
YUV pointer, past frame YUV pointer and future frame YUV pointer, data written in succession
to MCCF is received, and upon receiving a maximum of eight command parameters (estimating
mode, vector, buffer RAM 1 address), processing is begun. In processing, data is read in
macroblock units (Y: 16 pixels
After setting, as initial values, the input frame width/height, input YUV padding size, output frame
YUV pointer, past frame YUV pointer and future frame YUV pointer, data written in succession
to MCCF is received, and upon receiving a maximum of eight command parameters (estimating
mode, vector, buffer RAM 1 address), processing is begun. In processing, data is read in
macroblock units (Y: 16 pixels
× 16 lines, U/V: 8 pixels × 8 lines), and estimated image
generation is performed within the module. The generated image by the amount of a macroblock
is output to the output destination.
is output to the output destination.
The MC can store four commands (in register MCCF) and does not accept the command for the
next frame when four commands are already stored (command FIFO full). A judgment as to
whether processing has ended can be made by using either an interrupt or the MC_END bit in
GACISR.
next frame when four commands are already stored (command FIFO full). A judgment as to
whether processing has ended can be made by using either an interrupt or the MC_END bit in
GACISR.
Figure 20.6 illustrates the processing of one Y macroblock in "forward macroblock processing".
(After three rounds of processing for Y, U, and V have been done, a processing-end interrupt is
generated by writing an end command.) On the other hand, in "reverse macroblock processing" the
address for reading data from DDR2, shown in figure 20.6, is changed to the future frame pointer,
and other processing is the same. Hence in "bidirectional macroblock processing", (half-pixel-
corrected past data + half-pixel-corrected future data + 1)/2 is taken to be the correction processing
result, and calculation is performed with the IDCT data.
(After three rounds of processing for Y, U, and V have been done, a processing-end interrupt is
generated by writing an end command.) On the other hand, in "reverse macroblock processing" the
address for reading data from DDR2, shown in figure 20.6, is changed to the future frame pointer,
and other processing is the same. Hence in "bidirectional macroblock processing", (half-pixel-
corrected past data + half-pixel-corrected future data + 1)/2 is taken to be the correction processing
result, and calculation is performed with the IDCT data.
Finally, in "intra macroblock processing" signed IDCT data (the sign of which is discriminated
using the uppermost bit (bit 15)) is converted into unsigned 8-bit data and written to the output
position.
using the uppermost bit (bit 15)) is converted into unsigned 8-bit data and written to the output
position.