Renesas SH7781 - Manual (Page 1056)

20. Graphics Data Translation Accelerator (GDTA)
Rev.1.00 Jan. 10, 2008 Page 1026 of 1658
REJ09B0261-0100
No. Operation Description
(6) Half-pixel
correction
processing
Half-pixel correction processing is performed for the data read in (5).
(For the Y pointer: 16 to the right and 16 downward in 4 pixel units, for a total
of 256 processed)
* The Recon_down value is used to perform half-pixel correction even/odd
decisions. Decision results are as follows.
(The Recon_right value is similarly used to perform half-pixel correction
even/odd decisions.)
Recon_down -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
Y even/odd even odd even odd even odd even odd even odd even odd even
U/V
even/odd odd even even odd odd even even even odd odd even even odd
Formulae for each correction case at "Input data for one macroblock" in figure
20.6 are as follows.
Correction case 1: right = even, down = even
-> (A + A + 1)/2 = A (digits below the decimal point are discarded (no
change))
Correction case 2: right = even, down = odd
-> (A + C + 1)/2 = A' (digits below the decimal point are discarded)
Correction case 3: right = odd, down = even
-> (A + B + 1)/2 = A' (digits below the decimal point are discarded)
Correction case 4: right = odd, down = odd
-> (A + D + B + C + 2)/4 = A' (digits below the decimal point are discarded)
In bidirectional macroblock processing, the result of computation of the
following formula using past data after half-pixel correction and future data
after half-pixel correction is taken to be the half-pixel correction result, and
computation processing with IDCT data is performed.
(past data after half-pixel correction + future data after half-pixel correction
+1)/2 = half-pixel correction result
There are cases in which, due to the settings, input macroblock data is not
within the padding area and the data exceeds the screen height (invalid
image data), but the hardware performs processing without modification.