DCT / iDCT公式。
通常要用浮點作，不過為了效率，通常硬體都用fix point做。

所以現在問題集中在『10bit Decode』上。
怎樣能夠『不需要軟體支援』、但是卻能夠『直接改善輸出品質』呢？

其實一個簡單的問題：
軟體不支援的時候，硬體會怎樣？
直接能夠發揮作用？這似乎不太可能。
這邊另一個明顯的例子就是AMD K8 vs 64bit OS。

不然就是有『能讓硬體在與軟體沒有直接關聯的狀況下、直接發揮作用』，比方說硬體主動增加一些resampler之類的設計，來處理8bit資訊。

或許我直接和Matrox要SDK會比較快？

[EDIT]
嗯，Matrox的文件的確有講到，所以看來疑問解除了。

http://www.matrox.com/mga/products/parhelia512/technology/gigacolor.cfm

PDF的page8。

[quote]Parhelia-512 also offers additional flexibility.
When the computation of a particular pixel is completed by the GPU, the final result can be written to the frame buffer at 10-bit precision [b]or in the format required by the application. [/b]
When an application asks for the image to be drawn in a lesser pixel format (e.g. five or eight bits per component) [b]the GPU will apply a dithering algorithm to the resulting image prior to writing it to the frame buffer.[/b]
Dithering is a process that improves the perceived quality of an image when the color palette is reduced.
It does so through an error-diffusion pattern that simulates the missing colors.
Although there is a loss from the original image at the higher precision, dithering enables the output to benefit from the original precision output by the engine when displayed at a lower bit-depth.[/quote]

軟體不支援10bit 的時候，會要求8bit Frame Buffer，
所以能不能有改善，端看GPU會不會把10bit的內容作Dithering到8bit，
所以看來Parhelia的DVD playback的確會比本來多些資訊；
但是這時候是8bit Frame Buffer….

剩下的就很明顯了。