The problem is that there will always be new hardware and new graphics standards to support, and having a small hardware-specific layer reduces the effort required to do that (ie gives the development community a chance of keeping up). If the driver requirements ever really became static (or even static-ish) then optimizing for performance at the expense of extensibility might make sense.
The current Gallium3D API level seems like a pretty reasonable compromise between potential performance and required development effort.


Reply With Quote
