OSDev.org

I have been thinking that drawing in a screen could be done extremely fast if the base address of the graphics card (the BAR) could be moved to arbitrary regions of memory at runtime instead of relying on double/multiple buffering.

In this way, we could implement a high level stack of buffer pointers, and also a heap of such buffers.

Then, we could start pushing just the address of certain screen-sized buffer and pass it as a parameter for a drawing function.

When we finish drawing, we could, instead of copying the secondary buffer to the screen's frame buffer, just pop it directly into the PCI Base Address Register (BAR) and thus updating the current screen immediately?

If the hardware supported it, we could also move, free up and even page the frame buffer region and probably use it for other purposes (for example when the screen is turned off or when there's a long time away from the computer).

tjmonk15 wrote:Seeing as hardware sees physical memory (and not paged/linear/whats the other one called?) this is already "easy" to do.

- Monk

neon wrote:Most modern video devices support page flipping which can be used to alternate between multiple buffers without the need to copy data. I am not aware of any standard protocol (not even firmware interfaces) that provides these facilities though. In other words, you may need to write custom drivers for these devices. Copying the back buffer is slower but more portable.

~ wrote:instead of copying the secondary buffer to the screen's frame buffer, just pop it directly into the PCI Base Address Register (BAR) and thus updating the current screen immediately?