Graphics Corruption with P96 3.3.2 RTG Library (43.399)

Why didn't you try to resolve the version discrepancy first? The "resident" lib is different from the file you've referenced in the first picture. This sounds like the lib can be found in two of the library paths.

Is the system the one that can be read in the second picture? Mediator/Voodoo 3DFX?

Talked to Thor, and we have a suspicion what is happening here:

This may be a side effect of aperture switching, meaning, the configuration of the crossbar that's between VGA memory and the Amiga bus. Typically, this allows reversing the byte order, so switching between ARGB and BGRA (or similar) is possible. From the GPU side, both modes are identical and only differ in the Amiga-point-of-view.

This aperture switching is used a lot more often with newer versions of P96, because P96 was updated to support keeping bitmaps with different apertures in memory at the same time. Aperture switching is done if a blit needs to take place between these different bitmaps.

Switching of this apertures always happened on the driver side in function SetMemoryMode(), which also switches between planar and chunky modes (that's the VGA Chain-4 bit).

For far for the preamble. The hypothesis is that the driver does not use SerMemoryMode(), but only sets the aperture when loading the video mode with SetGC(), Bitmaps with an aperture different from the current one will be blitted with the wrong aperture.

In other words: We assume that Elbox did not work according to the P96 specs. This is a logical explanation, as Elbox never purchased the developer package back in the days, but developed against an API that wasn't documented for the public. Please point Elbox to our Wiki: When iComp took over the P96 package, I decided to make the API free for all interested developers, including those that have been in a legal grey zone prior to our takeover. I specifically made this move to improve 3rd party drivers. However, these 3rd parties need to accept the gift. We can't possibly do their work, especially given the fact that Elbox is still attempting to keep the Mediator registers a secret.

Possible workaround: Only use video modes that make use of the VGA-default-aperture, which is BGRA, Chunky and Hi-Colour PC modes. This way, P96 does not need to make Aperture switches, likely hiding the driver bug. Again, this is just a hunch, not pointing fingers. Please try the workaround and report back.

Quote from Drewlio77

I am curious what changed in 43,399 of the RTG Lib that has revealed the Mediator driver bug.

Like I wrote: P96 uses aperture switching a lot more often, so it requires the driver to be developed according to spec.

The specification is now open in our Wiki, not even a registration is required. Also, I can confirm that Elbox has bought a license of P96 (though some time ago, they currently don't have access to the latest version), so there is still *someone* who is checking things every once in a while. It is probably just a matter of calling SetMemoryMode() a few more times in select places, and the bug is gone. It may be a bit more elaborate on a PCI board with Ethernet, as the Ethernet board uses the VGA memory as a trampoline to relay packet data. They might require more memory aperture settings to catch DMA'd data.

Given that you report that 3.2.1 is not affected, it may be a different reason - in addition to the semi-confirmation you've found with regard to the apertures. Thor is cirrently building all intermediate versions that I'd like to send you as a packet - please eMail me, or mention your order ID; so I can send the archive to you via eMail.

eMail sent with a total of seven versions of the P96 libs (21 files: 3 libs per version).

Quote from Drewlio77

Does "APERTURESWITCHINOPERABLE" require an ON or OFF qualifier?

Almost - please try "APERTURESWITCHINOPERABLE=Yes".