On 2 April 2012 01:15, Gaurav Jain gaurav@gauravjain.org wrote:
Sounds good, so then are you okay with the patch with only the 6770M, since that has a different device ID?
The first step should really be getting the description string right. After that you should verify that that PCI ID is really unique for the 6770M, as opposed to having e.g. 4 or 5 different IDs that are used by all the 6600M/6700M cards.
Fair enough, the comment in the code states that wine estimates the *minimum* video memory. Is there any issues with wine reporting more? If so, then there are cases where certain models within a series may have less memory than others. Thus wine could improve it's guess by having a different entry for each card (if necessary). That way the guess would never over-estimate the memory (assuming that's an issue).
The comment is probably wrong, it's generally better to err on the side of reporting too much memory than too little.
Are there any problem with wine using the value from here instead of hard coding the video memory? Can bogus values show up? Can wine use GL_ATI_meminfo as well?
The issue with GL_NVX_gpu_memory_info is that because it's experimental it can change (and if you look at the spec revision history, it has), so we can't really rely on it to do anything at all. The problem with ATI_meminfo is that it doesn't allow you to query the total amount of video memory. There's an undocumented hack to do this anyway, but since that's not in the spec we can't use it either.