Re: wined3d: Added PCI device IDs for AMD Radeon 6750M & 6770M
On 31 March 2012 21:20, Gaurav Jain <gaurav(a)gauravjain.org> wrote:
CARD_AMD_RADEON_HD6600M = 0x6741, + CARD_AMD_RADEON_HD6750M = 0x6741,
If those are actually supposed to be the same, I don't think there's much of a point in this patch.
On 2012-03-31, at 4:49 PM, Henri Verbeet wrote:
On 31 March 2012 21:20, Gaurav Jain <gaurav(a)gauravjain.org> wrote:
CARD_AMD_RADEON_HD6600M = 0x6741, + CARD_AMD_RADEON_HD6750M = 0x6741,
If those are actually supposed to be the same, I don't think there's much of a point in this patch.
Ahh... I didn't catch that the 6600M & 6750M was identical. Is there even a 6600M? My google search only found a 6630M, 6650M & 6670M. I believe 6600M is the series. I can remove the 6750M from the patch if you'd like. Please note that there is another duplicate entry for 5700 & 5750. The 6770M did have a different PCI ID, would you like the patch modified with only changes for the 6770M?
On 1 April 2012 01:55, Gaurav Jain <gaurav(a)gauravjain.org> wrote:
Ahh... I didn't catch that the 6600M & 6750M was identical. Is there even a 6600M? My google search only found a 6630M, 6650M & 6670M. I believe 6600M is the series. I can remove the 6750M from the patch if you'd like. Please note that there is another duplicate entry for 5700 & 5750.
The 6770M did have a different PCI ID, would you like the patch modified with only changes for the 6770M?
Actually, I didn't spot this yesterday, but the description strings look suspect as well. Does the Windows driver really distinguish between these? Most of the r600 cards only distinguish the general series in the driver description string, and you can't necessarily tell them apart from the PCI IDs either. Unless the Windows drivers actually distinguish between these cards, they probably shouldn't have separate entries at all.
On 2012-04-01, at 6:57 AM, Henri Verbeet wrote:
Does the Windows driver really distinguish between these?
I'm not sure if it does, if the matching is only on the vendor and device ID, then it's unlikely. However isn't it important for wine to still distinguish between models (if possible) if they have different minimum memory sizes across models? Are there any other significant differences across models that wine need to be aware of? Gaurav
On 1 April 2012 19:08, Gaurav Jain <gaurav(a)gauravjain.org> wrote:
However isn't it important for wine to still distinguish between models (if possible) if they have different minimum memory sizes across models? Are there any other significant differences across models that wine need to be aware of?
In general it isn't very important to report the exact model to the application, as long as it's roughly the same class of hardware wrt e.g. supported shader model. Ideally applications wouldn't care at all about this and use the caps instead, though in practice applications might use the PCI IDs for workarounds for driver bugs or for capability detection. The video memory is even more of a guess than everything else, in particular because you can have the same model with different amounts of video memory. The GL_NVX_gpu_memory_info extension would be very useful for this if it wasn't experimental.
On 2012-04-01, at 6:06 PM, Henri Verbeet wrote:
Ideally applications wouldn't care at all about this and use the caps instead, though in practice applications might use the PCI IDs for workarounds for driver bugs or for capability detection.
Sounds good, so then are you okay with the patch with only the 6770M, since that has a different device ID?
The video memory is even more of a guess than everything else, in particular because you can have the same model with different amounts of video memory.
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 GL_NVX_gpu_memory_info extension would be very useful for this if it wasn't experimental.
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?
On 2 April 2012 01:15, Gaurav Jain <gaurav(a)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.
participants (2)
-
Gaurav Jain -
Henri Verbeet