On Monday, October 17, 2016 4:46:54 PM PDT Lionel Landwerlin wrote:
> Hi,
> Spoiler alert, this series doesn't actually make the border colors work on
> IvyBridge and Haswell. It still brings some useful changes (I think).
> On Haswell, it seems the hardware always reads the border color from the
> offset 0 of the "Dynamic State Memory Address", regardless of what offset is
> programmed in the "Border Color Pointer" field of the SAMPLER_STATE
> structure. Maybe some kind of cache invalidation is missing in this series?
> Any idea/pointer is more than welcome!
> Cheers,
> Lionel Landwerlin

Ouch.  I'm pretty sure you've hit the undocumented hardware bug I
discovered a few years ago.

The STATE_BASE_ADDRESS "Dynamic State Access Upper Bound" field is
supposed to be ignored when it is zero, but the sampler border color
access apparently fails to do so.  So the address gets clamped.

Program that to a proper value (or 0xFFFFFFFF) and I bet it will work.

See c25e5300cba7628b58df93ead14ebc3cc32f338c in the GL driver.

Sorry for not thinking to warn you sooner :(

Attachment: signature.asc
Description: This is a digitally signed message part.

mesa-dev mailing list

Reply via email to