[Mesa-dev] [RFC] [PATCH 0/4] New coordinate normalization patches with bind flags

[Luca Barbieri]

> Because there is some expectation (with the exception of the private
> flags) this is state set by the state tracker.

Yes, currently it's this way, but I think that overall, this is suboptimal.
For instance, I think that drivers should override the bind flags,
setting them to what the resource actually supports.
Right now, after creation the bind flags aren't very useful, since
they represent an arbitrary subset of what the resource actually
supports.

For instance, this would allow the driver to decide between using the
blitter module or the 2D engine just by checking the bind flags
instead of requiring internal flags or duplicated logic, and would
allow an arbitrary util module to choose between a fast path that
requires the ability to render and a software fallback, even if the
state tracker did not anticipate the need to render and thus didn't
ask for it.

In fact it's quite possible that right now some parts of the codebase
might be unnecessarily using fallbacks just because only a few bind
flags were specified at creation, even though most resources actually
support any kind of operation.