I would love to have something like this working. However, everyone
I have spoken to tells me that the protected memory architecture and
driver model currently used forbids this sort of sharing. Im not
smart enough/knowledgeable enough to know all the details of why,
but im also stubborn enough to know there *has* to be a work around.
i wrote that weird little plugin as a one potential (but kind of an
ugly hack ish) work around.
If it really is driver/VM (virtual memory) deep, the only workarounds
would be possibly Shared Memory (/usr/include/sys/shm.h) (not sure how
that would work with non-system memory resources though) or a kext
that does the shim thing you mention later...
Id imagine some sort of API would be required to know the memory
locations of the resources on the GPU (ie, this offset / location
for texture a, etc), and that information would have to be
constantly synced between apps - but I have no idea even where to
begin for those sorts of endeavors.
Perhaps a standalone shim app that acts as a GPU server/resource
allocator, which handles 'locking' issues of who has read/write to a
texture/buffer to alleviate potential conflicts? No app would then
actually own the resources, but would send them and recieve them
from the shim layer? That sounds awefully complicated, and would
require re-thinking of the whole pipeline...
Rather than a shim, the drivers could be modified (with another gl
extension, for example) to abstract this -- they already manage
resources in exactly the same way, so I'm guessing you'd just add some
flags for shareability, and calls to set/control those flags, and be
done with it. I have no if this is reasonable, or if there's already
something like this in place.... </wildSpeculation>
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Quartzcomposer-dev mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/quartzcomposer-dev/email@hidden
This email sent to email@hidden
