Until now, we would never attempt to suballocate small image heaps.
Some games, at least Nixxes ports, tend to create lots of smaller heaps.
The intent seems to be that the game can use fine-grained residency
priority. On implementations without EXT_pageable (e.g. RADV currently),
we assume a naive allocation model that cannot take advantage of
priorities, so there is no need to consider priority set by
applications. In future, should RADV get support for this EXT,
we may want to revisit these assumptions.
To make image suballocation work better, normalize heap type in allocations.
Fixes a potential problem where two different CUSTOM heaps
can end up allocating from each other. Normalize the CUSTOM heap type
to a standard type and compare those.
Ensure that all suballocated heaps have a GLOBAL buffer, but just
change the explicit usage. This way we don't have to check the heap
flags and RTV heaps and resource heaps can share chunks as long as
their memory types match.
Until now, we would never attempt to suballocate small image heaps. Some games, at least Nixxes ports, tend to create lots of smaller heaps. The intent seems to be that the game can use fine-grained residency priority. On implementations without EXT_pageable (e.g. RADV currently), we assume a naive allocation model that cannot take advantage of priorities, so there is no need to consider priority set by applications. In future, should RADV get support for this EXT, we may want to revisit these assumptions.
To make image suballocation work better, normalize heap type in allocations.
Fixes a potential problem where two different CUSTOM heaps can end up allocating from each other. Normalize the CUSTOM heap type to a standard type and compare those.
Ensure that all suballocated heaps have a GLOBAL buffer, but just change the explicit usage. This way we don't have to check the heap flags and RTV heaps and resource heaps can share chunks as long as their memory types match.