gfx-rs/wgpu

Split `binding_array` Limits into Separate Limit

Opened this issue · 0 comments

Two platforms have vastly different bindless and bindful limits.

Metal has low limits, <128 for directly bounded resources, but tier 2 argument buffers can touch 1M total resources.

Intel pre-xe supports <2000 regularly bound descriptors of each type, but 1M UPDATE_AFTER_BIND descriptors.

We need new limits so that binding arrays can be larger, but we continue enforcing bindful limits.

The exact shape of these limits is up to debate.