Merely porting the existing WR device to gfx-hal is just a start. The real fruits will come once we make deeper changes to the WR architecture to take the benefit of modern graphics APIs. This meta-issue is tracking those architectural changes:
[ ] Parallel recording - #188
[ ] Specialization - #211 (Most of the features are behind a spec const now expect dual source blending which requires to many shader changes)
[ ] In-place instance data - #219
[ ] Share descriptor sets between instances and frames - #236 (This is mostly done, which we miss from the issue is upstreaming the concept to WR)
[ ] Pipeline cache - #241 (Vulkan backend already uses this. Metal backend can't write/read a paritcular cache, but internally can manage it between resizes. DX12 backend lacks this feature. )
[x] Better memory management - #255
[ ] Array render passes - #269
[ ] Uber-pass frame rendering - #273
[x] Persistently mapped GPU cache - #271
[ ] Manage descriptor sets on the render backend - #334
Merely porting the existing WR device to gfx-hal is just a start. The real fruits will come once we make deeper changes to the WR architecture to take the benefit of modern graphics APIs. This meta-issue is tracking those architectural changes: