Follow-up from "Draft: panvk: Implement queue_submit"
The following discussions from !64 should be addressed:
-
@nanokatze started a discussion: (+1 comment) Looks like we're not using pool counters yet but I'm wondering do we actually need this? Vulkan allows allocations past requested size to succeed (and it's not undefined behavior so well-written apps can opportunistically use this) so pPoolSizes in practice is really just a really gymnastical way of specifying pool size in bytes.
-
@nanokatze started a discussion: (+1 comment) I'd like second opinion from e.g. Boris on this but perhaps this would be nicer:
cmdbuf->streams[i].builder = malloc(sizeof *cmdbuf->streams[i].builder);
-
@nanokatze started a discussion: (+1 comment) struct panvk_descriptor_set_binding_layout bindings[];
-
@nanokatze started a discussion: (+1 comment) unsigned char sha1[SHA1_DIGEST_LEN];
but also I think we should prefer to use BLAKE3 in all contexts where we'd previously use SHA1, as BLAKE3 is literally free performance, especially when it comes to hashing shaders, so at some point we'll want to upgrade this to BLAKE3 and do
blake3_hash hash;