Previews, as seen when this build job started (3ffa4b0):
WebGPU | IDL
WGSL
Explainer

Are we sure this won't run into limits on D3D12 and Metal?

D3D12_REQ_CONSTANT_BUFFER_ELEMENT_COUNT is 4096, which means 16K byte size. Has the interpretation of this limit changed?

D3D12_REQ_CONSTANT_BUFFER_ELEMENT_COUNT is 4096, which means 16K byte size. Has the interpretation of this limit changed?

I thought this wasn't an issue from the conclusion here:
#803 (comment)

... but I guess that was a wrong conclusion? (If so the error message is kind of misleading.)

Just tried 4096 elements of vec4, and it seems to pass FXC.

... but I guess that was a wrong conclusion? (If so the error message is kind of misleading.)

Actually, if this stackoverflow answer is correct: https://stackoverflow.com/a/18260016
In D3D11, D3D11_REQ_CONSTANT_BUFFER_ELEMENT_COUNT is actually describing the number of 16-byte elements specifically, not the number of actual individual elements, so it is actually 65536 bytes. (Also, this means the D3D12 error message is not misleading, only the constant name is misleading.)

Double-checked that we can create more than 4K "elements" in a constant buffer - http://shader-playground.timjones.io/09fe5a6eabe2ada04525de0f67f2159f

FXC doesn't seem sufficient to check this though. For example if I add one more field to that struct (so it goes over 64k) it still doesn't complain.

@RafaelCintron can you provide input from the D3D12 side?

@RafaelCintron can you provide input from the D3D12 side?

Yes, this should work from the D3D12 side.

[Editors]
With confirmation from D3D12 and Vulkan (and the fact that Metal doesn't have uniform buffers, or binding limits - only "Maximum buffer length"), this should be safe to land.

(Old metal feature set tables show "Maximum buffer length" as no less than 256MB. Now it's queried dynamically: https://developer.apple.com/documentation/metal/mtldevice/2966563-maxbufferlength)