David Neto
David Neto
Other things in the same situation (look at the description of OpSpecConstantOp, in the "If the Kernel...." section): - Bitcast - FAdd, FSub, FNegate, FMul, FDiv, FRem, FMod (The other...
This change allows things like 5000000000
To be clear, the single-letter form is intended to be supported already. The multi-letter swizzle as l-value is explicitly called out in a note as not supported. I think WGSL...
I separated out the shuffle primitive to #748
> "Those writes are visible to other threads via the usual memory consistency and coherency rules (at present, only atomics will be coherent, and only relaxed consistency is supported)." I...
I looked at how the WASM spec did it but it looked too complicated to do quickly. I'm not sure we have to be that complicated about it.
Right, it's a very strong tradition that tokenization is always greedy. However, with context-aware tokenization (#2717), I think it will be valid for the parser-tokenizer interaction to be like this:...
Heh. It should be tokenized as `1.f` `oo`
If WGSL supports empty structures, someone would have to check the layout rules still make sense, or update them.
SPIR-V has a few [universal limits](https://www.khronos.org/registry/spir-v/specs/unified1/SPIRV.html#_a_id_limits_a_universal_limits) that would clearly translate to the WGSL level. They adjudicate between "author problem vs. implementation bug". It's a way to cut off useless arguing...