Arseny Kapoulkine

@mingqiusun For my workload, byte-shift is "intuitively" a correct operation - it's just a natural implementation of the algorithm. Word-shift is a workaround for the lack of byte-shifts. I don't...

@abrown Yeah - this is a valid concern. A couple of thoughts on this: I$/D$ split is only at L1 level, so if you store the constants inline, and we...

> on the one hand, the V8 10-instruction code sequence seems bad, but it is predictably bad > on the other hand, the sequences you wrote out above [are...] occasionally...

Speaking of shifts by constants, I ran into an issue with v8 lowering non-byte shifts, which is very naive right now. I filed https://bugs.chromium.org/p/v8/issues/detail?id=10115 for this. I am assuming this...

Since this is fixed in v8 I think we can close this?

@dtig Absolutely, I will note this in the Instructions.md list - in a lot of cases the codegen is probably close to optimal but there are definitely some fallouts that...

I also like the current conventions just fine, but will also be happy with explicit `_s` suffix. However note that the intrinsics also include things like `v8x16` prefixes that are...

> Provide samples with Basis/KTX2 for color textures. Note mipmaps will have to be done manually for now. Can you expand on what you mean by "manually"? I can generate...

> My understanding is that the ktx2 branch of KTX-Software doesn't yet generate mipmaps Ah - I see. gltfpack doesn't use toktx/libktx because honestly these tools didn't seem ready. It...

Here's an example model, converted using `-tc` command line flag (needs `basisu` executable in path): [Avocado.zip](https://github.com/KhronosGroup/glTF/files/4132437/Avocado.zip)