Ingvar Stepanyan

I suspect this has to do with our checks where we tried to ensure the JsValue specific conversion only works with serde-wasm-bindgen and doesn't try to produce gibberish with other...

Converting to draft because this might prevent some simplifications in the generics PR that we discussed with @guybedford, going to look at it again after that one lands.

Definitely not merging for now, looks like might need more work here.

Out of curiosity also did tests with SAME_LOWER padding and that one works as expected, so it's only SAME_UPPER that is broken in WebGPU & WebGL: wasmwebglwebgpuStride 1, Pad SAME_UPPER(reference)Range:...

> Here's the absolutely minimal onnxscript-based repro. Well, "absolutely" has been a bit of an overstatement. Played a bit more with reducing dimensions so now it's a simple 1x1x3x3 convolution...

Could someone take a look at this? Unlike some other issues I reported, this one is an actual bug not just potential performance improvement.

Just to add - it's not just endianness, probably any customization. I spent a lot of time debugging an issue that stemmed from me only having `ctx = "order: Order"`...

@anchpop A workaround that works today is to convert and store your object as a `JsValue` ahead of time. Since it becomes a regular JS value, it preserves referential equality...

Sure, although that would be more expensive than statically storing converted value only where you know it can be reused.

> ```ts > doProbability(Math.random() * 200); // Well, we tried. > ``` FWIW I'd expect this to fail to be useful for uint8 / ... / uint32 scenarios. As in,...