Yan, Shaobo

@vladmandic I've created a simple sample to try to reproduce the perf difference but the perf result shows that: - average 'await tensor.data()'on 500 runs: ~53.23ms - average 'tensor.dataSync()' on...

@vladmandic Thanks for trying and the detailed data! But I still has concern due to the previous model average time. I'm a bit suspect that the async download might cause...

>b) its due to webgl pipeline also performing timer checks which interferes with browser handling of ticks. @vladmandic Thanks. I misunderstand your tests. I think this reason resolves my concern.

Hi @cvan, I'd like to translate English content to Chinese. And if WebVR Explainer need similar translation, I'd like to offer my help too :)

Although ICC profiles are arbitrary complex, do we have chance to list all of steps A-B color space conversion requires? Per my understanding, color space conversion always use a reference...

> @kainino0x just reminded me, we already have GPUQueue.copyExternalImageToTexture(). The problem of `GPUQueue.copyExternalImageToTexture()` is that we might lose the 1-copy chance due to implementation(Assume it could accept HTMLVideoElement directly while...

I'm slighty favour of `tryRefresh`. >What happens is that when an external texture is first imported, or after a tryRefresh, a new task is enqueued on a high-priority queue to...

@msiglreith thx for your feedback! Here're some comments: > How is this different from Vulkan here? For Vulkan, if high level shader language is GLSL, the push constants is default...

@msiglreith Thx for explaination.

The discussion about the max size of push constants reminds me the shared memory and L1/L2/L3 cache in GPU. I agree with started from a small size and that's how...