Ailing Zhang

@maximilianmandl What's your pytorch version btw? I've tried it on Colab and it works as expected. https://colab.research.google.com/github/pytorch/pytorch.github.io/blob/master/assets/hub/nvidia_deeplearningexamples_ssd.ipynb#scrollTo=H8caObhcgPqK

@Yihao-Shi On my computer the CPU memory used increased from around 14.6G-> 14.8G in a few minutes, does this match what you see on your end? If so I believe...

@Yihao-Shi Just to double check, what taichi version do you have?

@Yihao-Shi I couldn't repro the stable memory consumption usage when replacing line 14 by `t += 1`, it error out in my case complaining about divide by zero. But I...

@python3kgae FYI we're going to merge https://github.com/taichi-dev/taichi/pull/5846 soon after branchcut for v1.1.3 tmr which might conflict with this PR, would you mind holding on this PR a bit and rebase...

Hey @lucifer1004 ! Thanks for the feature request! IIUC pprintpp-like support is feasible on cpu backend, but for backends like cuda/vulkan printing actually happens on device and we'll mostly limit...

FYI this feature can be added without too much hassel since it's supported on both llvm and spirv based backends. reference: - https://github.com/KhronosGroup/Vulkan-ValidationLayers/blob/master/docs/debug_printf.md - https://docs.nvidia.com/cuda/ptx-writers-guide-to-interoperability/index.html This issue can be a...

Thanks! I've verified that this PR enables loading vulkan validation layer on macos, although printing still doesn't work as my M1 machine doesn't support non semantic info extension :(

Hey @bsavery thanks for reporting the perf regression! We should take a look and bisect the offending commit. cc: @turbo0628

@helloworldstone Yea it's likely related to Vulkan driver shipped in Harmony OS. We've also seen some issues with HarmonyOS so we only tested on android OS for Huawei phones...