NVIDIA
ScatterND or ScatterElements - Myelin error: myelinTargetSetPropertyMemorySize called with invalid memory size (0)
Description
[W] [TRT] Skipping tactic 0 due to Myelin error: myelinTargetSetPropertyMemorySize called with invalid memory size (0).
[E] Error[10]: [optimizer.cpp::computeCosts::2011] Error Code 10: Internal Error (Could not find any implementation for node {ForeignNode[data...scatternd_1]}.)
[E] Error[2]: [builder.cpp::buildSerializedNetwork::609] Error Code 2: Internal Error (Assertion enginePtr != nullptr failed. )
[E] Engine could not be created from network
Environment
TensorRT Version: TensorRT-8.4.1.5 NVIDIA GPU: GeForce RTX 3080 NVIDIA Driver Version: 460.106.00 CUDA Version: 11.6 CUDNN Version: 8.4 Operating System: Ubuntu 20.04 Python Version (if applicable): Tensorflow Version (if applicable): PyTorch Version (if applicable): Baremetal or Container (if so, version):
Relevant Files
Steps To Reproduce
mv scatternd.onnx.zip scatternd.onnx
./TensorRT-8.4.1.5/bin/trtexec --onnx=scatternd.onnx --saveEngine=scatternd.plan --fp16 --workspace=512 --dumpProfile --dumpLayerInfo --profilingVerbosity=detailed --verbose
I can not reproduce this on official docker image: nvcr.io/nvidia/tensorrt:22.07-py3
[08/15/2022-12:40:45] [I] Starting inference
[08/15/2022-12:40:48] [I] The e2e network timing is not reported since it is inaccurate due to the extra synchronizations when the profiler is enabled.
[08/15/2022-12:40:48] [I] To show e2e network timing report, add --separateProfileRun to profile layer timing in a separate run or remove --dumpProfile to disable the profiler.
[08/15/2022-12:40:48] [I]
[08/15/2022-12:40:48] [I] === Profile (14894 iterations ) ===
[08/15/2022-12:40:48] [I] Layer Time (ms) Avg. Time (ms) Median Time (ms) Time %
[08/15/2022-12:40:48] [I] {ForeignNode[data...scatternd_1]} 256.49 0.0172 0.0165 100.0
[08/15/2022-12:40:48] [I] Total 256.49 0.0172 0.0165 100.0
[08/15/2022-12:40:48] [I]
&&&& PASSED TensorRT.trtexec [TensorRT v8401] # trtexec --onnx=scatternd.onnx --saveEngine=scatternd.plan --fp16 --workspace=512 --dumpProfile --dumpLayerInfo --profilingVerbosity=detailed --verbose
Can you try using our docker images or upgrade your Nvidia driver first？
When I use the docker image nvcr.io/nvidia/tensorrt:22.07-py3, the trtexec works fine. Thank you for your suggestion, maybe the upgrade from TensorRT8.2 to TensorRT8.4 still needs some dependency upgrades. But I don't know which dependencies will affect Myelin Optimization.
closing since no activity for more than 14 days, please reopen if you still have question, thanks!
