error in ms_deformable_col2im_cuda: an illegal memory access was encountered

[makifozkanoglu]

Describe the bug
I'm getting the following error when trying to run deformable_detr

Reproduction

1. What command or script did you run?
   I tried to train the config file below

https://github.com/open-mmlab/mmdetection/blob/7a9bc498d5cc972171ec4f7332afcd70bb50e60e/configs/deformable_detr/deformable_detr_r50_16x2_50e_coco.py

2. Did you make any modifications on the code or config? Did you understand what you have modified?
   No I did not modify
3. What dataset did you use?

Environment
sys.platform: linux
Python: 3.8.12 (default, Oct 12 2021, 13:49:34) [GCC 7.5.0]
CUDA available: True
GPU 0: TITAN RTX
CUDA_HOME: /usr/local/cuda-11.0
NVCC: Build cuda_11.0_bu.TC445_37.28845127_0
GCC: gcc (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0
PyTorch: 1.7.0
PyTorch compiling details: PyTorch built with:

• GCC 7.3
• C++ Version: 201402
• Intel(R) Math Kernel Library Version 2020.0.0 Product Build 20191122 for Intel(R) 64 architecture applications
• Intel(R) MKL-DNN v1.6.0 (Git Hash 5ef631a030a6f73131c77892041042805a06064f)
• OpenMP 201511 (a.k.a. OpenMP 4.5)
• NNPACK is enabled
• CPU capability usage: AVX2
• CUDA Runtime 10.2
• NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75
• CuDNN 7.6.5
• Magma 2.5.2
• Build settings: BLAS=MKL, BUILD_TYPE=Release, CXX_FLAGS= -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DUSE_VULKAN_WRAPPER -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-sign-compare -Wno-unused-parameter -Wno-unused-variable -Wno-unused-function -Wno-unused-result -Wno-unused-local-typedefs -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-stringop-overflow, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, USE_CUDA=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON,

TorchVision: 0.8.0
OpenCV: 4.5.5
MMCV: 1.4.4
MMCV Compiler: GCC 7.5
MMCV CUDA Compiler: 11.0
MMDetection: 2.20.0+

torch is installed by pip

Error traceback

error in ms_deformable_col2im_cuda: an illegal memory access was encountered
Traceback (most recent call last):
  File "tools/train.py", line 200, in <module>
    main()
  File "tools/train.py", line 188, in main
    train_detector(
  File "/cta/users/mehmet/CenterNetMMCV/ssod/apis/train.py", line 206, in train_detector
    runner.run(data_loaders, cfg.workflow)
  File "/cta/users/mehmet/CenterNetMMCV/thirdparty/mmcv/mmcv/runner/epoch_based_runner.py", line 127, in run
    epoch_runner(data_loaders[i], **kwargs)
  File "/cta/users/mehmet/CenterNetMMCV/thirdparty/mmcv/mmcv/runner/epoch_based_runner.py", line 51, in train
    self.call_hook('after_train_iter')
  File "/cta/users/mehmet/CenterNetMMCV/thirdparty/mmcv/mmcv/runner/base_runner.py", line 309, in call_hook
    getattr(hook, fn_name)(self)
  File "/cta/users/mehmet/CenterNetMMCV/thirdparty/mmcv/mmcv/runner/hooks/optimizer.py", line 56, in after_train_iter
    runner.outputs['loss'].backward()
  File "/cta/users/mehmet/.conda/envs/centernetmmcv/lib/python3.8/site-packages/torch/tensor.py", line 221, in backward
    torch.autograd.backward(self, gradient, retain_graph, create_graph)
  File "/cta/users/mehmet/.conda/envs/centernetmmcv/lib/python3.8/site-packages/torch/autograd/__init__.py", line 130, in backward
    Variable._execution_engine.run_backward(
RuntimeError: CUDA error: an illegal memory access was encountered.

Bug fix
If you have already identified the reason, you can provide the information here. If you are willing to create a PR to fix it, please also leave a comment here and that would be much appreciated!

[PeterVennerstrom]

Experienced the same issue and tested a few environments and GPU models.

Fixed by using an earlier version of mmcv-full. 1.4.2 is the latest version of mmcv-full that worked for me.

CUDA_LAUNCH_BLOCKING=1 python ./tools/train.py configs/config.....

[imkzh]

Exactly same error:

error in ms_deformable_col2im_cuda: an illegal memory access was encountered
Traceback (most recent call last):
  File "./mmdetection/tools/train.py", line 209, in <module>
    main()
  File "./mmdetection/tools/train.py", line 198, in main
    train_detector(
  File "/home/user/.local/lib/python3.8/site-packages/mmdet/apis/train.py", line 208, in train_detector
    runner.run(data_loaders, cfg.workflow)
  File "/home/user/.local/lib/python3.8/site-packages/mmcv/runner/epoch_based_runner.py", line 127, in run
    epoch_runner(data_loaders[i], **kwargs)
  File "/home/user/.local/lib/python3.8/site-packages/mmcv/runner/epoch_based_runner.py", line 51, in train
    self.call_hook('after_train_iter')
  File "/home/user/.local/lib/python3.8/site-packages/mmcv/runner/base_runner.py", line 309, in call_hook
    getattr(hook, fn_name)(self)
  File "/home/user/.local/lib/python3.8/site-packages/mmcv/runner/hooks/optimizer.py", line 56, in after_train_iter
    runner.outputs['loss'].backward()
  File "/home/user/.local/lib/python3.8/site-packages/torch/_tensor.py", line 255, in backward
    torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs)
  File "/home/user/.local/lib/python3.8/site-packages/torch/autograd/__init__.py", line 147, in backward
    Variable._execution_engine.run_backward(
RuntimeError: CUDA error: an illegal memory access was encountered

I'm on:

• Ubuntu 20.04
• CUDA 11.2 (RTX3090)
• torch 1.9.0+cu111
• mmcv-full 1.4.7
• mmdet 2.22.0
• python 3.8.10
• nvcc V11.2.67
• gcc (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0

P.S.: downgrading mmcv-full to 1.4.2 solved the problem as @PeterVennerstrom mentioned above.

[Manningchan]

i met the same question, and in my environment, there is 8 gpus, if i use 0, it will not happened and if i use other gpus, it will occured

[xuqingyu26]

Hello, i met the same question as you. Have you solved this question?

[imkzh]

@xuqingyu26 a workaround is downgrading mmcv-full to 1.4.2 which solved the problem in my case, as mentioned in my comment.

[xbkaishui]

hi, any update on this?

[PeterVennerstrom]

It was fixed. Here's a link to the issue with a link to the PR.