🐛 Bug
In #49980 "unused" imports were removed, which also killed import pycuda.autoinit.
If I understand the docs correctly, the import itself is setting up the context and thus this import might have been detected as unused by the used tool.
Currently the test breaks with:
======================================================================
ERROR: test_alexnet (__main__.Test_PT_ONNX_TRT)
----------------------------------------------------------------------
Traceback (most recent call last):
File "/opt/pytorch/pytorch/caffe2/python/trt/test_pt_onnx_trt.py", line 112, in test_alexnet
self._test_model("alexnet", (3, 227, 227))
File "/opt/pytorch/pytorch/caffe2/python/trt/test_pt_onnx_trt.py", line 91, in _test_model
h_input, d_input, h_output, d_output, stream = allocate_buffers(engine)
File "/opt/pytorch/pytorch/caffe2/python/trt/test_pt_onnx_trt.py", line 30, in allocate_buffers
h_input = cuda.pagelocked_empty(trt.volume(engine.get_binding_shape(0)),
pycuda._driver.LogicError: explicit_context_dependent failed: invalid device context - no currently active context?
======================================================================
ERROR: test_googlenet (__main__.Test_PT_ONNX_TRT)
----------------------------------------------------------------------
Traceback (most recent call last):
File "/opt/pytorch/pytorch/caffe2/python/trt/test_pt_onnx_trt.py", line 138, in test_googlenet
self._test_model("googlenet")
File "/opt/pytorch/pytorch/caffe2/python/trt/test_pt_onnx_trt.py", line 91, in _test_model
h_input, d_input, h_output, d_output, stream = allocate_buffers(engine)
File "/opt/pytorch/pytorch/caffe2/python/trt/test_pt_onnx_trt.py", line 30, in allocate_buffers
h_input = cuda.pagelocked_empty(trt.volume(engine.get_binding_shape(0)),
pycuda._driver.LogicError: explicit_context_dependent failed: invalid device context - no currently active context?
[...]for all models.
To Reproduce
# install pycuda for the test
pip install pycuda
# building OSS release of ONNX parser on top of just installed
git clone https://github.com/onnx/onnx-tensorrt.git
cd onnx-tensorrt/
git submodule update --init --recursive
mkdir build
cd build
cmake ..
CPLUS_INCLUDE_PATH=/usr/local/cuda/include make -j$(nproc)
cp libnvonnxparser.so* /usr/lib/x86_64-linux-gnu
cd ../../
rm -rf onnx-tensorrt/
python caffe2/python/trt/test_pt_onnx_trt.py
Environment
PyTorch version: 1.8.0a0+52ea372
Is debug build: False
CUDA used to build PyTorch: 11.2
ROCM used to build PyTorch: N/A
OS: Ubuntu 20.04.1 LTS (x86_64)
GCC version: (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0
Clang version: Could not collect
CMake version: version 3.19.3
Python version: 3.8 (64-bit runtime)
Is CUDA available: True
CUDA runtime version: Could not collect
GPU models and configuration:
GPU 0: A100-SXM4-40GB
GPU 1: A100-SXM4-40GB
GPU 2: A100-SXM4-40GB
GPU 3: A100-SXM4-40GB
GPU 4: A100-SXM4-40GB
GPU 5: A100-SXM4-40GB
GPU 6: A100-SXM4-40GB
GPU 7: A100-SXM4-40GB
Nvidia driver version: 460.32.03
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.8.1.0
/usr/lib/x86_64-linux-gnu/libcudnn_adv_infer.so.8.1.0
/usr/lib/x86_64-linux-gnu/libcudnn_adv_train.so.8.1.0
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_infer.so.8.1.0
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_train.so.8.1.0
/usr/lib/x86_64-linux-gnu/libcudnn_ops_infer.so.8.1.0
/usr/lib/x86_64-linux-gnu/libcudnn_ops_train.so.8.1.0
HIP runtime version: N/A
MIOpen runtime version: N/A
Versions of relevant libraries:
[pip3] numpy==1.19.2
[pip3] pytorch-transformers==1.1.0
[pip3] torch==1.8.0a0+52ea372
[pip3] torchtext==0.9.0a0
[pip3] torchvision==0.9.0a0
[conda] magma-cuda110 2.5.2 5 local
[conda] mkl 2019.4 243
[conda] mkl-include 2019.4 243
[conda] nomkl 3.0 0
[conda] numpy 1.19.2 py38h6163131_0
[conda] numpy-base 1.19.2 py38h75fe3a5_0
[conda] pytorch-transformers 1.1.0 pypi_0 pypi
[conda] torch 1.8.0a0+52ea372 pypi_0 pypi
[conda] torchtext 0.9.0a0 pypi_0 pypi
[conda] torchvision 0.9.0a0 pypi_0 pypi
CC @r-barnes
I'll create a PR to add this import back, as it fixes the issue locally, but let me know if this is not the right approach (I'm not deeply familiar with pycuda).
cc @houseroad @spandantiwari @lara-hdr @BowenBao @neginraoof
🐛 Bug
In #49980 "unused" imports were removed, which also killed
import pycuda.autoinit.If I understand the docs correctly, the import itself is setting up the context and thus this import might have been detected as unused by the used tool.
Currently the test breaks with:
for all models.
To Reproduce
Environment
CC @r-barnes
I'll create a PR to add this import back, as it fixes the issue locally, but let me know if this is not the right approach (I'm not deeply familiar with
pycuda).cc @houseroad @spandantiwari @lara-hdr @BowenBao @neginraoof