[Bug] SDK无法支持多卡GPU

[ChrisKong93]

Checklist

• I have searched related issues but cannot get the expected help.
• 2. I have read the FAQ documentation but cannot get the expected help.
• 3. The bug has not been fixed in the latest version.

Describe the bug

我想在一个python代码中，将模型同时加载到两个GPU上，循环让两个GPU进行推理，第一次可以推理成功，第二次就报错了

Reproduction

主要代码如下：
`gpu_count = len(gpus_id)

try:
for i in range(len(gpus_id)):
gpu_id = int(gpus_id[i])
print(gpu_id)
model_path = "'./resnet50{}'".format(i)
exec('classifier{} = Classifier(model_path= {} ,device_name= {}, device_id = {})'.format(
i, model_path, "'cuda'", gpu_id))
except RuntimeError as e:
classifier = Classifier(model_path='./resnet50', device_name='cpu', device_id=0)`

Environment

08/28 09:30:11 - mmengine - INFO - 

08/28 09:30:11 - mmengine - INFO - **********Environmental information**********
08/28 09:30:11 - mmengine - INFO - sys.platform: linux
08/28 09:30:11 - mmengine - INFO - Python: 3.8.17 (default, Jul  5 2023, 21:04:15) [GCC 11.2.0]
08/28 09:30:11 - mmengine - INFO - CUDA available: True
08/28 09:30:11 - mmengine - INFO - GPU 0,1: NVIDIA GeForce RTX 3090
08/28 09:30:11 - mmengine - INFO - CUDA_HOME: /usr/local/cuda-11.3
08/28 09:30:11 - mmengine - INFO - NVCC: Cuda compilation tools, release 11.3, V11.3.109
08/28 09:30:11 - mmengine - INFO - GCC: gcc (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0
08/28 09:30:11 - mmengine - INFO - PyTorch: 1.12.1
08/28 09:30:11 - mmengine - INFO - PyTorch compiling details: PyTorch built with:
  - GCC 9.3
  - C++ Version: 201402
  - Intel(R) oneAPI Math Kernel Library Version 2021.4-Product Build 20210904 for Intel(R) 64 architecture applications
  - Intel(R) MKL-DNN v2.6.0 (Git Hash 52b5f107dd9cf10910aaa19cb47f3abf9b349815)
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - LAPACK is enabled (usually provided by MKL)
  - NNPACK is enabled
  - CPU capability usage: AVX2
  - CUDA Runtime 11.3
  - 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_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_37,code=compute_37
  - CuDNN 8.3.2  (built against CUDA 11.5)
  - Magma 2.5.2
  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.3, CUDNN_VERSION=8.3.2, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -fabi-version=11 -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-unused-parameter -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 -Werror=cast-function-type -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.12.1, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=OFF, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, 

08/28 09:30:11 - mmengine - INFO - TorchVision: 0.13.1
08/28 09:30:11 - mmengine - INFO - OpenCV: 4.8.0
08/28 09:30:11 - mmengine - INFO - MMCV: 1.5.1
08/28 09:30:11 - mmengine - INFO - MMCV Compiler: GCC 7.5
08/28 09:30:11 - mmengine - INFO - MMCV CUDA Compiler: 11.3
08/28 09:30:11 - mmengine - INFO - MMDeploy: 1.2.0+553f9b8
08/28 09:30:11 - mmengine - INFO - 

08/28 09:30:11 - mmengine - INFO - **********Backend information**********
08/28 09:30:11 - mmengine - INFO - tensorrt:    None
08/28 09:30:11 - mmengine - INFO - ONNXRuntime: None
08/28 09:30:11 - mmengine - INFO - ONNXRuntime-gpu:     1.8.1
08/28 09:30:11 - mmengine - INFO - ONNXRuntime custom ops:      Available
08/28 09:30:11 - mmengine - INFO - pplnn:       None
08/28 09:30:11 - mmengine - INFO - ncnn:        None
08/28 09:30:11 - mmengine - INFO - snpe:        None
08/28 09:30:11 - mmengine - INFO - openvino:    None
08/28 09:30:11 - mmengine - INFO - torchscript: 1.12.1
08/28 09:30:11 - mmengine - INFO - torchscript custom ops:      NotAvailable
08/28 09:30:11 - mmengine - INFO - rknn-toolkit:        None
08/28 09:30:11 - mmengine - INFO - rknn-toolkit2:       None
08/28 09:30:11 - mmengine - INFO - ascend:      None
08/28 09:30:11 - mmengine - INFO - coreml:      None
08/28 09:30:11 - mmengine - INFO - tvm: None
08/28 09:30:11 - mmengine - INFO - vacc:        None
08/28 09:30:11 - mmengine - INFO - 

08/28 09:30:11 - mmengine - INFO - **********Codebase information**********
08/28 09:30:11 - mmengine - INFO - mmdet:       None
08/28 09:30:11 - mmengine - INFO - mmseg:       None
08/28 09:30:11 - mmengine - INFO - mmpretrain:  None
08/28 09:30:11 - mmengine - INFO - mmocr:       None
08/28 09:30:11 - mmengine - INFO - mmagic:      None
08/28 09:30:11 - mmengine - INFO - mmdet3d:     None
08/28 09:30:11 - mmengine - INFO - mmpose:      None
08/28 09:30:11 - mmengine - INFO - mmrotate:    None
08/28 09:30:11 - mmengine - INFO - mmaction:    None
08/28 09:30:11 - mmengine - INFO - mmrazor:     None
08/28 09:30:11 - mmengine - INFO - mmyolo:      None

Error traceback

[ERROR][2023-08-28 09:35:33.983][resize.cu:1202] CUDA error: invalid resource handle
Aborted (core dumped)

[irexyc]

@ChrisKong93

device 管理的有点问题，我们会在下一版修复，目前的话，你可以先用 cudaSetDevice 绑一下线程和device.

如果你用多线程的，并且线程不会切换device使用的话，绑一次就可以了

pip install cuda-python==11.5

from mmdeploy_runtime import Classifier
import cv2
import numpy as np
from cuda import cudart

img = cv2.imread('/root/workspace/mmpretrain/demo/demo.JPEG')

model = []
for i in range(2):
    model.append(Classifier('/root/workspace/mmdeploy/work-dir/ort', 'cuda', i))

while True:
    for i in range(2):
        cudart.cudaSetDevice(i)
        res = model[i](img)
        print(res)

[ChrisKong93]

@ChrisKong93

device 管理的有点问题，我们会在下一版修复，目前的话，你可以先用 cudaSetDevice 绑一下线程和device.

如果你用多线程的，并且线程不会切换device使用的话，绑一次就可以了

pip install cuda-python==11.5

from mmdeploy_runtime import Classifier
import cv2
import numpy as np
from cuda import cudart

img = cv2.imread('/root/workspace/mmpretrain/demo/demo.JPEG')

model = []
for i in range(2):
    model.append(Classifier('/root/workspace/mmdeploy/work-dir/ort', 'cuda', i))

while True:
    for i in range(2):
        cudart.cudaSetDevice(i)
        res = model[i](img)
        print(res)

好的，感谢，我试一下这个方法，我用这个办法解决的，效果是达到了，但不知道解决方法是不是正确的

[github-actions]

This issue is closed because it has been stale for 5 days. Please open a new issue if you have similar issues or you have any new updates now.