关于深度学习:3090编译Ocean可变形卷积-Compile-Ocean-DCN-on-3090

see https://github.com/Soarkey/tr…

Ubuntu 20
Python 3.7
CUDA 11.4
PyTorch 1.9.0+cu111

找到 lib/models/dcn/src/deform_conv.py 文件, 将所有的 AT_CHECK 替换为 TORCH_CHECK, 同时将所有的.view 操作替换为 .reshape
操作, 之后从新执行编译python setup.py develop.

Found file lib/models/dcn/src/deform_conv.py, replace all AT_CHECK and .view operations with TORCH_CHECK and .reshape, then recompile python setup.py develop.

解决 / Solution

将 conda 环境所在文件夹中的 cpp_extension.py 内容从:

named_arches = collections.OrderedDict([('Kepler+Tesla', '3.7'),
    ('Kepler', '3.5+PTX'),
    ('Maxwell+Tegra', '5.3'),
    ('Maxwell', '5.0;5.2+PTX'),
    ('Pascal', '6.0;6.1+PTX'),
    ('Volta', '7.0+PTX'),
    ('Turing', '7.5+PTX'),
])
supported_arches = ['3.5', '3.7', '5.0', '5.2', '5.3', '6.0', '6.1', '6.2',
                    '7.0', '7.2', '7.5']

改为

named_arches = collections.OrderedDict([('Kepler+Tesla', '3.7'),
    ('Kepler', '3.5+PTX'),
    ('Maxwell+Tegra', '5.3'),
    ('Maxwell', '5.0;5.2+PTX'),
    ('Pascal', '6.0;6.1+PTX'),
    ('Volta', '7.0+PTX'),
    ('Turing', '7.5+PTX'),
    ('Ampere', '8.0;8.6+PTX'),
])
supported_arches = ['3.5', '3.7', '5.0', '5.2', '5.3', '6.0', '6.1', '6.2',
                    '7.0', '7.2', '7.5', '8.0', '8.6']

区别在于：减少了 8.6 的反对, 3090 就是属于 sm86 架构.

see solution: https://blog.csdn.net/ng323/a…)

起因：在导入某些和 pytorch 无关的第三方包时，如果先导入第三方包，容易产生这种谬误，正确的做法是首先导入 pytorch。
see solution: https://blog.csdn.net/slow122…

关于深度学习:3090编译Ocean可变形卷积-Compile-Ocean-DCN-on-3090

1. 环境 / Environment

2. 批改 / Modification

3. 可能碰到的问题及解决 / References

> ValueError: Unknown CUDA arch (8.6) or GPU not supported

> undefined symbol: THPVariableClass