python benchmark.py try to allocate cuda memory 9177984325.83 GiB

when I try to run benchmark.py, it raise an error says that RuntimeError: CUDA out of memory. Tried to allocate 9177984325.83 GiB (GPU 0; 5.93 GiB total capacity; 78.25 MiB already allocated; 4.44 GiB free; 0 bytes cached) (malloc at /pytorch/aten/src/THC/THCCachingAllocator.cpp:231)

here is my environments: pytorch 1.0 cuda9.0

Is something wrong? thanks a lot.

[hailsham]

hello! sorry for delay. I find the solution to problem of python setup.py build, according to this, I comment #include <torch/extension.h> in Correlation_Module/correlation_cuda_kernel.cu and replace it with #include <torch/types.h>, then it works fine. but it still try to allocate a lot gigabytes when I run python benchmark.py cuda, I will keep looking for the solution. thanks.

[XiaoXueHou]

@hailsham 朋友你好！很抱歉时隔这么久还在这个Issue中打扰你~我在安装“spatial-correlation-sampler”库时遇到了比较棘手的问题，然后我发现你当时的environments和我的很相似。所以冒昧打扰，希望能够得到你的指点和帮助。我的environments是：Python3.6、Pytorch1.0.1、CUDA9.0、gcc5.4.0，当我使用“pip install spatial-correlation-sampler”进行安装时，会报错如图1。当我使用"python setup.py install"进行安装时，会报错如图2。我也发布了新的Issue以期获得作者的帮助，但是可能是由于问题表达不够清晰，作者提出的方案未能解决我的问题。在寻找解决办法时，无意发现了我们environments比较相似，而且国内朋友可以更清晰地阐明问题。期待能够得到你的帮助~ 图①： 图②：