Can this version be compiled by CUDA 8.0? about pspnet HOT 12 CLOSED

Hello, I try to modify this repository to support cuDNN v5 and CUDA 8.0.
Here is my modified code: https://github.com/BassyKuo/PSPNET-cudnn5
It can be built on my server environment, so maybe you can try to install this one.
If you have any trouble when you build it, please feel free to tell me. Thank you:)

from pspnet.

@bearpaw I have the same problem as follows:

./include/caffe/common.cuh(9): error: function "atomicAdd(double *, double)" has already been defined

1 error detected in the compilation of "/tmp/tmpxft_000057f4_00000000-5_interp.cpp4.ii".
Makefile:588: recipe for target '.build_release/cuda/src/caffe/util/interp.o' failed
make: *** [.build_release/cuda/src/caffe/util/interp.o] Error 1
make: *** Waiting for unfinished jobs....
nvcc warning : The 'compute_20', 'sm_20', and 'sm_21' architectures are deprecated, and may be removed in a future release (Use -Wno-deprecated-gpu-targets to suppress warning).
./include/caffe/common.cuh(9): error: function "atomicAdd(double *, double)" has already been defined

1 error detected in the compilation of "/tmp/tmpxft_00005856_00000000-5_domain_transform_forward_only_layer.cpp4.ii".
Makefile:588: recipe for target '.build_release/cuda/src/caffe/layers/domain_transform_forward_only_layer.o' failed
make: *** [.build_release/cuda/src/caffe/layers/domain_transform_forward_only_layer.o] Error 1
nvcc warning : The 'compute_20', 'sm_20', and 'sm_21' architectures are deprecated, and may be removed in a future release (Use -Wno-deprecated-gpu-targets to suppress warning).
./include/caffe/common.cuh(9): error: function "atomicAdd(double *, double)" has already been defined

1 error detected in the compilation of "/tmp/tmpxft_00005814_00000000-5_domain_transform_layer.cpp4.ii".
Makefile:588: recipe for target '.build_release/cuda/src/caffe/layers/domain_transform_layer.o' failed
make: *** [.build_release/cuda/src/caffe/layers/domain_transform_layer.o] Error 1

have you solved it?

from pspnet.

@IvyGongoogle Hi. I just went back to CUDA-7.5. I think this code is not ready for CUDA-8.0.

from pspnet.

@bearpaw
It sounds bad. thank you very much~

from pspnet.

@BassyKuo nice! thank you.

from pspnet.

Hi @BassyKuo
Thanks for your work!

Few more questions: Have you tried to train or test with CUDA 8.0/cudnn v5? Are the results consistent with CUDA 7.5/cudnn v4?

from pspnet.

@bearpaw
I have not tested the consistency because I haven't installed CUDA7.5 and cudnn v4 currently.
Maybe I can compare the result of original repository with the version I modified by turning off the cudnn usage.
Thanks for your feedback! :D

from pspnet.

good work!!!

from pspnet.

@BassyKuo your work helps me a lot,thank you~

from pspnet.

I also got this problem, nvcc8 is not support:
#include <math.h> or
#include <cmath>
using std:isnan
using std:isinf

error:
NVCC src/caffe/layers/contrastive_loss_layer.cu
nvcc warning : The 'compute_20', 'sm_20', and 'sm_21' architectures are deprecated, and may be removed in a future release (Use -Wno-deprecated-gpu-targets to suppress warning).
nvcc warning : The 'compute_20', 'sm_20', and 'sm_21' architectures are deprecated, and may be removed in a future release (Use -Wno-deprecated-gpu-targets to suppress warning).
./include/caffe/common.hpp(85): error: namespace "std" has no member "isnan"

./include/caffe/common.hpp(86): error: namespace "std" has no member "isinf"

nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2016 NVIDIA Corporation
Built on Tue_Jan_10_13:22:03_CST_2017
Cuda compilation tools, release 8.0, V8.0.61

uname -a
Linux cuda 4.13.0-21-generic #24-Ubuntu SMP Mon Dec 18 17:29:16 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux

gcc version 5.4.1 20171010 (Ubuntu 5.5.0-1ubuntu1)

from pspnet.

@bearpaw 您好！请问您当时复现PSP用的是cuda7.5+cudnn4.0嘛？我现在想复现PSAnet,一直在报各种错误。。

from pspnet.

@EchoAmor Yes. But it's a too long time ago. I can't remember the details. You should follow #46 (comment) to solve your problem.

from pspnet.