build error using gcc 5

[Jacob]

I am using GCC 5.x to build caffe. I have used the work around in https://gist.github.com/wangruohui/679b05fcd1466bb0937f

I get the following error message below. I have added cudnn and python layer support from the default makefile. When I make cafe with debug I get no error message and I am able to train networks successfully.

...

NVCC src/caffe/layers/scale_layer.cu

NVCC src/caffe/layers/cudnn_sigmoid_layer.cu

NVCC src/caffe/layers/base_data_layer.cu

NVCC src/caffe/layers/sigmoid_cross_entropy_loss_layer.cu

NVCC src/caffe/layers/split_layer.cu

NVCC src/caffe/layers/hdf5_output_layer.cu

NVCC src/caffe/layers/crop_layer.cu

NVCC src/caffe/layers/hdf5_data_layer.cu

/usr/include/string.h: In function ‘void* __mempcpy_inline(void*, const void*, size_t)’:

/usr/include/string.h:652:42: error: ‘memcpy’ was not declared in this scope

   return (char *) memcpy (__dest, __src, __n) + __n;

                                          ^

Makefile:586: recipe for target '.build_release/cuda/src/caffe/layers/cudnn_sigmoid_layer.o' failed

make: *** [.build_release/cuda/src/caffe/layers/cudnn_sigmoid_layer.o] Error 1

make: *** Waiting for unfinished jobs....

/usr/include/string.h: In function ‘void* __mempcpy_inline(void*, const void*, size_t)’:

/usr/include/string.h:652:42: error: ‘memcpy’ was not declared in this scope

   return (char *) memcpy (__dest, __src, __n) + __n;

                                          ^

Makefile:586: recipe for target '.build_release/cuda/src/caffe/layers/scale_layer.o' failed

make: *** [.build_release/cuda/src/caffe/layers/scale_layer.o] Error 1

/usr/include/string.h: In function ‘void* __mempcpy_inline(void*, const void*, size_t)’:

/usr/include/string.h:652:42: error: ‘memcpy’ was not declared in this scope

   return (char *) memcpy (__dest, __src, __n) + __n;

                                          ^

Makefile:586: recipe for target '.build_release/cuda/src/caffe/layers/sigmoid_cross_entropy_loss_layer.o' failed

make: *** [.build_release/cuda/src/caffe/layers/sigmoid_cross_entropy_loss_layer.o] Error 1

/usr/include/string.h: In function ‘void* __mempcpy_inline(void*, const void*, size_t)’:

/usr/include/string.h:652:42: error: ‘memcpy’ was not declared in this scope

   return (char *) memcpy (__dest, __src, __n) + __n;

                                          ^

Makefile:586: recipe for target '.build_release/cuda/src/caffe/layers/base_data_layer.o' failed

make: *** [.build_release/cuda/src/caffe/layers/base_data_layer.o] Error 1

/usr/include/string.h: In function ‘void* __mempcpy_inline(void*, const void*, size_t)’:

/usr/include/string.h:652:42: error: ‘memcpy’ was not declared in this scope

   return (char *) memcpy (__dest, __src, __n) + __n;

                                          ^

Makefile:586: recipe for target '.build_release/cuda/src/caffe/layers/split_layer.o' failed

make: *** [.build_release/cuda/src/caffe/layers/split_layer.o] Error 1

/usr/include/string.h: In function ‘void* __mempcpy_inline(void*, const void*, size_t)’:

/usr/include/string.h:652:42: error: ‘memcpy’ was not declared in this scope

   return (char *) memcpy (__dest, __src, __n) + __n;

                                          ^

Makefile:586: recipe for target '.build_release/cuda/src/caffe/layers/hdf5_output_layer.o' failed

make: *** [.build_release/cuda/src/caffe/layers/hdf5_output_layer.o] Error 1

/usr/include/string.h: In function ‘void* __mempcpy_inline(void*, const void*, size_t)’:

/usr/include/string.h:652:42: error: ‘memcpy’ was not declared in this scope

   return (char *) memcpy (__dest, __src, __n) + __n;

                                          ^

Makefile:586: recipe for target '.build_release/cuda/src/caffe/layers/crop_layer.o' failed

make: *** [.build_release/cuda/src/caffe/layers/crop_layer.o] Error 1

/usr/include/string.h: In function ‘void* __mempcpy_inline(void*, const void*, size_t)’:

/usr/include/string.h:652:42: error: ‘memcpy’ was not declared in this scope

   return (char *) memcpy (__dest, __src, __n) + __n;

                                          ^

Makefile:586: recipe for target '.build_release/cuda/src/caffe/layers/hdf5_data_layer.o' failed

make: *** [.build_release/cuda/src/caffe/layers/hdf5_data_layer.o] Error 1

[Jicheng Wang]

I have the same problems. But when I Add set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -D_FORCE_INLINES") to the front of CMakeLists.txt, It works!!!

在 2016年4月12日星期二 UTC+8上午11:54:58，Jacob写道：

[Haoyu Y]

hi Jacob,

I encountered the same problem in the latest ubuntu 16.04 with gcc version 5.3. Have you solved the problem?

[Jicheng Wang]

Try to add this line  "set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -D_FORCE_INLINES")" to the front of CMakefile.txt and use the CMake to build caffe. It should solve this problem. I also use the Ubuntu 16.04 and gcc 5.3

在 2016年4月22日星期五 UTC+8上午1:03:27，Haoyu Y写道：

[Jacob]

I initially thought Jicheng solution worked but it was a false alarm. I have yet to find a way around it.

A simple hack is to comment out all the debug print statement in the .cpp layers before the build and compile it with debug mode.

[Luke Yeager]

Worked for me too, thanks!

[Haoyu Y]

Thanks Jicheng, that works. I think currently we can only use cmake to compile caffe with gcc > 5.0.

[Osman Tursun]

I also the Ubuntu 16.04 and gcc 5.3, but the solution did not work. Is there a any other solution ?

[Kujtim Rahmani]

try this http://stackoverflow.com/questions/37083936/keras-failed-to-compile-with-theano-backend 

you have to cmment some lines of code in /usr/include/string.h

[Werner Van Belle]

Hello,

The FORCE_INLINES idea was a good one. However, to get it to work without using cmake you have to modify the Makefile instead:

Add to the end of COMMON_FLAGS (Makefile line ~42): -D_FORCE_INLINES

With kind regards,

Werner,-

[Shane nunya]

Thanks Werner, I only have a Makefile to deal with and your fix was what I needed.

For anyone else finding this on google
NVIDIA_CUDA-7.5_Samples/6_Advanced/shfl_scan/Makefile:149

CCFLAGS += -D_FORCE_INLINES

 

[Fra Rossi]

It works for me too!

(Ubuntu 16 gcc 5.5)

Thanks