Error when compiling the parellel branch

[perfect1]

I'm getting an error when attempting to compile the Parallel branch.
I'm running ubuntu 14.04 x86_64. I'm able to compile the master branch fine, I only have this issue when trying to compile the parallel branch. I've got cudnn enabled, but everything else is pretty standard.

make all
All appears fine until:

/usr/local/cuda/bin/nvcc -ccbin=g++ -Xcompiler -fPIC -DNDEBUG -O2 -DUSE_CUDNN -I/usr/include/python2.7 -I/usr/local/lib/python2.7/dist-packages/numpy/core/include -I/usr/local/include -I.build_release/src -I./src -I./include -I/usr/local/cuda/include -gencode arch=compute_20,code=sm_20 -gencode arch=compute_20,code=sm_21 -gencode arch=compute_30,code=sm_30 -gencode arch=compute_35,code=sm_35  -c src/caffe/parallel.cu -o .build_release/src/caffe/parallel.cuo 2> .build_release/src/caffe/parallel.cuo.warnings.txt \
        || (cat .build_release/src/caffe/parallel.cuo.warnings.txt; exit 1)
src/caffe/parallel.cu(14): error: name followed by "::" must be a class or namespace name
          detected during instantiation of "void caffe::sync_master_kernel(Dtype *, Dtype **, size_t *, int, int, const cudaStream_t &, size_t) [with Dtype=float]" 
(30): here

src/caffe/parallel.cu(48): error: name followed by "::" must be a class or namespace name
          detected during instantiation of "void caffe::sync_worker_kernel(Dtype *, Dtype *, Dtype **, size_t *, Dtype **, uint8_t *, int, int, const cudaStream_t &, size_t) [with Dtype=float]" 
(72): here

2 errors detected in the compilation of "/tmp/tmpxft_00005afb_00000000-12_parallel.compute_35.cpp1.ii".
make: *** [.build_release/src/caffe/parallel.cuo] Error 1

What am I missing?

[melgor]

I have that problem too with parallel branch. I resolve it by changing boost library to 1.57.
Using other version of boost cause similar problem of different from the presented.

[perfect1]

Hmm. That didn't work for me. Did you upgrade any other libraries?
What version is your nvcc?
/usr/local/cuda/bin/nvcc --version

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2014 NVIDIA Corporation
Built on Wed_Aug_27_10:36:36_CDT_2014
Cuda compilation tools, release 6.5, V6.5.16

[melgor]

I have got the same version.
Right, I now remember what I done with that error. I change the source code of src/caffe/parallel.cu
In line 14 and 48 there is "IBChannel::FRAMES", and that variable cause problem. I do not why, I can not get that static value from IBChannel class.
So, I replace that two value by value of FRAMES variable, which is defined in parallel,h
static const int FRAMES = 1024;

After that change I could compile it, but there were errors in runtest. I have not tried to run parallel learning so far too.

[perfect1]

That worked for me, Thanks a lot!
I changed line 14 and 48 to

int q = (batch_start + b) & (1023);

I also got an errors on make runtest,

In file included from src/caffe/test/test_internal_thread.cpp:2:0:
src/caffe/test/test_internal_thread.cpp: In member function ‘virtual void caffe::InternalThreadTest_TestStartAndExit_Test::TestBody()’:
src/caffe/test/test_internal_thread.cpp:18:22: error: ‘class caffe::InternalThread’ has no member named ‘WaitForInternalThreadToExit’
   EXPECT_TRUE(thread.WaitForInternalThreadToExit());
                      ^
./src/gtest/gtest.h:7859:34: note: in definition of macro ‘GTEST_TEST_BOOLEAN_’
       ::testing::AssertionResult(expression)) \
                                  ^
src/caffe/test/test_internal_thread.cpp:18:3: note: in expansion of macro ‘EXPECT_TRUE’
   EXPECT_TRUE(thread.WaitForInternalThreadToExit());
   ^
make: *** [.build_release/src/caffe/test/test_internal_thread.o] Error 1

I'm also getting errors with gpus.bin. But I believe it's probably my configuration.
Don't close this yet, I'll try again tomorrow. I need a break for now. :)

[perfect1]

I'm not having any luck. Is there any form of documentation hiding anywhere?(for parallel)

[melgor]

I have no luck too. I asked the creator of parallel version here:
#1629

He said, that GPU parallel example may be broken.

[shelhamer]

Please comment on the parallel branch on the parallel pull request #1148 to keep the discussion focused.