[Caffe2] compiling error with gcc-6

[CosmicWebCreator]

If you have a question or would like help and support, please ask at our
forums.

If you are submitting a feature request, please preface the title with [feature request].
If you are submitting a bug report, please fill in the following details.

Issue description

When compiling caffe2, the following error occurs

[ 72%] Linking CXX shared library ../lib/libcaffe2.so
[ 72%] Built target caffe2
[ 72%] Building NVCC (Device) object caffe2/CMakeFiles/caffe2_gpu.dir/sgd/caffe2_gpu_generated_yellowfin_op_gpu.cu.o
gcc-6: error: unrecognized command line option ‘-faligned-new’; did you mean ‘-falign-jumps’?
CMake Error at caffe2_gpu_generated_yellowfin_op_gpu.cu.o.Release.cmake:219 (message):
Error generating
/home/mburon/pytorch/build/caffe2/CMakeFiles/caffe2_gpu.dir/sgd/./caffe2_gpu_generated_yellowfin_op_gpu.cu.o

caffe2/CMakeFiles/caffe2_gpu.dir/build.make:917: recipe for target 'caffe2/CMakeFiles/caffe2_gpu.dir/sgd/caffe2_gpu_generated_yellowfin_op_gpu.cu.o' failed
make[2]: *** [caffe2/CMakeFiles/caffe2_gpu.dir/sgd/caffe2_gpu_generated_yellowfin_op_gpu.cu.o] Error 1
CMakeFiles/Makefile2:1442: recipe for target 'caffe2/CMakeFiles/caffe2_gpu.dir/all' failed
make[1]: *** [caffe2/CMakeFiles/caffe2_gpu.dir/all] Error 2
Makefile:140: recipe for target 'all' failed
make: *** [all] Error 2

Provide a short description.

Code example

Please try to provide a minimal example to repro the bug.
Error messages and stack traces are also helpful.

System Info

Please copy and paste the output from our
environment collection script
(or fill out the checklist below manually).

You can get the script and run it with:

wget https://raw.githubusercontent.com/pytorch/pytorch/master/torch/utils/collect_env.py
# For security purposes, please check the contents of collect_env.py before running it.
python collect_env.py

• PyTorch or Caffe2: Caffe2
• How you installed PyTorch (conda, pip, source): cmake of caffe2
• Build command you used (if compiling from source): sudo make install
• OS: Ubuntu 18
• PyTorch version:
• Python version: 2.7
• CUDA/cuDNN version: 9.0 / 7.0.5
• GPU models and configuration: GTX 1070
• GCC version (if compiling from source): gcc-6
• CMake version: 3.10.2
• Versions of any other relevant libraries:

[timsschoen]

I can replicate this issue on Ubuntu 17.10 with CUDA 9.1, compiling with gcc-6.

[BrettRyland]

This issue also occurs on Debian/unstable with CUDA 9.2 compiling with gcc-7 (7.3.0-26) as the main compiler and -DCUDA_HOST_COMPILER=/usr/bin/gcc-6 due to the flag -faligned-new not existing in gcc-6 (6.4.0-18).
Commenting out the appropriate lines in the main CMakeLists.txt allows compilation to proceed:

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 1cb5d13bc..3ae2dcaa8 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -229,11 +229,11 @@ if(NOT MSVC)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-constexpr-not-const")
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-missing-braces")
   endif()
-  if ((APPLE AND (NOT ("${CLANG_VERSION_STRING}" VERSION_LESS "9.0")))
-    OR (CMAKE_COMPILER_IS_GNUCXX 
-    AND (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER 7.0 AND NOT APPLE)))
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -faligned-new")
-  endif()
+  #if ((APPLE AND (NOT ("${CLANG_VERSION_STRING}" VERSION_LESS "9.0")))
+    #OR (CMAKE_COMPILER_IS_GNUCXX 
+    #AND (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER 7.0 AND NOT APPLE)))
+    #set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -faligned-new") # This breaks nvcc using gcc-6
+  #endif()
   if ($ENV{WERROR})
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Werror")
   endif($ENV{WERROR})

[Roger-luo]

Get this error with Ubuntu 18.04 with DCUDA_HOST_COMPILER too

[eLvErDe]

Hello,

I can confirm this issue, the broken snippet is

  if ((APPLE AND (NOT ("${CLANG_VERSION_STRING}" VERSION_LESS "9.0")))
    OR (CMAKE_COMPILER_IS_GNUCXX
    AND (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER 7.0 AND NOT APPLE)))
    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -faligned-new")
  endif()

The test is incorrect because it's CUTA_HOST_CC which is used and not defaults's CC when building with CUDA.