mingw build error

[jeremych1000]

Hi,

I've just tried to build using mingw with opencl and tbb included in the library search paths, but have stumbled across this error. Has anyone seen this?

env: mingw64

Happens with CUDA 7.0/lib/Win32, 7.0/lib/x64, 8.0/lib/Win32, and 8.0/lib/x64.

Jeremy@JEREMY-PC /c/Users/Jeremy/Documents/GitHub/hpce-2016-cw5-cyk113
$ make -B all
cd provider && make all
make[1]: Entering directory `/c/Users/Jeremy/Documents/GitHub/hpce-2016-cw5-cyk113/provider'
g++  -std=c++11 -W -Wall  -g -O3 -I ../include -I C:/Program\ Files/NVIDIA\ GPU\ Computing\ Toolkit/CUDA/v8.0/include -I C:/tbb2017_20161004oss/include -ltbb  -c -o puzzles.o puzzles.cpp
In file included from user_random_walk.hpp:10:0,
                 from puzzles.cpp:3:
../include/CL/cl.hpp:242:28: warning: type qualifiers ignored on function return type [-Wignored-qualifiers]
     const cl_int err(void) const { return err_; }
                            ^~~~~
../include/CL/cl.hpp:2421:28: warning: ignoring attributes on template argument 'cl_int {aka int}' [-Wignored-attributes]
         VECTOR_CLASS<cl_int>* binaryStatus = NULL,
                            ^
../include/CL/cl.hpp: In member function 'cl_int cl::CommandQueue::enqueueNativeKernel(void (*)(void*), std::pair<void*, unsigned int>, const std::vector<cl::Memory>*, const std::vector<const void*>*, const std::vector<cl::Event>*, cl::Event*) const':
../include/CL/cl.hpp:2996:34: error: invalid conversion from 'void (*)(void*)' to 'void (__attribute__((__stdcall__)) *)(void*)' [-fpermissive]
                 (cl_event*) event),
                                  ^
<..........>
make[1]: *** [puzzles.o] Error 1
make: *** [lib/libpuzzler.a] Error 2

[m8pple]

You've got a conflict between version of the opencl C++ API - this has been a pain for a while now (see previous complains about OpenCL2 and NVidia).

The problem is the addition of __stdcall, going from here:

void (*)(void*)

to

void (__attribute__((__stdcall__)) *)(void*)

which means that it is specifying a difference calling convention: __stdcall is a different way of pushing function parameters on a stack.

Your solutions are to: 1 - Switch from the version of cl.hpp in the include directory to the one that comes with the CUDA sdk

2 - Switch from the headers that come with OpenCL to the ones that were included in CW3 (copy the opencl_sdk file over).

I would suggest option 2, as that way you get to control the headers that are available in both you MinGW environment, and in AWS.

Note that for linking, I will be doing the compilation under my AWS runs, so it is my job to make sure that -lOpenCL is used and that it will be available.