Re: OpenCL C++ API

Altera_Forum · ‎09-03-2014

We are attempting to port our existing OpenCL application from Windows/Linux/Macintosh to the Cyclone 5 FPGA Development Board. We are using the ARM based cross compiler to compile our source code.

Our source code uses the OpenCL C++ API. I am getting compilation errors that I need to resolve. This is what I have tried so far:


// filename, "main.cpp"# if defined(OPENCL_CPP_API)# include <CL/cl.hpp># else# include <CL/cl.h># endif
int main(void)
{
    return 0;
}

When I issue the following command (i.e., main.cpp uses the OpenCL C API) at the command line, it successfully compiles and produces an object file.

--- Quote Start ---

arm-linux-gnueabihf-g++ -IC:/altera/14.0/hld/host/include -c main.cpp

--- Quote End ---

However, when I issue the following command (i.e., main.cpp uses the OpenCL C++ API) at the command line, the compilation fails.

--- Quote Start ---

arm-linux-gnueabihf-g++ -IC:/altera/14.0/hld/host/include -c main.cpp -DOPENCL_CPP_API

In file included from main.cpp:2:0:

C:/altera/14.0/hld/host/include/CL/cl.hpp:160:19: fatal error: GL/gl.h: No such file or directory

compilation terminated.

--- Quote End ---

Is there a compiler flags that I need to include? Or do I need to install another library? OpenGL perhaps? When we took the Altera training class for OpenCL optimization, the instructor assured me that the OpenCL C++ API is supported for the SoC platform.

Any help would be appreciated.

Altera_Forum · ‎09-03-2014

I found the following "gl.h" files installed on my computer.

C:\altera\14.0\quartus\bin64\cygwin\usr\include\w32api\GL\gl.h

C:\altera\14.0\modelsim_ase\gcc-4.2.1-mingw32vc9\include\GL\gl.h

C:\altera\14.0\embedded\host_tools\cygwin\usr\include\w32api\GL\gl.h

C:\altera\14.0\embedded\host_tools\cygwin\usr\x86_64-w64-mingw32\sys-root\mingw\include\GL\gl.h

Do I include any of these directories when compiling? And if so, which one? I am using the following compiler:

--- Quote Start ---

C:\>arm-linux-gnueabihf-g++ --version

arm-linux-gnueabihf-g++.exe (crosstool-NG linaro-1.13.1-4.7-2013.03-20130313 - Linaro GCC 2013.03) 4.7.3 20130226 (prerelease)

This is free software; see the source for copying conditions. There is NO

warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

--- Quote End ---

Altera_Forum · ‎09-03-2014

I got a lot further in the process. I still have some outstanding questions I need answered but I figured I will document where I am at right now for someone else having my same problems.

Here is some sample code that I am compiling.


# include <string>
# include <iostream>
# include <sstream>
# include <CL/cl.hpp>
template<typename QueryType, typename ValueType>
void Log(QueryType& a_openclObject, cl_int a_propertyId, const std::string& a_propertyName, ValueType& a_value)
{
    a_openclObject.getInfo(a_propertyId, &a_value);
    std::stringstream os;
    std::string spaces(40 - a_propertyName.size(), ' ');
    os << a_propertyName << ":" << spaces <<  a_value << std::endl;
    std::cout << os.str();
}
# define LOG(openclObject, openclProperty, value) Log(openclObject, openclProperty,# openclProperty, value);
int main(void)
{
    // Find all the platforms on this computer...
    std::vector<cl::Platform> platforms;
    cl::Platform::get(&platforms);
    for (size_t i=0; i<platforms.size(); ++i)
    {
        std::string valueString;
        int valueInt;
        unsigned int valueUint;
        long long valueLongLong;
        LOG(platforms, CL_PLATFORM_VENDOR, valueString);
        // Now find out all the OpenCL devices supported for this platform...
        std::vector<cl::Device> devices;
        platforms.getDevices(CL_DEVICE_TYPE_ALL,&devices);
        for (size_t j=0; j<devices.size(); ++j)
        {
            // Dump out the specific information for this device...
            LOG(devices, CL_DEVICE_VERSION,                     valueString);
            LOG(devices, CL_DRIVER_VERSION,                     valueString);
            LOG(devices, CL_DEVICE_OPENCL_C_VERSION,            valueString);
            LOG(devices, CL_DEVICE_MAX_COMPUTE_UNITS,           valueInt);
            LOG(devices, CL_DEVICE_GLOBAL_MEM_SIZE,             valueLongLong);
            LOG(devices, CL_DEVICE_GLOBAL_MEM_CACHELINE_SIZE,   valueLongLong);
            LOG(devices, CL_DEVICE_LOCAL_MEM_SIZE,              valueLongLong);
            LOG(devices, CL_DEVICE_MAX_CONSTANT_BUFFER_SIZE,    valueLongLong);
            LOG(devices, CL_DEVICE_MAX_CONSTANT_ARGS,           valueLongLong);
            LOG(devices, CL_DEVICE_PREFERRED_VECTOR_WIDTH_CHAR, valueUint);
            LOG(devices, CL_DEVICE_NATIVE_VECTOR_WIDTH_CHAR,    valueUint);
            LOG(devices, CL_DEVICE_ENDIAN_LITTLE,               valueUint);
        }
    }
    
    return 0;
}

Here is the makefile that I used.


all: main.exe
main.exe: main.o
	arm-linux-gnueabihf-g++ -LC:\altera\14.0\hld\board\c5soc\arm32\lib -LC:/altera/14.0/hld/host/arm32/lib -lalteracl  -lalterahalmmd -lalterammdpcie -lelf -lrt -lstdc++ -o main.exe main.o
main.o: main.cpp
	arm-linux-gnueabihf-g++ -IC:/altera/14.0/hld/host/include -IC:\altera\14.0\quartus\bin64\cygwin\usr\include\w32api -c main.cpp
clean:
	rm main.o main.exe

I was able to get passed the compilation error by including this line in the compilation step.

--- Quote Start ---

-IC:\altera\14.0\quartus\bin64\cygwin\usr\include\w32api

--- Quote End ---

Next step was to download the executable to the development board. However, when I ran the executable, I got this error.

--- Quote Start ---

./main.exe: error while loading shared libraries: libalteracl.so: cannot open shared object file: No such file or directory

--- Quote End ---

Oops! I ran the initialization script again.

--- Quote Start ---

root@socfpga:~# source ./init_opencl.sh

Error: could not insert module /home/root/opencl_arm32_rte/board/c5soc/driver/aclsoc_drv.ko: File exists

--- Quote End ---

It seems to fail BUT now my executable runs! Yeah!

--- Quote Start ---

root@socfpga:~# ./main.exe

CL_PLATFORM_VENDOR: Altera Corporation

CL_DEVICE_VERSION: OpenCL 1.0 Altera SDK for OpenCL, Version 14.0

CL_DRIVER_VERSION: 14.0

CL_DEVICE_OPENCL_C_VERSION: OpenCL C 1.0

CL_DEVICE_MAX_COMPUTE_UNITS: 1

CL_DEVICE_GLOBAL_MEM_SIZE: 1073741824

CL_DEVICE_GLOBAL_MEM_CACHELINE_SIZE: 0

CL_DEVICE_LOCAL_MEM_SIZE: 16384

CL_DEVICE_MAX_CONSTANT_BUFFER_SIZE: 268435456

CL_DEVICE_MAX_CONSTANT_ARGS: 8

CL_DEVICE_PREFERRED_VECTOR_WIDTH_CHAR: 4

CL_DEVICE_NATIVE_VECTOR_WIDTH_CHAR: 4

CL_DEVICE_ENDIAN_LITTLE: 1

--- Quote End ---

So my question before is still outstanding. What directory do I include in my compilation step so it pulls in the correct "gl.h" file?

Also, I could not find any of this information in the documentation. I apologize if it is there but I couldn't find it. If it isn't there, you may want to include it for other developers in the future.

Altera_Forum · ‎09-04-2014

dan.j.williams, the Linux distribution that is on the SoC board does not have OpenGL installed. Also, the ARM cross-compiler tool chain that we ship does not have OpenGL support. That's why gl.h is not found. Of course, you probably don't care about OpenGL at all, only C++ OpenCL interface.

The best thing I can think of is to modify cl.hpp to not include gl.h (cl.hpp is created by the Khronos group and we do not modify it). Give it a quick try. I will try to do the same to see if I can get you through quickly.

We did use cl.hpp on non-SoC platform and it works. So that's the only issue you should see with the C++ interface.

Altera_Forum · ‎09-04-2014

OK, I fixed cl.hpp file to not include gl.h. We do not support OpenCL-OpenGL interoperability anyway, so there is no need to include that. Overwrite C:/altera/14.0/hld/host/include/CL/cl.hpp with attached cl.hpp.txt (make sure it's still called cl.hpp, I couldn't attach a file with .hpp extension).

I tested a simple program that includes cl.hpp and it failed as you described before the change and compiles with the change.

You can also use a random gl.hpp that you found, as long as you don't actually use any of the GL-related functions.

We'll fix this in the new release of the tool. Thanks a lot for letting us know!

Altera_Forum · ‎09-05-2014

dsk,

I followed your instructions of replacing the installed "cl.hpp" file with the one you posted here. I was able to compile, link, and run the sample code I have in this post successfully on the FPGA development board.

Thank you.

Here is the modified makefile for those interested.


all: main.exe
main.exe: main.o
	arm-linux-gnueabihf-g++ -LC:\altera\14.0\hld\board\c5soc\arm32\lib -LC:/altera/14.0/hld/host/arm32/lib -lalteracl  -lalterahalmmd -lalterammdpcie -lelf -lrt -lstdc++ -o main.exe main.o
main.o: main.cpp
	arm-linux-gnueabihf-g++ -IC:/altera/14.0/hld/host/include -c main.cpp
clean:
	rm main.o main.exe