i'm working on a ray tracing project that uses runtime compiled C++ kernels on CUDA and OpenCL. To support C++ (or a subset of it) in kernel code, we've used successfully AMD APP SDK (providing the static C++ extension) on windows and linux. While everything works fine on Windows, clBuildProgram under Linux now fails after upgrading to the latest linux64 SDK version:
error: unknown argument: '--c++'
We use "-x clc++" as build option to denote that it is C++ kernel code, we also tried different opencl kernel language versions (see AMD OpenCL 2.0 with C++ static kernel extension?) which did not resolve the problem. With the AMD-APP-SDKInstaller-v126.96.36.199-GA-linux64 package everything was working fine, with AMD-APP-SDKInstaller-v188.8.131.52-GA-linux64.tar.bz2 the C++ compilation does not work anymore.
We now have the following questions:
- Is this a SDK bug or a deliberate removal of a feature?
- Is there any way to get C++ code compiled with the new SDK version on linux64?
- What is AMD's way to go for C++ kernels in the future? OpenCL 2.2 with C++? ROCm · GitHub & HCC?
I remember struggling with a similar driver problem with very limited success (Static C++ support in Current OpenCL Drivers) and was wondering if or when there is new information available...
thanks and lg,