Printable View
Does anyone can put a bit light on Cuda and Boost libraries function. Here I am trying to work on a host function and kernel where I want integrate the same with a similar project. I am not able to figure out the working of the project where I am getting error. In this I had added some functionality by using Boost Libraries. But I am not able to compile a program for the same.
Compiling Boost requires the use of bjam, downloadable from Soundforge. The executable must be accessible from the command line. Once downloaded, the archive must be decompressed Boost, for example in / home / user / local / src or c: \. Using the command line, move into that folder. Boost can be compiled with several tools, including gcc and Microsoft Visual Studio. To do this we must add the option - toolset = msvc Windows, Linux, gcc is the compiler default.
The main use of Boost is the form of header files it offers. These files are in the subfolder boost. If you have installed Boost in another sub-folder (Windows or Linux), you must add this folder include subfolder in the environment variable INCLUDE. If you are using Visual Studio, add it in the folder list header options in the IDE. If you use the Boost libraries compiled, do the same with the bin to add to the PATH on Windows and add the lib to LD_LIBRARY_PATH under Linux.
NVIDIA has started to distribute its CUDA Toolkit 3.0 and CUDA SDK 3.0 its developers. It seems to have two objectives: improve support for OpenCL and start preparing for the arrival of products from architectural Fermi . Here in that version you will get a opportunity to learn more about 195.39 drivers that NVIDIA had recently online, including the improvements OpenCL. Try that out.
CUDA API is kind of high level, that is to say it is globally well as hardware abstraction take into account its specification is required to obtain an attractive return. For its part, with the MLC, AMD offers a low-level API. Roughly CUDA makes it possible to develop more easily, whereas MLC can better optimize the code. The CUDA driver handles the role of intermediary between the compiled code and the GPU.
The CUDA runtime is itself an intermediary between the developer and the driver that facilitates the development by hiding some details. CUDA offers either through the API or runtime direct access to the API driver. You can see the runtime API as high-level language API and driver as an intermediary between the high and low to optimize the code manually in more depth. In the opposite direction, AMD can write kernels in HLSL instead of a machine language, so as to facilitate development.
There might be some issue related to the compilation of the program. Try to get some latest toolkit and do a bit study on the same. The 3.0 toolkit of CUDA has the ability to control compilation optimizations via cl_nv_compiler_options. It supports 32-bit atomic operations and 1.0.48 of Khronos specifications OpenCL. So there might be something lacking in your system.