#7 [SOLVED] Double precision on AMD GPUs in version 1.2.0

closed-fixed
nobody
Other/Misc (7)
5
2012-08-20
2012-02-13
Karl Rupp
No

On some AMD GPUs, the double precision enabling pragma is set incorrectly.
In order to fix it, replace the function make_double_kernel() in viennacl/tools/tools.hpp, around line 150, by the following:

inline std::string make_double_kernel(std::string const & source,
std::string const & fp_extension)
{
std::stringstream ss;
ss << "#pragma OPENCL EXTENSION " << fp_extension << " : enable\n\n";

std::string result = ss.str();
result.append(strReplace(source, "float", "double"));
return result;
}

Discussion

  • Karl Rupp
    Karl Rupp
    2012-02-13

    • summary: Double precision on AMD GPUs in version 1.2.0 --> [SOLVED] Double precision on AMD GPUs in version 1.2.0
    • status: open --> open-fixed
     
  • Karl Rupp
    Karl Rupp
    2012-08-20

    • status: open-fixed --> closed-fixed