internal ManagedCuda.NVRTC.nvrtcResult LoadKernel(out string log)
{
string path = "MyKernels.c";
ManagedCuda.NVRTC.nvrtcResult result;
using (var rtc = new ManagedCuda.NVRTC.CudaRuntimeCompiler(File.ReadAllText(path), Path.GetFileName(path)))
{
try
{
rtc.Compile(new string[0]); // see http://docs.nvidia.com/cuda/nvrtc/index.html for usage and options
result = ManagedCuda.NVRTC.nvrtcResult.Success;
} catch (ManagedCuda.NVRTC.NVRTCException ex)
{
result = ex.NVRTCError;
}
log = rtc.GetLogAsString();
if (result == ManagedCuda.NVRTC.nvrtcResult.Success)
{
byte[] ptx = rtc.GetPTX();
multiply = ctx.LoadKernelFatBin(ptx, "Multiply"); // hard-coded method name from the CUDA kernel
}
}
return(result);
}
private byte[] DoCompile(string fullSource)
{
var rtc = new ManagedCuda.NVRTC.CudaRuntimeCompiler(fullSource, null);
try
{
rtc.Compile(new string[0]);
}
catch
{
throw new ApplicationException("Error compiling CUDA code: " + rtc.GetLogAsString());
}
return(rtc.GetPTX());
}
private byte[] DoCompile(string fullSource)
{
ManagedCuda.NVRTC.CudaRuntimeCompiler rtc = new ManagedCuda.NVRTC.CudaRuntimeCompiler(fullSource, null);
try
{
if (m_options == null || m_options.Length == 0)
{
rtc.Compile(new string[] { });
}
else
{
Logger.WriteLine($"Compiler Options: {string.Join(" ", m_options)}");
rtc.Compile(m_options);
//rtc.Compile(new string[] { "--use_fast_math", "--gpu-architecture=compute_60" });
}
}
catch
{
throw new ApplicationException("Error compiling CUDA code: " + rtc.GetLogAsString());
}
return(rtc.GetPTX());
}