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
findFirst = ctx.LoadKernelFatBin(ptx, "FindFirst"); // MY method name from the CUDA kernel
parallelFor = ctx.LoadKernelFatBin(ptx, "ParallelFor"); // MY 2nd method name from the CUDA kernel
}
}
return(result);
}
internal nvrtcResult LoadKernel(out string log)
{
nvrtcResult result;
using (var rtc = new CudaRuntimeCompiler(File.ReadAllText(path), Path.GetFileName(path)))
{
try
{
rtc.Compile(Array.Empty <string>());
result = nvrtcResult.Success;
}
catch (NVRTCException ex)
{
result = ex.NVRTCError;
}
log = rtc.GetLogAsString();
if (result == nvrtcResult.Success)
{
byte[] ptx = rtc.GetPTX();
multiply = ctx.LoadKernelFatBin(ptx, methodName);
}
}
return(result);
}
internal void LoadKernel()
{
ctx = new CudaContext(0, true);
nvrtcResult result;
using var rtc = new CudaRuntimeCompiler(File.ReadAllText(path), Path.GetFileName(path));
rtc.Compile(Array.Empty <string>());
result = nvrtcResult.Success;
if (result == nvrtcResult.Success)
{
byte[] ptx = rtc.GetPTX();
kernel = ctx.LoadKernelFatBin(ptx, methodName);
}
}