private void LoadKernels()
{
if (_kernelLoader == null)
{
_kernelLoader = new KernelLoader <float>(this._context);
var assembly = Assembly.GetExecutingAssembly();
using (Stream stream = assembly.GetManifestResourceStream("ConvNetSharp.Volume.GPU.Single.Kernels.log.cu"))
{
_kernelLoader.LoadKernel("Log", stream);
}
using (Stream stream = assembly.GetManifestResourceStream("ConvNetSharp.Volume.GPU.Single.Kernels.exp.cu"))
{
_kernelLoader.LoadKernel("Exp", stream);
}
using (Stream stream = assembly.GetManifestResourceStream("ConvNetSharp.Volume.GPU.Single.Kernels.div.cu"))
{
_kernelLoader.LoadKernel("Div", stream);
}
using (Stream stream = assembly.GetManifestResourceStream("ConvNetSharp.Volume.GPU.Single.Kernels.leakyrelu.cu"))
{
_kernelLoader.LoadKernel("LeakyRelu", stream);
}
using (Stream stream = assembly.GetManifestResourceStream("ConvNetSharp.Volume.GPU.Single.Kernels.leakyrelu_gradient.cu"))
{
_kernelLoader.LoadKernel("LeakyReluGradient", stream);
}
}
}
public static void init(int maxCnt)
{
_gpuVelocity = KernelLoader.load_kernel("update_velocities");
_gpuUpdate = KernelLoader.load_kernel("update_particles");
_dt = ISF.properties.dt;
torus_d = new float[3] {
ISF.properties.dx, ISF.properties.dy, ISF.properties.dz
};
torus_res = new int[3] {
ISF.properties.resx, ISF.properties.resy, ISF.properties.resz
};
torus_size = new int[3] {
ISF.properties.sizex, ISF.properties.sizey, ISF.properties.sizez
};
d_k1x = new CudaDeviceVariable <float>(maxCnt);
d_k1y = new CudaDeviceVariable <float>(maxCnt);
d_k1z = new CudaDeviceVariable <float>(maxCnt);
d_k2x = new CudaDeviceVariable <float>(maxCnt);
d_k2y = new CudaDeviceVariable <float>(maxCnt);
d_k2z = new CudaDeviceVariable <float>(maxCnt);
d_k3x = new CudaDeviceVariable <float>(maxCnt);
d_k3y = new CudaDeviceVariable <float>(maxCnt);
d_k3z = new CudaDeviceVariable <float>(maxCnt);
d_k4x = new CudaDeviceVariable <float>(maxCnt);
d_k4y = new CudaDeviceVariable <float>(maxCnt);
d_k4z = new CudaDeviceVariable <float>(maxCnt);
}
static TensorOpGpu()
{
_CudaContext = new CudaContext(0, true);
_CudaBlasHandle = new CudaBlas();
_CudaStream = new CudaStream();
_CudnnContext = new CudaDNNContext();
_KernelLoader = new KernelLoader();
}
private void LoadKernels()
{
if (_kernelLoader != null)
{
return;
}
_kernelLoader = new KernelLoader <double>(this._context);
var assembly = Assembly.GetExecutingAssembly();
// Retrieve all kernels from resources
var regex = new Regex(@"ConvNetSharp\.Volume\.GPU\.Double\.Kernels\.(.*)\.cu", RegexOptions.Compiled);
var tuples = assembly.GetManifestResourceNames().Where(o => regex.IsMatch(o)).Select(o => new Tuple <string, string>(o, regex.Match(o).Groups[1].Value));
foreach (var t in tuples)
{
using var stream = assembly.GetManifestResourceStream(t.Item1);
_kernelLoader.LoadKernel(t.Item2, stream);
}
}
public static void Init(SpaceProperties properties)
{
staggered = KernelLoader.load_kernel("staggered_sharp");
staggered.BlockDimensions = new dim3(1, 1, 1);
staggered.GridDimensions = new dim3(properties.num, 1, 1);
div = KernelLoader.load_kernel("div");
div.BlockDimensions = new dim3(1, 1, 1);
div.GridDimensions = new dim3(properties.num, 1, 1);
velocity_one = KernelLoader.load_kernel("velocity_one");
velocity_one.BlockDimensions = new dim3(1, 1, 1);
velocity_one.GridDimensions = new dim3(properties.num, 1, 1);
normalize = KernelLoader.load_kernel("normalize");
normalize.BlockDimensions = new dim3(1, 1, 1);
normalize.GridDimensions = new dim3(properties.num, 1, 1);
gauge = KernelLoader.load_kernel("gauge");
gauge.BlockDimensions = new dim3(1, 1, 1);
gauge.GridDimensions = new dim3(properties.num, 1, 1);
shift = KernelLoader.load_kernel("shift");
shift.BlockDimensions = new dim3(1, 1, 1);
shift.GridDimensions = new dim3(properties.num, 1, 1);
mul_each = KernelLoader.load_kernel("mul_each");
mul_each.BlockDimensions = new dim3(1, 1, 1);
mul_each.GridDimensions = new dim3(properties.num, 1, 1);
copy = KernelLoader.load_kernel("copy");
copy.BlockDimensions = new dim3(1, 1, 1);
copy.GridDimensions = new dim3(properties.num, 1, 1);
fft_norm = KernelLoader.load_kernel("fft_norm");
fft_norm.BlockDimensions = new dim3(1, 1, 1);
fft_norm.GridDimensions = new dim3(properties.num, 1, 1);
}
private void LoadKernels()
{
if (_kernelLoader == null)
{
_kernelLoader = new KernelLoader <double>(this._context);
var assembly = Assembly.GetExecutingAssembly();
using (Stream stream = assembly.GetManifestResourceStream("ConvNetSharp.Volume.GPU.Double.Kernels.log.cu"))
{
_kernelLoader.LoadKernel("Log", stream);
}
using (Stream stream = assembly.GetManifestResourceStream("ConvNetSharp.Volume.GPU.Double.Kernels.exp.cu"))
{
_kernelLoader.LoadKernel("Exp", stream);
}
using (Stream stream = assembly.GetManifestResourceStream("ConvNetSharp.Volume.GPU.Double.Kernels.div.cu"))
{
_kernelLoader.LoadKernel("Div", stream);
}
}
}
public new static void init()
{
_gpu = KernelLoader.load_kernel("add_particles");
}
