/// <summary>
/// Creates a 3D plan.
/// </summary>
/// <param name="fftType">Type of FFT.</param>
/// <param name="dataType">Data type.</param>
/// <param name="nx">The number of samples in x dimension.</param>
/// <param name="ny">The number of samples in y dimension.</param>
/// <param name="nz">The number of samples in z dimension.</param>
/// <param name="batchSize">Size of batch.</param>
/// <returns>Plan.</returns>
public override FFTPlan3D Plan3D(eFFTType fftType, eDataType dataType, int nx, int ny, int nz, int batchSize)
{
int insize, outsize;
CUFFTType cuFFTType = VerifyTypes(fftType, dataType, out insize, out outsize);
cufftHandle handle = new cufftHandle();
CUFFTResult res;
if (batchSize <= 1)
{
res = _driver.cufftPlan3d(ref handle, nx, ny, nz, cuFFTType);
}
else
{
res = _driver.cufftPlanMany(ref handle, 3, new int[] { nx, ny, nz }, null, 1, 0, null, 1, 0, cuFFTType, batchSize);
}
if (res != CUFFTResult.Success)
{
throw new CudafyHostException(res.ToString());
}
FFTPlan3D plan = new FFTPlan3D(nx, ny, nz, batchSize, this);
FFTPlan3DEx planEx = new FFTPlan3DEx(plan)
{
CudaFFTHandle = handle, CudaFFTType = cuFFTType, DataType = dataType
};
Plans.Add(plan, planEx);
return(plan);
}
private SizeT InitFFT(int i, int width, int height, int tileSize, int maxShift)
{
int blockSize = tileSize + 2 * maxShift;
int blockCountX = (width - maxShift * 2) / tileSize;
int blockCountY = (height - maxShift * 2) / tileSize;
int fftWidth = blockSize / 2 + 1;
int n = 2;
int[] dims = new int[] { blockSize, blockSize };
int batches = blockCountX * blockCountY;
int[] inembed = new int[] { 1, blockSize };
int[] onembed = new int[] { 1, fftWidth };
int idist = blockSize * blockSize;
int odist = blockSize * fftWidth;
int istride = 1;
int ostride = 1;
cufftHandle handleForward = cufftHandle.Create();
cufftHandle handleBackward = cufftHandle.Create();
SizeT sizeForward = new SizeT();
SizeT sizeBackward = new SizeT();
forward[i] = new CudaFFTPlanMany(handleForward, n, dims, batches, cufftType.R2C, inembed, istride, idist, onembed, ostride, odist, ref sizeForward, false);
backward[i] = new CudaFFTPlanMany(handleBackward, n, dims, batches, cufftType.C2R, onembed, ostride, odist, inembed, istride, idist, ref sizeBackward, false);
Console.WriteLine("Size FFT forward: " + sizeForward.ToString() + " backward: " + sizeBackward.ToString());
return(sizeForward > sizeBackward ? sizeForward : sizeBackward);
}
public PreAlignment(NPPImage_32fC1 img, CudaContext ctx)
{
width = img.WidthRoi;
height = img.HeightRoi;
imgToTrackRotated = new NPPImage_32fC1(width, height);
CUmodule mod = ctx.LoadModule("kernel.ptx");
int fftWidth = width / 2 + 1;
conjKernel = new conjugateComplexMulKernel(ctx, mod);
fourierFilterKernel = new fourierFilterKernel(ctx, mod);
fftshiftKernel = new fftshiftKernel(ctx, mod);
squaredSumKernel = new squaredSumKernel(ctx, mod);
boxFilterXKernel = new boxFilterWithBorderXKernel(ctx, mod);
boxFilterYKernel = new boxFilterWithBorderYKernel(ctx, mod);
normalizedCCKernel = new normalizedCCKernel(ctx, mod);
findMinimumKernel = new findMinimumKernel(ctx, mod);
int n = 2;
int[] dims = new int[] { height, width };
int batches = 1;
int[] inembed = new int[] { 1, imgToTrackRotated.Pitch / 4 };
int[] onembed = new int[] { 1, fftWidth };
int idist = height * imgToTrackRotated.Pitch / 4;
int odist = height * fftWidth;
int istride = 1;
int ostride = 1;
cufftHandle handleForward = cufftHandle.Create();
cufftHandle handleBackward = cufftHandle.Create();
SizeT sizeForward = new SizeT();
SizeT sizeBackward = new SizeT();
forward = new CudaFFTPlanMany(handleForward, n, dims, batches, cufftType.R2C, inembed, istride, idist, onembed, ostride, odist, ref sizeForward, false);
backward = new CudaFFTPlanMany(handleBackward, n, dims, batches, cufftType.C2R, onembed, ostride, odist, inembed, istride, idist, ref sizeBackward, false);
FFTBufferSize = sizeForward > sizeBackward ? sizeForward : sizeBackward;
}
public override FFTPlan1D Plan1D(eFFTType fftType, eDataType dataType, int nx, int batchSize, int istride, int idist, int ostride, int odist)
{
int insize, outsize;
CUFFTType cuFFTType = VerifyTypes(fftType, dataType, out insize, out outsize);
cufftHandle handle = new cufftHandle();
CUFFTResult res;
if (batchSize <= 1)
{
res = _driver.cufftPlan1d(ref handle, nx, cuFFTType, batchSize);
}
else
{
res = _driver.cufftPlanMany(ref handle, 1, new int[] { nx },
new int[] { idist }, //inembed
istride, //istride
idist, //idist
new int[] { odist }, //onembed
ostride, //ostride
odist, //odist
cuFFTType,
batchSize);
}
if (res != CUFFTResult.Success)
{
throw new CudafyHostException(res.ToString());
}
FFTPlan1D plan = new FFTPlan1D(nx, batchSize, this);
FFTPlan1DEx planEx = new FFTPlan1DEx(plan)
{
CudaFFTHandle = handle, CudaFFTType = cuFFTType, DataType = dataType
};
Plans.Add(plan, planEx);
return(plan);
}
private static extern CUFFTResult cufftExecZ2D_ext(cufftHandle plan, CUdeviceptr idata, CUdeviceptr odata);
public CUFFTResult cufftSetCompatibilityMode(cufftHandle plan, CUFFTCompatibility mode)
{
return(cufftSetCompatibilityMode_ext(plan, mode));
}
public static extern CUFFTResult cufftSetStream(cufftHandle p, cudaStream stream);
public CUFFTResult cufftPlan1d(ref cufftHandle plan, int nx, CUFFTType type, int batch)
{
return(cufftPlan1d_ext(ref plan, nx, type, batch));
}
public CUFFTResult cufftExecZ2D(cufftHandle plan, CUdeviceptr idata, CUdeviceptr odata)
{
return(cufftExecZ2D_ext(plan, idata, odata));
}
public void ExecuteRealToComplex(cufftHandle plan, CUdeviceptr input, CUdeviceptr output)
{
this.LastError = _driver.cufftExecR2C(plan, input, output);
}
private static extern CUFFTResult cufftPlan3d_ext(ref cufftHandle plan, int nx, int ny, int nz, CUFFTType type);
public void ExecuteComplexToComplex(cufftHandle plan, CUdeviceptr input, CUdeviceptr output, CUFFTDirection direction)
{
this.LastError = _driver.cufftExecC2C(plan, input, output, direction);
}
static void Main(string[] args)
{
// Init and select 1st device.
CUDA cuda = new CUDA(0, true);
// load module
//cuda.LoadModule(Path.Combine(Environment.CurrentDirectory, "simpleCUFFT.ptx"));
CUfunction func = new CUfunction();// cuda.GetModuleFunction("ComplexPointwiseMulAndScale");
// The filter size is assumed to be a number smaller than the signal size
const int SIGNAL_SIZE = 50;
const int FILTER_KERNEL_SIZE = 11;
// Allocate host memory for the signal
Float2[] h_signal = new Float2[SIGNAL_SIZE];
// Initalize the memory for the signal
Random r = new Random();
for (int i = 0; i < SIGNAL_SIZE; ++i)
{
h_signal[i].x = r.Next() / (float)int.MaxValue;
h_signal[i].y = 0;
}
// Allocate host memory for the filter
Float2[] h_filter_kernel = new Float2[FILTER_KERNEL_SIZE];
// Initalize the memory for the filter
for (int i = 0; i < FILTER_KERNEL_SIZE; ++i)
{
h_filter_kernel[i].x = r.Next() / (float)int.MaxValue;
h_filter_kernel[i].y = 0;
}
// Pad signal and filter kernel
Float2[] h_padded_signal;
Float2[] h_padded_filter_kernel;
int new_size = PadData(h_signal, out h_padded_signal, SIGNAL_SIZE,
h_filter_kernel, out h_padded_filter_kernel, FILTER_KERNEL_SIZE);
// Allocate device memory for signal
// Copy host memory to device
CUdeviceptr d_signal = cuda.CopyHostToDevice <Float2>(h_padded_signal);
// Allocate device memory for filter kernel
// Copy host memory to device
CUdeviceptr d_filter_kernel = cuda.CopyHostToDevice <Float2>(h_padded_filter_kernel);
// CUFFT plan
CUFFT fft = new CUFFT(cuda);
cufftHandle handle = new cufftHandle();
CUFFTResult fftres = CUFFTDriver.cufftPlan1d(ref handle, new_size, CUFFTType.C2C, 1);
//fft.Plan1D(new_size, CUFFTType.C2C, 1);
return;
// Transform signal and kernel
fft.ExecuteComplexToComplex(d_signal, d_signal, CUFFTDirection.Forward);
fft.ExecuteComplexToComplex(d_filter_kernel, d_filter_kernel, CUFFTDirection.Forward);
// Multiply the coefficients together and normalize the result
// ComplexPointwiseMulAndScale<<<32, 256>>>(d_signal, d_filter_kernel, new_size, 1.0f / new_size);
cuda.SetFunctionBlockShape(func, 256, 1, 1);
cuda.SetParameter(func, 0, (uint)d_signal.Pointer);
cuda.SetParameter(func, IntPtr.Size, (uint)d_filter_kernel.Pointer);
cuda.SetParameter(func, IntPtr.Size * 2, (uint)new_size);
cuda.SetParameter(func, IntPtr.Size * 2 + 4, 1.0f / new_size);
cuda.SetParameterSize(func, (uint)(IntPtr.Size * 2 + 8));
cuda.Launch(func, 32, 1);
// Transform signal back
fft.ExecuteComplexToComplex(d_signal, d_signal, CUFFTDirection.Inverse);
// Copy device memory to host
Float2[] h_convolved_signal = h_padded_signal;
cuda.CopyDeviceToHost <Float2>(d_signal, h_convolved_signal);
// Allocate host memory for the convolution result
Float2[] h_convolved_signal_ref = new Float2[SIGNAL_SIZE];
// Convolve on the host
Convolve(h_signal, SIGNAL_SIZE,
h_filter_kernel, FILTER_KERNEL_SIZE,
h_convolved_signal_ref);
// check result
bool res = cutCompareL2fe(h_convolved_signal_ref, h_convolved_signal, 2 * SIGNAL_SIZE, 1e-5f);
Console.WriteLine("Test {0}", (true == res) ? "PASSED" : "FAILED");
//Destroy CUFFT context
fft.Destroy();
// cleanup memory
cuda.Free(d_signal);
cuda.Free(d_filter_kernel);
}
public void Destroy(cufftHandle plan)
{
this.LastError = _driver.cufftDestroy(plan);
}
public cufftHandle Plan3D(int nx, int ny, int nz, CUFFTType type)
{
this.plan = new cufftHandle();
this.LastError = _driver.cufftPlan3d(ref this.plan, nx, ny, nz, type);
return(this.plan);
}
public cufftHandle Plan1D(int nx, CUFFTType type, int batch)
{
this.plan = new cufftHandle();
this.LastError = _driver.cufftPlan1d(ref this.plan, nx, type, batch);
return(this.plan);
}
private static extern CUFFTResult cufftExecZ2Z_ext(cufftHandle plan, CUdeviceptr idata, CUdeviceptr odata, CUFFTDirection direction);
private static extern CUFFTResult cufftPlanMany_ext(ref cufftHandle plan, int rank, [In, Out] int[] n, [In, Out] int[] inembed, int istride, int idist, [In, Out] int[] onembed, int ostride, int odist, CUFFTType type, int batch);
private static extern CUFFTResult cufftPlan1d_ext(ref cufftHandle plan, int nx, CUFFTType type, int batch);
private static extern CUFFTResult cufftSetCompatibilityMode_ext(cufftHandle plan, CUFFTCompatibility mode);
private static extern CUFFTResult cufftPlanMany_ext(ref cufftHandle plan, int rank, IntPtr n, IntPtr inembed, int istride, int idist, IntPtr onembed, int ostride, int odist, CUFFTType type, int batch);
public void ExecuteComplexToReal(cufftHandle plan, CUdeviceptr input, CUdeviceptr output)
{
this.LastError = _driver.cufftExecC2R(plan, input, output);
}
private static extern CUFFTResult cufftSetStream_ext(cufftHandle p, cudaStream stream);
private static extern CUFFTResult cufftDestroy_ext(cufftHandle plan);
public CUFFTResult cufftDestroy(cufftHandle plan)
{
return(cufftDestroy_ext(plan));
}
public CUFFTResult cufftPlan3d(ref cufftHandle plan, int nx, int ny, int nz, CUFFTType type)
{
return(cufftPlan3d_ext(ref plan, nx, ny, nz, type));
}
public CUFFTResult cufftExecZ2Z(cufftHandle plan, CUdeviceptr idata, CUdeviceptr odata, CUFFTDirection direction)
{
return(cufftExecZ2Z_ext(plan, idata, odata, direction));
}
public CUFFTResult cufftPlanMany(ref cufftHandle plan, int rank, int[] n, int[] inembed, int istride, int idist, int[] onembed, int ostride, int odist, CUFFTType type, int batch)
{
return(cufftPlanMany_ext(ref plan, rank, n, inembed, istride, idist, onembed, ostride, odist, type, batch));
}
public CUFFTResult cufftSetStream(cufftHandle plan, cudaStream stream)
{
return(cufftSetStream_ext(plan, stream));
}
public static extern CUFFTResult cufftPlan3d(ref cufftHandle plan, int nx, int ny, int nz, CUFFTType type);
