protected void LoadNextField()
{
// Keep t1 timestep as new t0. Update mapping on device side.
_t0X = _t1X;
new CudaTextureArray2D(_advectParticlesKernel, "vX_t0", CUAddressMode.Wrap, CUFilterMode.Linear, CUTexRefSetFlags.None, _t0X);
_t0Y = _t1Y;
new CudaTextureArray2D(_advectParticlesKernel, "vY_t0", CUAddressMode.Wrap, CUFilterMode.Linear, CUTexRefSetFlags.None, _t0Y);
CurrentTime++;
// Load new t1.
LoaderNCF ncFile = RedSea.Singleton.GetLoaderNCF(CurrentTime);
ScalarField t1X = ncFile.LoadFieldSlice(_ensembleRanges[0]);
ScalarField t1Y = ncFile.LoadFieldSlice(_ensembleRanges[1]);
ncFile.Close();
// All members are above each other.
int vHeight = _height * _numMembers;
// vX, t=1
_t1X = new CudaArray2D(CUArrayFormat.Float, _width, vHeight, CudaArray2DNumChannels.One);
_t1X.CopyFromHostToThis(t1X.Data);
new CudaTextureArray2D(_advectParticlesKernel, "vX_t1", CUAddressMode.Wrap, CUFilterMode.Linear, CUTexRefSetFlags.None, _t1X);
// vY, t=1
_t1Y = new CudaArray2D(CUArrayFormat.Float, _width, vHeight, CudaArray2DNumChannels.One);
_t1Y.CopyFromHostToThis(t1Y.Data);
new CudaTextureArray2D(_advectParticlesKernel, "vY_t1", CUAddressMode.Wrap, CUFilterMode.Linear, CUTexRefSetFlags.None, _t1Y);
}
public void Step(float stepSize)
{
// Load the next vector fields to device memory.
LoadNextField();
_cudaDxMapper.MapAllResources();
CudaArray2D lastFlowMap = _cudaDxMapper[0].GetMappedArray2D(0, 0);
// Advect from each member to each member. In each block, the same configuration is choosen.
dim3 grid = new dim3((int)((float)_width / BLOCK_SIZE + 0.5f), (int)((float)_height / BLOCK_SIZE + 0.5f), _numMembers);
// Advect a block in each member-member combination.
dim3 threads = new dim3(BLOCK_SIZE, BLOCK_SIZE);
_advectParticlesKernel.GridDimensions = grid;
_advectParticlesKernel.BlockDimensions = threads;
// (float* mapT1, const int width, const int height, const int numMembers, /*float timeScale, */ float stepSize, float minDensity, float invalid)
_advectParticlesKernel.Run(_pongFlowMap.DevicePointer, _width, _height, _numMembers, stepSize, 0.000001f, _texInvalidValue);
// Swap the Texture2D handles.
CudaSurfObject surf = new CudaSurfObject(lastFlowMap);
grid.z = 1;
_copyMapDataKernel.GridDimensions = grid;
_copyMapDataKernel.BlockDimensions = threads;
_copyMapDataKernel.Run(surf.SurfObject, _pongFlowMap.DevicePointer, _width, _height);
_cudaDxMapper.UnmapAllResources();
}
/// <summary>
/// Creates a surface object. <c>ResDesc</c> describes
/// the data to perform surface load/stores on. <c>ResDesc.resType</c> must be
/// <see cref="CUResourceType.Array"/> and <c>ResDesc.hArray</c>
/// must be set to a valid CUDA array handle.
/// </summary>
/// <param name="array">CudaArray2D</param>
public CudaSurfObject(CudaArray2D array)
{
_resDesc = new CudaResourceDesc(array);
_surfObject = new CUsurfObject();
res = DriverAPINativeMethods.SurfaceObjects.cuSurfObjectCreate(ref _surfObject, ref _resDesc);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cuSurfObjectCreate", res));
if (res != CUResult.Success) throw new CudaException(res);
}
private void InitializeResources()
{
int vHeight = _height * _numMembers;
_t0Y = new CudaArray2D(CUArrayFormat.Float, _width, vHeight, CudaArray2DNumChannels.One);
_t0X = new CudaArray2D(CUArrayFormat.Float, _width, vHeight, CudaArray2DNumChannels.One);
_t1Y = new CudaArray2D(CUArrayFormat.Float, _width, vHeight, CudaArray2DNumChannels.One);
_t1X = new CudaArray2D(CUArrayFormat.Float, _width, vHeight, CudaArray2DNumChannels.One);
_initialized = true;
}
protected void LoadNextField()
{
// Keep t1 timestep as new t0. Update mapping on device side.
var tmp = _t0X;
_t0X = _t1X;
_t1X = tmp;
new CudaTextureArray2D(_advectParticlesKernel, "vX_t0", CUAddressMode.Wrap, CUFilterMode.Linear, CUTexRefSetFlags.None, _t0X);
tmp = _t0X;
_t0Y = _t1Y;
_t1Y = tmp;
new CudaTextureArray2D(_advectParticlesKernel, "vY_t0", CUAddressMode.Wrap, CUFilterMode.Linear, CUTexRefSetFlags.None, _t0Y);
CurrentTime++;
// Load new t1.
LoaderNCF ncFile = RedSea.Singleton.GetLoaderNCF(CurrentTime);
ScalarField t1X = ncFile.LoadFieldSlice(_ensembleRanges[0]);
ScalarField t1Y = ncFile.LoadFieldSlice(_ensembleRanges[1]);
ncFile.Close();
// All members are above each other.
int vHeight = _height * _numMembers;
float[] paddedArray = new float[_t1X.Width * _t1X.Height];
Array.Copy(t1X.Data, paddedArray, t1X.Data.Length);
// vX, t=1
_t1X.CopyFromHostToThis <float>(paddedArray);
new CudaTextureArray2D(_advectParticlesKernel, "vX_t1", CUAddressMode.Wrap, CUFilterMode.Linear, CUTexRefSetFlags.None, _t1X);
Array.Copy(t1Y.Data, paddedArray, t1Y.Data.Length);
// vY, t=1
_t1Y.CopyFromHostToThis <float>(paddedArray);
new CudaTextureArray2D(_advectParticlesKernel, "vY_t1", CUAddressMode.Wrap, CUFilterMode.Linear, CUTexRefSetFlags.None, _t1Y);
}
static void Main(string[] args)
{
const int nx = 2048;
const int ny = 2048;
// shifts applied to x and y data
const int x_shift = 5;
const int y_shift = 7;
ShrQATest.shrQAStart(args);
if ((nx%TILE_DIM != 0) || (ny%TILE_DIM != 0))
{
Console.Write("nx and ny must be multiples of TILE_DIM\n");
ShrQATest.shrQAFinishExit(args, ShrQATest.eQAstatus.QA_WAIVED);
}
// execution configuration parameters
dim3 grid = new dim3(nx/TILE_DIM, ny/TILE_DIM, 1);
dim3 threads = new dim3(TILE_DIM, TILE_DIM, 1);
// This will pick the best possible CUDA capable device
int devID = findCudaDevice(args);
//Load Kernel image from resources
string resName;
if (IntPtr.Size == 8)
resName = "simplePitchLinearTexture_x64.ptx";
else
resName = "simplePitchLinearTexture.ptx";
string resNamespace = "simplePitchLinearTexture";
string resource = resNamespace + "." + resName;
Stream stream = Assembly.GetExecutingAssembly().GetManifestResourceStream(resource);
if (stream == null) throw new ArgumentException("Kernel not found in resources.");
byte[] kernels = new byte[stream.Length];
int bytesToRead = (int)stream.Length;
while (bytesToRead > 0)
{
bytesToRead -= stream.Read(kernels, (int)stream.Position, bytesToRead);
}
CudaKernel PLKernel = ctx.LoadKernelPTX(kernels, "shiftPitchLinear");
CudaKernel ArrayKernel = ctx.LoadKernelPTX(kernels, "shiftArray");
CudaStopWatch stopwatch = new CudaStopWatch();
// ----------------------------------
// Host allocation and initialization
// ----------------------------------
float[] h_idata = new float[nx * ny];
float[] h_odata = new float[nx * ny];
float[] gold = new float[nx * ny];
for (int i = 0; i < nx * ny; ++i) h_idata[i] = (float)i;
// ------------------------
// Device memory allocation
// ------------------------
// Pitch linear input data
CudaPitchedDeviceVariable<float> d_idataPL = new CudaPitchedDeviceVariable<float>(nx, ny);
// Array input data
CudaArray2D d_idataArray = new CudaArray2D(CUArrayFormat.Float, nx, ny, CudaArray2DNumChannels.One);
// Pitch linear output data
CudaPitchedDeviceVariable<float> d_odata = new CudaPitchedDeviceVariable<float>(nx, ny);
// ------------------------
// copy host data to device
// ------------------------
// Pitch linear
d_idataPL.CopyToDevice(h_idata);
// Array
d_idataArray.CopyFromHostToThis<float>(h_idata);
// ----------------------
// Bind texture to memory
// ----------------------
// Pitch linear
CudaTextureLinearPitched2D<float> texRefPL = new CudaTextureLinearPitched2D<float>(PLKernel, "texRefPL", CUAddressMode.Wrap, CUFilterMode.Point, CUTexRefSetFlags.NormalizedCoordinates, CUArrayFormat.Float, d_idataPL);
CudaTextureArray2D texRefArray = new CudaTextureArray2D(ArrayKernel, "texRefArray", CUAddressMode.Wrap, CUFilterMode.Point, CUTexRefSetFlags.NormalizedCoordinates, d_idataArray);
// ---------------------
// reference calculation
// ---------------------
for (int j = 0; j < ny; j++)
{
int jshift = (j + y_shift) % ny;
for (int i = 0; i < nx; i++)
{
int ishift = (i + x_shift) % nx;
gold[j * nx + i] = h_idata[jshift * nx + ishift];
}
}
// ----------------
// shiftPitchLinear
// ----------------
ctx.ClearMemory(d_odata.DevicePointer, 0, d_odata.TotalSizeInBytes);
PLKernel.BlockDimensions = threads;
PLKernel.GridDimensions = grid;
stopwatch.Start();
for (int i=0; i < NUM_REPS; i++)
{
PLKernel.Run(d_odata.DevicePointer, (int)(d_odata.Pitch/sizeof(float)), nx, ny, x_shift, y_shift);
}
stopwatch.Stop();
stopwatch.StopEvent.Synchronize();
float timePL = stopwatch.GetElapsedTime();
// check results
d_odata.CopyToHost(h_odata);
bool res = cutComparef(gold, h_odata);
bool success = true;
if (res == false) {
Console.Write("*** shiftPitchLinear failed ***\n");
success = false;
}
// ----------
// shiftArray
// ----------
ctx.ClearMemory(d_odata.DevicePointer, 0, d_odata.TotalSizeInBytes);
ArrayKernel.BlockDimensions = threads;
ArrayKernel.GridDimensions = grid;
stopwatch.Start();
for (int i=0; i < NUM_REPS; i++) {
ArrayKernel.Run(d_odata.DevicePointer, (int)(d_odata.Pitch/sizeof(float)), nx, ny, x_shift, y_shift);
}
stopwatch.Stop();
stopwatch.StopEvent.Synchronize();
float timeArray = stopwatch.GetElapsedTime();
// check results
d_odata.CopyToHost(h_odata);
res = cutComparef(gold, h_odata);
if (res == false) {
Console.Write("*** shiftArray failed ***\n");
success = false;
}
float bandwidthPL = 2.0f*1000.0f*nx*ny*sizeof(float)/(1e+9f)/(timePL/NUM_REPS);
float bandwidthArray = 2.0f*1000.0f*nx*ny*sizeof(float)/(1e+9f)/(timeArray/NUM_REPS);
Console.Write("\nBandwidth (GB/s) for pitch linear: {0}; for array: {1}\n",
bandwidthPL, bandwidthArray);
float fetchRatePL = nx*ny/1e+6f/(timePL/(1000.0f*NUM_REPS));
float fetchRateArray = nx*ny/1e+6f/(timeArray/(1000.0f*NUM_REPS));
Console.Write("\nTexture fetch rate (Mpix/s) for pitch linear: {0}; for array: {1}\n\n",
fetchRatePL, fetchRateArray);
// cleanup
texRefPL.Dispose();
texRefArray.Dispose();
d_idataPL.Dispose();
d_idataArray.Dispose();
d_odata.Dispose();
stopwatch.Dispose();
ctx.Dispose();
ShrQATest.shrQAFinishExit(args, (success == true) ? ShrQATest.eQAstatus.QA_PASSED : ShrQATest.eQAstatus.QA_FAILED);
}
/// <summary>
///
/// </summary>
/// <param name="var"></param>
public CudaResourceDesc(CudaArray2D var)
{
resType = CUResourceType.Array;
flags = 0;
res = new CudaResourceDescUnion();
res.hMipmappedArray = new CUmipmappedArray();
res.linear = new CudaResourceDescLinear();
res.pitch2D = new CudaResourceDescPitch2D();
res.hArray = var.CUArray;
}
static void Main(string[] args)
{
const int nx = 2048;
const int ny = 2048;
// shifts applied to x and y data
const int x_shift = 5;
const int y_shift = 7;
ShrQATest.shrQAStart(args);
if ((nx % TILE_DIM != 0) || (ny % TILE_DIM != 0))
{
Console.Write("nx and ny must be multiples of TILE_DIM\n");
ShrQATest.shrQAFinishExit(args, ShrQATest.eQAstatus.QA_WAIVED);
}
// execution configuration parameters
dim3 grid = new dim3(nx / TILE_DIM, ny / TILE_DIM, 1);
dim3 threads = new dim3(TILE_DIM, TILE_DIM, 1);
// This will pick the best possible CUDA capable device
int devID = findCudaDevice(args);
//Load Kernel image from resources
string resName;
if (IntPtr.Size == 8)
{
resName = "simplePitchLinearTexture_x64.ptx";
}
else
{
resName = "simplePitchLinearTexture.ptx";
}
string resNamespace = "simplePitchLinearTexture";
string resource = resNamespace + "." + resName;
Stream stream = Assembly.GetExecutingAssembly().GetManifestResourceStream(resource);
if (stream == null)
{
throw new ArgumentException("Kernel not found in resources.");
}
byte[] kernels = new byte[stream.Length];
int bytesToRead = (int)stream.Length;
while (bytesToRead > 0)
{
bytesToRead -= stream.Read(kernels, (int)stream.Position, bytesToRead);
}
CudaKernel PLKernel = ctx.LoadKernelPTX(kernels, "shiftPitchLinear");
CudaKernel ArrayKernel = ctx.LoadKernelPTX(kernels, "shiftArray");
CudaStopWatch stopwatch = new CudaStopWatch();
// ----------------------------------
// Host allocation and initialization
// ----------------------------------
float[] h_idata = new float[nx * ny];
float[] h_odata = new float[nx * ny];
float[] gold = new float[nx * ny];
for (int i = 0; i < nx * ny; ++i)
{
h_idata[i] = (float)i;
}
// ------------------------
// Device memory allocation
// ------------------------
// Pitch linear input data
CudaPitchedDeviceVariable <float> d_idataPL = new CudaPitchedDeviceVariable <float>(nx, ny);
// Array input data
CudaArray2D d_idataArray = new CudaArray2D(CUArrayFormat.Float, nx, ny, CudaArray2DNumChannels.One);
// Pitch linear output data
CudaPitchedDeviceVariable <float> d_odata = new CudaPitchedDeviceVariable <float>(nx, ny);
// ------------------------
// copy host data to device
// ------------------------
// Pitch linear
d_idataPL.CopyToDevice(h_idata);
// Array
d_idataArray.CopyFromHostToThis <float>(h_idata);
// ----------------------
// Bind texture to memory
// ----------------------
// Pitch linear
CudaTextureLinearPitched2D <float> texRefPL = new CudaTextureLinearPitched2D <float>(PLKernel, "texRefPL", CUAddressMode.Wrap, CUFilterMode.Point, CUTexRefSetFlags.NormalizedCoordinates, CUArrayFormat.Float, d_idataPL);
CudaTextureArray2D texRefArray = new CudaTextureArray2D(ArrayKernel, "texRefArray", CUAddressMode.Wrap, CUFilterMode.Point, CUTexRefSetFlags.NormalizedCoordinates, d_idataArray);
// ---------------------
// reference calculation
// ---------------------
for (int j = 0; j < ny; j++)
{
int jshift = (j + y_shift) % ny;
for (int i = 0; i < nx; i++)
{
int ishift = (i + x_shift) % nx;
gold[j * nx + i] = h_idata[jshift * nx + ishift];
}
}
// ----------------
// shiftPitchLinear
// ----------------
ctx.ClearMemory(d_odata.DevicePointer, 0, d_odata.TotalSizeInBytes);
PLKernel.BlockDimensions = threads;
PLKernel.GridDimensions = grid;
stopwatch.Start();
for (int i = 0; i < NUM_REPS; i++)
{
PLKernel.Run(d_odata.DevicePointer, (int)(d_odata.Pitch / sizeof(float)), nx, ny, x_shift, y_shift);
}
stopwatch.Stop();
stopwatch.StopEvent.Synchronize();
float timePL = stopwatch.GetElapsedTime();
// check results
d_odata.CopyToHost(h_odata);
bool res = cutComparef(gold, h_odata);
bool success = true;
if (res == false)
{
Console.Write("*** shiftPitchLinear failed ***\n");
success = false;
}
// ----------
// shiftArray
// ----------
ctx.ClearMemory(d_odata.DevicePointer, 0, d_odata.TotalSizeInBytes);
ArrayKernel.BlockDimensions = threads;
ArrayKernel.GridDimensions = grid;
stopwatch.Start();
for (int i = 0; i < NUM_REPS; i++)
{
ArrayKernel.Run(d_odata.DevicePointer, (int)(d_odata.Pitch / sizeof(float)), nx, ny, x_shift, y_shift);
}
stopwatch.Stop();
stopwatch.StopEvent.Synchronize();
float timeArray = stopwatch.GetElapsedTime();
// check results
d_odata.CopyToHost(h_odata);
res = cutComparef(gold, h_odata);
if (res == false)
{
Console.Write("*** shiftArray failed ***\n");
success = false;
}
float bandwidthPL = 2.0f * 1000.0f * nx * ny * sizeof(float) / (1e+9f) / (timePL / NUM_REPS);
float bandwidthArray = 2.0f * 1000.0f * nx * ny * sizeof(float) / (1e+9f) / (timeArray / NUM_REPS);
Console.Write("\nBandwidth (GB/s) for pitch linear: {0}; for array: {1}\n",
bandwidthPL, bandwidthArray);
float fetchRatePL = nx * ny / 1e+6f / (timePL / (1000.0f * NUM_REPS));
float fetchRateArray = nx * ny / 1e+6f / (timeArray / (1000.0f * NUM_REPS));
Console.Write("\nTexture fetch rate (Mpix/s) for pitch linear: {0}; for array: {1}\n\n",
fetchRatePL, fetchRateArray);
// cleanup
texRefPL.Dispose();
texRefArray.Dispose();
d_idataPL.Dispose();
d_idataArray.Dispose();
d_odata.Dispose();
stopwatch.Dispose();
ctx.Dispose();
ShrQATest.shrQAFinishExit(args, (success == true) ? ShrQATest.eQAstatus.QA_PASSED : ShrQATest.eQAstatus.QA_FAILED);
}
/// <summary>
/// Creates a new 2D texture from array memory. Allocates a new 2D array.
/// </summary>
/// <param name="kernel"></param>
/// <param name="texName"></param>
/// <param name="addressMode0"></param>
/// <param name="addressMode1"></param>
/// <param name="filterMode"></param>
/// <param name="flags"></param>
/// <param name="format"></param>
/// <param name="height">In elements</param>
/// <param name="width">In elements</param>
/// <param name="numChannels">1,2 or 4</param>
public CudaTextureArray2D(CudaKernel kernel, string texName, CUAddressMode addressMode0, CUAddressMode addressMode1, CUFilterMode filterMode, CUTexRefSetFlags flags, CUArrayFormat format, SizeT width, SizeT height, CudaArray2DNumChannels numChannels)
{
_texref = new CUtexref();
res = DriverAPINativeMethods.ModuleManagement.cuModuleGetTexRef(ref _texref, kernel.CUModule, texName);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}, Texture name: {3}", DateTime.Now, "cuModuleGetTexRef", res, texName));
if (res != CUResult.Success) throw new CudaException(res);
res = DriverAPINativeMethods.TextureReferenceManagement.cuTexRefSetAddressMode(_texref, 0, addressMode0);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cuTexRefSetAddressMode", res));
if (res != CUResult.Success) throw new CudaException(res);
res = DriverAPINativeMethods.TextureReferenceManagement.cuTexRefSetAddressMode(_texref, 1, addressMode1);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cuTexRefSetAddressMode", res));
if (res != CUResult.Success) throw new CudaException(res);
res = DriverAPINativeMethods.TextureReferenceManagement.cuTexRefSetFilterMode(_texref, filterMode);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cuTexRefSetFilterMode", res));
if (res != CUResult.Success) throw new CudaException(res);
res = DriverAPINativeMethods.TextureReferenceManagement.cuTexRefSetFlags(_texref, flags);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cuTexRefSetFlags", res));
if (res != CUResult.Success) throw new CudaException(res);
res = DriverAPINativeMethods.TextureReferenceManagement.cuTexRefSetFormat(_texref, format, (int)numChannels);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cuTexRefSetFormat", res));
if (res != CUResult.Success) throw new CudaException(res);
_filtermode = filterMode;
_flags = flags;
_addressMode0 = addressMode0;
_addressMode1 = addressMode1;
_format = format;
_height = height;
_width = width;
_numChannels = (int)numChannels;
_name = texName;
_module = kernel.CUModule;
_cufunction = kernel.CUFunction;
_channelSize = CudaHelperMethods.GetChannelSize(format);
_dataSize = height * width * _numChannels * _channelSize;
_array = new CudaArray2D(format, width, height, numChannels);
res = DriverAPINativeMethods.TextureReferenceManagement.cuTexRefSetArray(_texref, _array.CUArray, CUTexRefSetArrayFlags.OverrideFormat);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cuTexRefSetArray", res));
if (res != CUResult.Success) throw new CudaException(res);
//res = DriverAPINativeMethods.ParameterManagement.cuParamSetTexRef(kernel.CUFunction, CUParameterTexRef.Default, _texref);
//Debug.WriteLine("{0:G}, {1}: {2}", DateTime.Now, "cuParamSetTexRef", res);
//if (res != CUResult.Success) throw new CudaException(res);
}
/// <summary>
/// Creates a new 2D texture from array memory
/// </summary>
/// <param name="kernel"></param>
/// <param name="texName"></param>
/// <param name="addressMode"></param>
/// <param name="filterMode"></param>
/// <param name="flags"></param>
/// <param name="array"></param>
public CudaTextureArray2D(CudaKernel kernel, string texName, CUAddressMode addressMode, CUFilterMode filterMode, CUTexRefSetFlags flags, CudaArray2D array)
: this(kernel, texName, addressMode, addressMode, filterMode, flags, array)
{
}
/// <summary>
/// Setup as empty map with only one value at 1.
/// </summary>
/// <param name="pos"></param>
/// <param name="fieldEnsemble"></param>
/// <param name="startTime"></param>
/// <param name="endTime"></param>
public void SetupPoint(Int2 pos, int startTime)
{
// ~~~~~~~~~~~~~~ Copy relevant data ~~~~~~~~~~~~~~ \\
// Count up when advection was executed.
CurrentTime = startTime;
_startTime = startTime;
// ~~~~~~~~~~~~ Load ensemble ~~~~~~~~~~~~ \\
// Load fields first to get the grid size.
//Loader ncFile = new Loader(RedSea.Singleton.DataFolder + (_startTime + 1) + RedSea.Singleton.FileName);
//ScalarField t0X = ncFile.LoadFieldSlice(_ensembleRanges[0]);
//ScalarField t0Y = ncFile.LoadFieldSlice(_ensembleRanges[1]);
//ncFile.Close();
LoaderNCF ncFile = RedSea.Singleton.GetLoaderNCF(_startTime);
ScalarField t1X = ncFile.LoadFieldSlice(_ensembleRanges[0]);
ScalarField t1Y = ncFile.LoadFieldSlice(_ensembleRanges[1]);
ncFile.Close();
// ~~~~~~~~~~~~~~ Copy relevant data ~~~~~~~~~~~~~~ \\
// Keep for plane creation and size reference.
_ensembleGrid = t1X.Grid as RectlinearGrid;
// Mapper for binding the SlimDX texture to CUDA easily.
_cudaDxMapper = new CudaGraphicsInteropResourceCollection();
// Tell CUDA which value is a border.
_texInvalidValue = t1X.InvalidValue ?? float.MaxValue;
// ~~~~~~~~~~~~ Fill CUDA resources ~~~~~~~~~~~~ \\
// All members are above each other.
int vHeight = _height * _numMembers;
//// vX, t=0
//_t0X = new CudaArray2D(CUArrayFormat.Float, _width, vHeight, CudaArray2DNumChannels.One);
//_t0X.CopyFromHostToThis<float>(t0X.Data);
//new CudaTextureArray2D(_advectParticlesKernel, "vX_t0", CUAddressMode.Wrap, CUFilterMode.Linear, CUTexRefSetFlags.None, _t0X);
//// vY, t=0
//_t0Y = new CudaArray2D(CUArrayFormat.Float, _width, vHeight, CudaArray2DNumChannels.One);
//_t0Y.CopyFromHostToThis<float>(t0Y.Data);
//new CudaTextureArray2D(_advectParticlesKernel, "vY_t0", CUAddressMode.Wrap, CUFilterMode.Linear, CUTexRefSetFlags.None, _t0Y);
// vX, t=1
_t1X = new CudaArray2D(CUArrayFormat.Float, _width, vHeight, CudaArray2DNumChannels.One);
_t1X.CopyFromHostToThis <float>(t1X.Data);
new CudaTextureArray2D(_advectParticlesKernel, "vX_t1", CUAddressMode.Wrap, CUFilterMode.Linear, CUTexRefSetFlags.None, _t1X);
// vY, t=1
_t1Y = new CudaArray2D(CUArrayFormat.Float, _width, vHeight, CudaArray2DNumChannels.One);
_t1Y.CopyFromHostToThis <float>(t1Y.Data);
new CudaTextureArray2D(_advectParticlesKernel, "vY_t1", CUAddressMode.Wrap, CUFilterMode.Linear, CUTexRefSetFlags.None, _t1Y);
// ~~~~~~~~~~~~~ Create texture ~~~~~~~~~~~~~~~~~~~~ \\
// Create texture. Completely zero, except for one point.
Texture2DDescription desc = new Texture2DDescription
{
ArraySize = 1,
BindFlags = BindFlags.ShaderResource,
CpuAccessFlags = CpuAccessFlags.None,
Format = Format.R32_Float,
Width = _width,
Height = _height,
MipLevels = 1,
OptionFlags = ResourceOptionFlags.None,
SampleDescription = new SampleDescription(1, 0),
Usage = ResourceUsage.Default
};
// Put field data into stream/rectangle object
float[] zeros = new float[_width * _height];
Array.Clear(zeros, 0, zeros.Length);
// Fill the empty texture.
DataRectangle texData = new DataRectangle(_width * sizeof(float), new DataStream(zeros, true, true));
_pongFlowMap = new CudaDeviceVariable <float>(_width * _height);//new Texture2D(_device, desc, texData);
// Magically, copy to device happens here.
_pongFlowMap = zeros;
// Add one pixel for integration.
zeros[pos.X + pos.Y * _width] = 1;
texData = new DataRectangle(_width * sizeof(float), new DataStream(zeros, true, true));
// Create texture.
FlowMap = new Texture2D(_device, desc, texData);
// ~~~~~~~~~ Make textures mappable to CUDA ~~~~~~~~~~ \\
_cudaDxMapper.Add(new CudaDirectXInteropResource(FlowMap.ComPointer, CUGraphicsRegisterFlags.None, CudaContext.DirectXVersion.D3D11));
_cudaDxMapper.MapAllResources();
CudaArray2D lastFlowMap = _cudaDxMapper[0].GetMappedArray2D(0, 0);
new CudaTextureArray2D(_advectParticlesKernel, "flowMap", CUAddressMode.Wrap, CUFilterMode.Linear, CUTexRefSetFlags.None, lastFlowMap);
_cudaDxMapper.UnmapAllResources();
}
public void SetImage(CudaPitchedDeviceVariable <uchar4> image, Rotation orientation)
{
bool sizeChanged = false;
if (_realImageWidth != image.Width || _realImageHeight != image.Height || _rotation != orientation)
{
sizeChanged = true;
}
//Always create a new texture to recreate the mipmaps...
//if (_realImageWidth != image.Width || _realImageHeight != image.Height || _rotation != orientation)
{
_rotation = orientation;
_realImageWidth = image.Width;
_realImageHeight = image.Height;
_imageWidth = image.Width;
_imageHeight = image.Height;
if (_rotation == Rotation._90 || _rotation == Rotation._270)
{
_imageWidth = _realImageHeight;
_imageHeight = _realImageWidth;
}
initTexture(image.Width, image.Height);
}
d3dimage.Lock();
_graphicsres.MapAllResources();
CudaArray2D arr = _graphicsres[0].GetMappedArray2D(0, 0);
arr.CopyFromDeviceToThis <uchar4>(image);
_graphicsres.UnmapAllResources();
arr.Dispose();
d3dimage.Unlock();
//if size didn't change, don't change geometry
if (sizeChanged)
{
SetProjectionTransform();
SlimDX.Matrix rotMat = new SlimDX.Matrix();
switch (_rotation)
{
case Rotation._0:
rotMat = SlimDX.Matrix.RotationZ(0);
break;
case Rotation._90:
rotMat = SlimDX.Matrix.RotationZ((float)(90.0 / 180.0 * Math.PI));
break;
case Rotation._180:
rotMat = SlimDX.Matrix.RotationZ((float)(180.0 / 180.0 * Math.PI));
break;
case Rotation._270:
rotMat = SlimDX.Matrix.RotationZ((float)(270.0 / 180.0 * Math.PI));
break;
}
_device.SetTransform(TransformState.View, rotMat);
}
updateFrame();
}
private void Image_MouseMove(object sender, MouseEventArgs e)
{
System.Windows.Point position = GetPositionWithDpi(e);
int pX = (int)position.X;
int pY = (int)position.Y;
Point pixel = GetImagePixelFromMouseCoordinate(position);
uchar4[] p = new uchar4[1];
if (!double.IsInfinity(pixel.X) && !double.IsInfinity(pixel.Y))
{
d3dimage.Lock();
_graphicsres.MapAllResources();
CudaArray2D arr = _graphicsres[0].GetMappedArray2D(0, 0);
CUDAMemCpy2D copy = new CUDAMemCpy2D();
GCHandle handle = GCHandle.Alloc(p, GCHandleType.Pinned);
copy.dstHost = handle.AddrOfPinnedObject();
copy.srcArray = arr.CUArray;
copy.srcMemoryType = CUMemoryType.Array;
copy.dstMemoryType = CUMemoryType.Host;
copy.Height = 1;
copy.WidthInBytes = 4;
copy.srcXInBytes = (int)pixel.X * 4;
copy.srcY = (int)pixel.Y;
arr.CopyData(copy);
_graphicsres.UnmapAllResources();
arr.Dispose();
handle.Free();
d3dimage.Unlock();
}
SetValue(ColorOfPixelProperty, Color.FromArgb(p[0].w, p[0].z, p[0].y, p[0].x));
SetValue(PixelCoordinateProperty, pixel);
if (_clicked)
{
_viewShiftX += (-_lastX + pX) / _scaleFac * _projFac;
_viewShiftY += (-_lastY + pY) / _scaleFac * _projFac;
_lastX = pX;
_lastY = pY;
SlimDX.Matrix matScale = SlimDX.Matrix.Scaling(_scaleFac, _scaleFac, 1);
float shiftScale = Math.Min((float)ActualWidthDpi, (float)ActualHeightDpi);
SlimDX.Matrix matTrans = SlimDX.Matrix.Translation(_viewShiftX / shiftScale * _scaleFac, _viewShiftY / shiftScale * _scaleFac, 0);
SlimDX.Matrix mat = matScale * matTrans;
SlimDX.Matrix rotMat = new SlimDX.Matrix();
switch (_rotation)
{
case Rotation._0:
rotMat = SlimDX.Matrix.RotationZ(0);
break;
case Rotation._90:
rotMat = SlimDX.Matrix.RotationZ((float)(90.0 / 180.0 * Math.PI));
break;
case Rotation._180:
rotMat = SlimDX.Matrix.RotationZ((float)(180.0 / 180.0 * Math.PI));
break;
case Rotation._270:
rotMat = SlimDX.Matrix.RotationZ((float)(270.0 / 180.0 * Math.PI));
break;
}
_device.SetTransform(TransformState.View, rotMat * mat);
updateFrame();
}
}
