/// <summary>
/// For IDisposable
/// </summary>
/// <param name="fDisposing"></param>
protected virtual void Dispose(bool fDisposing)
{
if (fDisposing && !disposed)
{
_devVar.Dispose();
disposed = true;
// the _texref reference is not destroyed explicitly, as it is done automatically when module is unloaded
}
if (!fDisposing && !disposed)
{
Debug.WriteLine(String.Format("ManagedCUDA not-disposed warning: {0}", this.GetType()));
}
}
//Clean up before closing
private void Form1_FormClosing(object sender, FormClosingEventArgs e)
{
isRunning = false;
isInit = false;
cuda_vbo_resource.Dispose();
texref.Dispose();
dvfield.Dispose();
vxfield.Dispose();
vyfield.Dispose();
planc2r.Dispose();
planr2c.Dispose();
GL.BindBuffer(BufferTarget.ArrayBuffer, 0);
GL.DeleteBuffers(1, ref vbo);
stopwatch.Dispose();
ctx.Dispose();
}
private void Form1_FormClosing(object sender, FormClosingEventArgs e)
{
isRunning = false;
//Cleanup
if (graphicsres != null)
{
graphicsres.Dispose();
}
if (g_mparticles != null)
{
g_mparticles.Dispose();
}
if (stopwatch != null)
{
stopwatch.Dispose();
}
if (texref != null)
{
texref.Dispose();
}
if (g_dvfield != null)
{
g_dvfield.Dispose();
}
if (g_vxfield != null)
{
g_vxfield.Dispose();
}
if (g_vyfield != null)
{
g_vyfield.Dispose();
}
if (g_planc2r != null)
{
g_planc2r.Dispose();
}
if (g_planr2c != null)
{
g_planr2c.Dispose();
}
if (g_pVB != null)
{
g_pVB.Dispose();
}
if (g_pTexture != null)
{
g_pTexture.Dispose();
}
if (device != null)
{
device.Dispose();
}
if (d3d != null)
{
d3d.Dispose();
}
if (ctx != null)
{
ctx.Dispose();
}
}
private void myWindow_Closing(object sender, System.ComponentModel.CancelEventArgs e)
{
//Stop render loop before closing
if (frameTimer != null)
{
frameTimer.Tick -= new EventHandler(frameTimer_Tick);
frameTimer.Stop();
}
//Cleanup
if (graphicsres != null)
{
graphicsres.Dispose();
}
if (g_mparticles != null)
{
g_mparticles.Dispose();
}
if (stopwatch != null)
{
stopwatch.Dispose();
}
if (texref != null)
{
texref.Dispose();
}
if (g_dvfield != null)
{
g_dvfield.Dispose();
}
if (g_vxfield != null)
{
g_vxfield.Dispose();
}
if (g_vyfield != null)
{
g_vyfield.Dispose();
}
if (g_planc2r != null)
{
g_planc2r.Dispose();
}
if (g_planr2c != null)
{
g_planr2c.Dispose();
}
if (g_pVB != null)
{
g_pVB.Dispose();
}
if (g_pTexture != null)
{
g_pTexture.Dispose();
}
if (device != null)
{
device.Dispose();
}
if (d3d != null)
{
d3d.Dispose();
}
if (ctx != null)
{
ctx.Dispose();
}
}
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);
}
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);
}
private void btn_openImg_Click(object sender, EventArgs e)
{
OpenFileDialog ofd = new OpenFileDialog();
ofd.Filter = "Images|*.bmp;*.jpg;*.jpeg;*.tiff;*.tif;*.png;*.gif";
if (ofd.ShowDialog() != System.Windows.Forms.DialogResult.OK)
{
return;
}
bmp_src = new Bitmap(ofd.FileName);
if (bmp_src.PixelFormat != PixelFormat.Format24bppRgb)
{
MessageBox.Show("Only 24-bit RGB images are supported!");
bmp_src = null;
bmp_mask = null;
bmp_res = null;
if (npp_bmp_src != null)
{
npp_bmp_src.Dispose();
}
if (npp_bmp_res != null)
{
npp_bmp_res.Dispose();
}
if (npp_bmp_mask != null)
{
npp_bmp_mask.Dispose();
}
if (d_bmp_src != null)
{
d_bmp_src.Dispose();
}
if (d_bmp_res != null)
{
d_bmp_res.Dispose();
}
if (d_bmp_mask != null)
{
d_bmp_mask.Dispose();
}
return;
}
width = bmp_src.Width;
height = bmp_src.Height;
marker = new int[width * height];
bmp_res = new Bitmap(width, height, PixelFormat.Format32bppArgb);
bmp_mask = new Bitmap(width, height, PixelFormat.Format8bppIndexed);
SetPalette(bmp_mask);
pictureBox_src.Image = bmp_src;
selection.x = (int)Math.Ceiling(width * 0.1);
selection.y = (int)Math.Ceiling(height * 0.1);
selection.width = width - 2 * selection.x;
selection.height = height - 2 * selection.y;
if (npp_bmp_src != null)
{
npp_bmp_src.Dispose();
}
if (npp_bmp_res != null)
{
npp_bmp_res.Dispose();
}
if (npp_bmp_mask != null)
{
npp_bmp_mask.Dispose();
}
if (d_bmp_src != null)
{
d_bmp_src.Dispose();
}
if (d_bmp_res != null)
{
d_bmp_res.Dispose();
}
if (d_bmp_mask != null)
{
d_bmp_mask.Dispose();
}
NPPImage_8uC3 npp_temp = new NPPImage_8uC3(width, height);
CudaPitchedDeviceVariable <uchar3> d_bmp_temp = new CudaPitchedDeviceVariable <uchar3>(npp_temp.DevicePointer, width, height, npp_temp.Pitch);
npp_temp.CopyToDevice(bmp_src);
npp_bmp_src = new NPPImage_8uC4(width, height);
npp_bmp_res = new NPPImage_8uC4(width, height);
npp_bmp_mask = new NPPImage_8uC1(width, height);
d_bmp_src = new CudaPitchedDeviceVariable <uchar4>(npp_bmp_src.DevicePointer, width, height, npp_bmp_src.Pitch);
d_bmp_res = new CudaPitchedDeviceVariable <uchar4>(npp_bmp_res.DevicePointer, width, height, npp_bmp_res.Pitch);
d_bmp_mask = new CudaPitchedDeviceVariable <byte>(npp_bmp_mask.DevicePointer, width, height, npp_bmp_mask.Pitch);
grabcut = new GrabCut(d_bmp_src, d_bmp_mask, width, height);
grabcut.grabCutUtils.convertRGBToRGBA(d_bmp_src, d_bmp_temp, width, height);
d_bmp_temp.Dispose();
npp_temp.Dispose();
}