private void initGLAndCuda()
{
//Create render target control
m_renderControl = new OpenTK.GLControl(GraphicsMode.Default, 1, 0, GraphicsContextFlags.Default);
m_renderControl.Dock = DockStyle.Fill;
m_renderControl.BackColor = Color.White;
m_renderControl.BorderStyle = BorderStyle.FixedSingle;
m_renderControl.KeyDown += new KeyEventHandler(m_renderControl_KeyDown);
m_renderControl.MouseMove += new MouseEventHandler(m_renderControl_MouseMove);
m_renderControl.MouseDown += new MouseEventHandler(m_renderControl_MouseDown);
m_renderControl.SizeChanged += new EventHandler(m_renderControl_SizeChanged);
panel1.Controls.Add(m_renderControl);
Console.WriteLine(" OpenGL device is Available");
int deviceID = CudaContext.GetMaxGflopsDeviceId();
ctx = CudaContext.CreateOpenGLContext(deviceID, CUCtxFlags.BlockingSync);
string console = string.Format("CUDA device [{0}] has {1} Multi-Processors", ctx.GetDeviceName(), ctx.GetDeviceInfo().MultiProcessorCount);
Console.WriteLine(console);
CUmodule module = ctx.LoadModulePTX("kernel.ptx");
addForces_k = new CudaKernel("addForces_k", module, ctx);
advectVelocity_k = new CudaKernel("advectVelocity_k", module, ctx);
diffuseProject_k = new CudaKernel("diffuseProject_k", module, ctx);
updateVelocity_k = new CudaKernel("updateVelocity_k", module, ctx);
advectParticles_k = new CudaKernel("advectParticles_OGL", module, ctx);
hvfield = new cData[DS];
dvfield = new CudaPitchedDeviceVariable<cData>(DIM, DIM);
tPitch = dvfield.Pitch;
dvfield.CopyToDevice(hvfield);
vxfield = new CudaDeviceVariable<cData>(DS);
vyfield = new CudaDeviceVariable<cData>(DS);
// Create particle array
particles = new cData[DS];
initParticles(particles, DIM, DIM);
// TODO: update kernels to use the new unpadded memory layout for perf
// rather than the old FFTW-compatible layout
planr2c = new CudaFFTPlan2D(DIM, DIM, cufftType.R2C, Compatibility.FFTWPadding);
planc2r = new CudaFFTPlan2D(DIM, DIM, cufftType.C2R, Compatibility.FFTWPadding);
GL.GenBuffers(1, out vbo);
GL.BindBuffer(BufferTarget.ArrayBuffer, vbo);
GL.BufferData<cData>(BufferTarget.ArrayBuffer, new IntPtr(cData.SizeOf * DS), particles, BufferUsageHint.DynamicDraw);
int bsize;
GL.GetBufferParameter(BufferTarget.ArrayBuffer, BufferParameterName.BufferSize, out bsize);
if (bsize != DS * cData.SizeOf)
throw new Exception("Sizes don't match.");
GL.BindBuffer(BufferTarget.ArrayBuffer, 0);
cuda_vbo_resource = new CudaGraphicsInteropResourceCollection();
cuda_vbo_resource.Add(new CudaOpenGLBufferInteropResource(vbo, CUGraphicsRegisterFlags.None));
texref = new CudaTextureArray2D(advectVelocity_k, "texref", CUAddressMode.Wrap, CUFilterMode.Linear, 0, CUArrayFormat.Float, DIM, DIM, CudaArray2DNumChannels.Two);
stopwatch = new CudaStopWatch(CUEventFlags.Default);
reshape();
isInit = true;
display();
}
private void initTexture(int width, int height)
{
if (_texture == null)
{
_texture = new Texture(_device, width, height, 0, Usage.AutoGenerateMipMap, Format.X8R8G8B8, Pool.Default);
}
else
{
_graphicsres[0].Unregister();
_graphicsres.Clear();
_texture.Dispose();
_texture = new Texture(_device, width, height, 0, Usage.AutoGenerateMipMap, Format.X8R8G8B8, Pool.Default);
}
CudaDirectXInteropResource resource = new CudaDirectXInteropResource(_texture.ComPointer, CUGraphicsRegisterFlags.None, CudaContext.DirectXVersion.D3D9, CUGraphicsMapResourceFlags.None);
_graphicsres.Add(resource);
if (_vertexBufferTexture == null)
{
_vertexBufferTexture = new VertexBuffer(_device, 4 * Marshal.SizeOf(typeof(vertex)), Usage.None, VertexFormat.Position | VertexFormat.Texture1, Pool.Default);
}
DataStream str = _vertexBufferTexture.Lock(0, 4 * Marshal.SizeOf(typeof(vertex)), LockFlags.None);
str.Position = 0;
str.WriteRange <vertex>(GetVertices(width, height));
_vertexBufferTexture.Unlock();
//_device.SetTexture(0, _texture);
}
private void InitializeVB()
{
g_pVB = new VertexBuffer(device, DS * Marshal.SizeOf(typeof(vertex)), Usage.None, VertexFormat.Position | VertexFormat.Diffuse, Pool.Default);
updateVB();
graphicsres.Add(new CudaDirectXInteropResource(g_pVB.ComPointer, CUGraphicsRegisterFlags.None, CudaContext.DirectXVersion.D3D9));
}
void m_renderControl_KeyDown(object sender, KeyEventArgs e)
{
if (e.KeyCode == Keys.Escape)
{
Close();
}
if (e.KeyCode == Keys.R)
{
Array.Clear(hvfield, 0, DS);
dvfield.CopyToDevice(hvfield);
initParticles(particles, DIM, DIM);
cuda_vbo_resource.Clear();
GL.BindBuffer(BufferTarget.ArrayBuffer, vbo);
GL.BufferData <cData>(BufferTarget.ArrayBuffer, new IntPtr(cData.SizeOf * DS), particles, BufferUsageHint.DynamicDraw);
GL.BindBuffer(BufferTarget.ArrayBuffer, 0);
cuda_vbo_resource.Add(new CudaOpenGLBufferInteropResource(vbo, CUGraphicsRegisterFlags.None));
}
}
/// <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();
}
private void InitializeVB()
{
g_pVB = new VertexBuffer(device, DS * Marshal.SizeOf(typeof(vertex)), Usage.None, VertexFormat.Position | VertexFormat.Diffuse, Pool.Default);
updateVB();
graphicsres = new CudaGraphicsInteropResourceCollection();
graphicsres.Add(new CudaDirectXInteropResource(g_pVB.ComPointer, CUGraphicsRegisterFlags.None, CudaContext.DirectXVersion.D3D9));
}
private void initGLAndCuda()
{
//Create render target control
m_renderControl = new OpenTK.GLControl(GraphicsMode.Default, 1, 0, GraphicsContextFlags.Default);
m_renderControl.Dock = DockStyle.Fill;
m_renderControl.BackColor = Color.White;
m_renderControl.BorderStyle = BorderStyle.FixedSingle;
m_renderControl.KeyDown += new KeyEventHandler(m_renderControl_KeyDown);
m_renderControl.MouseMove += new MouseEventHandler(m_renderControl_MouseMove);
m_renderControl.MouseDown += new MouseEventHandler(m_renderControl_MouseDown);
m_renderControl.SizeChanged += new EventHandler(m_renderControl_SizeChanged);
panel1.Controls.Add(m_renderControl);
Console.WriteLine(" OpenGL device is Available");
int deviceID = CudaContext.GetMaxGflopsDeviceId();
ctx = CudaContext.CreateOpenGLContext(deviceID, CUCtxFlags.BlockingSync);
string console = string.Format("CUDA device [{0}] has {1} Multi-Processors", ctx.GetDeviceName(), ctx.GetDeviceInfo().MultiProcessorCount);
Console.WriteLine(console);
CUmodule module = ctx.LoadModulePTX("kernel.ptx");
addForces_k = new CudaKernel("addForces_k", module, ctx);
advectVelocity_k = new CudaKernel("advectVelocity_k", module, ctx);
diffuseProject_k = new CudaKernel("diffuseProject_k", module, ctx);
updateVelocity_k = new CudaKernel("updateVelocity_k", module, ctx);
advectParticles_k = new CudaKernel("advectParticles_OGL", module, ctx);
hvfield = new cData[DS];
dvfield = new CudaPitchedDeviceVariable <cData>(DIM, DIM);
tPitch = dvfield.Pitch;
dvfield.CopyToDevice(hvfield);
vxfield = new CudaDeviceVariable <cData>(DS);
vyfield = new CudaDeviceVariable <cData>(DS);
// Create particle array
particles = new cData[DS];
initParticles(particles, DIM, DIM);
// TODO: update kernels to use the new unpadded memory layout for perf
// rather than the old FFTW-compatible layout
planr2c = new CudaFFTPlan2D(DIM, DIM, cufftType.R2C, Compatibility.FFTWPadding);
planc2r = new CudaFFTPlan2D(DIM, DIM, cufftType.C2R, Compatibility.FFTWPadding);
GL.GenBuffers(1, out vbo);
GL.BindBuffer(BufferTarget.ArrayBuffer, vbo);
GL.BufferData <cData>(BufferTarget.ArrayBuffer, new IntPtr(cData.SizeOf * DS), particles, BufferUsageHint.DynamicDraw);
int bsize;
GL.GetBufferParameter(BufferTarget.ArrayBuffer, BufferParameterName.BufferSize, out bsize);
if (bsize != DS * cData.SizeOf)
{
throw new Exception("Sizes don't match.");
}
GL.BindBuffer(BufferTarget.ArrayBuffer, 0);
cuda_vbo_resource = new CudaGraphicsInteropResourceCollection();
cuda_vbo_resource.Add(new CudaOpenGLBufferInteropResource(vbo, CUGraphicsRegisterFlags.None));
texref = new CudaTextureArray2D(advectVelocity_k, "texref", CUAddressMode.Wrap, CUFilterMode.Linear, 0, CUArrayFormat.Float, DIM, DIM, CudaArray2DNumChannels.Two);
stopwatch = new CudaStopWatch(CUEventFlags.Default);
reshape();
isInit = true;
display();
}