public static void WriterValues(this Context oclContext, CommandQueue oclCQ, CL.Mem oclBuff, int[] values)
{
oclCQ.EnqueueWriteBuffer(oclBuff, true, 0, values.Length, values.ToIntPtr());
oclCQ.EnqueueBarrier();
oclCQ.Finish();
return;
}
/* возвращает буфер */
public unsafe Mem CreateBuffer()
{
if (buffer == null)
{
/* указатели в C# используются с fixed (противодействие для GC) */
fixed(short *dataptr = data)
{
buffer = manager_this.Context.CreateBuffer(MemFlags.COPY_HOST_PTR, data.Count() * 2, new IntPtr(dataptr));
}
}
return(buffer);
}
public static int[] ToInts(this Context oclContext, CommandQueue oclCQ, CL.Mem oclBuff, int Len)
{
int[] Ints = new int[Len];
Mem buffer = oclContext.CreateBuffer((MemFlags.WRITE_ONLY | MemFlags.USE_HOST_PTR), Len * 4, Ints.ToIntPtr());
oclCQ.EnqueueCopyBuffer(oclBuff, buffer, 0, 0, Len);
//oclCQ.EnqueueReadBuffer(oclBuff, true, 0, Len, Ints.ToIntPtr());
oclCQ.EnqueueBarrier();
IntPtr p = oclCQ.EnqueueMapBuffer(buffer, true, MapFlags.READ, 0, Len);
oclCQ.EnqueueUnmapMemObject(buffer, p);
oclCQ.Finish();
buffer.Dispose();
return(Ints);
}
/*
private unsafe Mem GetCounter()
{
fixed (ulong* dataptr = &voxelctr)
{
counter = manager.Context.CreateBuffer(MemFlags.READ_WRITE, 8, new IntPtr(dataptr));
}
return counter;
}*/
private unsafe void DoRayCasting(BitmapData output)
{
try
{
int deviceIndex = 0;
outputBuffer = manager.Context.CreateBuffer(MemFlags.USE_HOST_PTR, output.Stride * output.Height, output.Scan0);
if (first || changeDistance)
{
// модель камеры UVN
camPos = new Float4()
{
S0 =
vol.GetSize() / 2 - (float)Math.Cos(camAngle * Math.PI / 180) * camDist,
S1 = vol.GetSize() / 2,
S2 = vol.GetSize() / 2 - (float)Math.Sin(camAngle * Math.PI / 180) * camDist,
S3 = 0
};
first = false;
changeDistance = false;
camPosOld = camPos;
}
else
{
// поворот вокруг оси куба визуализации
if (angleChange && leftChange)
{
camPosOld.S0 -= camLookAt.S0;
camPosOld.S2 -= camLookAt.S2;
camPos.S0 = (float)Math.Cos(camAngle * Math.PI / 180) * camPosOld.S0 + (float)Math.Sin(camAngle * Math.PI / 180) * camPosOld.S2;
camPos.S1 = vol.GetSize() / 2;
camPos.S2 = -(float)Math.Sin(camAngle * Math.PI / 180) * camPosOld.S0 + (float)Math.Cos(camAngle * Math.PI / 180) * camPosOld.S2;
camPos.S3 = 0;
camPos.S0 += camLookAt.S0;
camPos.S2 += camLookAt.S2;
camPosOld = camPos;
angleChange = false;
leftChange = false;
}
if (angleChange && rightChange)
{
camPosOld.S0 -= camLookAt.S0;
camPosOld.S2 -= camLookAt.S2;
camPos.S0 = (float)Math.Cos(camAngle * Math.PI / 180) * camPosOld.S0 - (float)Math.Sin(camAngle * Math.PI / 180) * camPosOld.S2;
camPos.S1 = vol.GetSize() / 2;
camPos.S2 = (float)Math.Sin(camAngle * Math.PI / 180) * camPosOld.S0 + (float)Math.Cos(camAngle * Math.PI / 180) * camPosOld.S2;
camPos.S3 = 0;
camPos.S0 += camLookAt.S0;
camPos.S2 += camLookAt.S2;
camPosOld = camPos;
angleChange = false;
leftChange = false;
}
}
camLookAt = new Float4()
{
S0 = vol.GetSize() / camfactorX,
S1 = vol.GetSize() / camfactorX,
S2 = vol.GetSize() / camfactorZ,
S3 = 0
};
//light = camPos;
// направление камеры, UVN модель
camForward = camLookAt.Sub(camPos).Normalize(); // направление просмотра
var up = new Float4(0.0f, 1.0f, 0.0f, 0.0f);
var right = MathClass.Cross(up, camForward).Normalize().Times(1.5f);
up = MathClass.Cross(camForward, right).Normalize().Times(-1.5f);
/* обработка выходного изображения BitmapData в OpenCl устройстве */
for (var x = 0; x < output.Width; x += blocksize)
{
for (var y = 0; y < output.Height; y += blocksize)
{
var rayTracingGlobalWorkSize = new IntPtr[2]; // work_dim = 2
rayTracingGlobalWorkSize[0] = (IntPtr)(output.Width - x > blocksize ? blocksize : output.Width - x);
rayTracingGlobalWorkSize[1] = (IntPtr)(output.Height - y > blocksize ? blocksize : output.Height - y);
var rayTracingGlobalOffset = new IntPtr[2];
rayTracingGlobalOffset[0] = (IntPtr)x;
rayTracingGlobalOffset[1] = (IntPtr)y;
float ka = (float)(Convert.ToDouble(kamb.Text));
float kd = (float)(Convert.ToDouble(kdiff.Text));
float ks = (float)(Convert.ToDouble(kspec.Text));
float exp = (float)(Convert.ToDouble(specexp.Text));
float kkc = (float)(Convert.ToDouble(this.kc.Text));
float kkl = (float)(Convert.ToDouble(this.kl.Text));
float kkq = (float)(Convert.ToDouble(this.kq.Text));
/* передали аргументы в kernel функцию */
kernel.SetArg(0, output.Width);
kernel.SetArg(1, output.Height);
kernel.SetArg(2, outputBuffer); // в ядре с global, поскольку для выполнения требуется доступ к output
kernel.SetArg(3, output.Stride);
kernel.SetArg(4, camPos);
kernel.SetArg(5, camForward);
kernel.SetArg(6, right);
kernel.SetArg(7, up);
kernel.SetArg(8, vol.CreateBuffer());
kernel.SetArg(9, vol.GetSize());
kernel.SetArg(10, light);
kernel.SetArg(11, boxMinCon);
kernel.SetArg(12, boxMaxCon);
kernel.SetArg(13, Convert.ToInt16(colorMi.Text));
kernel.SetArg(14, Convert.ToInt16(colorMa.Text));
kernel.SetArg(15, _cutArrea.Checked ? (short)1 : (short)0);
kernel.SetArg(16, _trilinear.Checked ? (short)1 : (short)0);
kernel.SetArg(17, tf.Checked ? (short)1: (short)0 );
kernel.SetArg(18, GetColors());
kernel.SetArg(19, winWidth_vox);
kernel.SetArg(20, winCentre_vox);
kernel.SetArg(21, form_this.knots_counter);
kernel.SetArg(22, Convert.ToInt16(colorMi2.Text));
kernel.SetArg(23, Convert.ToInt16(colorMa2.Text));
kernel.SetArg(24, GetOpacity());
kernel.SetArg(25, ka);
kernel.SetArg(26, kd);
kernel.SetArg(27, ks);
kernel.SetArg(28, exp);
kernel.SetArg(29, kkc);
kernel.SetArg(30, kkl);
kernel.SetArg(31, kkq);
//kernel.SetArg(32, GetCounter());
/* Ставит в очередь команду для исполнения kernel на устройстве */
/*
rayTracingGlobalOffset -
* globalWorkOffset: может использоваться для указания массива значений
* размерности work_dim unsigned который описывает смещение используемое для расчета global ID work-item
* вместо того чтобы global IDs всегда начинался со смещение (0, 0,... 0).
* rayTracingGlobalWorkSize -
* globalWorkSize: общее число global work-items вычисляется как global_work_size[0] *...* global_work_size[work_dim - 1].
*
*/
manager.CQ[deviceIndex].EnqueueNDRangeKernel(kernel, 2, rayTracingGlobalOffset, rayTracingGlobalWorkSize, null);
}
}
/* подождали пока все work-items выполнятся */
manager.CQ[deviceIndex].EnqueueBarrier();
/* для того чтобы получить доступ к памяти и записать в выходное изображение мы просим у OpenCL *наложить* данные в хост-устройство */
IntPtr p = manager.CQ[deviceIndex].EnqueueMapBuffer(outputBuffer, true, MapFlags.READ, IntPtr.Zero, (IntPtr)(output.Stride * output.Height));
//IntPtr z = manager.CQ[deviceIndex].EnqueueMapBuffer(counter, true, MapFlags.READ_WRITE, IntPtr.Zero, (IntPtr)(sizeof(ulong)));
/* когда мы заканчиваем работать с буфером надо вызвать эту функцию */
manager.CQ[deviceIndex].EnqueueUnmapMemObject(outputBuffer, p);
//manager.CQ[deviceIndex].EnqueueUnmapMemObject(counter, z);
manager.CQ[deviceIndex].Finish();
realctr += voxelctr;
voxelCounter.Text = Convert.ToString(realctr);
}
catch (Exception ex)
{
MessageBox.Show("Ray casting exception:" + ex.Message, "Exception");
//Environment.Exit(-1);
}
finally
{
if (outputBuffer != null)
{
outputBuffer.Dispose();
}
}
}
public void SetMemArg(int argIndex, Mem mem)
{
SetIntPtrArg(argIndex, mem.MemID);
}
unsafe public FLACCLTask(Program _openCLProgram, int channelsCount, int channels, uint bits_per_sample, int max_frame_size, FLACCLWriter writer, int groupSize, bool gpuOnly, bool gpuRice)
{
this.UseGPUOnly = gpuOnly;
this.UseGPURice = gpuOnly && gpuRice;
this.UseMappedMemory = writer._settings.MappedMemory || writer._settings.DeviceType == OpenCLDeviceType.CPU;
this.groupSize = groupSize;
this.channels = channels;
this.channelsCount = channelsCount;
this.writer = writer;
openCLProgram = _openCLProgram;
#if DEBUG
var prop = CommandQueueProperties.PROFILING_ENABLE;
#else
var prop = CommandQueueProperties.NONE;
#endif
openCLCQ = openCLProgram.Context.CreateCommandQueue(openCLProgram.Context.Devices[0], prop);
int MAX_ORDER = this.writer.eparams.max_prediction_order;
int MAX_FRAMES = this.writer.framesPerTask;
int MAX_CHANNELSIZE = MAX_FRAMES * ((writer.eparams.block_size + 3) & ~3);
residualTasksLen = sizeof(FLACCLSubframeTask) * 32 * channelsCount * MAX_FRAMES;
bestResidualTasksLen = sizeof(FLACCLSubframeTask) * channels * MAX_FRAMES;
int samplesBufferLen = writer.PCM.BlockAlign * MAX_CHANNELSIZE * channelsCount;
int residualBufferLen = sizeof(int) * MAX_CHANNELSIZE * channels; // need to adjust residualOffset?
int partitionsLen = sizeof(int) * ((writer.PCM.BitsPerSample > 16 ? 31 : 15) * 2 << 8) * channels * MAX_FRAMES;
int riceParamsLen = sizeof(int) * (4 << 8) * channels * MAX_FRAMES;
int autocorLen = sizeof(float) * (MAX_ORDER + 1) * lpc.MAX_LPC_WINDOWS * channelsCount * MAX_FRAMES;
int lpcDataLen = autocorLen * 32;
int resOutLen = sizeof(int) * channelsCount * (lpc.MAX_LPC_WINDOWS * lpc.MAX_LPC_ORDER + 8) * MAX_FRAMES;
int wndLen = sizeof(float) * MAX_CHANNELSIZE /** 2*/ * lpc.MAX_LPC_WINDOWS;
int selectedLen = sizeof(int) * 32 * channelsCount * MAX_FRAMES;
int riceLen = sizeof(int) * channels * MAX_CHANNELSIZE;
if (!this.UseMappedMemory)
{
clSamplesBytes = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, samplesBufferLen / 2);
clResidual = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, residualBufferLen);
clBestRiceParams = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, riceParamsLen / 4);
clResidualTasks = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, residualTasksLen);
clBestResidualTasks = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, bestResidualTasksLen);
clWindowFunctions = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, wndLen);
clSelectedTasks = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, selectedLen);
clRiceOutput = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, riceLen);
clSamplesBytesPinned = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, samplesBufferLen / 2);
clResidualPinned = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, residualBufferLen);
clBestRiceParamsPinned = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, riceParamsLen / 4);
clResidualTasksPinned = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, residualTasksLen);
clBestResidualTasksPinned = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, bestResidualTasksLen);
clWindowFunctionsPinned = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, wndLen);
clSelectedTasksPinned = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, selectedLen);
clRiceOutputPinned = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, riceLen);
clSamplesBytesPtr = openCLCQ.EnqueueMapBuffer(clSamplesBytesPinned, true, MapFlags.READ_WRITE, 0, samplesBufferLen / 2);
clResidualPtr = openCLCQ.EnqueueMapBuffer(clResidualPinned, true, MapFlags.READ_WRITE, 0, residualBufferLen);
clBestRiceParamsPtr = openCLCQ.EnqueueMapBuffer(clBestRiceParamsPinned, true, MapFlags.READ_WRITE, 0, riceParamsLen / 4);
clResidualTasksPtr = openCLCQ.EnqueueMapBuffer(clResidualTasksPinned, true, MapFlags.READ_WRITE, 0, residualTasksLen);
clBestResidualTasksPtr = openCLCQ.EnqueueMapBuffer(clBestResidualTasksPinned, true, MapFlags.READ_WRITE, 0, bestResidualTasksLen);
clWindowFunctionsPtr = openCLCQ.EnqueueMapBuffer(clWindowFunctionsPinned, true, MapFlags.READ_WRITE, 0, wndLen);
clSelectedTasksPtr = openCLCQ.EnqueueMapBuffer(clSelectedTasksPinned, true, MapFlags.READ_WRITE, 0, selectedLen);
clRiceOutputPtr = openCLCQ.EnqueueMapBuffer(clRiceOutputPinned, true, MapFlags.READ_WRITE, 0, riceLen);
}
else
{
clSamplesBytes = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, (uint)samplesBufferLen / 2);
clResidual = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, residualBufferLen);
clBestRiceParams = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, riceParamsLen / 4);
clResidualTasks = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, residualTasksLen);
clBestResidualTasks = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, bestResidualTasksLen);
clWindowFunctions = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, wndLen);
clSelectedTasks = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, selectedLen);
clRiceOutput = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, riceLen);
clSamplesBytesPtr = openCLCQ.EnqueueMapBuffer(clSamplesBytes, true, MapFlags.READ_WRITE, 0, samplesBufferLen / 2);
clResidualPtr = openCLCQ.EnqueueMapBuffer(clResidual, true, MapFlags.READ_WRITE, 0, residualBufferLen);
clBestRiceParamsPtr = openCLCQ.EnqueueMapBuffer(clBestRiceParams, true, MapFlags.READ_WRITE, 0, riceParamsLen / 4);
clResidualTasksPtr = openCLCQ.EnqueueMapBuffer(clResidualTasks, true, MapFlags.READ_WRITE, 0, residualTasksLen);
clBestResidualTasksPtr = openCLCQ.EnqueueMapBuffer(clBestResidualTasks, true, MapFlags.READ_WRITE, 0, bestResidualTasksLen);
clWindowFunctionsPtr = openCLCQ.EnqueueMapBuffer(clWindowFunctions, true, MapFlags.READ_WRITE, 0, wndLen);
clSelectedTasksPtr = openCLCQ.EnqueueMapBuffer(clSelectedTasks, true, MapFlags.READ_WRITE, 0, selectedLen);
clRiceOutputPtr = openCLCQ.EnqueueMapBuffer(clRiceOutput, true, MapFlags.READ_WRITE, 0, riceLen);
//clSamplesBytesPtr = clSamplesBytes.HostPtr;
//clResidualPtr = clResidual.HostPtr;
//clBestRiceParamsPtr = clBestRiceParams.HostPtr;
//clResidualTasksPtr = clResidualTasks.HostPtr;
//clBestResidualTasksPtr = clBestResidualTasks.HostPtr;
//clWindowFunctionsPtr = clWindowFunctions.HostPtr;
}
clSamples = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, samplesBufferLen);
clLPCData = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, lpcDataLen);
clAutocorOutput = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, autocorLen);
clSelectedTasksSecondEstimate = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, selectedLen);
clSelectedTasksBestMethod = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, selectedLen);
if (UseGPUOnly)
{
clPartitions = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, partitionsLen);
clRiceParams = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, riceParamsLen);
}
//openCLCQ.EnqueueMapBuffer(clSamplesBytes, true, MapFlags.WRITE, 0, samplesBufferLen / 2);
clComputeAutocor = openCLProgram.CreateKernel("clComputeAutocor");
clStereoDecorr = openCLProgram.CreateKernel("clStereoDecorr");
//cudaChannelDecorr = openCLProgram.CreateKernel("clChannelDecorr");
clChannelDecorr2 = openCLProgram.CreateKernel("clChannelDecorr2");
clChannelDecorrX = openCLProgram.CreateKernel("clChannelDecorrX");
clFindWastedBits = openCLProgram.CreateKernel("clFindWastedBits");
clComputeLPC = openCLProgram.CreateKernel("clComputeLPC");
clQuantizeLPC = openCLProgram.CreateKernel("clQuantizeLPC");
//cudaComputeLPCLattice = openCLProgram.CreateKernel("clComputeLPCLattice");
clSelectStereoTasks = openCLProgram.CreateKernel("clSelectStereoTasks");
clEstimateResidual = openCLProgram.CreateKernel("clEstimateResidual");
clChooseBestMethod = openCLProgram.CreateKernel("clChooseBestMethod");
if (UseGPUOnly)
{
clEncodeResidual = openCLProgram.CreateKernel("clEncodeResidual");
if (openCLCQ.Device.DeviceType != DeviceType.CPU)
{
clCalcPartition = openCLProgram.CreateKernel("clCalcPartition");
clCalcPartition16 = openCLProgram.CreateKernel("clCalcPartition16");
}
clSumPartition = openCLProgram.CreateKernel("clSumPartition");
clFindRiceParameter = openCLProgram.CreateKernel("clFindRiceParameter");
clFindPartitionOrder = openCLProgram.CreateKernel("clFindPartitionOrder");
if (UseGPURice)
{
clCalcOutputOffsets = openCLProgram.CreateKernel("clCalcOutputOffsets");
clRiceEncoding = openCLProgram.CreateKernel("clRiceEncoding");
}
}
samplesBuffer = new int[MAX_CHANNELSIZE * channelsCount];
outputBuffer = new byte[max_frame_size * MAX_FRAMES + 1];
frame = new FlacFrame(channelsCount);
frame.writer = new BitWriter(outputBuffer, 0, outputBuffer.Length);
if (writer._settings.DoVerify)
{
verify = new FlakeReader(new AudioPCMConfig((int)bits_per_sample, channels, 44100));
verify.DoCRC = false;
}
}
public Mandelbrot( Platform platform, int width, int height )
{
openCLPlatform = platform;
openCLDevices = openCLPlatform.QueryDevices(DeviceType.ALL);
openCLContext = openCLPlatform.CreateDefaultContext();
openCLCQ = openCLContext.CreateCommandQueue(openCLDevices[0], CommandQueueProperties.PROFILING_ENABLE);
mandelBrotProgram = openCLContext.CreateProgramWithSource(File.ReadAllText("Mandelbrot.cl"));
try
{
mandelBrotProgram.Build();
}
catch (OpenCLException)
{
string buildLog = mandelBrotProgram.GetBuildLog(openCLDevices[0]);
MessageBox.Show(buildLog,"Build error(64 bit debug sessions in vs2008 always fail like this - debug in 32 bit or use vs2010)");
Application.Exit();
}
mandelbrotKernel = mandelBrotProgram.CreateKernel("Mandelbrot");
Left = -2.0f;
Top = 2.0f;
Right = 2.0f;
Bottom = -2.0f;
BitmapWidth = width;
BitmapHeight = height;
mandelbrotMemBuffer = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.WRITE_ONLY), width*height*4, IntPtr.Zero);
}
public Core(int Nxp,int Nyp, int Nzp, int Ntm, double Bbeta, double Flux)
{
Nx = Nxp; Ny = Nyp; Nz = Nzp; Nt = Ntm; betagauge = (floattype)Bbeta; flux = (floattype)Flux;
N = Nx * Ny * Nz * Nt; Nspace = Nx * Ny * Nz;
string strforcompiler = "-D Nt=" + Nt.ToString() + " -D Nxyz=" + (Nx * Ny * Nz).ToString() + " -D Nxy=" + (Nx*Ny).ToString() +
" -D Nx="+(Nx).ToString()+" -D Ny="+(Ny).ToString()+" -D Nz="+(Nz).ToString();
strforcompiler += typeof(floattype) == typeof(double) ? " -D floattype=double -D floattype2=double2 -D floattype4=double4" :
" -D floattype=float -D floattype2=float2 -D floattype4=float4";
strforcompiler += " -D phi=" + flux.ToString().Replace(',', '.') + " -D KAPPA=" + kappa.ToString().Replace(',', '.');
string fp64support = "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n";
Plocalsize = AdjustLocalSize(Nspace);
Slocalsize = AdjustLocalSize(N / 2);
XhermYlocalsize = AdjustLocalSize(4 * N);
// Plocalsize = 16; Slocalsize = 16;
PNumGroups = Nx * Ny * Nz / Plocalsize;
SNumGroups = N/2 / Slocalsize;
XhermYNumGroups = 4*4*N / XhermYlocalsize;
BufferLength = N * 4 * 9 * 2 * sizeof(floattype);
SeedBufLen = N * sizeof(Int32)/2 * 4;
AllocBuffers();
openCLPlatform = OpenCL.GetPlatform(0);
openCLDevices = openCLPlatform.QueryDevices(DeviceType.ALL);
openCLContext = openCLPlatform.CreateDefaultContext();
openCLCQ = openCLContext.CreateCommandQueue(openCLDevices[0], CommandQueueProperties.PROFILING_ENABLE);
MyKernelProgram = openCLContext.CreateProgramWithSource(
(typeof(floattype)==typeof(double)?fp64support:"") + File.ReadAllText("MyKernel.cl")+File.ReadAllText("dirak_mul.cl"));
try
{
MyKernelProgram.Build(openCLDevices, strforcompiler, null, IntPtr.Zero);
}
catch (OpenCLException)
{
string buildLog = MyKernelProgram.GetBuildLog(openCLDevices[0]);
MessageBox.Show(buildLog, "Build error(64 bit debug sessions in vs2008 always fail like this - debug in 32 bit or use vs2010)");
// Application.Exit();
}
MyKernelKernel = MyKernelProgram.CreateKernel("MyKernel");
PReductionKernel = MyKernelProgram.CreateKernel("PLoop");
SReductionKernel = MyKernelProgram.CreateKernel("CalcS");
DiralMulKernel = MyKernelProgram.CreateKernel("dirakMatrMul");
FillWithKernel = MyKernelProgram.CreateKernel("FillWith");
FillLinkWithKernel = MyKernelProgram.CreateKernel("FillLinkWith");
FillWithRandomKernel = MyKernelProgram.CreateKernel("FillWithRandom");
AXPYKernel = MyKernelProgram.CreateKernel("AXPY");
XhermYKernel = MyKernelProgram.CreateKernel("XhermY");
BackupLinkKernel = MyKernelProgram.CreateKernel("BackupLink");
RestoreLinkKernel = MyKernelProgram.CreateKernel("RestoreLink");
SeedMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), SeedBufLen, IntPtr.Zero);
LinkMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), BufferLength, IntPtr.Zero);
PGroupMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize * PNumGroups, IntPtr.Zero);
PResMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize, IntPtr.Zero);
SGroupMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize * SNumGroups, IntPtr.Zero);
SResMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize, IntPtr.Zero);
XhermYGroupMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize * 2*XhermYNumGroups, IntPtr.Zero);
XhermYresMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize * 2, IntPtr.Zero);
XhermYrespointer = System.Runtime.InteropServices.Marshal.AllocHGlobal(floatsize * 2);
SeedVectorMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), SeedVectorBuf.Length * sizeof(int), IntPtr.Zero);
StorageMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), linksize, IntPtr.Zero);
dSmem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize, IntPtr.Zero);
dSpointer = System.Runtime.InteropServices.Marshal.AllocHGlobal(floatsize);
MyKernelKernel.SetArg(0, (byte)EvenOdd);
MyKernelKernel.SetArg(1, (floattype)betagauge);
MyKernelKernel.SetArg(2, (floattype)flux);
MyKernelKernel.SetArg(3, SeedMem);
MyKernelKernel.SetArg(4, LinkMem);
PReductionKernel.SetArg(0, LinkMem);
PReductionKernel.SetArg(1, PGroupMem);
PReductionKernel.SetArg(2, PResMem);
IntPtr ptr = new IntPtr(Plocalsize * floatsize);
PReductionKernel.SetArg(3, ptr, IntPtr.Zero);
SReductionKernel.SetArg(0, LinkMem);
SReductionKernel.SetArg(1, SGroupMem);
SReductionKernel.SetArg(2, SResMem);
IntPtr ptr1 = new IntPtr(Slocalsize * floatsize);
SReductionKernel.SetArg(3, ptr1, IntPtr.Zero);
XhermYKernel.SetArg(2, XhermYresMem);
XhermYKernel.SetArg(3, XhermYGroupMem);
XhermYKernel.SetArg(4, new IntPtr(XhermYlocalsize*floatsize*2),IntPtr.Zero);
openCLCQ.EnqueueWriteBuffer(SeedMem, true, 0, SeedBufLen, ipseed);
openCLCQ.EnqueueWriteBuffer(LinkMem, true, 0, BufferLength, ip);
openCLCQ.EnqueueWriteBuffer(SeedVectorMem, true, 0, SeedVectorBuf.Length*sizeof(int), ipseedvector);
rhat0 = new Vector();
//init BICGStab vectors
phi = new Vector();
r0 = new Vector();
//rprev = new Vector();
pi = new Vector();
vi = new Vector();
t = new Vector();
s = new Vector();
// xprev = new Vector();
// vprev = new Vector();
// pprev = new Vector();
temp = new Vector();
ri = new Vector();
x = new Vector();
//for fermion update
chi = new Vector();
CalculateS();
double s1 = S[0];
BackupLink(0, 0,1, 0, 1);
CalculateS();
double s2 = S[0];
RestoreLink(0, 0, 1, 0, 1);
CalculateS();
double s3 = S[0];
//MessageBox.Show(s1.ToString() + s2.ToString() + s3.ToString());
}
public unsafe void CopyOCLBitmapToBitmap(Mem oclBitmap, Bitmap bitmap)
{
IntPtr[] origin = new IntPtr[3];
IntPtr[] region = new IntPtr[3];
Mem buffer;
BitmapData bd = bitmap.LockBits(new Rectangle(0, 0, bitmap.Width, bitmap.Height), ImageLockMode.WriteOnly, PixelFormat.Format32bppArgb);
buffer = oclContext.CreateBuffer((MemFlags)((long)MemFlags.WRITE_ONLY | (long)MemFlags.USE_HOST_PTR), bd.Height * bd.Stride, bd.Scan0);
origin[0] = (IntPtr)0;
origin[1] = (IntPtr)0;
origin[2] = (IntPtr)0;
region[0] = (IntPtr)bd.Width;
region[1] = (IntPtr)bd.Height;
region[2] = (IntPtr)1;
oclCQ.EnqueueCopyImageToBuffer(oclBitmap, buffer, origin, region, IntPtr.Zero);
oclCQ.EnqueueBarrier();
IntPtr p = oclCQ.EnqueueMapBuffer(buffer, true, MapFlags.READ, IntPtr.Zero, (IntPtr)(bd.Height * bd.Stride));
oclCQ.EnqueueUnmapMemObject(buffer, p);
oclCQ.Finish();
buffer.Dispose();
bitmap.UnlockBits(bd);
}
public void EnqueueCopyBufferRect(Mem src_buffer,
Mem dst_buffer,
IntPtr[] src_origin,
IntPtr[] dst_origin,
IntPtr[] region,
IntPtr src_row_pitch,
IntPtr src_slice_pitch,
IntPtr dst_row_pitch,
IntPtr dst_slice_pitch)
{
ErrorCode result;
result = OpenCL.EnqueueCopyBufferRect(CommandQueueID,
src_buffer.MemID,
dst_buffer.MemID,
src_origin,
dst_origin,
region,
src_row_pitch,
src_slice_pitch,
dst_row_pitch,
dst_slice_pitch,
0,
null,
null);
if (result != ErrorCode.SUCCESS)
throw new OpenCLException("EnqueueCopyBufferRect failed with error code " + result, result);
}
public void EnqueueCopyBufferRect(Mem src_buffer, Mem dst_buffer, long[] src_origin, long[] dst_origin, long[] region, long src_row_pitch, long src_slice_pitch, long dst_row_pitch, long dst_slice_pitch, int num_events_in_wait_list, Event[] event_wait_list)
{
ErrorCode result;
IntPtr* pSrcOrigin = stackalloc IntPtr[3];
IntPtr* pDstOrigin = stackalloc IntPtr[3];
IntPtr* pRegion = stackalloc IntPtr[3];
IntPtr* pRepackedEvents = stackalloc IntPtr[num_events_in_wait_list];
if (num_events_in_wait_list == 0)
pRepackedEvents = null;
InteropTools.A3ToIntPtr3(src_origin, pSrcOrigin);
InteropTools.A3ToIntPtr3(dst_origin, pDstOrigin);
InteropTools.A3ToIntPtr3(region, pRegion);
InteropTools.ConvertEventsToEventIDs(num_events_in_wait_list, event_wait_list, pRepackedEvents);
result = OpenCL.EnqueueCopyBufferRect(CommandQueueID,
src_buffer.MemID,
dst_buffer.MemID,
pSrcOrigin,
pDstOrigin,
pRegion,
src_row_pitch,
src_slice_pitch,
dst_row_pitch,
dst_slice_pitch,
num_events_in_wait_list,
pRepackedEvents,
null);
if (result != ErrorCode.SUCCESS)
throw new OpenCLException("EnqueueCopyBufferRect failed with error code " + result, result);
}
public void EnqueueCopyBufferRect(Mem src_buffer,
Mem dst_buffer,
IntPtr[] src_origin,
IntPtr[] dst_origin,
IntPtr[] region,
IntPtr src_row_pitch,
IntPtr src_slice_pitch,
IntPtr dst_row_pitch,
IntPtr dst_slice_pitch,
uint num_events_in_wait_list,
Event[] event_wait_list)
{
ErrorCode result;
result = OpenCL.EnqueueCopyBufferRect(CommandQueueID,
src_buffer.MemID,
dst_buffer.MemID,
src_origin,
dst_origin,
region,
src_row_pitch,
src_slice_pitch,
dst_row_pitch,
dst_slice_pitch,
num_events_in_wait_list,
InteropTools.ConvertEventsToEventIDs(event_wait_list),
null);
if (result != ErrorCode.SUCCESS)
throw new OpenCLException("EnqueueCopyBufferRect failed with error code " + result, result);
}
internal NativeKernelCallbackData(NativeKernel nk,CommandQueue cq, object o, Mem[] buffers)
{
NativeKernel = nk;
CQ = cq;
O = o;
Buffers = buffers;
}
private static int AddNativeKernelParams( NativeKernel nk, CommandQueue cq, object o, Mem[] buffers)
{
int callbackId;
NativeKernelCallbackData callbackData = new NativeKernelCallbackData(nk, cq, o, buffers);
bool gotMutex = false;
try
{
gotMutex = NativeKernelParamsMutex.WaitOne();
do
{
callbackId = NativeKernelParamsId++;
} while (NativeKernelDispatch.ContainsKey(callbackId));
NativeKernelDispatch.Add(callbackId, callbackData);
}
finally
{
if (gotMutex)
NativeKernelParamsMutex.ReleaseMutex();
}
return callbackId;
}
public static float[] ReadFloatValues(this Context oclContext, CommandQueue oclCQ, CL.Mem oclBuff, int Len)
{
float[] floats = new float[Len];
oclCQ.EnqueueReadBuffer(oclBuff, true, 0, Len, floats.ToIntPtr());
oclCQ.EnqueueBarrier();
oclCQ.Finish();
return(floats);
}
private unsafe Mem GetOpacity()
{
fixed (float* dataptr = form_this.opacity)
{
opacity = manager.Context.CreateBuffer(MemFlags.COPY_HOST_PTR, form_this.opacity.Count(), new IntPtr(dataptr));
}
return opacity;
}
/// <summary>
/// OpenCL 1.1
/// </summary>
/// <param name="buffer"></param>
/// <param name="flags"></param>
/// <param name="buffer_create_info"></param>
/// <param name="errcode_ret"></param>
/// <returns></returns>
public Mem CreateSubBuffer(Mem buffer, MemFlags flags, BufferRegion buffer_create_info, out ErrorCode errcode_ret)
{
IntPtr memID = OpenCL.CreateSubBuffer(buffer.MemID, flags, buffer_create_info, out errcode_ret);
return new Mem(buffer.Context, memID);
}
public void EnqueueCopyBufferRect(Mem src_buffer, Mem dst_buffer, long[] src_origin, long[] dst_origin, long[] region, long src_row_pitch, long src_slice_pitch, long dst_row_pitch, long dst_slice_pitch)
{
ErrorCode result;
IntPtr* pSrcOrigin = stackalloc IntPtr[3];
IntPtr* pDstOrigin = stackalloc IntPtr[3];
IntPtr* pRegion = stackalloc IntPtr[3];
InteropTools.A3ToIntPtr3(src_origin, pSrcOrigin);
InteropTools.A3ToIntPtr3(dst_origin, pDstOrigin);
InteropTools.A3ToIntPtr3(region, pRegion);
result = OpenCL.EnqueueCopyBufferRect(CommandQueueID,
src_buffer.MemID,
dst_buffer.MemID,
pSrcOrigin,
pDstOrigin,
pRegion,
src_row_pitch,
src_slice_pitch,
dst_row_pitch,
dst_slice_pitch,
0,
null,
null);
if (result != ErrorCode.SUCCESS)
throw new OpenCLException("EnqueueCopyBufferRect failed with error code " + result, result);
}
private static void Main()
{
OpenCL.GetPlatformIDs(32, new IntPtr[32], out uint num_platforms);
var devices = new List <Device>();
for (int i = 0; i < num_platforms; i++)
{
devices.AddRange(OpenCL.GetPlatform(i).QueryDevices(DeviceType.ALL));
}
int device = SelectForm.Show((from Device d in devices select d.Name).ToArray());
if (device == -1)
{
return;
}
platform = devices[device].Platform; //平台
oclDevice = devices[device]; //选中运算设备
oclContext = platform.CreateContext(new[] { (IntPtr)ContextProperties.PLATFORM, platform.PlatformID, IntPtr.Zero, IntPtr.Zero }, new[] { oclDevice }, new ContextNotify(OpenCLContextNotifyCallBack), IntPtr.Zero); //根据配置建立上下文
oclCQ = oclContext.CreateCommandQueue(oclDevice, CommandQueueProperties.PROFILING_ENABLE); //创建请求队列
if (!oclDevice.ImageSupport)
{
return; //如果失败返回
}
if (!oclContext.SupportsImageFormat(MemFlags.READ_WRITE, MemObjectType.IMAGE2D, ChannelOrder.RGBA, ChannelType.UNSIGNED_INT8))
{
return;
}
sampler = oclContext.CreateSampler(true, AddressingMode.NONE, FilterMode.NEAREST);
FilterKernel = oclContext.MakeCode("FilterImage", CLCode1);
Kernel K2 = oclContext.MakeCode("vector_add_gpu", CLCode2);
int aaa = K2.PECount(oclDevice);
aaa = aaa;
#region 试一下用GPU做运算
int[] a = new[] { 1, 2, 3, 1722 };
int[] b = new[] { 456, 2, 1, 56 };
int[] c = new[] { 0, 0, 0, 0 };
CL.Mem n1 = oclContext.CreateBuffer(MemFlags.READ_WRITE | MemFlags.COPY_HOST_PTR, a.Length * sizeof(int), a.ToIntPtr());
CL.Mem n2 = oclContext.CreateBuffer(MemFlags.READ_WRITE | MemFlags.COPY_HOST_PTR, b.Length * sizeof(int), b.ToIntPtr());
CL.Mem n3 = null;
unchecked { n3 = oclContext.CreateBuffer(MemFlags.READ_WRITE, b.Length * sizeof(int), IntPtr.Zero); }
K2.SetArg(0, n1);
K2.SetArg(1, n2);
K2.SetArg(2, n3);
K2.SetArg(3, (int)c.Length);
oclCQ.EnqueueNDRangeKernel(K2, 1, null, new[] { c.Length, 0 }, null);
oclCQ.EnqueueBarrier();
oclCQ.Finish();
// */
//oclContext.WriterValues(oclCQ, n3, B);
c = oclContext.ReadIntValues(oclCQ, n3, c.Length);
c = c;
// */
#endregion
var sdi = new ShowDeviceInfo();
var lb = sdi.listBox1.Items;
lb.Add($"Name:{oclDevice.Name}");
lb.Add($"DeviceType:{oclDevice.DeviceType.ToString()}");
lb.Add($"MaxComputeUnits(最大计算单元):{oclDevice.MaxComputeUnits}");
lb.Add($"ImageSupport:{oclDevice.ImageSupport}");
lb.Add($"AddressBits:{oclDevice.AddressBits}");
lb.Add($"DriverVersion:{oclDevice.DriverVersion}");
lb.Add($"MaxClockFrequency(最大时钟频率):{oclDevice.MaxClockFrequency}MHz");
lb.Add($"MaxMemAllocSize(最大内存):{oclDevice.MaxMemAllocSize / 1024 / 1024 / 1024}GB");
lb.Add($"MaxWorkItemDimensions(最大工作维度):{oclDevice.MaxWorkItemDimensions}");
lb.Add($"MaxWorkGroupSize(最大工作组数量):{oclDevice.MaxWorkGroupSize }");
lb.Add($"Version(OpenCL版本):{oclDevice.Version}");
lb.Add($"GlobalMemSize(显存):{oclDevice.GlobalMemSize / 1024 / 1024 / 1024}GB");
lb.Add($"GlobalMemCacheSize(显存缓存):{oclDevice.GlobalMemCacheSize / 1024}KB");
lb.Add($"GlobalMemCacheLineSize:{oclDevice.GlobalMemCacheLineSize}");
lb.Add($"Vendor(厂商):{oclDevice.Vendor}");
lb.Add($"HostUnifiedMemory(是否和Host共用内存):{oclDevice.HostUnifiedMemory}");
sdi.ShowDialog();
#region 调用编译好的生命游戏程序
//if (oclDevice.DeviceType == DeviceType.GPU)
{
OutImage1 = oclContext.CreateImage2D(MemFlags.READ_WRITE, CL.ImageFormat.RGBA8U, TestImage.Width, TestImage.Height, 0, IntPtr.Zero);
OutImage2 = oclContext.CreateImage2D(MemFlags.READ_WRITE, CL.ImageFormat.RGBA8U, TestImage.Width, TestImage.Height, 0, IntPtr.Zero);
FilterKernel.SetArg(0, 1.0f);
FilterKernel.SetArg(1, 1.0f);
FilterKernel.SetArg(2, oclContext.ToCLImage(TestImage));
FilterKernel.SetArg(3, OutImage1);
FilterKernel.SetArg(4, sampler);
oclCQ.EnqueueNDRangeKernel(FilterKernel, 2, null, new IntPtr[] { OutImage1.Width, OutImage1.Height, IntPtr.Zero }, null);
oclCQ.EnqueueBarrier();
oclCQ.Finish();
new MainForm().ShowDialog();
}
// */
#endregion
Application.Exit();
}
public void EnqueueCopyImage(Mem src_image, Mem dst_image, IntPtr[] src_origin, IntPtr[] dst_origin, IntPtr[] region, int num_events_in_wait_list, Event[] event_wait_list, out Event _event)
{
ErrorCode result;
IntPtr tmpEvent;
result = OpenCL.EnqueueCopyImage(CommandQueueID,
src_image.MemID,
dst_image.MemID,
src_origin,
dst_origin,
region,
(uint)num_events_in_wait_list,
InteropTools.ConvertEventsToEventIDs(event_wait_list),
&tmpEvent);
_event = new Event(Context, this, tmpEvent);
if (result != ErrorCode.SUCCESS)
throw new OpenCLException("EnqueueCopyImage failed with error code " + result, result);
}
private unsafe Mem GetCounter()
{
fixed (ulong* dataptr = &voxelctr)
{
counter = manager.Context.CreateBuffer(MemFlags.READ_WRITE, 8, new IntPtr(dataptr));
}
return counter;
}
public void EnqueueCopyImage(Mem src_image, Mem dst_image, long[] src_origin, long[] dst_origin, long[] region, int num_events_in_wait_list, Event[] event_wait_list, out Event _event)
{
ErrorCode result;
IntPtr tmpEvent;
IntPtr* repackedSrcOrigin = stackalloc IntPtr[3];
IntPtr* repackedDstOrigin = stackalloc IntPtr[3];
IntPtr* repackedRegion = stackalloc IntPtr[3];
IntPtr* repackedEvents = stackalloc IntPtr[num_events_in_wait_list];
if (num_events_in_wait_list == 0)
repackedEvents = null;
InteropTools.A3ToIntPtr3(src_origin, repackedSrcOrigin);
InteropTools.A3ToIntPtr3(dst_origin, repackedDstOrigin);
InteropTools.A3ToIntPtr3(region, repackedRegion);
InteropTools.ConvertEventsToEventIDs(num_events_in_wait_list, event_wait_list, repackedEvents);
result = OpenCL.EnqueueCopyImage(CommandQueueID, src_image, dst_image, repackedSrcOrigin, repackedDstOrigin, repackedRegion, num_events_in_wait_list, repackedEvents, &tmpEvent);
_event = new Event(Context, this, tmpEvent);
if (result != ErrorCode.SUCCESS)
throw new OpenCLException("EnqueueCopyImage failed with error code " + result, result);
}
/// <summary>
/// Set argument argIndex to mem
/// </summary>
/// <param name="argIndex"></param>
/// <param name="mem"></param>
public void SetArg(Int32 argIndex, Mem mem)
{
this.SetArg(argIndex, mem.MemID);
}
public void EnqueueCopyImage(Mem src_image, Mem dst_image, IntPtr[] src_origin, IntPtr[] dst_origin, IntPtr[] region)
{
ErrorCode result;
result = OpenCL.EnqueueCopyImage(CommandQueueID,
src_image.MemID,
dst_image.MemID,
src_origin,
dst_origin,
region,
(uint)0,
null,
null);
if (result != ErrorCode.SUCCESS)
throw new OpenCLException("EnqueueCopyImage failed with error code " + result, result);
}
/* возвращает буфер */
public unsafe Mem CreateBuffer()
{
if (buffer == null)
{
/* указатели в C# используются с fixed (противодействие для GC) */
fixed (short* dataptr = data)
{
buffer = manager_this.Context.CreateBuffer(MemFlags.COPY_HOST_PTR, data.Count() * 2, new IntPtr(dataptr));
}
}
return buffer;
}
public void EnqueueCopyImage(Mem src_image, Mem dst_image, long[] src_origin, long[] dst_origin, long[] region)
{
ErrorCode result;
IntPtr* repackedSrcOrigin = stackalloc IntPtr[3];
IntPtr* repackedDstOrigin = stackalloc IntPtr[3];
IntPtr* repackedRegion = stackalloc IntPtr[3];
InteropTools.A3ToIntPtr3(src_origin, repackedSrcOrigin);
InteropTools.A3ToIntPtr3(dst_origin, repackedDstOrigin);
InteropTools.A3ToIntPtr3(region, repackedRegion);
result = OpenCL.EnqueueCopyImage(CommandQueueID, src_image, dst_image, repackedSrcOrigin, repackedDstOrigin, repackedRegion, 0, null, null);
if (result != ErrorCode.SUCCESS)
throw new OpenCLException("EnqueueCopyImage failed with error code " + result, result);
}
/// <summary>
/// Set argument argIndex to mem
/// </summary>
/// <param name="argIndex"></param>
/// <param name="mem"></param>
public void SetArg(int argIndex, Mem mem)
{
SetArg(argIndex, mem.MemID);
}
public void EnqueueCopyImageToBuffer(Mem src_image, Mem dst_buffer, IntPtr[] src_origin, IntPtr[] region, IntPtr dst_offset, int num_events_in_wait_list, Event[] event_wait_list)
{
ErrorCode result;
result = OpenCL.EnqueueCopyImageToBuffer(CommandQueueID,
src_image.MemID,
dst_buffer.MemID,
src_origin,
region,
dst_offset,
(uint)num_events_in_wait_list,
InteropTools.ConvertEventsToEventIDs(event_wait_list),
null);
if (result != ErrorCode.SUCCESS)
throw new OpenCLException("EnqueueCopyImageToBuffer failed with error code " + result, result);
}
public void EnqueueCopyImageToBuffer(Mem src_image, Mem dst_buffer, long[] src_origin, long[] region, long dst_offset, int num_events_in_wait_list, Event[] event_wait_list)
{
ErrorCode result;
IntPtr* repackedSrcOrigin = stackalloc IntPtr[3];
IntPtr* repackedRegion = stackalloc IntPtr[3];
IntPtr* repackedEvents = stackalloc IntPtr[num_events_in_wait_list];
if (num_events_in_wait_list == 0)
repackedEvents = null;
InteropTools.A3ToIntPtr3(src_origin, repackedSrcOrigin);
InteropTools.A3ToIntPtr3(region, repackedRegion);
InteropTools.ConvertEventsToEventIDs(num_events_in_wait_list, event_wait_list, repackedEvents);
result = OpenCL.EnqueueCopyImageToBuffer(CommandQueueID, src_image, dst_buffer, repackedSrcOrigin, repackedRegion, dst_offset, num_events_in_wait_list, repackedEvents, null);
if (result != ErrorCode.SUCCESS)
throw new OpenCLException("EnqueueCopyImageToBuffer failed with error code " + result, result);
}
/// <summary>
/// Loads two bitmaps and locks them for the duration of the program.
/// Also creates two OpenCL buffers that map to the locked images
/// </summary>
private void SetupBitmaps()
{
InputBitmap0 = (Bitmap)Bitmap.FromFile(@"Input0.png");
InputBitmap1 = (Bitmap)Bitmap.FromFile(@"Input1.png");
if (InputBitmap1.Size != InputBitmap0.Size)
InputBitmap1 = new Bitmap(InputBitmap1, InputBitmap0.Size);
OutputBitmap = new Bitmap(InputBitmap0);
InputBitmapData0 = InputBitmap0.LockBits(new Rectangle(0, 0, InputBitmap0.Width, InputBitmap0.Height), ImageLockMode.ReadOnly, PixelFormat.Format32bppArgb);
InputBitmapData1 = InputBitmap1.LockBits(new Rectangle(0, 0, InputBitmap1.Width, InputBitmap1.Height), ImageLockMode.ReadOnly, PixelFormat.Format32bppArgb);
InputBuffer0 = OCLMan.Context.CreateBuffer(MemFlags.USE_HOST_PTR, InputBitmapData0.Stride * InputBitmapData0.Height, InputBitmapData0.Scan0);
InputBuffer1 = OCLMan.Context.CreateBuffer(MemFlags.USE_HOST_PTR, InputBitmapData1.Stride * InputBitmapData1.Height, InputBitmapData1.Scan0);
}
private unsafe Mem GetColors()
{
fixed (uint* dataptr = form_this.colors)
{
color_buffer = manager.Context.CreateBuffer(MemFlags.COPY_HOST_PTR, form_this.colors.Count() * 4, new IntPtr(dataptr));
}
return color_buffer;
}
/// <summary>
/// Launch the CrossFadeKernel
///
/// First we set its arguments,
/// then we enqueue the kernel using EnqueueNDRangeKernel,
/// and finally, map the buffer for reading to make sure
/// there aren't any cache issues when OpenCL completes.
/// </summary>
/// <param name="ratio"></param>
/// <param name="width"></param>
/// <param name="height"></param>
/// <param name="input0"></param>
/// <param name="inputStride0"></param>
/// <param name="input1"></param>
/// <param name="inputStride1"></param>
/// <param name="output"></param>
private void DoCrossFade( float ratio,
int width, int height,
Mem input0, int inputStride0,
Mem input1, int inputStride1,
BitmapData output )
{
Mem outputBuffer;
int deviceIndex = comboBoxDeviceSelector.SelectedIndex;
outputBuffer = OCLMan.Context.CreateBuffer(MemFlags.USE_HOST_PTR, output.Stride * output.Height, output.Scan0);
CrossFadeGlobalWorkSize[0] = (IntPtr)width;
CrossFadeGlobalWorkSize[1] = (IntPtr)height;
CrossFadeKernel.SetArg(0, ratio);
CrossFadeKernel.SetArg(1, width);
CrossFadeKernel.SetArg(2, height);
CrossFadeKernel.SetArg(3, input0);
CrossFadeKernel.SetArg(4, inputStride0);
CrossFadeKernel.SetArg(5, input1);
CrossFadeKernel.SetArg(6, inputStride1);
CrossFadeKernel.SetArg(7, outputBuffer);
CrossFadeKernel.SetArg(8, output.Stride);
OCLMan.CQ[deviceIndex].EnqueueNDRangeKernel(CrossFadeKernel, 2, null, CrossFadeGlobalWorkSize, null);
OCLMan.CQ[deviceIndex].EnqueueBarrier();
IntPtr p = OCLMan.CQ[deviceIndex].EnqueueMapBuffer(outputBuffer, true, MapFlags.READ, IntPtr.Zero, (IntPtr)(output.Stride * output.Height));
OCLMan.CQ[deviceIndex].EnqueueUnmapMemObject(outputBuffer, p);
OCLMan.CQ[deviceIndex].Finish();
outputBuffer.Dispose();
}
public VoxelImage()
{
InitializeComponent();
first = true;
angleChange = false;
changeDistance = false;
leftChange = false;
rightChange = false;
voxelctr = 0;
realctr = 0;
this.lx.Text = Convert.ToString(light.S0);
this.ly.Text = Convert.ToString(light.S1);
this.lz.Text = Convert.ToString(light.S2);
this.trackminX.SetRange(0, 512);
this.trackminY.SetRange(0, 512);
this.trackminZ.SetRange(0, 512);
this.trackmaxX.SetRange(0, 512);
this.trackmaxY.SetRange(0, 512);
this.trackmaxZ.SetRange(0, 512);
this.trackminX.Value = (int)(boxMinCon.S0);
this.trackminY.Value = (int)(boxMinCon.S1);
this.trackminZ.Value = (int)(boxMinCon.S2);
this.trackmaxX.Value = (int)(boxMaxCon.S0);
this.trackmaxY.Value = (int)(boxMaxCon.S1);
this.trackmaxZ.Value = (int)(boxMaxCon.S2);
this.minX.Text = Convert.ToString(boxMinCon.S0);
this.minY.Text = Convert.ToString(boxMinCon.S1);
this.minZ.Text = Convert.ToString(boxMinCon.S2);
this.maxX.Text = Convert.ToString(boxMaxCon.S0);
this.maxY.Text = Convert.ToString(boxMaxCon.S1);
this.maxZ.Text = Convert.ToString(boxMaxCon.S2);
this.kamb.Text = Convert.ToString(0.1);
this.kdiff.Text = Convert.ToString(0.5);
this.kspec.Text = Convert.ToString(0.4);
this.tracklx.SetRange(-1000,1000);
this.trackly.SetRange(-1000, 1000);
this.tracklz.SetRange(-1000, 1000);
this.tracklx.Value = (int)light.S0;
this.trackly.Value = (int)light.S1;
this.tracklz.Value = (int)light.S2;
this.kc.Text = "1,0";
this.kl.Text = "0,0";
this.kq.Text = "0,0";
this.specexp.Text = "1,0";
outputBuffer = null;
color_buffer =null;
opacity = null;
clock = new Stopwatch();
}
public void SetMemArg(int argIndex, Mem mem)
{
SetIntPtrArg(argIndex, mem.MemID);
}
public TextureInfo(Mem mem)
{
Mem = mem;
}
/// <summary>
/// Set argument argIndex to mem
/// </summary>
/// <param name="argIndex"></param>
/// <param name="mem"></param>
public void SetArg(int argIndex, Mem mem)
{
SetArg(argIndex, mem.MemID);
}
private static void Main()
{
OpenCL.GetPlatformIDs(32, new IntPtr[32], out uint num_platforms);
List <Device> pt = new List <Device>();
for (int i = 0; i < num_platforms; pt.AddRange(OpenCL.GetPlatform(i++).QueryDevices(DeviceType.ALL)))
{
;
}
int PT = 0; // SelectForm.Show((from Device d in pt select d.Name).ToArray());
if (PT == -1)
{
return;
}
platform = pt[PT].Platform; //平台
oclDevice = pt[PT]; //选中运算设备
oclContext = platform.CreateContext(new[] { (IntPtr)ContextProperties.PLATFORM, platform.PlatformID, IntPtr.Zero, IntPtr.Zero }, new[] { oclDevice }, new ContextNotify(OpenCLContextNotifyCallBack), IntPtr.Zero); //根据配置建立上下文
oclCQ = oclContext.CreateCommandQueue(oclDevice, CommandQueueProperties.PROFILING_ENABLE); //创建请求队列
if (!oclDevice.ImageSupport)
{
return; //如果失败返回
}
if (!oclContext.SupportsImageFormat(MemFlags.READ_WRITE, MemObjectType.IMAGE2D, ChannelOrder.RGBA, ChannelType.UNSIGNED_INT8))
{
return;
}
sampler = oclContext.CreateSampler(true, AddressingMode.NONE, FilterMode.NEAREST);
FilterKernel = oclContext.MakeCode("FilterImage", CLCode1);
Kernel K2 = oclContext.MakeCode("vector_add_gpu", CLCode2);
#region
int[] A = new[] { 1, 2, 3, 1722 };
int[] B = new[] { 456, 2, 1, 56 };
int[] C = new[] { 0, 0, 0, 0 };
CL.Mem n1 = oclContext.CreateBuffer(MemFlags.READ_WRITE | MemFlags.COPY_HOST_PTR, A.Length * sizeof(int), A.ToIntPtr());
CL.Mem n2 = oclContext.CreateBuffer(MemFlags.READ_WRITE | MemFlags.COPY_HOST_PTR, B.Length * sizeof(int), B.ToIntPtr());
CL.Mem n3 = oclContext.CreateBuffer(MemFlags.READ_WRITE, B.Length * sizeof(int), IntPtr.Zero);
K2.SetArg(0, n1);
K2.SetArg(1, n2);
K2.SetArg(2, n3);
K2.SetArg(3, (int)C.Length);
oclCQ.EnqueueNDRangeKernel(K2, 1, null, new[] { C.Length, 0 }, null);
oclCQ.EnqueueBarrier();
oclCQ.Finish();
// */
//oclContext.WriterValues(oclCQ, n3, B);
C = oclContext.ReadIntValues(oclCQ, n3, C.Length);
C = C;
// */
#endregion
#region 调用编译好的程序
OutImage1 = oclContext.CreateImage2D(MemFlags.READ_WRITE, CL.ImageFormat.RGBA8U, TestImage.Width, TestImage.Height, 0, IntPtr.Zero);
OutImage2 = oclContext.CreateImage2D(MemFlags.READ_WRITE, CL.ImageFormat.RGBA8U, TestImage.Width, TestImage.Height, 0, IntPtr.Zero);
FilterKernel.SetArg(0, 1.0f);
FilterKernel.SetArg(1, 1.0f);
FilterKernel.SetArg(2, oclContext.ToCLImage(TestImage));
FilterKernel.SetArg(3, OutImage1);
FilterKernel.SetArg(4, sampler);
oclCQ.EnqueueNDRangeKernel(FilterKernel, 2, null, new IntPtr[] { OutImage1.Width, OutImage1.Height, IntPtr.Zero }, null);
oclCQ.EnqueueBarrier();
oclCQ.Finish();
// */
#endregion
new MainForm().ShowDialog();
Application.Exit();
}
private unsafe void DoRayCasting(BitmapData output)
{
try{
int deviceIndex = 0;
outputBuffer = manager.Context.CreateBuffer(MemFlags.USE_HOST_PTR, output.Stride * output.Height, output.Scan0);
if (first || changeDistance)
{
// camera model UVN
camPos = new Float4()
{
S0 =
vol.GetSize() / 2 - (float)Math.Cos(camAngle * Math.PI / 180) * camDist,
S1 = vol.GetSize() / 2,
S2 = vol.GetSize() / 2 - (float)Math.Sin(camAngle * Math.PI / 180) * camDist,
S3 = 0
};
first = false;
changeDistance = false;
camPosOld = camPos;
}
else{
// rotation around the axis of the cube visualization
if (angleChange && leftChange){
camPosOld.S0 -= camLookAt.S0;
camPosOld.S2 -= camLookAt.S2;
camPos.S0 = (float)Math.Cos(camAngle * Math.PI / 180) * camPosOld.S0 + (float)Math.Sin(camAngle * Math.PI / 180) * camPosOld.S2;
camPos.S1 = vol.GetSize() / 2;
camPos.S2 = -(float)Math.Sin(camAngle * Math.PI / 180) * camPosOld.S0 + (float)Math.Cos(camAngle * Math.PI / 180) * camPosOld.S2;
camPos.S3 = 0;
camPos.S0 += camLookAt.S0;
camPos.S2 += camLookAt.S2;
camPosOld = camPos;
angleChange = false;
leftChange = false;
}
if (angleChange && rightChange){
camPosOld.S0 -= camLookAt.S0;
camPosOld.S2 -= camLookAt.S2;
camPos.S0 = (float)Math.Cos(camAngle * Math.PI / 180) * camPosOld.S0 - (float)Math.Sin(camAngle * Math.PI / 180) * camPosOld.S2;
camPos.S1 = vol.GetSize() / 2;
camPos.S2 = (float)Math.Sin(camAngle * Math.PI / 180) * camPosOld.S0 + (float)Math.Cos(camAngle * Math.PI / 180) * camPosOld.S2;
camPos.S3 = 0;
camPos.S0 += camLookAt.S0;
camPos.S2 += camLookAt.S2;
camPosOld = camPos;
angleChange = false;
leftChange = false;
}
}
camLookAt = new Float4(){
S0 = vol.GetSize() / camfactorX,
S1 = vol.GetSize() / camfactorX,
S2 = vol.GetSize() / camfactorZ,
S3 = 0
};
// direction of the camera, UVN model
camForward = camLookAt.Sub(camPos).Normalize(); // viewing direction
var up = new Float4(0.0f, 1.0f, 0.0f, 0.0f);
var right = MathClass.Cross(up, camForward).Normalize().Times(1.5f);
up = MathClass.Cross(camForward, right).Normalize().Times(-1.5f);
/* processing of the output image in BitmapData OpenCl device */
for (var x = 0; x < output.Width; x += blocksize){
for (var y = 0; y < output.Height; y += blocksize){
var rayTracingGlobalWorkSize = new IntPtr[2]; // work_dim = 2
rayTracingGlobalWorkSize[0] = (IntPtr)(output.Width - x > blocksize ? blocksize : output.Width - x);
rayTracingGlobalWorkSize[1] = (IntPtr)(output.Height - y > blocksize ? blocksize : output.Height - y);
var rayTracingGlobalOffset = new IntPtr[2];
rayTracingGlobalOffset[0] = (IntPtr)x;
rayTracingGlobalOffset[1] = (IntPtr)y;
float ka = (float)(Convert.ToDouble(kamb.Text));
float kd = (float)(Convert.ToDouble(kdiff.Text));
float ks = (float)(Convert.ToDouble(kspec.Text));
float exp = (float)(Convert.ToDouble(specexp.Text));
float kkc = (float)(Convert.ToDouble(this.kc.Text));
float kkl = (float)(Convert.ToDouble(this.kl.Text));
float kkq = (float)(Convert.ToDouble(this.kq.Text));
/* arguments passed to kernel function */
kernel.SetArg(0, output.Width);
kernel.SetArg(1, output.Height);
kernel.SetArg(2, outputBuffer); // in the nucleus with global, as is required for access to the output
kernel.SetArg(3, output.Stride);
kernel.SetArg(4, camPos);
kernel.SetArg(5, camForward);
kernel.SetArg(6, right);
kernel.SetArg(7, up);
kernel.SetArg(8, vol.CreateBuffer());
kernel.SetArg(9, vol.GetSize());
kernel.SetArg(10, light);
kernel.SetArg(11, boxMinCon);
kernel.SetArg(12, boxMaxCon);
kernel.SetArg(13, Convert.ToInt16(colorMi.Text));
kernel.SetArg(14, Convert.ToInt16(colorMa.Text));
kernel.SetArg(15, _cutArrea.Checked ? (short)1 : (short)0);
kernel.SetArg(16, _trilinear.Checked ? (short)1 : (short)0);
kernel.SetArg(17, tf.Checked ? (short)1: (short)0 );
kernel.SetArg(18, GetColors());
kernel.SetArg(19, winWidth_vox);
kernel.SetArg(20, winCentre_vox);
kernel.SetArg(21, form_this.knots_counter);
kernel.SetArg(22, Convert.ToInt16(colorMi2.Text));
kernel.SetArg(23, Convert.ToInt16(colorMa2.Text));
kernel.SetArg(24, GetOpacity());
kernel.SetArg(25, ka);
kernel.SetArg(26, kd);
kernel.SetArg(27, ks);
kernel.SetArg(28, exp);
kernel.SetArg(29, kkc);
kernel.SetArg(30, kkl);
kernel.SetArg(31, kkq);
kernel.SetArg(32, GetCounter());
/* Queues a command for kernel execution on the device */
/* rayTracingGlobalOffset -
* globalWorkOffset: can be used to specify an array of values
* work_dim unsigned dimension that describes the offset used to calculate global ID work-item
* instead of global IDs always start at offset (0, 0,... 0).
* rayTracingGlobalWorkSize -
* globalWorkSize: общее число global work-items вычисляется как global_work_size[0] *...* global_work_size[work_dim - 1].*/
manager.CQ[deviceIndex].EnqueueNDRangeKernel(kernel, 2, rayTracingGlobalOffset, rayTracingGlobalWorkSize, null);
}
}
/*wait until all the work-items executed */
manager.CQ[deviceIndex].EnqueueBarrier();
/* to get access to memory and written to the output image we ask OpenCL * impose * data to the host device */
IntPtr p = manager.CQ[deviceIndex].EnqueueMapBuffer(outputBuffer, true, MapFlags.READ, IntPtr.Zero, (IntPtr)(output.Stride * output.Height));
IntPtr z = manager.CQ[deviceIndex].EnqueueMapBuffer(counter, true, MapFlags.READ_WRITE, IntPtr.Zero, (IntPtr)(sizeof(ulong)));
/* when we finish working with the buffer should call this function */
manager.CQ[deviceIndex].EnqueueUnmapMemObject(outputBuffer, p);
manager.CQ[deviceIndex].EnqueueUnmapMemObject(counter, z);
manager.CQ[deviceIndex].Finish();
realctr += voxelctr;
voxelCounter.Text = Convert.ToString(realctr);
}
catch (Exception ex){
MessageBox.Show("Ray casting exception:" + ex.Message, "Exception");
Environment.Exit(-1);
}
finally{
if (outputBuffer != null){
outputBuffer.Dispose();
}
}
}
public static IntPtr[] ConvertMemToMemIDs(Mem[] mems)
{
IntPtr[] memIDs;
if (mems == null)
return null;
memIDs = new IntPtr[mems.Length];
for (int i = 0; i < mems.Length; i++)
memIDs[i] = mems[i].MemID;
return memIDs;
}
public static int[] ReadIntValues(this Context oclContext, CommandQueue oclCQ, CL.Mem oclBuff, int Len)
{
int[] values = new int[Len];
oclCQ.EnqueueReadBuffer(oclBuff, true, 0, Len * sizeof(int), values.ToIntPtr());
oclCQ.EnqueueBarrier();
oclCQ.Finish();
return(values);
}
