public void CreateImageTests()
{
ErrorCode error;
if (Cl.GetDeviceInfo(_device, DeviceInfo.ImageSupport, out error).CastTo <Bool>() == Bool.False)
{
Console.WriteLine("No image support");
return;
}
{
var image2DData = new float[200 * 200 * sizeof(float)];
IMem image2D = Cl.CreateImage2D(_context, MemFlags.CopyHostPtr | MemFlags.ReadOnly, new ImageFormat(ChannelOrder.RGBA, ChannelType.Float),
(IntPtr)200, (IntPtr)200, (IntPtr)0, image2DData, out error);
Assert.AreEqual(error, ErrorCode.Success);
Assert.AreEqual(Cl.GetImageInfo(image2D, ImageInfo.Width, out error).CastTo <uint>(), 200);
Assert.AreEqual(Cl.GetImageInfo(image2D, ImageInfo.Height, out error).CastTo <uint>(), 200);
image2D.Dispose();
}
{
var image3DData = new float[200 * 200 * 200 * sizeof(float)];
IMem image3D = Cl.CreateImage3D(_context, MemFlags.CopyHostPtr | MemFlags.ReadOnly, new ImageFormat(ChannelOrder.RGBA, ChannelType.Float),
(IntPtr)200, (IntPtr)200, (IntPtr)200, IntPtr.Zero, IntPtr.Zero, image3DData, out error);
Assert.AreEqual(error, ErrorCode.Success);
Assert.AreEqual(Cl.GetImageInfo(image3D, ImageInfo.Width, out error).CastTo <uint>(), 200);
Assert.AreEqual(Cl.GetImageInfo(image3D, ImageInfo.Height, out error).CastTo <uint>(), 200);
Assert.AreEqual(Cl.GetImageInfo(image3D, ImageInfo.Depth, out error).CastTo <uint>(), 200);
image3D.Dispose();
}
}
public void MemBufferTests()
{
const int bufferSize = 100;
ErrorCode error;
Random random = new Random();
float[] values = (from value in Enumerable.Range(0, bufferSize) select(float) random.NextDouble()).ToArray();
IMem buffer = Cl.CreateBuffer(_context, MemFlags.CopyHostPtr | MemFlags.ReadOnly, (IntPtr)(sizeof(float) * bufferSize), values, out error);
Assert.AreEqual(error, ErrorCode.Success);
Assert.AreEqual(Cl.GetMemObjectInfo(buffer, MemInfo.Type, out error).CastTo <MemObjectType>(), MemObjectType.Buffer);
Assert.AreEqual(Cl.GetMemObjectInfo(buffer, MemInfo.Size, out error).CastTo <uint>(), values.Length * sizeof(float));
// TODO: Verify values
//int index = 0;
//foreach (float value in Cl.GetMemObjectInfo(buffer, Cl.MemInfo.HostPtr, out error).CastToEnumerable<float>(Enumerable.Range(0, 100)))
//{
// Assert.AreEqual(values[index], value);
// index++;
//}
buffer.Dispose();
}
// IMem
public static KernelArgChain SetKernelArg(this Kernel kernel, IMem buffer)
{
Cl.SetKernelArg(kernel, 0, TypeSize <IntPtr> .Size, buffer).Check();
return(new KernelArgChain {
Kernel = kernel, Count = 0
});
}
// images cannot be read_write... so let's continue using plain buffers
// should implement this in a way that allows imgfAccu to be loaded only once
// should test for image size consistency
public void Accumulate(FloatMap imgfAccu, FloatMap imgfSrc, float k)
{
var kernel = _kernels["accumulate"];
// Creation of on-device memory objects
IMem <float> accuMapBuffer = Cl.CreateBuffer <float>(_context, MemFlags.ReadWrite, imgfAccu.Size, out err); // why MemFlags.CopyHostPtr doesn't work here (and forces me to manual copy) ???
assert(err, "accu buf creation");
IMem <float> srcMapBuffer = Cl.CreateBuffer <float>(_context, MemFlags.WriteOnly, imgfSrc.Size, out err);
assert(err, "src buf creation");
// Set memory objects as parameters to kernel
err = Cl.SetKernelArg(kernel, 0, intPtrSize, accuMapBuffer);
assert(err, "accu map setKernelArg");
err = Cl.SetKernelArg(kernel, 1, intPtrSize, srcMapBuffer);
assert(err, "src map setKernelArg");
err = Cl.SetKernelArg(kernel, 2, intSize, imgfAccu.Stride);
assert(err, "in stride setKernelArg");
err = Cl.SetKernelArg(kernel, 3, intSize, imgfSrc.Stride);
assert(err, "out stride setKernelArg");
err = Cl.SetKernelArg(kernel, 4, floatSize, k);
assert(err, "out stride setKernelArg");
// write actual data into memory object
Event clevent;
err = Cl.EnqueueWriteBuffer <float>(_commandsQueue, accuMapBuffer, Bool.True, 0, imgfAccu.Size, imgfAccu._buf, 0, null, out clevent);
clevent.Dispose();
assert(err, "write accu buffer");
err = Cl.EnqueueWriteBuffer <float>(_commandsQueue, srcMapBuffer, Bool.True, 0, imgfSrc.Size, imgfSrc._buf, 0, null, out clevent);
clevent.Dispose();
assert(err, "write src buffer");
// execute
err = Cl.EnqueueNDRangeKernel(_commandsQueue, kernel, 2,
new[] { (IntPtr)0, (IntPtr)0, (IntPtr)0 }, // offset
new[] { new IntPtr(imgfAccu.W), new IntPtr(imgfAccu.H), (IntPtr)1 }, // range
null, 0, null, out clevent);
clevent.Dispose();
assert(err, "Cl.EnqueueNDRangeKernel");
// sync
Cl.Finish(_commandsQueue);
// read from output memory object into actual buffer
err = Cl.EnqueueReadBuffer <float>(_commandsQueue, accuMapBuffer, Bool.True, imgfAccu._buf, 0, null, out clevent);
clevent.Dispose();
assert(err, "read output buffer");
Cl.ReleaseMemObject(srcMapBuffer);
Cl.ReleaseMemObject(accuMapBuffer); // maybe i could return this without disposing; would affect non-OpenCl implementation
}
public override bool DeleteUser(string username, bool deleteAllRelatedData)
{
var user = IMem.GetUser8Name(username);
user.Frozen = true;
IMem.UpdateUser(user);
return(true);
}
public static long CrackHigh(int[] sequence, long low)
{
if (!crackHighProgram.HasValue || !crackHighKernel.HasValue || !initialized || sequence.Length != 16)
{
return(-1);
}
ErrorCode error;
IMem <int> sequence_dev = Cl.CreateBuffer <int>(context, MemFlags.CopyHostPtr | MemFlags.ReadOnly, sequence, out error);
ErrorCheck(error, "CrackHigh(): Cl.CreateBuffer");
long[] seeds = new long[1];
IMem <long> seed_dev = Cl.CreateBuffer <long>(context, MemFlags.CopyHostPtr | MemFlags.WriteOnly, seeds, out error);
ErrorCheck(error, "InitializeParameters(): Cl.CreateBuffer");
error = Cl.SetKernelArg(crackHighKernel.Value, 0, sequence_dev);
ErrorCheck(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(crackHighKernel.Value, 1, seed_dev);
ErrorCheck(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(crackHighKernel.Value, 2, (IntPtr)sizeof(long), low);
ErrorCheck(error, "Cl.SetKernelArg");
CommandQueue cmdQueue = Cl.CreateCommandQueue(context, device, 0, out error);
ErrorCheck(error, "Cl.CreateCommandQueue");
int maxGroupWorkSize = Cl.GetKernelWorkGroupInfo(crackHighKernel.Value, device, KernelWorkGroupInfo.WorkGroupSize, out error).CastTo <int>();
ErrorCheck(error, "Cl.GetKernelWorkGroupInfo");
Event e;
int threads = maxGroupWorkSize * 12;
IntPtr[] workSize = new IntPtr[] { (IntPtr)threads };
error = Cl.EnqueueNDRangeKernel(cmdQueue, crackHighKernel.Value, 1, null, workSize, null, 0, null, out e);
ErrorCheck(error, "Cl.EnqueueNDRangeKernel");
error = Cl.Finish(cmdQueue);
ErrorCheck(error, "Cl.Finish");
long[] seed_host = new long[1];
error = Cl.EnqueueReadBuffer(cmdQueue, seed_dev, Bool.True, (IntPtr)0, (IntPtr)(sizeof(long) * 1), seed_host, 0, null, out e);
ErrorCheck(error, "CL.EnqueueReadBuffer");
//Dispose your shit
error = Cl.ReleaseCommandQueue(cmdQueue);
ErrorCheck(error, "CL.ReleaseCommandQueue");
error = Cl.ReleaseMemObject(sequence_dev);
ErrorCheck(error, "CL.ReleaseMemObject");
error = Cl.ReleaseMemObject(seed_dev);
ErrorCheck(error, "CL.ReleaseMemObject");
return(seed_host[0]);
}
public void singlePass(Kernel kernel, FloatMap inMap, FloatMap outMap)
{
var clInImageFormat = new OpenCL.Net.ImageFormat(ChannelOrder.Luminance, ChannelType.Float);
IMem inputMapBuffer = Cl.CreateImage2D(_context, MemFlags.CopyHostPtr | MemFlags.ReadOnly, clInImageFormat,
(IntPtr)inMap.W, (IntPtr)inMap.H, new IntPtr(inMap.Stride * sizeof(float)),
inMap._buf, out err);
assert(err, "input img creation");
IMem outputMapBuffer = Cl.CreateImage2D(_context, MemFlags.WriteOnly, clInImageFormat,
(IntPtr)outMap.W, (IntPtr)outMap.H, new IntPtr(outMap.Stride * sizeof(float)),
outMap._buf, out err);
assert(err, "output img creation");
// Set memory objects as parameters to kernel
err = Cl.SetKernelArg(kernel, 0, intPtrSize, inputMapBuffer);
assert(err, "input map setKernelArg");
err = Cl.SetKernelArg(kernel, 1, intPtrSize, outputMapBuffer);
assert(err, "output map setKernelArg");
// write actual data into memory object
IntPtr[] originPtr = new IntPtr[] { (IntPtr)0, (IntPtr)0, (IntPtr)0 }; //x, y, z
IntPtr[] inRegionPtr = new IntPtr[] { (IntPtr)inMap.W, (IntPtr)inMap.H, (IntPtr)1 }; //x, y, z
IntPtr[] outRegionPtr = new IntPtr[] { (IntPtr)outMap.W, (IntPtr)outMap.H, (IntPtr)1 }; //x, y, z
IntPtr[] workGroupSizePtr = new IntPtr[] { (IntPtr)outMap.W, (IntPtr)outMap.H, (IntPtr)1 };
Event clevent;
//err = Cl.EnqueueWriteImage(_commandsQueue, inputMapBuffer, Bool.True, originPtr, inRegionPtr, (IntPtr)0, (IntPtr)0, inMap._buf, 0, null, out clevent);
//clevent.Dispose();
//assert(err, "write input img");
// execute
err = Cl.EnqueueNDRangeKernel(_commandsQueue, kernel, 2,
originPtr,
workGroupSizePtr,
null, 0, null, out clevent);
clevent.Dispose();
assert(err, "Cl.EnqueueNDRangeKernel");
// sync
Cl.Finish(_commandsQueue);
// read from output memory object into actual buffer
err = Cl.EnqueueReadImage(_commandsQueue, outputMapBuffer, Bool.True, originPtr, outRegionPtr, new IntPtr(outMap.Stride * sizeof(float)), (IntPtr)0, outMap._buf, 0, null, out clevent);
clevent.Dispose();
assert(err, "read output buffer");
Cl.ReleaseMemObject(inputMapBuffer);
Cl.ReleaseMemObject(outputMapBuffer);
}
public KernelImage(int index, MemFlags flags, byte[] bytes, int width, int height, ImageFormat pixelFormat)
: this(index, flags, width, height, pixelFormat)
{
Source = bytes;
GpuBuffer = Cl.CreateImage2D(EasyClCompiler.Context, (OpenCL.Net.MemFlags)Flags, ImageFormat, (IntPtr)Width, (IntPtr)Height, (IntPtr)0, Source, out ErrorCode error);
if (error != ErrorCode.Success)
{
throw new GPUException("SetImage2D", error.ToString());
}
}
public static ErrorCode EnqueueWriteBuffer <T>(CommandQueue commandQueue,
IMem <T> buffer,
Bool blockingWrite,
T[] data,
int numEventsInWaitList,
Event[] eventWaitList,
out Event e) where T : struct
{
return(EnqueueWriteBuffer(commandQueue, buffer, blockingWrite, 0, data.Length, data, numEventsInWaitList, eventWaitList, out e));
}
public static void WriteToBuffer <T>(this CommandQueue commandQueue, IMem buffer, T[] data, int offset = 0, long length = -1, params Event[] waitFor)
where T : struct
{
Event e;
var elemSize = TypeSize <T> .SizeInt;
Cl.EnqueueWriteBuffer(commandQueue, buffer, Bool.True, (IntPtr)(offset * elemSize), (IntPtr)((length == -1 ? data.Length : length) * elemSize), data,
(uint)waitFor.Length, waitFor.Length == 0 ? null : waitFor, out e)
.Check();
e.Dispose();
}
public override bool ValidateUser(string username, string password)
{
MODEL.User user = IMem.GetUser8Name(username);
if (user == null)
{
return(false);
}
else
{
return(user.Password == password);
}
}
public ClBuffer(Context context, int size, out ErrorCode error)
{
ready = false;
this.Size = size;
buffer = new T[size];
clMem = Cl.CreateBuffer <T>(context, MemFlags.UseHostPtr, buffer, out error);
if (error == ErrorCode.Success)
{
ready = true;
}
}
public static ErrorCode EnqueueWriteBuffer <T>(CommandQueue commandQueue,
IMem <T> buffer,
Bool blockingWrite,
int offset,
int length,
T[] data,
int numEventsInWaitList,
Event[] eventWaitList,
out Event e) where T : struct
{
var elementSize = (int)TypeSize <T> .Size;
return(EnqueueWriteBuffer(commandQueue, buffer, blockingWrite, (IntPtr)(offset * elementSize), (IntPtr)(length * elementSize), data, (uint)numEventsInWaitList, eventWaitList, out e));
}
public override MembershipUser GetUser(object providerUserKey, bool userIsOnline)
{
var user = IMem.GetUser8Id((Guid)providerUserKey);
if (user == null)
{
return(null);
}
MembershipUser mu = new MembershipUser("memProvider", user.Name,
user.ID, user.Email, string.Empty, string.Empty, true, true, user.CreateTime, DateTime.Now,
DateTime.Now, DateTime.Now, DateTime.Now);
return(mu);
}
public GpuSimiulation(IStockData stockData, int holding = 10, float ann = 5.0990f)
{
_stockData = stockData;
_holding = holding;
_ann = ann;
_programSource = File.ReadAllText("Kernels.cl");
_env = "*AMD*".CreateCLEnvironment(DeviceType.Gpu, CommandQueueProperties.ProfilingEnable);
_h_output = new float[(stockData.QuotesPerStock - holding)];
_d_stocksAndPrices = Cl.CreateBuffer <float>(_env.Context, MemFlags.ReadOnly | MemFlags.CopyHostPtr, _stockData.RawData, out ErrorCode err);
_d_portfolioStockMv = Cl.CreateBuffer <float>(_env.Context, MemFlags.ReadOnly | MemFlags.CopyHostPtr, _stockData.MarketValues, out err);
_d_output = Cl.CreateBuffer <float>(_env.Context, MemFlags.ReadWrite, stockData.StocksCount * (stockData.QuotesPerStock - holding), out err);
}
public override MembershipUser GetUser(string username, bool userIsOnline)
{
MODEL.User user = IMem.GetUser8Name(username);
if (user == null)
{
return(null);
}
MembershipUser mu = new MembershipUser("memProvider",
username, user.ID, user.Email, string.Empty, string.Empty,
true, true, user.CreateTime,
DateTime.Now, DateTime.Now, DateTime.Now, DateTime.Now);
return(mu);
}
public override bool ChangePassword(string username, string oldPassword, string newPassword)
{
var user = IMem.GetUser8Name(username);
if (user.Password == oldPassword)
{
user.Password = newPassword;
IMem.UpdateUser(user);
return(true);
}
else
{
return(false);
}
}
public void EnqueueWriteBuffer <T>(IMem <T> buffer, int offset, int length, T[] data)
where T : struct
{
Cl.EnqueueWriteBuffer(
CommandQueue,
buffer,
Bool.True,
offset,
length,
data,
0, //_awaitEvent == null ? 0 : 1,
null, //_awaitEvent == null ? new Event[0] : new Event[]{ _awaitEvent.Value },
out var awaitEvent
);
//_awaitEvent = awaitEvent;
}
public static InfoBuffer GetMemObjectInfo(IMem mem, MemInfo paramName, out ErrorCode error)
{
if (paramName == MemInfo.HostPtr) // Handle special case
{
IntPtr size = GetInfo(Cl.GetMemObjectInfo, mem, MemInfo.Size, out error).CastTo <IntPtr>();
var buffer = new InfoBuffer(size);
error = GetMemObjectInfo(mem, paramName, size, buffer, out size);
if (error != ErrorCode.Success)
{
return(InfoBuffer.Empty);
}
return(buffer);
}
return(GetInfo(GetMemObjectInfo, mem, paramName, out error));
}
public KernelImage(int index, MemFlags flags, Bitmap bmp) : this(index, flags, bmp.Width, bmp.Height, bmp.PixelFormat)
{
Rectangle bmpRect = new Rectangle(Point.Empty, bmp.Size);
SystemGDI.BitmapData bmpData = bmp.LockBits(bmpRect, SystemGDI.ImageLockMode.ReadWrite, bmp.PixelFormat);
IntPtr bmpPtr = bmpData.Scan0;
int byteCount = bmpData.Stride * Height;
Source = new byte[byteCount];
Marshal.Copy(bmpPtr, Source, 0, byteCount);
GpuBuffer = Cl.CreateImage2D(EasyClCompiler.Context, (OpenCL.Net.MemFlags)flags, ImageFormat, (IntPtr)Width, (IntPtr)Height, (IntPtr)0, Source, out ErrorCode error);
if (error != ErrorCode.Success)
{
throw new GPUException("SetImage2D", error.ToString());
}
bmp.UnlockBits(bmpData);
}
public static void ImportFile(string filename, IMem memobject)
{
var parts = Path.GetFileNameWithoutExtension(filename).Split('-').Select(x => Convert.ToInt32(x)).ToArray();
Uri fname = new Uri(filename, UriKind.RelativeOrAbsolute);
BmpBitmapDecoder bd = new BmpBitmapDecoder(fname, BitmapCreateOptions.PreservePixelFormat, BitmapCacheOption.None);
int width = parts[1];
int height = parts[2];
int size = width * height;
var data = new byte[size];
bd.Frames.Single().CopyPixels(data, width, 0);
memobject.writepixels(data, parts[0], size);
}
public void Dispose()
{
UnlockBits();
original = null;
if (clMem != null)
{
Cl.ReleaseMemObject(clMem);
clMem = null;
}
if (bitmapData != null)
{
bitmapData = null;
}
ready = false;
}
/// <summary>
/// Writes an image to GPU memeory
/// </summary>
/// <param name="image">the image to write</param>
/// <param name="memFlags">the memory flags</param>
public void WriteImage(Bitmap image, MemFlags memFlags)
{
ErrorCode error;
_image = Cl.CreateImage2D(_handle.Context, memFlags, new OpenCL.Net.ImageFormat(ChannelOrder.RGBA, ChannelType.Unorm_Int8),
new IntPtr(image.Width), new IntPtr(image.Height), new IntPtr(0), new IntPtr(0), out error);
CLException.CheckException(error);
byte[] data = BmpToBytes(image);
IntPtr[] originArray = new IntPtr[] { new IntPtr(0), new IntPtr(0), new IntPtr(0) };
IntPtr[] sizeArray = new IntPtr[] { new IntPtr(image.Width), new IntPtr(image.Height), new IntPtr(1) };
Event e;
error = Cl.EnqueueWriteImage(_handle.Queue, _image, Bool.True,
originArray, sizeArray, new IntPtr(0), new IntPtr(0),
data, 0, null, out e);
CLException.CheckException(error);
}
public T[] ReadBuffer <T>(IMem <T> buffer, int offset, int length)
where T : struct
{
var data = new T[length];
Cl.EnqueueReadBuffer(
CommandQueue,
buffer,
Bool.True,
offset,
length,
data,
0, //_awaitEvent == null ? 0 : 1,
null, //_awaitEvent == null ? new Event[0] : new Event[]{ _awaitEvent.Value },
out var awaitEvent
);
//s_awaitEvent = awaitEvent;
return(data);
}
public override MembershipUser CreateUser(string username, string password, string email, string passwordQuestion, string passwordAnswer, bool isApproved, object providerUserKey, out MembershipCreateStatus status)
{
var exsit = GetUser(username, true);
var user = new MODEL.User()
{
Name = username,
Password = password,
Opentype = MODEL.OpenType.localuser,
CreateTime = DateTime.Now,
LastLoginTime = DateTime.Now
};
IMem.CreateUser(user);
MembershipUser mu = new MembershipUser("memProvider",
username, user.ID, email, "", "", true, true, DateTime.Now, DateTime.Now, DateTime.Now, DateTime.Now, DateTime.Now);
status = MembershipCreateStatus.Success;
return(mu);
}
public FloatMap GaussianBlur(FloatMap inMap, float sigma)
{
var k = _kernels["convolveImg"];
FloatMap outMap = new FloatMap(inMap.W, inMap.H);
FloatMap mask = createBlurMask(sigma);
IMem <float> maskBuf = Cl.CreateBuffer <float>(_context, MemFlags.CopyHostPtr | MemFlags.ReadOnly, mask._buf, out err);
assert(err, "capholes mask, mem object creation");
err = Cl.SetKernelArg(k, 2, intPtrSize, maskBuf);
assert(err, "capholes mask, setKernelArg");
err = Cl.SetKernelArg(k, 3, intSize, (mask.W - 1) / 2);
assert(err, "capholes mask, setKernelArg");
singlePass(k, inMap, outMap);
return(outMap);
}
private FloatMap capHoles(FloatMap inMap, int filterHalfSize)
{
var k = _kernels["capHolesImg"];
FloatMap outMap = new FloatMap(inMap.W, inMap.H);
FloatMap mask = getDistanceWeightMap(filterHalfSize);
IMem <float> maskBuf = Cl.CreateBuffer <float>(_context, MemFlags.CopyHostPtr | MemFlags.ReadOnly, mask._buf, out err);
assert(err, "capholes mask, mem object creation");
err = Cl.SetKernelArg(k, 2, intPtrSize, maskBuf);
assert(err, "capholes mask, setKernelArg");
err = Cl.SetKernelArg(k, 3, intSize, filterHalfSize);
assert(err, "capholes mask, setKernelArg");
singlePass(k, inMap, outMap);
Cl.ReleaseMemObject(maskBuf);
return(outMap);
}
public WriteableBitmap GenerateSerialNxMWritableBitmap(IMem pixels, int loffset, int width, int height, int tileSizeX, int tileSizeY, BitmapPalette palette)
{
var wb = new WriteableBitmap(width, height, 96, 96, PixelFormats.Indexed8, palette);
int tileLinearLength = tileSizeX * tileSizeY;
int tilewidth = width / tileSizeX;
int tileheight = height / tileSizeY;
for (int tj = 0; tj < tileheight; tj++)
{
for (int ti = 0; ti < tilewidth; ti++)
{
var left = ti * tileSizeX;
var top = tj * tileSizeY;
var offset = loffset + (ti * tileLinearLength) + (tj * tileLinearLength * tilewidth);
var arr = pixels.readpixels(offset, tileLinearLength);
//// Reverse X
//var ar1 = arr.Select((x, idx) => new { x, idx })
// .GroupBy(x => x.idx / tileSizeX)
// .Select(x => x.Select(y => y.x).Reverse());
//// Reverse Y
//arr = ar1.SelectMany(x => x);
var ar3 = arr.Select(x => Convert.ToByte(x)).ToArray();
wb.WritePixels(new Int32Rect(left, top, tileSizeX, tileSizeY), ar3, tileSizeX, 0);
}
}
return(wb);
}
private void ready()
{
ErrorCode error;
context = Cl.CreateContext(null, 1, new[] { device }, null, IntPtr.Zero, out error);
string source = System.IO.File.ReadAllText("kernels.cl");
program = Cl.CreateProgramWithSource(context, 1, new[] { source }, null, out error);
error = Cl.BuildProgram(program, 1, new[] { device }, string.Empty, null, IntPtr.Zero);
InfoBuffer buildStatus = Cl.GetProgramBuildInfo(program, device, ProgramBuildInfo.Status, out error);
if (buildStatus.CastTo <BuildStatus>() != BuildStatus.Success)
{
throw new Exception($"OpenCL could not build the kernel successfully: {buildStatus.CastTo<BuildStatus>()}");
}
allGood(error);
Kernel[] kernels = Cl.CreateKernelsInProgram(program, out error);
kernel = kernels[0];
allGood(error);
queue = Cl.CreateCommandQueue(context, device, CommandQueueProperties.None, out error);
allGood(error);
dataOut = Cl.CreateBuffer(context, MemFlags.WriteOnly, (IntPtr)(globalSize * sizeof(int)), out error);
allGood(error);
var intSizePtr = new IntPtr(Marshal.SizeOf(typeof(int)));
error |= Cl.SetKernelArg(kernel, 2, new IntPtr(Marshal.SizeOf(typeof(IntPtr))), dataOut);
error |= Cl.SetKernelArg(kernel, 3, intSizePtr, new IntPtr(worldSeed));
error |= Cl.SetKernelArg(kernel, 4, intSizePtr, new IntPtr(globalSize));
allGood(error);
}
public static void ReadFromBuffer <T>(this CommandQueue commandQueue, IMem <T> buffer, T[] array, int offset = 0, long length = -1, params Event[] waitFor)
where T : struct
{
ReadFromBuffer(commandQueue, (IMem)buffer, array, offset, length, waitFor);
}
public void sortKeysOnly(IMem input, IMem output,
int numElements)
{
debugRead = new int[Math.Max(numElements,numCounters)];
OpenCL.Net.ErrorCode error;
Event eve;
mCounters = Cl.CreateBuffer(cxGPUContext, MemFlags.ReadWrite, gpuConstants.numGroupsPerBlock * gpuConstants.numRadices * gpuConstants.numBlocks * sizeof(int),
out error);
CheckErr(error, "Cl.CreateBuffer");
mRadixPrefixes = Cl.CreateBuffer(cxGPUContext, MemFlags.ReadWrite, gpuConstants.numRadices * sizeof(int), out error);
CheckErr(error, "Cl.CreateBuffer");
gpuConstants.numElementsPerGroup = (numElements/(gpuConstants.numBlocks*gpuConstants.numGroupsPerBlock)) +1 ;
gpuConstants.numTotalElements = numElements;
int i;
for (i = 0; i < 8; i++)
{
error = Cl.EnqueueWriteBuffer(cqCommandQueue, mCounters, Bool.True, IntPtr.Zero, (IntPtr)(numCounters * 4),
counters, 0, null, out eve);
CheckErr(error, "Cl.EnqueueWriteBuffer Counter initialize");
if (i%2 == 0)
{
DateTime before = DateTime.Now;
SetupAndCount(input, 4 * i);
Console.WriteLine("Setup and Count =" + (DateTime.Now - before).TotalMilliseconds);
before = DateTime.Now;
SumIt(input, 4 * i);
Console.WriteLine("SumIt =" + (DateTime.Now - before).TotalMilliseconds);
before = DateTime.Now;
ReorderingKeysOnly(input, output, 4 * i);
Console.WriteLine("Reorder =" + (DateTime.Now - before).TotalMilliseconds);
}
else
{
SetupAndCount(output, 4 * i);
SumIt(output, 4 * i);
ReorderingKeysOnly(output, input, 4 * i);
}
}
if (i%2 != 0)
{
error= Cl.EnqueueCopyBuffer(cqCommandQueue, input, output, IntPtr.Zero, IntPtr.Zero, (IntPtr)(numElements * 4), 0, null, out eve);
CheckErr(error, "Cl.EnqueueCopyBuffer");
error = Cl.Finish(cqCommandQueue);
CheckErr(error, "Cl.Finish Copybuffer");
}
error = Cl.ReleaseMemObject(mRadixPrefixes);
CheckErr(error, "Cl.ReleaseMemObj");
error = Cl.ReleaseMemObject(mCounters);
CheckErr(error, "Cl.ReleaseMemObj");
}
public void sortKeysValue(IMem key, IMem value,
int numElements)
{
debugRead = new int[Math.Max(numElements, numCounters)];
OpenCL.Net.ErrorCode error;
Event eve;
/*
error = Cl.EnqueueReadBuffer(cqCommandQueue, input, Bool.True, IntPtr.Zero, (IntPtr)(numElements * 4),
debugRead, 0, null, out eve);
CheckErr(error, "Cl.EnqueueReadBuffer");
*/
mCounters = Cl.CreateBuffer(cxGPUContext, MemFlags.ReadWrite, gpuConstants.numGroupsPerBlock * gpuConstants.numRadices * gpuConstants.numBlocks * sizeof(int),
out error);
CheckErr(error, "Cl.CreateBuffer");
mRadixPrefixes = Cl.CreateBuffer(cxGPUContext, MemFlags.ReadWrite, gpuConstants.numRadices * sizeof(int), out error);
CheckErr(error, "Cl.CreateBuffer");
IMem outputKey = Cl.CreateBuffer(cxGPUContext, MemFlags.ReadWrite, (IntPtr)(4 * numElements),
out error);
CheckErr(error, "Cl.CreateBuffer");
IMem outputValue = Cl.CreateBuffer(cxGPUContext, MemFlags.ReadWrite, (IntPtr)(8 * numElements),
out error);
CheckErr(error, "Cl.CreateBuffer");
gpuConstants.numElementsPerGroup = (numElements / (gpuConstants.numBlocks * gpuConstants.numGroupsPerBlock)) + 1;
gpuConstants.numTotalElements = numElements;
int i;
for (i = 0; i < 8; i++)
{
error = Cl.EnqueueWriteBuffer(cqCommandQueue, mCounters, Bool.True, IntPtr.Zero, (IntPtr)(numCounters * 4),
counters, 0, null, out eve);
CheckErr(error, "Cl.EnqueueWriteBuffer Counter initialize");
if (i % 2 == 0)
{
SetupAndCount(key, 4 * i);
SumIt(key, 4 * i);
ReorderingKeyValue(key, outputKey, value, outputValue, 4 * i);
}
else
{
SetupAndCount(outputKey, 4 * i);
SumIt(outputKey, 4 * i);
ReorderingKeyValue(outputKey, key, outputValue, value, 4 * i);
}
}
if (i % 2 == 0)
{
error = Cl.EnqueueCopyBuffer(cqCommandQueue, outputKey, key, IntPtr.Zero, IntPtr.Zero, (IntPtr)(numElements * 4), 0, null, out eve);
CheckErr(error, "Cl.EnqueueCopyBuffer");
error = Cl.Finish(cqCommandQueue);
CheckErr(error, "Cl.Finish Copybuffer");
error = Cl.EnqueueCopyBuffer(cqCommandQueue, outputValue, value, IntPtr.Zero, IntPtr.Zero, (IntPtr)(numElements * 8), 0, null, out eve);
CheckErr(error, "Cl.EnqueueCopyBuffer");
error = Cl.Finish(cqCommandQueue);
CheckErr(error, "Cl.Finish Copybuffer");
}
error = Cl.ReleaseMemObject(outputKey);
CheckErr(error, "Cl.ReleaseMemObj");
error = Cl.ReleaseMemObject(outputValue);
CheckErr(error, "Cl.ReleaseMemObj");
error = Cl.ReleaseMemObject(mRadixPrefixes);
CheckErr(error, "Cl.ReleaseMemObj");
error = Cl.ReleaseMemObject(mCounters);
CheckErr(error, "Cl.ReleaseMemObj");
}
private void ReorderingKeyValue(IMem inputKey, IMem outputKey,IMem inputValue, IMem outputValue, int bitOffset)
{
OpenCL.Net.ErrorCode error;
IntPtr agentPtrSize = (IntPtr)0;
agentPtrSize = (IntPtr)Marshal.SizeOf(typeof(IntPtr));
var ptrSize = (IntPtr)Marshal.SizeOf(typeof(Mem));
int globalWorkSize = gpuConstants.numThreadsPerBlock * gpuConstants.numBlocks;
int localWorkSize = gpuConstants.numThreadsPerBlock;
IntPtr[] workGroupSizePtr = new IntPtr[] { (IntPtr)globalWorkSize };
IntPtr[] localWorkGroupSizePtr = new IntPtr[] { (IntPtr)localWorkSize };
Event clevent;
error = Cl.SetKernelArg(ckReorderingKeyValue, 0, ptrSize, inputKey);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckReorderingKeyValue, 1, ptrSize, outputKey);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckReorderingKeyValue, 2, ptrSize, inputValue);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckReorderingKeyValue, 3, ptrSize, outputValue);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckReorderingKeyValue, 4, ptrSize, mCounters);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckReorderingKeyValue, 5, ptrSize, mRadixPrefixes);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckReorderingKeyValue, 6, (IntPtr)(gpuConstants.numGroupsPerBlock * gpuConstants.numBlocks * gpuConstants.numRadicesPerBlock * 4), null);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckReorderingKeyValue, 7, (IntPtr)(Marshal.SizeOf(typeof(GPUConstants))), gpuConstants);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckReorderingKeyValue, 8, (IntPtr)4, bitOffset);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.EnqueueNDRangeKernel(cqCommandQueue, ckReorderingKeyValue, 1, null, workGroupSizePtr, localWorkGroupSizePtr, 0, null, out clevent);
CheckErr(error, "Cl.EnqueueNDRangeKernel");
error = Cl.Finish(cqCommandQueue);
CheckErr(error, "Cl.Finish");
if (DEBUG)
{
Console.WriteLine("-------------------------------Reordering-------------------------------------------------");
Event eve;
Console.WriteLine(" Input ");
Cl.EnqueueReadBuffer(cqCommandQueue, inputKey, Bool.True, IntPtr.Zero, (IntPtr)(gpuConstants.numTotalElements * 4), debugRead, 0,
null, out eve);
CheckErr(error, "Cl.EnqueueReadBuffer");
PrintElementBuffer(debugRead, gpuConstants.numTotalElements, "Reordering -> Input -> bitoffset = " + bitOffset);
Cl.EnqueueReadBuffer(cqCommandQueue, inputValue, Bool.True, IntPtr.Zero, (IntPtr)(gpuConstants.numTotalElements * 4), debugRead, 0,
null, out eve);
CheckErr(error, "Cl.EnqueueReadBuffer");
PrintElementBuffer(debugRead, gpuConstants.numTotalElements, "Reordering -> InputValues -> bitoffset = " + bitOffset);
Console.WriteLine(" Counters ");
Cl.EnqueueReadBuffer(cqCommandQueue, mCounters, Bool.True, IntPtr.Zero, (IntPtr)(numCounters * sizeof(int)), debugRead, 0,
null, out eve);
CheckErr(error, "Cl.EnqueueReadBuffer");
PrintCounterBuffer(debugRead, "Reordering -> bitoffset = " + bitOffset);
Console.WriteLine(" Counters ");
Cl.EnqueueReadBuffer(cqCommandQueue, mRadixPrefixes, Bool.True, IntPtr.Zero, (IntPtr)(gpuConstants.numRadices * sizeof(int)), debugRead, 0,
null, out eve);
CheckErr(error, "Cl.EnqueueReadBuffer");
PrintElementBuffer(debugRead, gpuConstants.numRadices, "Reordering -> RadixPrefixe -> bitoffset = " + bitOffset);
Console.WriteLine(" Output ");
Cl.EnqueueReadBuffer(cqCommandQueue, outputKey, Bool.True, IntPtr.Zero, (IntPtr)(gpuConstants.numTotalElements * 4), debugRead, 0,
null, out eve);
CheckErr(error, "Cl.EnqueueReadBuffer");
PrintElementBuffer(debugRead, gpuConstants.numTotalElements, "Reordering -> Output -> bitoffset = " + bitOffset);
Cl.EnqueueReadBuffer(cqCommandQueue, outputValue, Bool.True, IntPtr.Zero, (IntPtr)(gpuConstants.numTotalElements * 4), debugRead, 0,
null, out eve);
CheckErr(error, "Cl.EnqueueReadBuffer");
PrintElementBuffer(debugRead, gpuConstants.numTotalElements, "Reordering -> OutputValue -> bitoffset = " + bitOffset);
Console.WriteLine("Reordering -> bitoffset = " + bitOffset);
Console.WriteLine();
};
}
private void SumIt(IMem input, int bitOffset)
{
OpenCL.Net.ErrorCode error;
IntPtr agentPtrSize = (IntPtr)0;
agentPtrSize = (IntPtr)Marshal.SizeOf(typeof(IntPtr));
var ptrSize = (IntPtr)Marshal.SizeOf(typeof(Mem));
int globalWorkSize = gpuConstants.numThreadsPerBlock * gpuConstants.numBlocks;
int localWorkSize = gpuConstants.numThreadsPerBlock;
IntPtr[] workGroupSizePtr = new IntPtr[] { (IntPtr)globalWorkSize };
IntPtr[] localWorkGroupSizePtr = new IntPtr[] { (IntPtr)localWorkSize };
Event clevent;
error = Cl.SetKernelArg(ckSumIt, 0, ptrSize, input);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckSumIt, 1, ptrSize, mCounters);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckSumIt, 2, ptrSize, mRadixPrefixes);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckSumIt, 3, (IntPtr)(Marshal.SizeOf(typeof(GPUConstants))), gpuConstants);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckSumIt, 4, (IntPtr)4, bitOffset);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckSumIt, 5, (IntPtr)(4* gpuConstants.numBlocks*gpuConstants.numGroupsPerBlock), null);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.EnqueueNDRangeKernel(cqCommandQueue, ckSumIt, 1, null, workGroupSizePtr, localWorkGroupSizePtr, 0, null, out clevent);
CheckErr(error, "Cl.EnqueueNDRangeKernel");
error = Cl.Finish(cqCommandQueue);
CheckErr(error, "Cl.Finish");
if (DEBUG)
{
Event eve;
error = Cl.EnqueueReadBuffer(cqCommandQueue, input, Bool.True, IntPtr.Zero, (IntPtr)(gpuConstants.numTotalElements * 4),
debugRead, 0, null, out eve);
CheckErr(error, "Cl.EnqueueReadBuffer");
PrintElementBuffer(debugRead, gpuConstants.numTotalElements, "SumIt -> Input -> bitoffset = " + bitOffset);
Console.WriteLine(" Counters ");
Cl.EnqueueReadBuffer(cqCommandQueue, mCounters, Bool.True, IntPtr.Zero, (IntPtr)(numCounters * sizeof(int)), debugRead, 0,
null, out eve);
CheckErr(error, "Cl.EnqueueReadBuffer");
PrintCounterBuffer(debugRead, "SumIt -> bitoffset = " + bitOffset);
Console.WriteLine("SumIt -> bitoffset = " + bitOffset);
Console.WriteLine();
};
}
public static ErrorCode EnqueueReleaseGLObjects(CommandQueue queue, IMem[] glObjects, uint waitListCount, OpenCL.Net.Event[] waitList, out OpenCL.Net.Event outEvent){
return clEnqueueReleaseGLObjects (queue, (uint)glObjects.Length, (from m in glObjects select (m as IHandleData).Handle).ToArray (), waitListCount, waitList, out outEvent);
}
private void InitOpenCl()
{
if (File.Exists(Path.Combine(System.Environment.CurrentDirectory, ".." + Path.DirectorySeparatorChar + ".." + Path.DirectorySeparatorChar + "log.txt")))
{
File.Delete(Path.Combine(System.Environment.CurrentDirectory, ".." + Path.DirectorySeparatorChar + ".." + Path.DirectorySeparatorChar + "log.txt"));
}
if (File.Exists(Path.Combine(System.Environment.CurrentDirectory, ".." + Path.DirectorySeparatorChar + ".." + Path.DirectorySeparatorChar + "sortLog.txt")))
{
File.Delete(Path.Combine(System.Environment.CurrentDirectory, ".." + Path.DirectorySeparatorChar + ".." + Path.DirectorySeparatorChar + "sortLog.txt"));
}
if (File.Exists(Path.Combine(System.Environment.CurrentDirectory, ".." + Path.DirectorySeparatorChar + ".." + Path.DirectorySeparatorChar + "OpenCLDebugLog.txt")))
{
File.Delete(Path.Combine(System.Environment.CurrentDirectory, ".." + Path.DirectorySeparatorChar + ".." + Path.DirectorySeparatorChar + "OpenCLDebugLog.txt"));
}
OpenCL.Net.ErrorCode error;
var platforms = Cl.GetPlatformIDs(out error);
List<Device> devicesList = new List<Device>();
CheckErr(error, "Cl.GetPlatformIDs");
foreach (Platform platform in platforms)
{
string platformName = Cl.GetPlatformInfo(platform, PlatformInfo.Name, out error).ToString();
Console.WriteLine("Platform: " + platformName);
CheckErr(error, "Cl.GetPlatformInfo");
//We will be looking only for GPU devices
foreach (Device device in Cl.GetDeviceIDs(platform, DeviceType.Gpu, out error))
{
CheckErr(error, "Cl.GetDeviceIDs");
var vendor = Cl.GetDeviceInfo(device, DeviceInfo.Vendor, out error);
var name = Cl.GetDeviceInfo(device, DeviceInfo.Name, out error);
var worksize = Cl.GetDeviceInfo(device, DeviceInfo.MaxWorkGroupSize, out error);
Console.WriteLine("Vendor: " + vendor + " , " + name);
Console.WriteLine("Device: " + device.GetType());
Console.WriteLine("Workgroupsize: " + worksize.CastTo<long>());
devicesList.Add(device);
}
}
if (devicesList.Count <= 0)
{
Console.WriteLine("No devices found.");
return;
}
_device = devicesList[0];
if (Cl.GetDeviceInfo(_device, DeviceInfo.ImageSupport, out
error).CastTo<Bool>() == Bool.False)
{
Console.WriteLine("No image support.");
return;
}
cxGPUContext
= Cl.CreateContext(null, 1, new[] { _device }, ContextNotify,
IntPtr.Zero, out error); //Second parameter is amount of devices
CheckErr(error, "Cl.CreateContext");
cqCommandQueue = Cl.CreateCommandQueue(cxGPUContext, _device, (CommandQueueProperties)0, out error);
CheckErr(error, "Cl.CreateCommandQueue");
sort = new GPURadixSort(cqCommandQueue, cxGPUContext, _device);
constants.numBlocks = numBlocks;
constants.numThreadsPerBlock = numThreadsPerBlock;
constants.xMax = (float)m_OuterBoundary.X;
constants.yMax = (float)m_OuterBoundary.Y;
constants.zMax = (float)m_OuterBoundary.Z;
constants.cellSizeX = (float)m_CellSize.X;
constants.cellSizeY = (float)m_CellSize.Y;
constants.cellSizeZ = (float)m_CellSize.Z;
constants.xGridBoundary = (int)(m_OuterBoundary.X / m_CellSize.X);
constants.yGridBoundary = (int)(m_OuterBoundary.Y / m_CellSize.Y);
constants.zGridBoundary = (int)(m_OuterBoundary.Z / m_CellSize.Z);
string programSource = System.IO.File.ReadAllText(Path.Combine(System.Environment.CurrentDirectory, ".." + Path.DirectorySeparatorChar + ".." + Path.DirectorySeparatorChar + "CollisionDetection.cl"));
IntPtr[] progSize = new IntPtr[] { (IntPtr)programSource.Length };
OpenCL.Net.Program clProgramCollision = Cl.CreateProgramWithSource(cxGPUContext, 1, new[] { programSource }, progSize,
out error);
CheckErr(error, "createProgramm");
string flags = "-cl-fast-relaxed-math";
error = Cl.BuildProgram(clProgramCollision, 1, new[] { _device }, flags, null, IntPtr.Zero);
CheckErr(error, "Cl.BuildProgram");
//Check for any compilation errors
if (Cl.GetProgramBuildInfo(clProgramCollision, _device, ProgramBuildInfo.Status, out error).CastTo<BuildStatus>()
!= BuildStatus.Success)
{
CheckErr(error, "Cl.GetProgramBuildInfo");
Console.WriteLine("Cl.GetProgramBuildInfo != Success");
Console.WriteLine(Cl.GetProgramBuildInfo(clProgramCollision, _device, ProgramBuildInfo.Log, out error));
using (StreamWriter sw = File.AppendText(Path.Combine(System.Environment.CurrentDirectory, ".." + Path.DirectorySeparatorChar + ".." + Path.DirectorySeparatorChar + "OpenCLDebugLog.txt")))
{
sw.WriteLine(Cl.GetProgramBuildInfo(clProgramCollision, _device, ProgramBuildInfo.Log, out error));
}
return;
}
int ciErrNum;
ckCreateCellIdArray = Cl.CreateKernel(clProgramCollision, "CreateCellId", out error);
CheckErr(error, "Cl.CreateKernel");
ckCreateCollisionList = Cl.CreateKernel(clProgramCollision, "CreateCollisionList", out error);
CheckErr(error, "Cl.CreateKernel");
ckCreateCollsionTuples = Cl.CreateKernel(clProgramCollision, "CreateCollsionTuples", out error);
CheckErr(error, "Cl.CreateKernel");
ckCheckCollsions = Cl.CreateKernel(clProgramCollision, "CheckCollsions", out error);
CheckErr(error, "Cl.CreateKernel");
freeCellList = new bool[constants.xGridBoundary*constants.yGridBoundary*constants.zGridBoundary];
freeCellMem = Cl.CreateBuffer(cxGPUContext, MemFlags.ReadWrite, (IntPtr)(freeCellList.Length *Marshal.SizeOf(typeof(bool))), sharedIdxMem,
out error);
CheckErr(error, "Createbuffer");
}
private void CheckCollsions()
{
ErrorCode error;
Event eve;
IntPtr agentPtrSize = (IntPtr)0;
var ptrSize = (IntPtr)Marshal.SizeOf(typeof(Mem));
int globalWorkSize = constants.numThreadsPerBlock * constants.numBlocks;
int localWorkSize = constants.numThreadsPerBlock;
IntPtr[] workGroupSizePtr = new IntPtr[] { (IntPtr)globalWorkSize };
IntPtr[] localWorkGroupSizePtr = new IntPtr[] { (IntPtr)localWorkSize };
Event clevent;
collisons = Cl.CreateBuffer(cxGPUContext, MemFlags.ReadWrite, (IntPtr)(sharedIdx[0] * Marshal.SizeOf(typeof(CollisionTupel))),
out error);
CheckErr(error, "Createbuffer");
collisionShapes = Cl.CreateBuffer(cxGPUContext, MemFlags.ReadWrite, (IntPtr)(sharedIdx[0] * Marshal.SizeOf(typeof(ShapeTupel))),
out error);
CheckErr(error, "Createbuffer");
collisionShapeTupels = new ShapeTupel[sharedIdx[0]];
for (int i = 0; i < sharedIdx[0]; i++)
{
collisionShapeTupels[i] = new ShapeTupel();
int keytmp = (int)(readCellData[tupelList[i].obj1] & 0xFFFFFFFF);
collisionShapeTupels[i].obj1 = objIdClShapeMap[keytmp];
keytmp = (int)(readCellData[tupelList[i].obj2] & 0xFFFFFFFF);
collisionShapeTupels[i].obj2 = objIdClShapeMap[keytmp];
}
// CheckCollsions
error = Cl.EnqueueWriteBuffer(cqCommandQueue, sharedIdxMem, Bool.True, IntPtr.Zero, (IntPtr)(4),
0, 0, null, out eve);
CheckErr(error, "EnqBuffer");
error = Cl.EnqueueWriteBuffer(cqCommandQueue, collisionShapes, Bool.True, IntPtr.Zero, (IntPtr)(sharedIdx[0] * Marshal.SizeOf(typeof(ShapeTupel))),
collisionShapeTupels, 0, null, out eve);
CheckErr(error, "EnqBuffer");
error = Cl.SetKernelArg(ckCheckCollsions, 0, ptrSize, cellData);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckCheckCollsions, 1, ptrSize, collisionTuples);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckCheckCollsions, 2, ptrSize, collisionShapes);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckCheckCollsions, 3, ptrSize, collisons);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckCheckCollsions, 4, ptrSize, sharedIdxMem);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckCheckCollsions, 5, (IntPtr)(Marshal.SizeOf(typeof(CollisionConstants))), constants);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckCheckCollsions, 6, (IntPtr)4, sharedIdx[0]);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.EnqueueNDRangeKernel(cqCommandQueue, ckCheckCollsions, 1, null, workGroupSizePtr, localWorkGroupSizePtr, 0, null, out clevent);
CheckErr(error, "Cl.EnqueueNDRangeKernel");
error = Cl.Finish(cqCommandQueue);
CheckErr(error, "Cl.Finish");
error = Cl.EnqueueReadBuffer(cqCommandQueue, sharedIdxMem, Bool.True, IntPtr.Zero, (IntPtr)(4),
sharedIdx, 0, null, out eve);
CheckErr(error, "Cl.EnqueueReadBuffer");
tupelList = new CollisionTupel[sharedIdx[0]];
error = Cl.EnqueueReadBuffer(cqCommandQueue, collisons, Bool.True, IntPtr.Zero, (IntPtr)(sharedIdx[0] * Marshal.SizeOf(typeof(CollisionTupel))),
tupelList, 0, null, out eve);
CheckErr(error, "Cl.EnqueueReadBuffer");
}
private void CreateBuffers()
{
ErrorCode error;
Event eve;
collisionList = Cl.CreateBuffer(cxGPUContext, MemFlags.ReadWrite, (IntPtr)( constants.numTotalElements * 4 * 8),
out error);
// Create Buffers
shapeMem = Cl.CreateBuffer(cxGPUContext, MemFlags.ReadWrite, (IntPtr)( constants.numTotalElements * Marshal.SizeOf(typeof(clShapeObject))), clShapeArray.ToArray(),
out error);
CheckErr(error, "Createbuffer");
sharedIdxMem = Cl.CreateBuffer(cxGPUContext, MemFlags.ReadWrite, (IntPtr)(4), sharedIdxMem,
out error);
CheckErr(error, "Createbuffer");
objIdMem = Cl.CreateBuffer(cxGPUContext, MemFlags.ReadWrite, (IntPtr)(clObjArray.Count * 8),
out error);
CheckErr(error, "Createbuffer");
cellIds = Cl.CreateBuffer(cxGPUContext, MemFlags.ReadWrite, (IntPtr)(clObjArray.Count * 4 * 8),
out error);
CheckErr(error, "Createbuffer");
cellData = Cl.CreateBuffer(cxGPUContext, MemFlags.ReadWrite, (IntPtr)(clObjArray.Count * 8 * 8),
out error);
CheckErr(error, "Createbuffer");
if (DEBUG)
{
DebugCollisionPosList = Cl.CreateBuffer(cxGPUContext, MemFlags.ReadWrite, (IntPtr)( constants.numTotalElements * 8 * Marshal.SizeOf(typeof(CollisionCell))),
out error);
}
error = Cl.EnqueueWriteBuffer(cqCommandQueue, shapeMem, Bool.True, IntPtr.Zero, (IntPtr)(clShapeArray.Count * Marshal.SizeOf(typeof(clShapeObject))),
clShapeArray.ToArray(), 0, null,
out eve);
CheckErr(error, "EnqBuffer");
error = Cl.EnqueueWriteBuffer(cqCommandQueue, objIdMem, Bool.True, IntPtr.Zero, (IntPtr)(clObjArray.Count * 8),
clObjArray.ToArray(), 0, null,
out eve);
CheckErr(error, "EnqBuffer");
error = Cl.EnqueueWriteBuffer(cqCommandQueue, sharedIdxMem, Bool.True, IntPtr.Zero, (IntPtr)(4),
0, 0, null,
out eve);
CheckErr(error, "EnqBuffer");
/* error = Cl.EnqueueWriteBuffer(cqCommandQueue, freeCellMem, Bool.True, IntPtr.Zero,(IntPtr)(freeCellList.Length * Marshal.SizeOf(typeof(bool))),
freeCellList, 0, null,
out eve);
CheckErr(error, "EnqBuffer");*/
error = Cl.Finish(cqCommandQueue);
CheckErr(error, "Cl.Finish");
}
private void CreateCollsionTuples()
{
ErrorCode error;
Event eve;
IntPtr agentPtrSize = (IntPtr)0;
var ptrSize = (IntPtr)Marshal.SizeOf(typeof(Mem));
int globalWorkSize = constants.numThreadsPerBlock * constants.numBlocks;
int localWorkSize = constants.numThreadsPerBlock;
IntPtr[] workGroupSizePtr = new IntPtr[] { (IntPtr)globalWorkSize };
IntPtr[] localWorkGroupSizePtr = new IntPtr[] { (IntPtr)localWorkSize };
Event clevent;
collisionTuples = Cl.CreateBuffer(cxGPUContext, MemFlags.ReadWrite, (IntPtr)( constants.numTotalElements * 8 * Marshal.SizeOf(typeof(CollisionTupel))),
out error);
CheckErr(error, "Createbuffer");
error = Cl.EnqueueWriteBuffer(cqCommandQueue, sharedIdxMem, Bool.True, IntPtr.Zero, (IntPtr)(4),
0, 0, null, out eve);
Console.WriteLine(" Num Elements {0} total {1}", sharedIdx[0], constants.numTotalElements);
CheckErr(error, "EnqBuffer");
error = Cl.SetKernelArg(ckCreateCollsionTuples, 0, ptrSize, cellData);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckCreateCollsionTuples, 1, ptrSize, collisionList);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckCreateCollsionTuples, 2, ptrSize, collisionTuples);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckCreateCollsionTuples, 3, ptrSize, sharedIdxMem);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckCreateCollsionTuples, 4, (IntPtr)(Marshal.SizeOf(typeof(CollisionConstants))), constants);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.SetKernelArg(ckCreateCollsionTuples, 5, (IntPtr)4, sharedIdx[0]);
CheckErr(error, "Cl.SetKernelArg");
error = Cl.EnqueueNDRangeKernel(cqCommandQueue, ckCreateCollsionTuples, 1, null, workGroupSizePtr, localWorkGroupSizePtr, 0, null, out clevent);
CheckErr(error, "Cl.EnqueueNDRangeKernel");
error = Cl.Finish(cqCommandQueue);
CheckErr(error, "Cl.Finish");
error = Cl.EnqueueReadBuffer(cqCommandQueue, sharedIdxMem, Bool.True, IntPtr.Zero, (IntPtr)(4),
sharedIdx, 0, null, out eve);
CheckErr(error, "Cl.EnqueueReadBuffer");
error = Cl.EnqueueReadBuffer(cqCommandQueue, collisionTuples, Bool.True, IntPtr.Zero, (IntPtr)(sharedIdx[0] * Marshal.SizeOf(typeof(CollisionTupel))),
tupelList, 0, null, out eve);
CheckErr(error, "Cl.EnqueueReadBuffer");
for (int i = 0; i < readCellIdList.Count(); i++)
{
if (cellMap.ContainsKey(readCellIdList[i]))
{
cellPosList[i] = cellMap[readCellIdList[i]];
}
else
{
cellPosList[i] = new CollisionCell();
}
}
if (DEBUG)
{
DebugHelper.PrintCollisionTuples(tupelList, readCellIdList, readCellData, cellPosList, "Collision tuples to check", sharedIdx[0]);
}
error = Cl.ReleaseMemObject(collisionList);
CheckErr(error, "Cl.ReleaseMem");
}