extern internal static ComputeErrorCode GetProgramBuildInfo(
CLProgramHandle program,
CLDeviceHandle device,
ComputeProgramBuildInfo param_name,
IntPtr param_value_size,
IntPtr param_value,
out IntPtr param_value_size_ret);
extern internal static ComputeErrorCode GetKernelWorkGroupInfo(
CLKernelHandle kernel,
CLDeviceHandle device,
ComputeKernelWorkGroupInfo param_name,
IntPtr param_value_size,
IntPtr param_value,
out IntPtr param_value_size_ret);
internal OpenCLDevice(OpenCLPlatform platform, CLDeviceHandle handle)
{
Handle = handle;
SetID(Handle.Value);
this.Platform = platform;
}
public GPURadixSort(
ComputeCommandQueue commandQue,
ComputeContext context,
ComputeDevice device
)
{
gpuConstants = new GPUConstants();
gpuConstants.L = radix_BitsL;
gpuConstants.numGroupsPerBlock = NumGroupsPerBlock;
gpuConstants.R = R;
gpuConstants.numThreadsPerGroup = numThreadsPerBlock / NumGroupsPerBlock;
gpuConstants.numThreadsPerBlock = numThreadsPerBlock;
gpuConstants.numBlocks = numBlocks;
gpuConstants.numRadices = num_Radices;
gpuConstants.numRadicesPerBlock = num_Radices / numBlocks;
gpuConstants.bitMask = BIT_MASK_START;
counters.Initialize();
ComputeErrorCode error;
cxGPUContext = context.Handle;
cqCommandQueue = commandQue.Handle;
_device = device.Handle;
//Create a command queue, where all of the commands for execution will be added
/*cqCommandQueue = CL10.CreateCommandQueue(cxGPUContext, _device, (CommandQueueProperties)0, out error);
* CheckErr(error, "CL10.CreateCommandQueue");*/
string programSource = System.IO.File.ReadAllText(programPath);
IntPtr[] progSize = new IntPtr[] { (IntPtr)programSource.Length };
string flags = "-cl-fast-relaxed-math";
ComputeProgram prog = new ComputeProgram(context, programSource);
prog.Build(new List <ComputeDevice>()
{
device
}, flags, null, IntPtr.Zero);
if (prog.GetBuildStatus(device) != ComputeProgramBuildStatus.Success)
{
Debug.WriteLine(prog.GetBuildLog(device));
throw new ArgumentException("UNABLE to build programm");
}
// ComputeProgram clProgramRadix = CL10.CreateProgramWithSource(cxGPUContext, 1, new[] { programSource },progSize,
// out error);
CLProgramHandle clProgramRadix = prog.Handle;
ckSetupAndCount = CL10.CreateKernel(clProgramRadix, "SetupAndCount", out error);
CheckErr(error, "CL10.CreateKernel");
ckSumIt = CL10.CreateKernel(clProgramRadix, "SumIt", out error);
CheckErr(error, "CL10.CreateKernel");
ckReorderingKeysOnly = CL10.CreateKernel(clProgramRadix, "ReorderingKeysOnly", out error);
CheckErr(error, "CL10.CreateKernel");
ckReorderingKeyValue = CL10.CreateKernel(clProgramRadix, "ReorderingKeyValue", out error);
CheckErr(error, "CL10.CreateKernel");
}
internal ComputeDevice(ComputePlatform platform, CLDeviceHandle handle)
{
Handle = handle;
SetID(Handle.Value);
addressBits = GetInfo <uint>(ComputeDeviceInfo.AddressBits);
available = GetBoolInfo(ComputeDeviceInfo.Available);
compilerAvailable = GetBoolInfo(ComputeDeviceInfo.CompilerAvailable);
driverVersion = GetStringInfo(ComputeDeviceInfo.DriverVersion);
endianLittle = GetBoolInfo(ComputeDeviceInfo.EndianLittle);
errorCorrectionSupport = GetBoolInfo(ComputeDeviceInfo.ErrorCorrectionSupport);
executionCapabilities = (ComputeDeviceExecutionCapabilities)GetInfo <long>(ComputeDeviceInfo.ExecutionCapabilities);
string extensionString = GetStringInfo(ComputeDeviceInfo.Extensions);
extensions = new ReadOnlyCollection <string>(extensionString.Split(new char[] { ' ' }, StringSplitOptions.RemoveEmptyEntries));
globalMemoryCachelineSize = GetInfo <uint>(ComputeDeviceInfo.GlobalMemoryCachelineSize);
globalMemoryCacheSize = (long)GetInfo <ulong>(ComputeDeviceInfo.GlobalMemoryCacheSize);
globalMemoryCacheType = (ComputeDeviceMemoryCacheType)GetInfo <long>(ComputeDeviceInfo.GlobalMemoryCacheType);
globalMemorySize = (long)GetInfo <ulong>(ComputeDeviceInfo.GlobalMemorySize);
image2DMaxHeight = (long)GetInfo <IntPtr>(ComputeDeviceInfo.Image2DMaxHeight);
image2DMaxWidth = (long)GetInfo <IntPtr>(ComputeDeviceInfo.Image2DMaxWidth);
image3DMaxDepth = (long)GetInfo <IntPtr>(ComputeDeviceInfo.Image3DMaxDepth);
image3DMaxHeight = (long)GetInfo <IntPtr>(ComputeDeviceInfo.Image3DMaxHeight);
image3DMaxWidth = (long)GetInfo <IntPtr>(ComputeDeviceInfo.Image3DMaxWidth);
imageSupport = GetBoolInfo(ComputeDeviceInfo.ImageSupport);
localMemorySize = (long)GetInfo <ulong>(ComputeDeviceInfo.LocalMemorySize);
localMemoryType = (ComputeDeviceLocalMemoryType)GetInfo <long>(ComputeDeviceInfo.LocalMemoryType);
maxClockFrequency = GetInfo <uint>(ComputeDeviceInfo.MaxClockFrequency);
maxComputeUnits = GetInfo <uint>(ComputeDeviceInfo.MaxComputeUnits);
maxConstantArguments = GetInfo <uint>(ComputeDeviceInfo.MaxConstantArguments);
maxConstantBufferSize = (long)GetInfo <ulong>(ComputeDeviceInfo.MaxConstantBufferSize);
maxMemAllocSize = (long)GetInfo <ulong>(ComputeDeviceInfo.MaxMemoryAllocationSize);
maxParameterSize = (long)GetInfo <IntPtr>(ComputeDeviceInfo.MaxParameterSize);
maxReadImageArgs = GetInfo <uint>(ComputeDeviceInfo.MaxReadImageArguments);
maxSamplers = GetInfo <uint>(ComputeDeviceInfo.MaxSamplers);
maxWorkGroupSize = (long)GetInfo <IntPtr>(ComputeDeviceInfo.MaxWorkGroupSize);
maxWorkItemDimensions = GetInfo <uint>(ComputeDeviceInfo.MaxWorkItemDimensions);
maxWorkItemSizes = new ReadOnlyCollection <long>(ComputeTools.ConvertArray(GetArrayInfo <CLDeviceHandle, ComputeDeviceInfo, IntPtr>(Handle, ComputeDeviceInfo.MaxWorkItemSizes, CL10.GetDeviceInfo)));
maxWriteImageArgs = GetInfo <uint>(ComputeDeviceInfo.MaxWriteImageArguments);
memBaseAddrAlign = GetInfo <uint>(ComputeDeviceInfo.MemoryBaseAddressAlignment);
minDataTypeAlignSize = GetInfo <uint>(ComputeDeviceInfo.MinDataTypeAlignmentSize);
name = GetStringInfo(ComputeDeviceInfo.Name);
this.platform = platform;
preferredVectorWidthChar = GetInfo <uint>(ComputeDeviceInfo.PreferredVectorWidthChar);
preferredVectorWidthFloat = GetInfo <uint>(ComputeDeviceInfo.PreferredVectorWidthFloat);
preferredVectorWidthInt = GetInfo <uint>(ComputeDeviceInfo.PreferredVectorWidthInt);
preferredVectorWidthLong = GetInfo <uint>(ComputeDeviceInfo.PreferredVectorWidthLong);
preferredVectorWidthShort = GetInfo <uint>(ComputeDeviceInfo.PreferredVectorWidthShort);
profile = GetStringInfo(ComputeDeviceInfo.Profile);
profilingTimerResolution = (long)GetInfo <IntPtr>(ComputeDeviceInfo.ProfilingTimerResolution);
queueProperties = (ComputeCommandQueueFlags)GetInfo <long>(ComputeDeviceInfo.CommandQueueProperties);
singleCapabilities = (ComputeDeviceSingleCapabilities)GetInfo <long>(ComputeDeviceInfo.SingleFPConfig);
type = (ComputeDeviceTypes)GetInfo <long>(ComputeDeviceInfo.Type);
vendor = GetStringInfo(ComputeDeviceInfo.Vendor);
vendorId = GetInfo <uint>(ComputeDeviceInfo.VendorId);
version = GetStringInfo(ComputeDeviceInfo.Version);
}
/// <summary>
/// Gets a read-only collection of available devices on the platform.
/// </summary>
/// <returns> A read-only collection of the available devices on the platform. </returns>
/// <remarks> This method resets the <c>ComputePlatform.Devices</c>. This is useful if one or more of them become unavailable (<c>ComputeDevice.Available</c> is <c>false</c>) after a device and command queues that use the device have been created and commands have been queued to them. Further calls will trigger an <c>OutOfResourcesComputeException</c> until this method is executed. You will also need to recreate any <see cref="ComputeResource"/> that was created on the no longer available device. </remarks>
public ReadOnlyCollection <IComputeDevice> QueryDevices()
{
CL10.GetDeviceIDsWrapper(Handle, ComputeDeviceTypes.All, 0, null, out int handlesLength);
var handles = new CLDeviceHandle[handlesLength];
CL10.GetDeviceIDsWrapper(Handle, ComputeDeviceTypes.All, handlesLength, handles, out handlesLength);
var devices = new ComputeDevice[handlesLength];
for (int i = 0; i < handlesLength; i++)
{
devices[i] = new ComputeDevice(this, handles[i]);
}
Devices = new ReadOnlyCollection <IComputeDevice>(devices);
return(Devices);
}
///<summary>
///<See cref = "ComputeDevice"/> s on the <see cref = "ComputePlatform"/>.
///</summary>
///<returns> A read-only collection of the available <see cref = "ComputeDevice"/> s on the <see cref = "ComputePlatform"/>. </returns>
///<c> ComputeDevice.Available </c> is <c> false </c> </remarks> This method resets the <c> ComputePlatform.Devices </c> after see <cref = "ComputeContext"/> and <see cref = "ComputeCommandQueue"/> s that use the <see cref = "ComputeDevice"/> have been created and commands have been queued to them. Further calls will trigger an <c> cref = "ComputeDevice"/>. </c> </c> After this method is executed. remarks>
public ReadOnlyCollection<ComputeDevice> QueryDevices()
{
int handlesLength = 0;
ComputeErrorCode error = CL12.GetDeviceIDs(Handle, ComputeDeviceTypes.All, 0, null, out handlesLength);
ComputeException.ThrowOnError(error);
CLDeviceHandle[] handles = new CLDeviceHandle[handlesLength];
error = CL12.GetDeviceIDs(Handle, ComputeDeviceTypes.All, handlesLength, handles, out handlesLength);
ComputeException.ThrowOnError(error);
ComputeDevice[] devices = new ComputeDevice[handlesLength];
for (int i = 0; i < handlesLength; i++)
devices[i] = new ComputeDevice(this, handles[i]);
this.devices = new ReadOnlyCollection<ComputeDevice>(devices);
return this.devices;
}
internal static CLDeviceHandle[] ExtractHandles(IList <OpenCLDevice> computeObjects, out int handleCount)
{
if (computeObjects == null || computeObjects.Count == 0)
{
handleCount = 0;
return(null);
}
CLDeviceHandle[] result = new CLDeviceHandle[computeObjects.Count];
int i = 0;
foreach (OpenCLDevice computeObj in computeObjects)
{
result[i] = computeObj.Handle;
i++;
}
handleCount = computeObjects.Count;
return(result);
}
/// <summary>
/// Extract device handles of <see cref="ComputeDevice"/>s
/// </summary>
/// <param name="computeObjects"></param>
/// <param name="handleCount"></param>
/// <returns></returns>
public static CLDeviceHandle[] ExtractHandles(ICollection <ComputeDevice> computeObjects, out int handleCount)
{
if (computeObjects == null || computeObjects.Count == 0)
{
handleCount = 0;
return(null);
}
CLDeviceHandle[] result = new CLDeviceHandle[computeObjects.Count];
int i = 0;
foreach (ComputeDevice computeObj in computeObjects)
{
result[i] = computeObj.Handle;
i++;
}
handleCount = computeObjects.Count;
return(result);
}
/// <summary>
/// Gets a read-only collection of available devices on the platform.
/// </summary>
/// <returns> A read-only collection of the available devices on the platform. </returns>
/// <remarks> This method resets the <c>ComputePlatform.Devices</c>. This is useful if one or more of them become unavailable (<c>ComputeDevice.Available</c> is <c>false</c>) after a device and command queues that use the device have been created and commands have been queued to them. Further calls will trigger an <c>OutOfResourcesComputeException</c> until this method is executed. You will also need to recreate any <see cref="ComputeResource"/> that was created on the no longer available device. </remarks>
public ReadOnlyCollection <IComputeDevice> QueryDevices()
{
var error = CL10.GetDeviceIDs(Handle, ComputeDeviceTypes.All, 0, null, out int handlesLength);
ComputeException.ThrowOnError(error);
var handles = new CLDeviceHandle[handlesLength];
error = CL10.GetDeviceIDs(Handle, ComputeDeviceTypes.All, handlesLength, handles, out handlesLength);
ComputeException.ThrowOnError(error);
var devices = new ComputeDevice[handlesLength];
for (int i = 0; i < handlesLength; i++)
{
devices[i] = new ComputeDevice(this, handles[i]);
}
Devices = new ReadOnlyCollection <IComputeDevice>(devices);
return(Devices);
}
/// <summary>
/// Gets a read-only collection of available <see cref="OpenCLDevice"/>s on the <see cref="OpenCLPlatform"/>.
/// </summary>
/// <returns> A read-only collection of the available <see cref="OpenCLDevice"/>s on the <see cref="OpenCLPlatform"/>. </returns>
/// <remarks> This method resets the <c>OpenCLPlatform.Devices</c>. This is useful if one or more of them become unavailable (<c>OpenCLDevice.Available</c> is <c>false</c>) after a <see cref="OpenCLContext"/> and <see cref="OpenCLCommandQueue"/>s that use the <see cref="OpenCLDevice"/> have been created and commands have been queued to them. Further calls will trigger an <c>OutOfResourcesOpenCLException</c> until this method is executed. You will also need to recreate any <see cref="OpenCLResource"/> that was created on the no longer available <see cref="OpenCLDevice"/>. </remarks>
public ReadOnlyCollection <OpenCLDevice> QueryDevices()
{
int handlesLength = 0;
OpenCLErrorCode error = CL10.GetDeviceIDs(Handle, OpenCLDeviceType.All, 0, null, out handlesLength);
OpenCLException.ThrowOnError(error);
CLDeviceHandle[] handles = new CLDeviceHandle[handlesLength];
error = CL10.GetDeviceIDs(Handle, OpenCLDeviceType.All, handlesLength, handles, out handlesLength);
OpenCLException.ThrowOnError(error);
OpenCLDevice[] devices = new OpenCLDevice[handlesLength];
for (int i = 0; i < handlesLength; i++)
{
devices[i] = new OpenCLDevice(this, handles[i]);
}
this.devices = new ReadOnlyCollection <OpenCLDevice>(devices);
return(this.devices);
}
public ReadOnlyCollection <ComputeDevice> CreateSubDevices(uint num_devices)
{
CLDeviceHandle[] handles = new CLDeviceHandle[num_devices];
cl_device_partition_property[] properties = new cl_device_partition_property[] {
cl_device_partition_property.CL_DEVICE_PARTITION_EQUALLY,
(cl_device_partition_property)(uint)(MaxComputeUnits / num_devices),
0
};
ComputeErrorCode error = CL12.CreateSubDevices(Handle, num_devices, properties, handles, out num_devices);
ComputeException.ThrowOnError(error);
ComputeDevice[] subDevices = new ComputeDevice[num_devices];
for (int i = 0; i < num_devices; i++)
{
subDevices[i] = new ComputeDevice(this.Platform, handles[i]);
}
this.subDevices = subDevices;
return(new ReadOnlyCollection <ComputeDevice> (this.subDevices));
}
extern internal static ComputeErrorCode SetDefaultDeviceCommandQueue(
CLContextHandle context,
CLDeviceHandle device,
CLCommandQueueHandle command_queue);
extern internal static CLCommandQueueHandle CreateCommandQueueWithProperties(
CLContextHandle context,
CLDeviceHandle device,
[MarshalAs(UnmanagedType.LPArray)] ComputeCommandQueueFlags[] properties,
out ComputeErrorCode errcode_ret);
extern internal static CLCommandQueueHandle CreateCommandQueue(
CLContextHandle context,
CLDeviceHandle device,
ComputeCommandQueueFlags properties,
out ComputeErrorCode errcode_ret);