/// <summary>
/// Creates OpenCL environment on this process
/// </summary>
/// <param name="_MpiEnv">MPI environment of this process</param>
public clEnvironment(MPIEnviroment _MpiEnv)
{
MpiEnv = _MpiEnv;
cl_platform_id[] platforms = cl.GetPlatformIDs();
cl_platform_info_return pinfo = cl.GetPlatformInfo(platforms[0]);
string vnd = pinfo.vendor;
switch (vnd)
{
case "NVIDIA Corporation":
_vendor = Vendor.NVIDIA;
break;
case "Advanced Micro Devices, Inc.":
_vendor = Vendor.AMD;
break;
default:
_vendor = Vendor.Other;
break;
}
Console.WriteLine(vnd + " : " + _vendor.ToString());
Console.WriteLine(pinfo.version);
if (_vendor == Vendor.AMD)
{
clVectorSource.source[0] = clVectorSource.source[0].Replace("cl_khr_fp64", "cl_amd_fp64");
clMatrixSource.source[0] = clMatrixSource.source[0].Replace("cl_khr_fp64", "cl_amd_fp64");
Console.WriteLine("Adjusted AMD fp64 extenstion");
}
cl_device_id[] devices = cl.GetDeviceIDs(platforms[0], cl_device_type.CL_DEVICE_TYPE_GPU);
int numDevices = devices.Length;
if (numDevices < _MpiEnv.ProcessesOnMySMP)
{
throw new ApplicationException("not enougth OpenCL devices; There must be at least one OpenCL device for each MPI process;");
}
device = devices[0];
context = cl.CreateContext(platforms[0], devices);
//cl_device_info_return dinfo = cl.GetDeviceInfo(device);
//Console.WriteLine("Max work group size: " + dinfo.max_work_group_size);
//Console.WriteLine("Process " + _MpiEnv.ProcessRankOnSMP + " running on device " + dinfo.name + ", " + dinfo.version);
//Console.WriteLine(dinfo.extensions);
}
public static void Main()
{
const int cnBlockSize = 4;
const int cnBlocks = 3;
IntPtr cnDimension = new IntPtr(cnBlocks * cnBlockSize);
string sProgramSource = @"
__kernel void
vectorAdd(__global const float * a,
__global const float * b,
__global float * c)
{
// Vector element index
int nIndex = get_global_id(0);
c[nIndex] = a[nIndex] + b[nIndex];
}
";
ErrorCode error;
// create OpenCL device & context
cl_context hContext;
unsafe { hContext = CL.CreateContextFromType((ContextProperties *)null, DeviceTypeFlags.DeviceTypeDefault, IntPtr.Zero, IntPtr.Zero, &error); }
// query all devices available to the context
IntPtr nContextDescriptorSize;
CL.GetContextInfo(hContext, ContextInfo.ContextDevices, IntPtr.Zero, IntPtr.Zero, out nContextDescriptorSize);
cl_device_id[] aDevices = new cl_device_id[nContextDescriptorSize.ToInt32()];
unsafe
{
fixed(cl_device_id *ptr = aDevices)
{
IntPtr ret;
CL.GetContextInfo(hContext, ContextInfo.ContextDevices, nContextDescriptorSize, new IntPtr(ptr), out ret);
}
}
// create a command queue for first device the context reported
cl_command_queue hCmdQueue = CL.CreateCommandQueue(hContext, aDevices[0], (CommandQueueFlags)0, out error);
// create & compile program
cl_program hProgram;
unsafe { hProgram = CL.CreateProgramWithSource(hContext, 1, new string[] { sProgramSource }, null, &error); }
CL.BuildProgram(hProgram, 0, (IntPtr[])null, null, IntPtr.Zero, IntPtr.Zero);
// create kernel
cl_kernel hKernel = CL.CreateKernel(hProgram, "vectorAdd", out error);
// allocate host vectors
float[] A = new float[cnDimension.ToInt32()];
float[] B = new float[cnDimension.ToInt32()];
float[] C = new float[cnDimension.ToInt32()];
// initialize host memory
Random rand = new Random();
for (int i = 0; i < A.Length; i++)
{
A[i] = rand.Next() % 256;
B[i] = rand.Next() % 256;
}
// allocate device memory
unsafe
{
fixed(float *pA = A)
fixed(float *pB = B)
fixed(float *pC = C)
{
cl_mem hDeviceMemA, hDeviceMemB, hDeviceMemC;
hDeviceMemA = CL.CreateBuffer(hContext,
MemFlags.MemReadOnly | MemFlags.MemCopyHostPtr,
new IntPtr(cnDimension.ToInt32() * sizeof(float)),
new IntPtr(pA),
out error);
hDeviceMemB = CL.CreateBuffer(hContext,
MemFlags.MemReadOnly | MemFlags.MemCopyHostPtr,
new IntPtr(cnDimension.ToInt32() * sizeof(float)),
new IntPtr(pA),
out error);
hDeviceMemC = CL.CreateBuffer(hContext,
MemFlags.MemWriteOnly,
new IntPtr(cnDimension.ToInt32() * sizeof(float)),
IntPtr.Zero,
out error);
// setup parameter values
CL.SetKernelArg(hKernel, 0, new IntPtr(sizeof(cl_mem)), new IntPtr(&hDeviceMemA));
CL.SetKernelArg(hKernel, 1, new IntPtr(sizeof(cl_mem)), new IntPtr(&hDeviceMemB));
CL.SetKernelArg(hKernel, 2, new IntPtr(sizeof(cl_mem)), new IntPtr(&hDeviceMemC));
// write data from host to device
CL.EnqueueWriteBuffer(hCmdQueue, hDeviceMemA, true, IntPtr.Zero,
new IntPtr(cnDimension.ToInt32() * sizeof(float)),
new IntPtr(pA), 0, null, (IntPtr[])null);
CL.EnqueueWriteBuffer(hCmdQueue, hDeviceMemB, true, IntPtr.Zero,
new IntPtr(cnDimension.ToInt32() * sizeof(float)),
new IntPtr(pB), 0, null, (IntPtr[])null);
// execute kernel
error = (ErrorCode)CL.EnqueueNDRangeKernel(hCmdQueue, hKernel, 1, null, &cnDimension, null, 0, null, null);
if (error != ErrorCode.Success)
{
throw new Exception(error.ToString());
}
// copy results from device back to host
IntPtr event_handle = IntPtr.Zero;
error = (ErrorCode)CL.EnqueueReadBuffer(hCmdQueue, hDeviceMemC, true, IntPtr.Zero,
new IntPtr(cnDimension.ToInt32() * sizeof(float)),
new IntPtr(pC), 0, null, (IntPtr[])null);
if (error != ErrorCode.Success)
{
throw new Exception(error.ToString());
}
CL.Finish(hCmdQueue);
CL.ReleaseMemObject(hDeviceMemA);
CL.ReleaseMemObject(hDeviceMemB);
CL.ReleaseMemObject(hDeviceMemC);
}
}
for (int i = 0; i < A.Length; i++)
{
System.Diagnostics.Trace.WriteLine(String.Format("{0} + {1} = {2}", A[i], B[i], C[i]));
}
}
public static void Main()
{
const int cnBlockSize = 4;
const int cnBlocks = 3;
IntPtr cnDimension = new IntPtr(cnBlocks * cnBlockSize);
string sProgramSource = @"
__kernel void
vectorAdd(__global const float * a,
__global const float * b,
__global float * c)
{
// Vector element index
int nIndex = get_global_id(0);
c[nIndex] = a[nIndex] + b[nIndex];
}
";
ErrorCode error;
// create OpenCL device & context
cl_context hContext;
unsafe { hContext = CL.CreateContextFromType((ContextProperties*)null, DeviceTypeFlags.DeviceTypeDefault, IntPtr.Zero, IntPtr.Zero, &error); }
// query all devices available to the context
IntPtr nContextDescriptorSize;
CL.GetContextInfo(hContext, ContextInfo.ContextDevices, IntPtr.Zero, IntPtr.Zero, out nContextDescriptorSize);
cl_device_id[] aDevices = new cl_device_id[nContextDescriptorSize.ToInt32()];
unsafe
{
fixed (cl_device_id* ptr = aDevices)
{
IntPtr ret;
CL.GetContextInfo(hContext, ContextInfo.ContextDevices, nContextDescriptorSize, new IntPtr(ptr), out ret);
}
}
// create a command queue for first device the context reported
cl_command_queue hCmdQueue = CL.CreateCommandQueue(hContext, aDevices[0], (CommandQueueFlags)0, out error);
// create & compile program
cl_program hProgram;
unsafe { hProgram = CL.CreateProgramWithSource(hContext, 1, new string[] { sProgramSource }, null, &error); }
CL.BuildProgram(hProgram, 0, (IntPtr[])null, null, IntPtr.Zero, IntPtr.Zero);
// create kernel
cl_kernel hKernel = CL.CreateKernel(hProgram, "vectorAdd", out error);
// allocate host vectors
float[] A = new float[cnDimension.ToInt32()];
float[] B = new float[cnDimension.ToInt32()];
float[] C = new float[cnDimension.ToInt32()];
// initialize host memory
Random rand = new Random();
for (int i = 0; i < A.Length; i++)
{
A[i] = rand.Next() % 256;
B[i] = rand.Next() % 256;
}
// allocate device memory
unsafe
{
fixed (float* pA = A)
fixed (float* pB = B)
fixed (float* pC = C)
{
cl_mem hDeviceMemA, hDeviceMemB, hDeviceMemC;
hDeviceMemA = CL.CreateBuffer(hContext,
MemFlags.MemReadOnly | MemFlags.MemCopyHostPtr,
new IntPtr(cnDimension.ToInt32() * sizeof(float)),
new IntPtr(pA),
out error);
hDeviceMemB = CL.CreateBuffer(hContext,
MemFlags.MemReadOnly | MemFlags.MemCopyHostPtr,
new IntPtr(cnDimension.ToInt32() * sizeof(float)),
new IntPtr(pA),
out error);
hDeviceMemC = CL.CreateBuffer(hContext,
MemFlags.MemWriteOnly,
new IntPtr(cnDimension.ToInt32() * sizeof(float)),
IntPtr.Zero,
out error);
// setup parameter values
CL.SetKernelArg(hKernel, 0, new IntPtr(sizeof(cl_mem)), new IntPtr(&hDeviceMemA));
CL.SetKernelArg(hKernel, 1, new IntPtr(sizeof(cl_mem)), new IntPtr(&hDeviceMemB));
CL.SetKernelArg(hKernel, 2, new IntPtr(sizeof(cl_mem)), new IntPtr(&hDeviceMemC));
// write data from host to device
CL.EnqueueWriteBuffer(hCmdQueue, hDeviceMemA, true, IntPtr.Zero,
new IntPtr(cnDimension.ToInt32() * sizeof(float)),
new IntPtr(pA), 0, null, (IntPtr[])null);
CL.EnqueueWriteBuffer(hCmdQueue, hDeviceMemB, true, IntPtr.Zero,
new IntPtr(cnDimension.ToInt32() * sizeof(float)),
new IntPtr(pB), 0, null, (IntPtr[])null);
// execute kernel
error = (ErrorCode)CL.EnqueueNDRangeKernel(hCmdQueue, hKernel, 1, null, &cnDimension, null, 0, null, null);
if (error != ErrorCode.Success)
throw new Exception(error.ToString());
// copy results from device back to host
IntPtr event_handle = IntPtr.Zero;
error = (ErrorCode)CL.EnqueueReadBuffer(hCmdQueue, hDeviceMemC, true, IntPtr.Zero,
new IntPtr(cnDimension.ToInt32() * sizeof(float)),
new IntPtr(pC), 0, null, (IntPtr[])null);
if (error != ErrorCode.Success)
throw new Exception(error.ToString());
CL.Finish(hCmdQueue);
CL.ReleaseMemObject(hDeviceMemA);
CL.ReleaseMemObject(hDeviceMemB);
CL.ReleaseMemObject(hDeviceMemC);
}
}
for (int i = 0; i < A.Length; i++)
{
System.Diagnostics.Trace.WriteLine(String.Format("{0} + {1} = {2}", A[i], B[i], C[i]));
}
}
internal static extern IntPtr clCreateCommandQueue(cl_context context, cl_device_id device, cl_command_queue_properties properties, [MarshalAs(UnmanagedType.I4)] out ErrorCode errcode_ret);
internal static extern ErrorCode clGetKernelWorkGroupInfo(cl_kernel kernel, cl_device_id device, cl_kernel_work_group_info param_name, IntPtr param_value_size, void *param_value, out IntPtr param_value_size_ret);
internal static extern cl_int clGetProgramBuildInfo(cl_program program, cl_device_id device, cl_program_build_info param_name, IntPtr param_value_size, void *param_value, out IntPtr param_value_size_ret);
internal static extern cl_int clGetDeviceInfo(cl_device_id device, cl_device_info param_name, IntPtr param_value_size, void *param_value, out IntPtr param_value_size_ret);
public static ErrorCode CreateSubDevicesEXT(cl_device_id in_device, byte[] properties, cl_uint num_entries, cl_device_id[] out_devices, [Out] cl_uint *num_devices)
{
return(OpenCLAPI.clCreateSubDevicesEXT(in_device, properties, num_entries, out_devices, num_devices));
}
public static ErrorCode RetainDeviceEXT(cl_device_id device)
{
return(OpenCLAPI.clRetainDeviceEXT(device));
}
public static ErrorCode ReleaseDeviceEXT(cl_device_id device) {
return OpenCLAPI.clReleaseDeviceEXT(device);
}
