private static void SimInit()
{
Console.WriteLine("Deserializing class");
CudafyModule km = CudafyModule.TryDeserialize(typeof(Program).Name);
Console.WriteLine("Got: " + km);
var tvc = km == null ? false : km.TryVerifyChecksums();
Console.WriteLine("TVC: " + tvc);
if (km == null || !tvc)
{
Console.WriteLine("Serializing");
km = CudafyTranslator.Cudafy(typeof(Program));
km.Serialize();
}
Console.WriteLine("Requesting device");
_gpu = CudafyHost.GetDevice(eGPUType.Cuda);
if (_gpu == null)
{
_gpu = CudafyHost.GetDevice(eGPUType.OpenCL);
if (_gpu == null)
{
_gpu = CudafyHost.GetDevice(eGPUType.Emulator);
if (_gpu == null)
{
Console.WriteLine("No deivce found!");
return;
}
}
else Console.WriteLine("Got OpenCL Device: " + _gpu.DeviceId);
}
else Console.WriteLine("Got CUDA Device: " + _gpu.DeviceId);
Console.WriteLine("Loading module");
_gpu.LoadModule(km);
}
/// <summary>
/// Removes the specified GPGPU from the cache.
/// </summary>
/// <param name="gpu">The gpu.</param>
/// <returns>True if gpu was removed, else false.</returns>
public static bool RemoveDevice(GPGPU gpu)
{
List <GPGPU> gpus = GPGPUs.Values.Where(v => v == gpu).ToList();
bool removed = gpus.Count > 0;
List <string> names = new List <string>();
for (int i = 0; i < gpus.Count; i++)
{
gpus[i].Dispose();
foreach (var v in GPGPUs)
{
if (v.Value == gpu)
{
names.Add(v.Key);
}
}
}
foreach (var s in names.Distinct())
{
GPGPUs.Remove(s);
}
return(removed);
}
public override bool CanAccessPeer(GPGPU peer)
{
lock (_peerAccessLock)
{
return(peer != this && peer is EmulatedGPU);
}
}
static Bitmap Render(GPGPU gpu, int frameNum)
{
uint[,] deviceImage = gpu.Allocate <uint>(width, height);
float[] pX1_gpu = gpu.CopyToDevice <float>(pX1);
float[] pY1_gpu = gpu.CopyToDevice <float>(pY1);
float[] pZ1_gpu = gpu.CopyToDevice <float>(pZ1);
float[] colorPosition_gpu = gpu.CopyToDevice <float>(colorPosition);
float[] currentTime_gpu = gpu.CopyToDevice <float>(currentTime);
dim3 threadsPerBlock = new dim3(8, 8);
dim3 numBlocks = new dim3(width / threadsPerBlock.x, height / threadsPerBlock.y);
gpu.Launch(numBlocks, threadsPerBlock).renderKernel(deviceImage, pX1_gpu, pY1_gpu, pZ1_gpu, colorPosition_gpu, currentTime_gpu);
uint[,] finalImage = new uint[width, height];
gpu.CopyFromDevice <uint>(deviceImage, finalImage);
gpu.Free(deviceImage);
gpu.Free(pX1_gpu);
gpu.Free(pY1_gpu);
gpu.Free(pZ1_gpu);
gpu.Free(colorPosition_gpu);
gpu.Free(currentTime_gpu);
GCHandle pixels = GCHandle.Alloc(finalImage, GCHandleType.Pinned);
Bitmap bmp = new Bitmap(width, height, width * sizeof(int), PixelFormat.Format32bppRgb, pixels.AddrOfPinnedObject());
bmp.Save("spring" + frameNum + ".png");
pixels.Free();
return(bmp);
}
public void InitGPU()
{
// Work around for bug in Cudafy trying to find the path..
var os64Bit = Environment.Is64BitOperatingSystem;
if (os64Bit)
{
var dir = Environment.GetEnvironmentVariable("ProgramFiles");
Environment.SetEnvironmentVariable("ProgramFiles", "C:\\Program Files\\");
dir = Environment.GetEnvironmentVariable("ProgramFiles");
}
if (Gpu == null)
{
Gpu = CudafyHost.GetDevice(_gpuType, 0);
//Blas = GPGPUBLAS.Create(Gpu);
if (_gpuType == eGPUType.Cuda)
{
Blas = new SharpBLAS(Gpu);
Rand = GPGPURAND.Create(Gpu, curandRngType.CURAND_RNG_PSEUDO_DEFAULT);
Rand.SetPseudoRandomGeneratorSeed((ulong)RandomHelpers.Next(9999));
}
CudafyTranslator.GenerateDebug = true;
Debug.WriteLine("CUDA workdir = " + CudafyTranslator.WorkingDirectory);
Console.WriteLine("Recompile module");
CudafyTranslator.Language = eLanguage.Cuda;
var km = CudafyTranslator.Cudafy(eArchitecture.sm_30);
km = CudafyTranslator.Cudafy();
km.Serialize();
Gpu.LoadModule(km);
}
}
static void Main(string[] args)
{
CudafyModule km = CudafyTranslator.Cudafy();
GPGPU gpu = CudafyHost.GetDevice(CudafyModes.Target, CudafyModes.DeviceId);
gpu.LoadModule(km);
int numFrames = numberOfSeconds * framesPerSecond;
InitializeParticles();
File.WriteAllText("length.txt", numFrames.ToString());
for (int i = 0; i < numFrames; i++)
{
DateTime frameStart = DateTime.Now;
Simulate(gpu);
Bitmap frame = Render(gpu, i);
TimeSpan frameTime = DateTime.Now - frameStart;
Console.WriteLine("Frame " + i + " complete. Time: " + frameTime.TotalMilliseconds + "ms");
}
}
public void SetUp()
{
//CudafyModes.Architecture = eArchitecture.sm_30;
_gpu = CudafyHost.GetDevice(eArchitecture.sm_30, CudafyModes.DeviceId);
Assert.IsFalse(_gpu is OpenCLDevice, "OpenCL devices are not supported.");
_cm = CudafyModule.TryDeserialize();
if (_cm == null || !_cm.TryVerifyChecksums())
{
_cm = CudafyTranslator.Cudafy(eArchitecture.sm_30);
Console.WriteLine(_cm.CompilerOutput);
_cm.TrySerialize();
}
_gpu.LoadModule(_cm);
inputIntArray = new int[] { 0x17, 0x01, 0x7f, 0xd1, 0xfe, 0x23, 0x2c, 0xa0, 0x00, 0xcf, 0xaa, 0x7a, 0x35, 0xf4, 0x04, 0xbc,
0xe9, 0x6d, 0xb2, 0x55, 0xb0, 0xc8, 0x10, 0x49, 0x76, 0x17, 0x92, 0xab, 0xf3, 0xf2, 0xab, 0xcb }; // arbitrary values
d_inputIntArray = _gpu.CopyToDevice(inputIntArray);
d_outputIntArray = _gpu.Allocate <int>(WARP_SIZE);
gpuIntResult = new int[WARP_SIZE];
cpuIntResult = new int[WARP_SIZE];
inputFloatArray = new float[] { 1.7f, -37.03f, 2147.6436f, -0.1f, 7.7f, 99.99f, -809.142f, -0.1115f,
1.0f, 2.0f, 3.0f, 5.0f, 7.5f, 0.1001f, 11.119f, -9.0f,
7749.9847f, -860249.118843f, 0.0f, -2727745.586215f, 12.0f, -11.0f, 77.77f, 22.0f,
377.1112f, -377.1112f, 0.12345f, -0.12345f, 0.11111f, -0.11111f, 700000f, -14f }; // arbitrary values
d_inputFloatArray = _gpu.CopyToDevice(inputFloatArray);
d_outputFloatArray = _gpu.Allocate <float>(WARP_SIZE);
gpuFloatResult = new float[WARP_SIZE];
cpuFloatResult = new float[WARP_SIZE];
}
public void InitGPU()
{
// Work around for bug in Cudafy trying to find the path..
var os64Bit = Environment.Is64BitOperatingSystem;
if (os64Bit)
{
var dir = Environment.GetEnvironmentVariable("ProgramFiles");
Environment.SetEnvironmentVariable("ProgramFiles", "C:\\Program Files\\");
dir = Environment.GetEnvironmentVariable("ProgramFiles");
}
if (Gpu == null)
{
Gpu = CudafyHost.GetDevice(_gpuType, 0);
//Blas = GPGPUBLAS.Create(Gpu);
if (_gpuType == eGPUType.Cuda)
{
Blas = new SharpBLAS(Gpu);
Rand = GPGPURAND.Create(Gpu, curandRngType.CURAND_RNG_PSEUDO_DEFAULT);
Rand.SetPseudoRandomGeneratorSeed((ulong)RandomHelpers.Next(9999));
}
CudafyTranslator.GenerateDebug = true;
Debug.WriteLine("CUDA workdir = " + CudafyTranslator.WorkingDirectory);
Console.WriteLine("Recompile module");
CudafyTranslator.Language = eLanguage.Cuda;
var km = CudafyTranslator.Cudafy(eArchitecture.sm_30);
km = CudafyTranslator.Cudafy();
km.Serialize();
Gpu.LoadModule(km);
}
}
public static float[] CallGPU()
{
CudafyModes.Target = eGPUType.OpenCL;
CudafyModes.DeviceId = 0;
CudafyTranslator.Language = eLanguage.OpenCL;
CudafyModule km = CudafyTranslator.Cudafy(ePlatform.Auto, eArchitecture.OpenCL, typeof(GPU));
GPGPU gpu = CudafyHost.GetDevice(eGPUType.OpenCL, 0);
gpu.LoadModule(km);
km.Serialize();
float[] input = Utils.GenerateRandomVector();
float[,,] NN = Utils.GenerateRandomMatrix().AsSingleDimension();
float[] output = new float[Utils.N];
Stopwatch gpuSW = new Stopwatch();
gpuSW.Start();
float[] dev_output = gpu.Allocate <float>(output);
float[] dev_input = gpu.CopyToDevice(input);
float[,,] dev_NN = gpu.CopyToDevice(NN);
gpu.Launch(Utils.GRID_SIZE, Utils.BLOCK_SIZE).CalculateNeuralNetwork(dev_input, dev_NN, dev_output);
gpu.CopyFromDevice(dev_output, output);
gpu.FreeAll();
gpuSW.Stop();
Console.WriteLine("GPU: " + gpuSW.ElapsedMilliseconds);
return(output);
}
public void SetUp()
{
//CudafyModes.Architecture = eArchitecture.sm_30;
_gpu = CudafyHost.GetDevice(eArchitecture.sm_30, CudafyModes.DeviceId);
Assert.IsFalse(_gpu is OpenCLDevice, "OpenCL devices are not supported.");
_cm = CudafyModule.TryDeserialize();
if (_cm == null || !_cm.TryVerifyChecksums())
{
_cm = CudafyTranslator.Cudafy(eArchitecture.sm_30);
Console.WriteLine(_cm.CompilerOutput);
_cm.TrySerialize();
}
_gpu.LoadModule(_cm);
inputIntArray = new int[] { 0x17, 0x01, 0x7f, 0xd1, 0xfe, 0x23, 0x2c, 0xa0, 0x00, 0xcf, 0xaa, 0x7a, 0x35, 0xf4, 0x04, 0xbc,
0xe9, 0x6d, 0xb2, 0x55, 0xb0, 0xc8, 0x10, 0x49, 0x76, 0x17, 0x92, 0xab, 0xf3, 0xf2, 0xab, 0xcb}; // arbitrary values
d_inputIntArray = _gpu.CopyToDevice(inputIntArray);
d_outputIntArray = _gpu.Allocate<int>(WARP_SIZE);
gpuIntResult = new int[WARP_SIZE];
cpuIntResult = new int[WARP_SIZE];
inputFloatArray = new float[] { 1.7f, -37.03f, 2147.6436f, -0.1f, 7.7f, 99.99f, -809.142f, -0.1115f,
1.0f, 2.0f, 3.0f, 5.0f, 7.5f, 0.1001f, 11.119f, -9.0f,
7749.9847f, -860249.118843f, 0.0f, -2727745.586215f, 12.0f, -11.0f, 77.77f, 22.0f,
377.1112f, -377.1112f, 0.12345f, -0.12345f, 0.11111f, -0.11111f, 700000f, -14f}; // arbitrary values
d_inputFloatArray = _gpu.CopyToDevice(inputFloatArray);
d_outputFloatArray = _gpu.Allocate<float>(WARP_SIZE);
gpuFloatResult = new float[WARP_SIZE];
cpuFloatResult = new float[WARP_SIZE];
}
public static void MyFirstBlasEmulatorTest()
{
Console.WriteLine("MyTest()");
// Get GPU device
CudafyModes.Target = eGPUType.Emulator;
GPGPU gpu = CudafyHost.GetDevice(CudafyModes.Target);
// Create GPGPUBLAS (CUBLAS Wrapper)
using (GPGPUBLAS blas = GPGPUBLAS.Create(gpu))
{
const int N = 100;
float[] a = new float[N];
float[] b = new float[N];
float[] c = new float[N];
float alpha = -1;
float beta = 0;
float[] device_a = gpu.CopyToDevice(a);
float[] device_b = gpu.CopyToDevice(b);
float[] device_c = gpu.CopyToDevice(c);
int m = 10;
int n = 10;
int k = 10;
cublasOperation Op = cublasOperation.N;
blas.GEMM(m, k, n, alpha, device_a, device_b, beta, device_c, Op);
gpu.CopyFromDevice <float>(device_c, c);
}
}
public static void Execute()
{
CudafyModule km = CudafyTranslator.Cudafy(typeof(ParamsStruct), typeof(ImpliedVolatile));
_gpu = CudafyHost.GetDevice(CudafyModes.Target);
_gpu.LoadModule(km);
ParamsStruct[] host_par = new ParamsStruct[1];
ParamsStruct[] result = new ParamsStruct[1];
host_par[0].OP = 96.95;
host_par[0].Price = 1332.24;
host_par[0].Strike = 1235;
host_par[0].TD = 31;
host_par[0].R = 0.0001355;
host_par[0].Q = 0.0166;
host_par[0].N = 100;// 1000;
host_par[0].kind = 1;
ParamsStruct[] dev_par = _gpu.CopyToDevice(host_par);
float[] PA = _gpu.Allocate<float>(1001);
_gpu.Launch(1,1, "impliedVolatile", dev_par, PA);
_gpu.CopyFromDevice(dev_par, 0, result, 0, 1);
Console.WriteLine("I={0}, B={1}", result[0].i, result[0].B);
//Console.ReadKey();
}
public void SetUp()
{
_gpu = CudafyHost.CreateDevice(CudafyModes.Target);
_hostInput = new double[N * BATCH];
_hostInputCplx = new ComplexD[N * BATCH];
_hostOutput = new double[N * BATCH];
_hostOutputCplx = new ComplexD[N * BATCH];
_devInput = _gpu.Allocate(_hostInput);
_devInputCplx = _gpu.Allocate(_hostInputCplx);
_devInter = _gpu.Allocate<double>(N * 2 * BATCH);
_devInterCplx = _gpu.Allocate<ComplexD>(N * BATCH);
_devOutput = _gpu.Allocate(_hostOutput);
_devOutputCplx = _gpu.Allocate(_hostOutputCplx);
_fft = GPGPUFFT.Create(_gpu);
for (int b = 0; b < BATCH; b++)
{
for (int i = 0; i < N; i++)
{
ComplexD cf = new ComplexD();
cf.x = (double)((10.0F * Math.Sin(100 * 2 * Math.PI * i / N * Math.PI / 180)));
cf.y = (double)((10.0F * Math.Sin(200 * 2 * Math.PI * i / N * Math.PI / 180)));
_hostInput[i + b * N] = cf.x;
_hostInputCplx[i + b * N] = cf;
}
}
}
public Layer(GPUModule gpuModule, Layer previousLayer = null,int size = 0, string id = "", int miniBatchSize = Int32.MinValue)
{
if (previousLayer != null) MinibatchSize = previousLayer.MinibatchSize;
if (miniBatchSize != Int32.MinValue) MinibatchSize = miniBatchSize;
LayerIndex = IdCounter++;
Id = id;
if (String.IsNullOrEmpty(Id))
{
Id = "ID" + LayerIndex.ToString().PadLeft(2, '0');
}
_gpuModule = gpuModule;
_gpu = _gpuModule.Gpu;
PreviousLayer = previousLayer;
if (size != 0)
{
this.Size = size;
AddArray(ArrayName.Outputs, MinibatchSize, this.Size);
}
if ((previousLayer != null) && (size > 0))
{
AddArray(ArrayName.Gradients, MinibatchSize, size);
}
}
public static void primaGPU()
{
CudafyModule modul_kernel = CudafyTranslator.Cudafy();
GPGPU vga = CudafyHost.GetDevice(CudafyModes.Target, CudafyModes.DeviceId);
vga.LoadModule(modul_kernel);
Stopwatch waktu = new Stopwatch();
waktu.Start();
int[] list_cpu = new int[KONSTANTA_THREAD];
int[] list_cpy = new int[KONSTANTA_THREAD];
int[] list = vga.Allocate <int>(KONSTANTA_THREAD);
vga.Launch(KONSTANTA_THREAD, 1).ModulAtomic(list);
vga.CopyFromDevice(list, list_cpy);
vga.FreeAll();
int index = 0;
for (int z = 0; z < list_cpy.Length; z++)
{
if (list_cpy[z] != -1)
{
list_cpu[index] = list_cpy[z];
//Console.WriteLine(list_cpu[index]);
index++;
}
}
waktu.Stop();
TimeSpan ts = waktu.Elapsed;
String total = ts.Seconds.ToString();
Console.WriteLine("Total GPU ------ {0} detik> ", total);
}
public static void cudaTranspose(ref MathNet.Numerics.LinearAlgebra.Double.DenseMatrix dm)
{
GPGPU gpu = CudafyHost.GetDevice(eGPUType.Cuda);
GPGPUBLAS blas = GPGPUBLAS.Create(gpu);
int cols = dm.ColumnCount, rows = dm.RowCount;
int restRows = rows - cols;
//double[] a = dm.Storage.ToColumnMajorArray();
double[] a = dm.SubMatrix(0, cols, 0, cols).Storage.ToColumnMajorArray();
double[] b = dm.SubMatrix(cols, restRows, 0, cols).Storage.ToColumnMajorArray();
dm = null;
double[] a_d = gpu.CopyToDevice <double>(a);
a = null;
double[] c_d = gpu.Allocate <double>(cols * cols);
double[] x_d = gpu.CopyToDevice <double>(new double[] { 1 });
blas.GEMV(cols, cols, 1, c_d, x_d, 0, x_d, Cudafy.Maths.BLAS.Types.cublasOperation.T);
a = new double[cols * rows];
gpu.CopyFromDevice <double>(c_d, 0, a, 0, cols * cols);
gpu.FreeAll();
a_d = gpu.CopyToDevice <double>(b);
b = null;
c_d = gpu.Allocate <double>(restRows * cols);
x_d = gpu.CopyToDevice <double>(new double[] { 1 });
blas.GEMV(restRows, cols, 1, c_d, x_d, 0, x_d, Cudafy.Maths.BLAS.Types.cublasOperation.T);
gpu.CopyFromDevice <double>(c_d, 0, a, cols * cols, restRows * cols);
gpu.FreeAll();
dm = new MathNet.Numerics.LinearAlgebra.Double.DenseMatrix(cols, rows, a);
}
public static void eksekusi()
{
CudafyModule kernel_modul = CudafyTranslator.Cudafy();
GPGPU vga = CudafyHost.GetDevice(CudafyModes.Target, CudafyModes.DeviceId);
vga.LoadModule(kernel_modul);
Stopwatch waktu = new Stopwatch();
waktu.Start();
int[] array_vga = vga.Allocate <int>(KONSTANTA_THREAD);
int[] array_hasil = new int[KONSTANTA_THREAD];
//long[] matriks1 = vga.Allocate<long>(KONSTANTA_THREAD);
//long[] matriks2 = vga.Allocate<long>(KONSTANTA_THREAD);//new int[KONSTANTA_THREAD];
//long[] matriks3 = vga.Allocate<long>(KONSTANTA_THREAD); //[KONSTANTA_THREAD];
vga.Launch(KONSTANTA_THREAD, 1).fungsiAtomic(array_vga);
vga.CopyFromDevice(array_vga, array_hasil);
vga.FreeAll();
//for(int z = 0; z < array_hasil.Length; z++)
//{
// Console.WriteLine("Hasil Ekstrak----" + array_hasil[z]);
//}
vga.FreeAll();
waktu.Stop();
TimeSpan ts = waktu.Elapsed;
String total = ts.Milliseconds.ToString();
Console.WriteLine("Total VGA ------ > " + total);
}
public static bool TestGpuDoublePrecision(int DeviceId)
{
if (DeviceId > CudafyHost.GetDeviceCount(eGPUType.OpenCL))
{
return(false);
}
try
{
CudafyModes.Target = eGPUType.OpenCL;
CudafyTranslator.Language = eLanguage.OpenCL;
CudafyModule km = CudafyTranslator.Cudafy();
GPGPU gpu = CudafyHost.GetDevice(eGPUType.OpenCL, DeviceId);
gpu.LoadModule(km);
double c;
double[] dev_c = gpu.Allocate <double>();
gpu.Launch().add_double(2.5d, 7.5d, dev_c);
gpu.CopyFromDevice(dev_c, out c);
gpu.Free(dev_c);
return(c == 10.0d);
}
catch
{ return(false); }
}
public static int MA(int[,] A, int[,] B, int[,] C, GPGPU gpu, int maxTheadBlockSize, int Size)
{
// allocate the memory on the GPU
int[,] GPU_A = gpu.Allocate<int>(A);
int[,] GPU_B = gpu.Allocate<int>(B);
int[,] GPU_C = gpu.Allocate<int>(C);
// copy the arrays 'a' and 'b' to the GPU
gpu.CopyToDevice(A, GPU_A);
gpu.CopyToDevice(B, GPU_B);
dim3 threadsPerBlock;
// find the number of threads and blocks
if (Size < maxTheadBlockSize)
{
threadsPerBlock = new dim3(Size, Size);
}
else
{
threadsPerBlock = new dim3(maxTheadBlockSize, maxTheadBlockSize);
}
dim3 block = new dim3(Size, Size);
// launch GPU_MA
gpu.Launch(block, threadsPerBlock, "GPU_MA", GPU_A, GPU_B, GPU_C, Size);
// copy the array 'c' back from the GPU to the CPU
gpu.CopyFromDevice(GPU_C, C);
gpu.Free(GPU_A);
gpu.Free(GPU_B);
gpu.Free(GPU_C);
return 1;
}
public void Test_TwoThreadCopy()
{
_gpu = CudafyHost.GetDevice(eGPUType.Cuda);
_gpuuintBufferIn3 = _gpu.Allocate(_uintBufferIn1);
_gpuuintBufferIn4 = _gpu.Allocate(_uintBufferIn1);
_gpu.EnableMultithreading();
bool j1 = false;
bool j2 = false;
for (int i = 0; i < 10; i++)
{
Console.WriteLine(i);
SetInputs();
ClearOutputs();
Thread t1 = new Thread(Test_TwoThreadCopy_Thread1);
Thread t2 = new Thread(Test_TwoThreadCopy_Thread2);
t1.Start();
t2.Start();
j1 = t1.Join(10000);
j2 = t2.Join(10000);
if (!j1 || !j2)
{
break;
}
}
_gpu.DisableMultithreading();
_gpu.FreeAll();
Assert.IsTrue(j1);
Assert.IsTrue(j2);
}
public void ExeTestKernel()
{
GPGPU gpu = CudafyHost.GetDevice(CudafyModes.Target, 0);
eArchitecture arch = gpu.GetArchitecture();
CudafyModule km = CudafyTranslator.Cudafy(arch);
gpu.LoadModule(km);
int[] host_results = new int[N];
// Either assign a new block of memory to hold results on device
var dev_results = gpu.Allocate <int>(N);
// Or fill your array with values first and then
for (int i = 0; i < N; i++)
{
host_results[i] = i * 3;
}
// Copy array with ints to device
var dev_filled_results = gpu.CopyToDevice(host_results);
// 64*16 = 1024 threads per block (which is max for sm_30)
dim3 threadsPerBlock = new dim3(64, 16);
// 8*8 = 64 blocks per grid , just for show so you get varying numbers
dim3 blocksPerGrid = new dim3(8, 8);
//var threadsPerBlock = 1024; // this will only give you blockDim.x = 1024, .y = 0, .z = 0
//var blocksPerGrid = 1; // just for show
gpu.Launch(blocksPerGrid, threadsPerBlock, "GenerateRipples", dev_results, dev_filled_results);
gpu.CopyFromDevice(dev_results, host_results);
}
public static uint[] Evaluate(ulong[] hands, int numCards)
{
// Translates this class to CUDA C and then compliles
CudafyModule km = CudafyTranslator.Cudafy();//eArchitecture.sm_20);
// Get the first GPU and load the module
GPGPU gpu = CudafyHost.GetDevice(CudafyModes.Target, CudafyModes.DeviceId);
gpu.LoadModule(km);
int blockSize = 256;
int blockx = hands.Length / blockSize;
if (hands.Length % blockSize != 0)
{
blockx++;
}
ulong[] dev_hands = gpu.Allocate <ulong>(hands.Length);
uint[] dev_ranks = gpu.Allocate <uint>(hands.Length);
gpu.CopyToDevice(hands, dev_hands);
gpu.StartTimer();
gpu.Launch(blockx, blockSize).evaluate(dev_hands, numCards, hands.Length, dev_ranks);
var ts = gpu.StopTimer();
uint[] toReturn = new uint[hands.Length];
gpu.CopyFromDevice(dev_ranks, toReturn);
return(toReturn);
}
public static void ClearGpuArray(GPGPU gpu, int[] gpuArray, int size)
{
var array = new int[size];
Array.Clear(array, 0, array.Length);
gpu.CopyToDevice(array, gpuArray);
}
public void SetUp()
{
_gpu = CudafyHost.GetDevice();
_sparse = GPGPUSPARSE.Create(_gpu);
_blas = GPGPUBLAS.Create(_gpu);
_solver = new Solver(_gpu, _blas, _sparse);
}
/// <summary>
/// Copies any reference type fields (e.g. arrays) of the object to the device.
/// </summary>
private static void CopyReferenceTypeFieldsToDevice <T>(GPGPU gpu, T hostObject)
{
var fields = DeviceClassHelper.GetFieldsStandardLayout(hostObject.GetType());
foreach (FieldInfo field in fields)
{
object fieldValue = field.GetValue(hostObject);
// Ignore if CudafyIgnore
if (field.GetCustomAttributes(typeof(CudafyIgnoreAttribute), true).Count() > 0)
{
continue;
}
// Only copy field to the device if this is not already done.
if (field.FieldType.IsArray && !deviceObjectFromHostObject[gpu].ContainsKey(fieldValue))
{
// If the elements of the array are value types, then we make a deep copy. Otherwise, we create an array of
// pointers to the objects.
if (field.FieldType.GetElementType().IsValueType)
{
Array hostArray = (Array)fieldValue;
CopyArrayToDevice(gpu, hostArray);
}
else
{
CopyArrayOfReferenceTypeToDevice(gpu, (Array)fieldValue);
}
}
else if (!field.FieldType.IsValueType && !deviceObjectFromHostObject[gpu].ContainsKey(fieldValue))
{
CreateDeviceObject(gpu, fieldValue);
}
}
}
private static void AssignPointerFields(GPGPU gpu, object hostObject, object deviceObject, List <FieldMapping> pointerFields)
{
foreach (FieldMapping mapping in pointerFields)
{
object fieldValue = mapping.HostObjectField.GetValue(hostObject);
// Get the IntPtr to the device memory for the array.
var devicePointer = TryGetDeviceMemoryFromHostObject(gpu, fieldValue);
if (devicePointer == null)
{
throw new ArgumentException("No device memory allocated for field " + mapping.Name);
}
// The device object contains this pointer.
mapping.DeviceObjectField.SetValue(deviceObject, devicePointer.Pointer);
// If the field is an array then set the dimension fields too.
if (mapping is ArrayFieldMapping)
{
ArrayFieldMapping arrayFieldMapping = (ArrayFieldMapping)mapping;
Array array = fieldValue as Array;
for (int i = 0; i < arrayFieldMapping.ArrayRank; ++i)
{
arrayFieldMapping.DeviceObjectDimensionFields[i].SetValue(deviceObject, array.GetLength(i));
}
}
}
}
public static void Execute()
{
CudafyModule km = CudafyTranslator.Cudafy(typeof(ParamsStruct), typeof(ImpliedVolatile));
_gpu = CudafyHost.GetDevice(CudafyModes.Target);
_gpu.LoadModule(km);
ParamsStruct[] host_par = new ParamsStruct[1];
ParamsStruct[] result = new ParamsStruct[1];
host_par[0].OP = 96.95;
host_par[0].Price = 1332.24;
host_par[0].Strike = 1235;
host_par[0].TD = 31;
host_par[0].R = 0.0001355;
host_par[0].Q = 0.0166;
host_par[0].N = 100;// 1000;
host_par[0].kind = 1;
ParamsStruct[] dev_par = _gpu.CopyToDevice(host_par);
float[] PA = _gpu.Allocate <float>(1001);
_gpu.Launch(1, 1, "impliedVolatile", dev_par, PA);
_gpu.CopyFromDevice(dev_par, 0, result, 0, 1);
Console.WriteLine("I={0}, B={1}", result[0].i, result[0].B);
//Console.ReadKey();
}
public static void Example2(GPGPU gpu)
{
ArrayView view1 = new ArrayView();
ArrayView view2 = new ArrayView();
float[] data = Enumerable.Range(0, 1000).Select(t => (float)t).ToArray();
// Two views of the array, simply applying an offset to the array; could slice instead for example.
view1.CreateView(data, 100);
view2.CreateView(data, 200);
for (int i = 0; i < 1000; ++i)
{
data[i] = data[i] * 10f;
}
// Should copy the 'large' array to the device only once; this is referenced by each ArrayView instance.
var dev_view1 = DeviceClassHelper.CreateDeviceObject(gpu, view1);
var dev_view2 = DeviceClassHelper.CreateDeviceObject(gpu, view2);
var dev_result = gpu.Allocate <float>(5);
var hostResult = new float[5];
gpu.Launch(1, 1).Test2(dev_view1, dev_view2, dev_result);
gpu.CopyFromDevice(dev_result, hostResult);
bool pass = (hostResult[0] == 1050f && hostResult[1] == 7f);
Console.WriteLine(pass ? "Pass" : "Fail");
}
public static void Execute()
{
CudafyModule km = CudafyModule.TryDeserialize();
if (km == null || !km.TryVerifyChecksums())
{
km = CudafyTranslator.Cudafy(typeof(Generic <ushort, ushort>), typeof(SimpleGeneric));
km.Serialize();
}
GPGPU gpu = CudafyHost.GetDevice(CudafyModes.Target);
gpu.LoadModule(km);
var input = new Generic <ushort, ushort>();
input.A = 187;
int[] devoutput = gpu.Allocate <int>(1);
gpu.Launch(1, 1, "Kernel", input, devoutput);
int output;
gpu.CopyFromDevice(devoutput, out output);
Console.WriteLine("Simple Generic: " + ((output == 1) ? "PASSED" : "FAILED"));
}
public void Test_copyOnHost()
{
int len = 35687;
int[] bufa = new int[len];
int[] bufb = new int[len];
Random r = new Random();
for (int i = 0; i < len; i++)
{
bufa[i] = r.Next() + 1;
}
IntPtr ha = _gpu.HostAllocate <int>(len);
ha.Write(bufa, 0, 0, len);
IntPtr hb = _gpu.HostAllocate <int>(len);
GPGPU.CopyMemory(hb, ha, (uint)len * sizeof(int));
hb.Read(bufb, 0, 0, len);
for (int i = 0; i < len; i++)
{
Assert.True(bufa[i] == bufb[i]);
Assert.False(bufa[i] == 0);
}
_gpu.HostFreeAll();
}
/// <summary>
/// Creates a SPARSE wrapper based on the specified gpu. Note only CudaGPU is supported.
/// </summary>
/// <param name="gpu">The gpu.</param>
/// <returns></returns>
public static GPGPUSPARSE Create(GPGPU gpu)
{
if (gpu is CudaGPU)
return new CudaSPARSE(gpu);
else
throw new NotImplementedException(gpu.ToString());
}
/// <summary>
/// Вызов и исполнение одной элементарной функции по имени функции
/// </summary>
/// <param name="function"></param>
public static void Execute(string function)
{
Debug.Assert(_indexes1.Last() == _sequencies1.Length);
Debug.Assert(_indexes2.Last() == _sequencies2.Length);
CudafyModule km = CudafyTranslator.Cudafy();
GPGPU gpu = CudafyHost.GetDevice();
gpu.LoadModule(km);
// copy the arrays 'a' and 'b' to the GPU
int[] devIndexes1 = gpu.CopyToDevice(_indexes1);
int[] devIndexes2 = gpu.CopyToDevice(_indexes2);
int[] devSequencies1 = gpu.CopyToDevice(_sequencies1);
int[] devSequencies2 = gpu.CopyToDevice(_sequencies2);
int[,] devMatrix = gpu.Allocate(_matrix);
int rows = _matrix.GetLength(0);
int columns = _matrix.GetLength(1);
dim3 gridSize = Math.Min(15, (int)Math.Pow((double)rows * columns, 0.33333333333));
dim3 blockSize = Math.Min(15, (int)Math.Pow((double)rows * columns, 0.33333333333));
gpu.Launch(gridSize, blockSize, function,
devSequencies1, devIndexes1,
devSequencies2, devIndexes2,
devMatrix);
// copy the array 'c' back from the GPU to the CPU
gpu.CopyFromDevice(devMatrix, _matrix);
// free the memory allocated on the GPU
gpu.FreeAll();
}
public void SetUp()
{
_gpu = CudafyHost.CreateDevice(CudafyModes.Target);
_hostInput = new double[N * BATCH];
_hostInputCplx = new ComplexD[N * BATCH];
_hostOutput = new double[N * BATCH];
_hostOutputCplx = new ComplexD[N * BATCH];
_devInput = _gpu.Allocate(_hostInput);
_devInputCplx = _gpu.Allocate(_hostInputCplx);
_devInter = _gpu.Allocate <double>(N * 2 * BATCH);
_devInterCplx = _gpu.Allocate <ComplexD>(N * BATCH);
_devOutput = _gpu.Allocate(_hostOutput);
_devOutputCplx = _gpu.Allocate(_hostOutputCplx);
_fft = GPGPUFFT.Create(_gpu);
for (int b = 0; b < BATCH; b++)
{
for (int i = 0; i < N; i++)
{
ComplexD cf = new ComplexD();
cf.x = (double)((10.0F * Math.Sin(100 * 2 * Math.PI * i / N * Math.PI / 180)));
cf.y = (double)((10.0F * Math.Sin(200 * 2 * Math.PI * i / N * Math.PI / 180)));
_hostInput[i + b * N] = cf.x;
_hostInputCplx[i + b * N] = cf;
}
}
}
public void Test_TwoThreadTwoGPU()
{
_gpu0 = CudafyHost.CreateDevice(CudafyModes.Target, 0);
_gpu1 = CudafyHost.CreateDevice(CudafyModes.Target, 1);
_gpu0.EnableMultithreading();
_gpu1.EnableMultithreading();
bool j1 = false;
bool j2 = false;
for (int i = 0; i < 10; i++)
{
Console.WriteLine(i);
Thread t1 = new Thread(Test_TwoThreadTwoGPU_Thread0);
Thread t2 = new Thread(Test_TwoThreadTwoGPU_Thread1);
t1.Start();
t2.Start();
j1 = t1.Join(10000);
j2 = t2.Join(10000);
if (!j1 || !j2)
{
break;
}
}
_gpu0.DisableMultithreading();
_gpu0.FreeAll();
_gpu1.DisableMultithreading();
_gpu1.FreeAll();
Assert.IsTrue(j1);
Assert.IsTrue(j2);
}
public void SetUp()
{
_gpu = CudafyHost.CreateDevice(CudafyModes.Target);
Console.WriteLine("CUDA driver version={0}", _gpu.GetDriverVersion());
_fft = GPGPUFFT.Create(_gpu);
_hostInput = new float[N * BATCH];
_hostInputCplx = new ComplexF[N * BATCH];
_hostOutput = new float[N * BATCH];
_hostOutputCplx = new ComplexF[N * BATCH];
_devInput = _gpu.Allocate(_hostInput);
_devInputCplx = _gpu.Allocate(_hostInputCplx);
_devInter = _gpu.Allocate<float>(N * 2 * BATCH);
_devInterCplx = _gpu.Allocate<ComplexF>(N * BATCH);
_devOutput = _gpu.Allocate(_hostOutput);
_devOutputCplx = _gpu.Allocate(_hostOutputCplx);
Console.WriteLine("CUFFT version={0}", _fft.GetVersion());
for (int b = 0; b < BATCH; b++)
{
for (int i = 0; i < N; i++)
{
ComplexF cf = new ComplexF();
cf.x = (float)((10.0F * Math.Sin(100 * 2 * Math.PI * i / N * Math.PI / 180)));
cf.y = (float)((10.0F * Math.Sin(200 * 2 * Math.PI * i / N * Math.PI / 180)));
_hostInput[i + b * N] = cf.x;
_hostInputCplx[i + b * N] = cf;
}
}
}
public void SetUp()
{
_gpu = CudafyHost.CreateDevice(CudafyModes.Target);
Console.WriteLine(_gpu.GetDriverVersion());
_fft = GPGPUFFT.Create(_gpu);
_hostInput = new float[N * BATCH];
_hostInputCplx = new ComplexF[N * BATCH];
_hostOutput = new float[N * BATCH];
_hostOutputCplx = new ComplexF[N * BATCH];
_devInput = _gpu.Allocate(_hostInput);
_devInputCplx = _gpu.Allocate(_hostInputCplx);
_devInter = _gpu.Allocate <float>(N * 2 * BATCH);
_devInterCplx = _gpu.Allocate <ComplexF>(N * BATCH);
_devOutput = _gpu.Allocate(_hostOutput);
_devOutputCplx = _gpu.Allocate(_hostOutputCplx);
Console.WriteLine(_fft.GetVersion());
for (int b = 0; b < BATCH; b++)
{
for (int i = 0; i < N; i++)
{
ComplexF cf = new ComplexF();
cf.x = (float)((10.0F * Math.Sin(100 * 2 * Math.PI * i / N * Math.PI / 180)));
cf.y = (float)((10.0F * Math.Sin(200 * 2 * Math.PI * i / N * Math.PI / 180)));
_hostInput[i + b * N] = cf.x;
_hostInputCplx[i + b * N] = cf;
}
}
}
public void SetUp()
{
_gpu = CudafyHost.GetDevice();
_sparse = GPGPUSPARSE.Create(_gpu);
_blas = GPGPUBLAS.Create(_gpu);
_solver = new Solver(_gpu, _blas, _sparse);
}
public void SetUp()
{
_gpu = CudafyHost.GetDevice(CudafyModes.Architecture, CudafyModes.DeviceId);
_byteBufferIn = new byte[N];
_byteBufferOut = new byte[N];
_sbyteBufferIn = new sbyte[N];
_sbyteBufferOut = new sbyte[N];
_ushortBufferIn = new ushort[N];
_ushortBufferOut = new ushort[N];
_uintBufferIn = new uint[N];
_uintBufferOut = new uint[N];
_ulongBufferIn = new ulong[N];
_ulongBufferOut = new ulong[N];
_cplxDBufferIn = new ComplexD[N];
_cplxDBufferOut = new ComplexD[N];
_cplxFBufferIn = new ComplexF[N];
_cplxFBufferOut = new ComplexF[N];
SetInputs();
ClearOutputsAndGPU();
}
//
// http://stackoverflow.com/questions/18628447/cudafy-throws-an-exception-while-testing
//
private static void BlasSample(int deviceId)
{
CudafyModes.Target = eGPUType.Emulator;
GPGPU gpu = CudafyHost.GetDevice(CudafyModes.Target, deviceId);
CudafyModes.DeviceId = deviceId;
eArchitecture arch = gpu.GetArchitecture();
CudafyModule km = CudafyTranslator.Cudafy(arch);
gpu.LoadModule(km);
GPGPUBLAS blas = GPGPUBLAS.Create(gpu);
const int N = 100;
float[] a = new float[N];
float[] b = new float[N];
float[] c = new float[N];
float alpha = -1;
float beta = 0;
float[] device_a = gpu.CopyToDevice(a);
float[] device_b = gpu.CopyToDevice(b);
float[] device_c = gpu.CopyToDevice(c);
int m = 10;
int n = 10;
int k = 10;
cublasOperation Op = cublasOperation.N;
blas.GEMM(m, k, n, alpha, device_a, device_b, beta, device_c, Op);
throw new NotImplementedException();
}
/// <summary>
/// Вызов и исполнение функции проверки что массив отсортирован
/// </summary>
public static void ExecuteSorted(int direction = 1)
{
CudafyModule km = CudafyTranslator.Cudafy();
GPGPU gpu = CudafyHost.GetDevice();
gpu.LoadModule(km);
int[] devA = gpu.Allocate(_a);
int[] devB = gpu.Allocate(_b);
int[] devC = gpu.Allocate(_c);
int[] devD = gpu.Allocate(D);
gpu.CopyToDevice(_a, devA);
gpu.Launch(1, 1).Split(devA, devB, devC, _middle);
gpu.Launch(_gridSize, _blockSize).Sorted(devA, devB, devC, devD, 0, direction);
gpu.Launch(1, 1).Sorted(devA, devB, devC, devD, 1, direction);
gpu.CopyFromDevice(devD, D);
// free the memory allocated on the GPU
gpu.FreeAll();
}
/// <summary>
/// Creates a BLAS wrapper based on the specified gpu.
/// </summary>
/// <param name="gpu">The gpu.</param>
/// <returns></returns>
public static GPGPUBLAS Create(GPGPU gpu)
{
if (gpu is CudaGPU)
return new CudaBLAS(gpu);
else
return new HostBLAS(gpu);
//throw new NotImplementedException(gpu.ToString());
}
public GPU_func()
{
km = CudafyTranslator.Cudafy();
gpu = CudafyHost.GetDevice(CudafyModes.Target, CudafyModes.DeviceId);
gpu.LoadModule(km);
GPU_prop = gpu.GetDeviceProperties();
}
public void SetUp()
{
_gpu = CudafyHost.CreateDevice(CudafyModes.Target);
_blas = GPGPUBLAS.Create(_gpu);
_hostInput = new float[ciROWS, ciCOLS];
_hostInput2 = new float[ciROWS, ciCOLS];
_hostOutput = new float[ciROWS, ciCOLS];
_devPtr = _gpu.Allocate<float>(_hostInput);
_devPtr2 = _gpu.Allocate<float>(_hostOutput);
}
public void SetUp()
{
//var x = CompilerHelper.Create(ePlatform.x64, eArchitecture.OpenCL, eCudafyCompileMode.Default);
var y = CompilerHelper.Create(ePlatform.x64, CudafyModes.Architecture, eCudafyCompileMode.DynamicParallelism);
_cm = CudafyTranslator.Cudafy(new CompileProperties[] {y}, this.GetType());
Console.WriteLine(_cm.CompilerOutput);
_cm.Serialize();
_gpu = CudafyHost.GetDevice(y.Architecture, CudafyModes.DeviceId);
_gpu.LoadModule(_cm);
}
public void SetUp()
{
_gpu = CudafyHost.CreateDevice(CudafyModes.Target);
_blas = GPGPUBLAS.Create(_gpu);
Console.Write("BLAS Version={0}", _blas.GetVersion());
_hostInput1 = new float[ciN];
_hostInput2 = new float[ciN];
_hostOutput1 = new float[ciN];
_hostOutput2 = new float[ciN];
_devPtr1 = _gpu.Allocate<float>(_hostInput1);
_devPtr2 = _gpu.Allocate<float>(_hostOutput1);
}
internal CudaRAND(GPGPU gpu, curandRngType rng_type)
{
_gpu = gpu;
if (IntPtr.Size == 8)
{
_driver = new CURANDDriver64();
}
else
{
_driver = new CURANDDriver32();
}
}
public void Initialize(int bytes)
{
CudafyModule km = CudafyTranslator.Cudafy();
_gpu = CudafyHost.GetDevice(CudafyModes.Target, CudafyModes.DeviceId);
_gpu.LoadModule(km);
_dev_bitmap = _gpu.Allocate<byte>(bytes);
_blocks = new dim3(DIM / 16, DIM / 16);
_threads = new dim3(16, 16);
}
public void SetUp()
{
CudafyTranslator.GenerateDebug = true;
_cm = CudafyModule.TryDeserialize();
_gpu = CudafyHost.GetDevice(CudafyModes.Architecture, CudafyModes.DeviceId);
if (_cm == null || !_cm.TryVerifyChecksums())
{
_cm = CudafyTranslator.Cudafy(_gpu.GetArchitecture(), this.GetType(), (_gpu is OpenCLDevice) ? null : typeof(StringConstClass));
_cm.TrySerialize();
}
_gpu.LoadModule(_cm);
}
internal CudaRAND(GPGPU gpu, curandRngType rng_type)
{
_gpu = gpu;
if (IntPtr.Size == 8)
{
_driver = new CURANDDriver64();
}
else
{
throw new NotSupportedException();
//_driver = new CURANDDriver32();
}
}
public virtual void SetUp()
{
_gpu = CudafyHost.GetDevice(CudafyModes.Architecture, CudafyModes.DeviceId);
var types = new List<Type>();
types.Add(this.GetType());
types.Add(typeof(MathSingleTest));
SupportsDouble = _gpu.GetDeviceProperties().SupportsDoublePrecision;
if (SupportsDouble)
types.Add(typeof(MathDoubleTest));
_cm = CudafyTranslator.Cudafy(CudafyModes.Architecture, types.ToArray());
Debug.WriteLine(_cm.SourceCode);
_gpu.LoadModule(_cm);
}
public Solver(GPGPU gpu, GPGPUBLAS blas, GPGPUSPARSE sparse)
{
this.gpu = gpu;
this.blas = blas;
this.sparse = sparse;
var km = CudafyModule.TryDeserialize();
if (km == null || !km.TryVerifyChecksums())
{
km = CudafyTranslator.Cudafy();
km.TrySerialize();
}
gpu.LoadModule(km);
}
public void SetUp()
{
_gpu = CudafyHost.GetDevice(CudafyModes.Architecture, CudafyModes.DeviceId);
_cm = CudafyModule.TryDeserialize();
if (_cm == null || !_cm.TryVerifyChecksums())
{
_cm = CudafyTranslator.Cudafy(CudafyModes.Architecture);//typeof(PrimitiveStruct), typeof(BasicFunctionTests));
Console.WriteLine(_cm.CompilerOutput);
_cm.TrySerialize();
}
_gpu.LoadModule(_cm);
}
public void SetUp()
{
_gpu = CudafyHost.CreateDevice(CudafyModes.Target);
_sparse = GPGPUSPARSE.Create(_gpu);
_hiMatrixMN = new double[M * N];
_hiMatrixMN2 = new double[M * N];
_hoMatrixMN = new double[M * N];
_hoPerVector = new int[M];
_hoPerVector2 = new int[N];
_diPerVector2 = _gpu.Allocate(_hoPerVector2);
_diMatrixMN = _gpu.Allocate(_hiMatrixMN);
_diMatrixMN2 = _gpu.Allocate(_hiMatrixMN2);
_diPerVector = _gpu.Allocate(_hoPerVector);
}
public void SetUp()
{
_gpu = CudafyHost.GetDevice();
_sparse = GPGPUSPARSE.Create(_gpu);
hiMatrixMN = new double[M * N];
hiMatrixMK = new double[M * K];
hiMatrixKM = new double[K * M];
hiMatrixKN = new double[K * N];
hiMatrixNN = new double[N * N];
hiVectorXM = new double[M];
hiVectorXN = new double[N];
hiVectorYM = new double[M];
hiVectorYN = new double[N];
gpuResultM = new double[M];
gpuResultN = new double[N];
gpuResultMN = new double[M * N];
}
public GpuRenderer()
{
var availableOpenCLDevices = CudafyHost.GetDeviceProperties(eGPUType.OpenCL);
if (availableOpenCLDevices.Any() == false)
{
throw new Exception("No OpenCL devices found...");
}
var device = availableOpenCLDevices.First();
Module = CudafyTranslator.Cudafy(eArchitecture.OpenCL12);
var blockSide =
Enumerable
.Range(1, 15)
.Reverse()
.First(count => count * count <= device.MaxThreadsPerBlock);
BlockSize = new dim3(blockSide, blockSide);
// Initialize gpu and load the module (avoids reloading every time)
gpu = CudafyHost.GetDevice(eGPUType.OpenCL);
gpu.LoadModule(Module);
}
public void SetUp()
{
_gpu = CudafyHost.GetDevice(CudafyModes.Target);
_blas = GPGPUBLAS.Create(_gpu);
hiMatrixAMM = new double[M * M];
hiMatrixANN = new double[N * N];
hiMatrixAMK = new double[M * K];
hiMatrixAKM = new double[K * M];
hiMatrixBMN = new double[M * N];
hiMatrixBKN = new double[K * N];
hiMatrixBNK = new double[N * K];
hiMatrixBMK = new double[M * K];
hiMatrixBKM = new double[K * M];
hiMatrixCMN = new double[M * N];
hiMatrixCKN = new double[K * N];
hiMatrixCMK = new double[M * K];
hiMatrixCMM = new double[M * M];
gpuResultMN = new double[M * N];
gpuResultMM = new double[M * M];
}
public void SetUp()
{
_gpu = CudafyHost.GetDevice(CudafyModes.Target);
_blas = GPGPUBLAS.Create(_gpu);
Console.Write("BLAS Version={0}", _blas.GetVersion());
// Initialize CPU Buffer
hiMatrixA = new double[M * N];
hiMatrixANN = new double[N * N];
hiMatrixACBC = new double[(KL + KU + 1) * N];
hiMatrixASCBC = new double[(K + 1) * N];
hiMatrixAPS = new double[(N * (N + 1)) / 2];
hiVectorXM = new double[M];
hiVectorXN = new double[N];
hiVectorYM = new double[M];
hiVectorYN = new double[N];
gpuResultM = new double[M];
gpuResultN = new double[N];
gpuResultMN = new double[M * N];
gpuResultNN = new double[N * N];
gpuResultP = new double[(N * (N + 1)) / 2];
}
public static void Execute()
{
_gpu = CudafyHost.GetDevice(eGPUType.Cuda);
CudafyModule km = CudafyTranslator.Cudafy(ePlatform.Auto, _gpu.GetArchitecture(), typeof(TextInsertion));
Console.WriteLine(km.CompilerOutput);
_gpu.LoadModule(km);
int[] data = new int[64];
int[] data_d = _gpu.CopyToDevice(data);
int[] res_d = _gpu.Allocate(data);
int[] res = new int[64];
_gpu.Launch(1, 1, "AHybridMethod", data_d, res_d);
_gpu.CopyFromDevice(data_d, res);
for(int i = 0; i < 64; i++)
if (data[i] != res[i])
{
Console.WriteLine("Failed");
break;
}
}
public static void Execute()
{
_gpu = CudafyHost.GetDevice(eGPUType.Cuda);
CudafyModule km = CudafyTranslator.Cudafy(ePlatform.Auto, _gpu.GetArchitecture(), typeof(SIMDFunctions));
//CudafyModule km = CudafyTranslator.Cudafy(ePlatform.Auto, eArchitecture.sm_12, typeof(SIMDFunctions));
_gpu.LoadModule(km);
int w = 1024;
int h = 1024;
for (int loop = 0; loop < 3; loop++)
{
uint[] a = new uint[w * h];
Fill(a);
uint[] dev_a = _gpu.CopyToDevice(a);
uint[] b = new uint[w * h];
Fill(b);
uint[] dev_b = _gpu.CopyToDevice(b);
uint[] c = new uint[w * h];
uint[] dev_c = _gpu.Allocate(c);
_gpu.StartTimer();
_gpu.Launch(h, w, "SIMDFunctionTest", dev_a, dev_b, dev_c);
_gpu.CopyFromDevice(dev_c, c);
float time = _gpu.StopTimer();
Console.WriteLine("Time: {0}", time);
if (loop == 0)
{
bool passed = true;
GThread thread = new GThread(1, 1, null);
for (int i = 0; i < w * h; i++)
{
uint exp = thread.vadd2(a[i], b[i]);
if (exp != c[i])
passed = false;
}
Console.WriteLine("Test {0}", passed ? "passed. " : "failed!");
}
_gpu.FreeAll();
}
}
public void SetUp()
{
_gpu = CudafyHost.GetDevice();
_sparse = GPGPUSPARSE.Create(_gpu);
_hiVectorX = new float[N];
_hiVectorY = new float[N];
_hoVectorY = new float[N];
FillBufferSparse(_hiVectorX, out NNZ);
FillBuffer(_hiVectorY);
_hiIndicesX = new int[NNZ];
_hoValsX = new float[NNZ];
_hiValsX = new float[NNZ];
GetSparseIndex(_hiVectorX, _hiValsX, _hiIndicesX);
_diValsX = _gpu.Allocate(_hiValsX);
_diIndicesX = _gpu.Allocate(_hiIndicesX);
_diVectorY = _gpu.Allocate(_hiVectorY);
}