public Core(int Nxp,int Nyp, int Nzp, int Ntm, double Bbeta, double Flux)
{
Nx = Nxp; Ny = Nyp; Nz = Nzp; Nt = Ntm; betagauge = (floattype)Bbeta; flux = (floattype)Flux;
N = Nx * Ny * Nz * Nt; Nspace = Nx * Ny * Nz;
string strforcompiler = "-D Nt=" + Nt.ToString() + " -D Nxyz=" + (Nx * Ny * Nz).ToString() + " -D Nxy=" + (Nx*Ny).ToString() +
" -D Nx="+(Nx).ToString()+" -D Ny="+(Ny).ToString()+" -D Nz="+(Nz).ToString();
strforcompiler += typeof(floattype) == typeof(double) ? " -D floattype=double -D floattype2=double2 -D floattype4=double4" :
" -D floattype=float -D floattype2=float2 -D floattype4=float4";
strforcompiler += " -D phi=" + flux.ToString().Replace(',', '.') + " -D KAPPA=" + kappa.ToString().Replace(',', '.');
string fp64support = "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n";
Plocalsize = AdjustLocalSize(Nspace);
Slocalsize = AdjustLocalSize(N / 2);
XhermYlocalsize = AdjustLocalSize(4 * N);
// Plocalsize = 16; Slocalsize = 16;
PNumGroups = Nx * Ny * Nz / Plocalsize;
SNumGroups = N/2 / Slocalsize;
XhermYNumGroups = 4*4*N / XhermYlocalsize;
BufferLength = N * 4 * 9 * 2 * sizeof(floattype);
SeedBufLen = N * sizeof(Int32)/2 * 4;
AllocBuffers();
openCLPlatform = OpenCL.GetPlatform(0);
openCLDevices = openCLPlatform.QueryDevices(DeviceType.ALL);
openCLContext = openCLPlatform.CreateDefaultContext();
openCLCQ = openCLContext.CreateCommandQueue(openCLDevices[0], CommandQueueProperties.PROFILING_ENABLE);
MyKernelProgram = openCLContext.CreateProgramWithSource(
(typeof(floattype)==typeof(double)?fp64support:"") + File.ReadAllText("MyKernel.cl")+File.ReadAllText("dirak_mul.cl"));
try
{
MyKernelProgram.Build(openCLDevices, strforcompiler, null, IntPtr.Zero);
}
catch (OpenCLException)
{
string buildLog = MyKernelProgram.GetBuildLog(openCLDevices[0]);
MessageBox.Show(buildLog, "Build error(64 bit debug sessions in vs2008 always fail like this - debug in 32 bit or use vs2010)");
// Application.Exit();
}
MyKernelKernel = MyKernelProgram.CreateKernel("MyKernel");
PReductionKernel = MyKernelProgram.CreateKernel("PLoop");
SReductionKernel = MyKernelProgram.CreateKernel("CalcS");
DiralMulKernel = MyKernelProgram.CreateKernel("dirakMatrMul");
FillWithKernel = MyKernelProgram.CreateKernel("FillWith");
FillLinkWithKernel = MyKernelProgram.CreateKernel("FillLinkWith");
FillWithRandomKernel = MyKernelProgram.CreateKernel("FillWithRandom");
AXPYKernel = MyKernelProgram.CreateKernel("AXPY");
XhermYKernel = MyKernelProgram.CreateKernel("XhermY");
BackupLinkKernel = MyKernelProgram.CreateKernel("BackupLink");
RestoreLinkKernel = MyKernelProgram.CreateKernel("RestoreLink");
SeedMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), SeedBufLen, IntPtr.Zero);
LinkMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), BufferLength, IntPtr.Zero);
PGroupMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize * PNumGroups, IntPtr.Zero);
PResMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize, IntPtr.Zero);
SGroupMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize * SNumGroups, IntPtr.Zero);
SResMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize, IntPtr.Zero);
XhermYGroupMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize * 2*XhermYNumGroups, IntPtr.Zero);
XhermYresMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize * 2, IntPtr.Zero);
XhermYrespointer = System.Runtime.InteropServices.Marshal.AllocHGlobal(floatsize * 2);
SeedVectorMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), SeedVectorBuf.Length * sizeof(int), IntPtr.Zero);
StorageMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), linksize, IntPtr.Zero);
dSmem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize, IntPtr.Zero);
dSpointer = System.Runtime.InteropServices.Marshal.AllocHGlobal(floatsize);
MyKernelKernel.SetArg(0, (byte)EvenOdd);
MyKernelKernel.SetArg(1, (floattype)betagauge);
MyKernelKernel.SetArg(2, (floattype)flux);
MyKernelKernel.SetArg(3, SeedMem);
MyKernelKernel.SetArg(4, LinkMem);
PReductionKernel.SetArg(0, LinkMem);
PReductionKernel.SetArg(1, PGroupMem);
PReductionKernel.SetArg(2, PResMem);
IntPtr ptr = new IntPtr(Plocalsize * floatsize);
PReductionKernel.SetArg(3, ptr, IntPtr.Zero);
SReductionKernel.SetArg(0, LinkMem);
SReductionKernel.SetArg(1, SGroupMem);
SReductionKernel.SetArg(2, SResMem);
IntPtr ptr1 = new IntPtr(Slocalsize * floatsize);
SReductionKernel.SetArg(3, ptr1, IntPtr.Zero);
XhermYKernel.SetArg(2, XhermYresMem);
XhermYKernel.SetArg(3, XhermYGroupMem);
XhermYKernel.SetArg(4, new IntPtr(XhermYlocalsize*floatsize*2),IntPtr.Zero);
openCLCQ.EnqueueWriteBuffer(SeedMem, true, 0, SeedBufLen, ipseed);
openCLCQ.EnqueueWriteBuffer(LinkMem, true, 0, BufferLength, ip);
openCLCQ.EnqueueWriteBuffer(SeedVectorMem, true, 0, SeedVectorBuf.Length*sizeof(int), ipseedvector);
rhat0 = new Vector();
//init BICGStab vectors
phi = new Vector();
r0 = new Vector();
//rprev = new Vector();
pi = new Vector();
vi = new Vector();
t = new Vector();
s = new Vector();
// xprev = new Vector();
// vprev = new Vector();
// pprev = new Vector();
temp = new Vector();
ri = new Vector();
x = new Vector();
//for fermion update
chi = new Vector();
CalculateS();
double s1 = S[0];
BackupLink(0, 0,1, 0, 1);
CalculateS();
double s2 = S[0];
RestoreLink(0, 0, 1, 0, 1);
CalculateS();
double s3 = S[0];
//MessageBox.Show(s1.ToString() + s2.ToString() + s3.ToString());
}
//y=MulD(x)
void MulD(Vector x, Vector result)
{
DiralMulKernel.SetArg(0, LinkMem);
DiralMulKernel.SetArg(1, x.buf);
DiralMulKernel.SetArg(2, result.buf);
openCLCQ.EnqueueNDRangeKernel(DiralMulKernel, 1, null, new int[1] { 4 * N }, null);
openCLCQ.Finish();
openCLCQ.Flush();
// result.Updatearray();
}
Complex V1hermV2(Vector v1, Vector v2)
{
floattype[] c = new floattype[2];
for (int i = 0; i < 2; i++)
{
XhermYKernel.SetArg(0, v1.buf);
XhermYKernel.SetArg(1, v2.buf);
openCLCQ.EnqueueNDRangeKernel(XhermYKernel, 1, null, new int[1] { 4 * N }, new int[1] { XhermYlocalsize });
openCLCQ.Finish();
openCLCQ.Flush();
openCLCQ.EnqueueReadBuffer(XhermYresMem, true, 0, floatsize * 2, XhermYrespointer);
System.Runtime.InteropServices.Marshal.Copy(XhermYrespointer, c, 0, 2);
}
return new Complex(c[0], c[1]);
}
void AXPY(Complex a, Vector x, Vector y,Vector result)
{
AXPYKernel.SetArg(0, new floattype2(a.x,a.y));
AXPYKernel.SetArg(1, x.buf);
AXPYKernel.SetArg(2, y.buf);
AXPYKernel.SetArg(3, result.buf);
openCLCQ.EnqueueNDRangeKernel(AXPYKernel, 1, null, new int[1] { 4 * N }, null);
openCLCQ.Finish();
//result.Updatearray();
}
