public static string Init(float[] inputs, uint span, uint blockSize)
{
_span = span;
_block_size = blockSize;
_area = _span * _span;
_blocks_per_span = span / blockSize;
_blockCount = _blocks_per_span * _blocks_per_span;
d_indexRands = new IntPtr();
d_grid = new IntPtr();
_cudaArray = new CudaArray();
_gridProcs = new GridProcs();
_randoProcs = new RandoProcs();
var strRet = _cudaArray.ResetDevice();
strRet = strRet + _randoProcs.MakeGenerator32(SEED);
strRet = strRet + _cudaArray.MallocFloatsOnDevice(ref d_grid, _area);
strRet = strRet + _cudaArray.CopyFloatsToDevice(inputs, d_grid, _area);
strRet = strRet + _cudaArray.MallocIntsOnDevice(ref d_indexRands, _blockCount);
return(strRet);
}
public static string CopyToDevice(out GpuMatrix gmOut, GpuMatrix gmIn)
{
if (gmIn.DevHostState == DevHostState.DeviceNotAllocated)
{
gmOut = null;
return("Device data pointer not allocated");
}
var aa = new CudaArray();
var strRet = aa.CopyFloatsToDevice(
gmIn.Matrix.Data.ToArray(),
gmIn.DevPtr, (uint)gmIn.Matrix.Data.Length);
if (!String.IsNullOrEmpty(strRet))
{
gmOut = null;
return(strRet);
}
gmOut = new GpuMatrix(
matrix: gmIn.Matrix,
devPtr: gmIn.DevPtr,
devHostState: DevHostState.Synched);
return(String.Empty);
}
public static string Init(float[] inputs, uint span)
{
_span = span;
_area = _span * _span;
d_gridA = new IntPtr();
d_gridB = new IntPtr();
_cudaArray = new CudaArray();
_gridProcs = new GridProcs();
_randoProcs = new RandoProcs();
var strRet = _cudaArray.ResetDevice();
strRet = strRet + _cudaArray.MallocFloatsOnDevice(ref d_gridA, _area);
strRet = strRet + _cudaArray.MallocFloatsOnDevice(ref d_gridB, _area);
strRet = strRet + _cudaArray.CopyFloatsToDevice(inputs, d_gridA, _area);
strRet = strRet + _cudaArray.CopyFloatsToDevice(inputs, d_gridB, _area);
return(strRet);
}
static string TestCopyFloatsDeviceToDevice()
{
string testName = "TestCopyFloatsDeviceToDevice";
uint arrayLen = 1000;
var alist = Enumerable.Range(4, (int)arrayLen).Select(t => (float)t).ToArray();
var aa = new CudaArray();
IntPtr devDataA = new System.IntPtr();
IntPtr devDataB = new System.IntPtr();
var retlist = new float[(int)arrayLen];
try
{
var res = aa.ResetDevice();
res = res + aa.MallocFloatsOnDevice(ref devDataA, arrayLen);
res = res + aa.MallocFloatsOnDevice(ref devDataB, arrayLen);
res = res + aa.CopyFloatsToDevice(alist, devDataA, arrayLen);
res = res + aa.CopyFloatsDeviceToDevice(devDataB, devDataA, arrayLen);
res = res + aa.CopyFloatsFromDevice(retlist, devDataB, arrayLen);
res = res + aa.ReleaseDevicePtr(devDataA);
res = res + aa.ReleaseDevicePtr(devDataB);
if (!alist.SequenceEqual(retlist))
{
return(testName + " fail: sequences do not match");
}
if (res != String.Empty)
{
return(testName + " fail: " + res);
}
return(testName + " pass");
}
catch (Exception ex)
{
return(testName + " exception " + ex.Message);
}
finally
{
aa.ReleaseDevicePtr(devDataA);
aa.ReleaseDevicePtr(devDataB);
aa.ResetDevice();
}
}
public static string Init(float[] temp_inputs, int[] flip_inputs, uint span, uint blockSize, int seed)
{
_span = span;
_block_size = blockSize;
_area = _span * _span;
_blocks_per_span = span / blockSize;
_blockCount = _blocks_per_span * _blocks_per_span;
d_flipData = new IntPtr();
d_tempData = new IntPtr();
d_flipRands = new IntPtr();
d_indexRands = new IntPtr();
d_threshes = new IntPtr();
_cudaArray = new CudaArray();
_gridProcs = new GridProcs();
_randoProcs = new RandoProcs();
var strRet = _cudaArray.ResetDevice();
strRet = strRet + _randoProcs.MakeGenerator32(seed);
strRet = strRet + _cudaArray.MallocFloatsOnDevice(ref d_tempData, _area);
strRet = strRet + _cudaArray.MallocFloatsOnDevice(ref d_heatBlocks, _area / 1024);
strRet = strRet + _cudaArray.MallocIntsOnDevice(ref d_flipData, _area);
strRet = strRet + _cudaArray.MallocIntsOnDevice(ref d_indexRands, _blockCount);
strRet = strRet + _cudaArray.MallocIntsOnDevice(ref d_flipRands, _blockCount);
strRet = strRet + _cudaArray.MallocFloatsOnDevice(ref d_threshes, _allTempSteps);
strRet = strRet + _cudaArray.CopyIntsToDevice(flip_inputs, d_flipData, _area);
strRet = strRet + _cudaArray.CopyFloatsToDevice(temp_inputs, d_tempData, _area);
var res9 = new int[_area];
strRet = strRet + _cudaArray.CopyIntsFromDevice(res9, d_flipData, _area);
return(strRet);
}
public static ProcResult UpdateE(int steps, float temp)
{
var strRet = String.Empty;
float t2 = (float)(1.0 / (1.0 + Math.Exp(2 * temp)));
float t4 = (float)(1.0 / (1.0 + Math.Exp(4 * temp)));
float[] thresh = new float[10];
//thresh[1] = 1.0f;
//thresh[3] = 1.0f;
thresh[1] = 1.0f - t4;
thresh[3] = 1.0f - t2;
thresh[5] = 0.5f;
thresh[7] = t2;
thresh[9] = t4;
strRet = strRet + _cudaArray.CopyFloatsToDevice(thresh, d_betas, 10);
IntPtr dSrc;
IntPtr dDest = IntPtr.Zero;
_stopwatch.Reset();
_stopwatch.Start();
float energyTot = 0;
int[] resB = new int[_area / 1024];
for (var s = 0; s < steps; s++)
{
if (_phase == 0)
{
dSrc = d_gridA;
dDest = d_gridB;
strRet = strRet + _randoProcs.MakeUniformRands(d_rands, _area);
_phase = 1;
}
else
{
dSrc = d_gridB;
dDest = d_gridA;
_phase = 0;
}
strRet = strRet + _gridProcs.Runk_Ising_dg(
destPtr: dDest,
energyPtr: d_energy,
srcPtr: dSrc,
randPtr: d_rands,
span: _span,
alt: _phase,
threshPtr: d_betas
);
strRet = strRet + _cudaArray.RunBlockAddInts_32_Kernel(
destPtr: d_energyBlocks,
srcPtr: d_energy,
span: _span
);
strRet = strRet + _cudaArray.CopyIntsFromDevice(resB, d_energyBlocks, _area / 1024);
energyTot += (float)resB.Sum() / (float)_area;
}
energyTot /= steps;
int[] res = new int[_area];
strRet = strRet + _cudaArray.CopyIntsFromDevice(res, dDest, _area);
_stopwatch.Stop();
var dRet = new Dictionary <string, object>();
dRet["Grid"] = new SimGrid <int>(name: "UpdateE",
width: _span,
height: _span,
data: res);
dRet["Energy"] = energyTot;
return(new ProcResult(data: dRet,
err: strRet,
stepsCompleted: steps,
time: _stopwatch.ElapsedMilliseconds));
}
public static ProcResult UpdateH(int steps, float qRate, float filpEnergy, float beta)
{
var strRet = String.Empty;
_stopwatch.Reset();
_stopwatch.Start();
for (var s = 0; s < steps; s++)
{
var res9 = new int[_area];
strRet = strRet + _cudaArray.CopyIntsFromDevice(res9, d_flipData, _area);
var bbs = FloatFuncs.Betas(_tempSteps, beta);
strRet = strRet + _cudaArray.CopyFloatsToDevice(bbs, d_threshes, _allTempSteps);
strRet = strRet + _randoProcs.MakeRandomInts(d_indexRands, _blockCount);
strRet = strRet + _randoProcs.MakeUniformRands(d_flipRands, _blockCount);
strRet = strRet + _gridProcs.Run_k_ThermoIsing_bp(
temp_data: d_tempData,
flip_data: d_flipData,
index_rands: d_indexRands,
flip_rands: d_flipRands,
threshes: d_threshes,
flip_energy: filpEnergy,
block_size: _block_size,
blocks_per_span: _blocks_per_span,
q_rate: qRate);
}
_stopwatch.Stop();
var dRet = new Dictionary <string, object>();
float[] bhts = new float[_area / 1024];
strRet = strRet + _cudaArray.RunBlockAddFloats_32_Kernel(
destPtr: d_heatBlocks,
srcPtr: d_tempData,
span: _span
);
strRet = strRet + _cudaArray.CopyFloatsFromDevice(bhts, d_heatBlocks, _area / 1024);
float tot = bhts.Sum();
tot /= _area;
dRet["TotalHeat"] = tot;
var tres = new float[_area];
strRet = strRet + _cudaArray.CopyFloatsFromDevice(tres, d_tempData, _area);
dRet["ThermGrid"] = new SimGrid <float>(name: "Therms",
width: _span,
height: _span,
data: tres);
var fres = new int[_area];
strRet = strRet + _cudaArray.CopyIntsFromDevice(fres, d_flipData, _area);
dRet["FlipGrid"] = new SimGrid <int>(name: "Flips",
width: _span,
height: _span,
data: fres);
return(new ProcResult(data: dRet,
err: strRet,
stepsCompleted: steps,
time: _stopwatch.ElapsedMilliseconds));
}
public static ProcResult ProcIsingRb(int steps, float temp)
{
var strRet = String.Empty;
float t2 = (float)(1.0 / (1.0 + Math.Exp(2 * temp)));
float t4 = (float)(1.0 / (1.0 + Math.Exp(4 * temp)));
float[] thresh = new float[5];
//thresh[0] = 1.0f;
//thresh[1] = 1.0f;
thresh[0] = 1.0f - t4;
thresh[1] = 1.0f - t2;
thresh[2] = 0.5f;
thresh[3] = t2;
thresh[4] = t4;
strRet = strRet + _cudaArray.CopyFloatsToDevice(thresh, d_betas, 5);
_stopwatch.Reset();
_stopwatch.Start();
for (var i = 0; i < steps; i++)
{
strRet = strRet + _randoProcs.MakeRandomInts(d_indexRands, _blockCount);
strRet = strRet + _randoProcs.MakeUniformRands(d_tempRands, _blockCount);
strRet = strRet + _gridProcs.Run_k_Ising_bp(
destPtr: d_grid,
energyPtr: d_energy,
indexRandPtr: d_indexRands,
tempRandPtr: d_tempRands,
block_size: _block_size,
blocks_per_span: _blocks_per_span,
threshPtr: d_betas
);
}
strRet = strRet + _gridProcs.Runk_Energy4(d_energy, d_grid, _span);
int[] mres = new int[_area / 1024];
strRet = strRet + _cudaArray.RunBlockAddInts_32_Kernel(
destPtr: d_energyBlocks,
srcPtr: d_energy,
span: _span);
strRet = strRet + _cudaArray.CopyIntsFromDevice(mres, d_energyBlocks, _area / 1024);
float tot = mres.Sum();
tot /= _area;
int[] res = new int[_area];
strRet = strRet + _cudaArray.CopyIntsFromDevice(res, d_grid, _area);
_stopwatch.Stop();
//uint[] res2 = new uint[_blockCount];
//strRet = strRet + _cudaArray.CopyUIntsFromDevice(res2, d_rands, _blockCount);
var dRet = new Dictionary <string, object>();
dRet["Grid"] = new SimGrid <int>(name: "Update",
width: _span,
height: _span,
data: res);
dRet["Energy"] = tot;
return(new ProcResult(data: dRet,
err: strRet,
stepsCompleted: steps,
time: _stopwatch.ElapsedMilliseconds));
}