static string TestCublasHandle()
{
CublasOp o = CublasClr.CublasOp.N;
string testName = "TestCublasHandle";
var cuby = new CublasClr.Cublas();
IntPtr devHandle = new IntPtr();
var aa = new CudaArray();
try
{
var res = aa.ResetDevice();
res = res + cuby.MakeCublasHandle(ref devHandle);
res = res + cuby.DestroyCublasHandle(devHandle);
if (res != String.Empty)
{
return(testName + " fail: " + res);
}
return(testName + " pass");
}
catch
{
return(testName + " fail");
}
finally
{
//aa.ReleaseDevicePtr(devData);
aa.ResetDevice();
}
}
public static string Init(int[] inputs, uint span)
{
_span = span;
_area = _span * _span;
d_rands = new IntPtr();
d_gridA = new IntPtr();
d_gridB = new IntPtr();
d_energy = new IntPtr();
d_energyBlocks = new IntPtr();
_cudaArray = new CudaArray();
_gridProcs = new GridProcs();
_randoProcs = new RandoProcs();
var strRet = _cudaArray.ResetDevice();
strRet = strRet + _randoProcs.MakeGenerator64(SEED);
strRet = strRet + _cudaArray.MallocIntsOnDevice(ref d_gridA, _area);
strRet = strRet + _cudaArray.MallocIntsOnDevice(ref d_energy, _area);
strRet = strRet + _cudaArray.MallocIntsOnDevice(ref d_energyBlocks, _area / 1024);
strRet = strRet + _cudaArray.MallocIntsOnDevice(ref d_gridB, _area);
strRet = strRet + _cudaArray.CopyIntsToDevice(inputs, d_gridA, _area);
strRet = strRet + _cudaArray.CopyIntsToDevice(inputs, d_gridB, _area);
strRet = strRet + _cudaArray.MallocFloatsOnDevice(ref d_rands, _area);
strRet = strRet + _cudaArray.MallocFloatsOnDevice(ref d_betas, 10);
return(strRet);
}
public static string Init(int[] 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.MallocIntsOnDevice(ref d_grid, _area);
strRet = strRet + _cudaArray.MallocIntsOnDevice(ref d_energy, _area);
strRet = strRet + _cudaArray.MallocIntsOnDevice(ref d_energyBlocks, _area / 1024);
strRet = strRet + _cudaArray.CopyIntsToDevice(inputs, d_grid, _area);
strRet = strRet + _cudaArray.MallocIntsOnDevice(ref d_indexRands, _blockCount);
strRet = strRet + _cudaArray.MallocFloatsOnDevice(ref d_tempRands, _blockCount);
strRet = strRet + _gridProcs.Runk_Energy4(d_energy, d_grid, _span);
strRet = strRet + _cudaArray.MallocFloatsOnDevice(ref d_betas, 5);
return(strRet);
}
static string TestcublasSgemm2()
{
string testName = "TestcublasSgemm2";
uint aw = 2;
uint bh = aw;
uint ah = 3;
uint bw = 3;
uint ch = ah;
uint cw = bw;
GpuMatrix gpuA;
GpuMatrix gpuB;
GpuMatrix gpuC;
var dataA = MatrixUtils.AA();
var dataB = MatrixUtils.BB();
var cRes = new float[ch * cw];
var cuby = new CublasClr.Cublas();
var aa = new CudaArray();
var res = aa.ResetDevice();
res = res + GpuMatrixOps.SetupGpuMatrix(
out gpuA,
new Matrix <float>(_rows: ch, _cols: cw,
host_data: ImmutableArray.Create(dataA),
matrixFormat: MatrixFormat.Column_Major));
res = res + GpuMatrixOps.SetupGpuMatrix(
out gpuB,
new Matrix <float>(_rows: bh, _cols: bw,
host_data: ImmutableArray.Create(dataB),
matrixFormat: MatrixFormat.Column_Major));
res = res + GpuMatrixOps.SetupGpuMatrix(
out gpuC,
new Matrix <float>(_rows: ch, _cols: cw,
host_data: ImmutableArray.Create(cRes),
matrixFormat: MatrixFormat.Column_Major));
IntPtr cublasHandle = new IntPtr();
res = res + cuby.MakeCublasHandle(ref cublasHandle);
GpuMatrix gpuProd;
res = res + GpuMatrixOps.Multiply(
gmOut: out gpuProd,
cublasHandle: cublasHandle,
gmA: gpuA, gmB: gpuB, gmC: gpuC);
GpuMatrix gpuSynched;
res = res + GpuMatrixOps.CopyToHost(out gpuSynched, gpuProd);
// GpuMatrixUtils.MatrixMult(C: cRes, A: dataA, B: dataB, wA: aw, hA: ah, wB: bw);
return(string.Empty);
}
public static string Init(int[] inputs, uint span)
{
// init libs
_cudaArray = new CudaArray();
_gridProcs = new GridProcs();
_randoProcs = new RandoProcs();
// Set grid sizes
_span = span;
_area = _span * _span;
// Set block and thread sizes
_blockSize = (_span < MAXTHREADS) ? _span : MAXTHREADS;
_gridSize = _area / _blockSize;
// Set memory sizes
_mem_N = sizeof(int) * (_area);
_mem_rand = sizeof(double) * (3 * _area);
_mem_1 = sizeof(int) * (1);
_mem_measured_quantity = sizeof(int) * (_gridSize);
_mem_measured_magnet = sizeof(int) * (_gridSize);
// Allocate device arrays
d_rands = new IntPtr();
d_grid = new IntPtr();
var strRet = _cudaArray.ResetDevice();
strRet = strRet + _randoProcs.MakeGenerator64(SEED);
strRet = strRet + _cudaArray.MallocIntsOnDevice(ref d_grid, _area);
strRet = strRet + _cudaArray.CopyIntsToDevice(inputs, d_grid, _area);
strRet = strRet + _cudaArray.MallocFloatsOnDevice(ref d_rands, _area);
return(strRet);
}
static string TestCopyIntsDeviceToDevice()
{
string testName = "TestCopyIntsDeviceToDevice";
uint arrayLen = 1000;
var alist = Enumerable.Range(4, (int)arrayLen).ToArray();
var aa = new CudaArray();
IntPtr devDataA = new System.IntPtr();
IntPtr devDataB = new System.IntPtr();
var retlist = new int[(int)arrayLen];
try
{
var res = aa.ResetDevice();
res = res + aa.MallocIntsOnDevice(ref devDataA, arrayLen);
res = res + aa.MallocIntsOnDevice(ref devDataB, arrayLen);
res = res + aa.CopyIntsToDevice(alist, devDataA, arrayLen);
res = res + aa.CopyIntsDeviceToDevice(devDataB, devDataA, arrayLen);
res = res + aa.CopyIntsFromDevice(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();
}
}
static string TestMakeNormalRands()
{
string testName = "TestMakeNormalRands";
var rdo = new RandoClr.RandoProcs();
var aa = new CudaArray();
uint arrayLen = 1000;
int seed = 1234;
IntPtr devRando = new IntPtr();
IntPtr devData = new IntPtr();
var retlist = new float[(int)arrayLen];
try
{
var res = aa.ResetDevice();
res = res + rdo.MakeGenerator64(ref devRando, seed);
res = res + aa.MallocFloatsOnDevice(ref devData, arrayLen);
res = res + rdo.MakeNormalRands(devData, devRando, arrayLen, 0.0f, 1.0f);
res = res + aa.CopyFloatsFromDevice(retlist, devData, arrayLen);
res = res + aa.ReleaseDevicePtr(devData);
res = res + rdo.DestroyGenerator(devRando);
if (res != String.Empty)
{
return(testName + " fail: " + res);
}
return(testName + " pass");
}
catch
{
return(testName + " fail");
}
finally
{
//rdo.DestroyGenerator(devRando);
aa.ReleaseDevicePtr(devData);
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 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 TestcublasSgemm1()
{
string testName = "TestcublasSgemm";
uint aw = 5;
uint bh = aw;
uint ah = 5;
uint bw = 5;
uint ch = ah;
uint cw = bw;
var cuby = new CublasClr.Cublas();
var aa = new CudaArray();
var res = aa.ResetDevice();
var dataA = MatrixUtils.MakeIdentity(rows: ah, cols: aw);
GpuMatrix gpuA;
res = res + GpuMatrixOps.SetupGpuMatrix(
out gpuA,
new Matrix <float>(_rows: ah, _cols: aw,
host_data: ImmutableArray.Create(dataA),
matrixFormat: MatrixFormat.Column_Major));
var dataB = MatrixUtils.MakeIdentiPoke(rows: bh, cols: bw);
GpuMatrix gpuB;
res = res + GpuMatrixOps.SetupGpuMatrix(
out gpuB,
new Matrix <float>(_rows: bh, _cols: bw,
host_data: ImmutableArray.Create(dataB),
matrixFormat: MatrixFormat.Column_Major));
var dataC = MatrixUtils.MakeZeroes(rows: bh, cols: bw);
GpuMatrix gpuC;
res = res + GpuMatrixOps.SetupGpuMatrix(
out gpuC,
new Matrix <float>(_rows: ch, _cols: cw,
host_data: ImmutableArray.Create(dataC),
matrixFormat: MatrixFormat.Column_Major));
IntPtr cublasHandle = new IntPtr();
res = res + cuby.MakeCublasHandle(ref cublasHandle);
GpuMatrix gpuProd;
res = res + GpuMatrixOps.Multiply(
gmOut: out gpuProd,
cublasHandle: cublasHandle,
gmA: gpuA, gmB: gpuB, gmC: gpuC);
GpuMatrix gpuSynched;
res = res + GpuMatrixOps.CopyToHost(out gpuSynched, gpuProd);
var cpuRes = new float[ah * bw];
MatrixUtils.RowMajorMatrixMult(C: cpuRes, A: dataA, B: dataB, wA: aw, hA: ah, wB: bw);
var cpuRes2 = new float[bh * aw];
MatrixUtils.RowMajorMatrixMult(C: cpuRes2, A: dataB, B: dataA, wA: bw, hA: bh, wB: aw);
return(res);
}