public override void Swap(VectorBase other)
{
if (!this.IsLocked || !other.IsLocked)
{
throw new ApplicationException("works only in locked mode");
}
CudaVector _other = other as CudaVector;
if (_other == null)
{
throw new ArgumentException("other must be of type CudaVector.", "other");
}
if (_other.Part.LocalLength != this.Part.LocalLength)
{
throw new ArgumentException("mismatch in vector size.");
}
CUdeviceptr temp = _other.d_data;
_other.d_data = this.d_data;
this.d_data = temp;
}
internal override void CallDriver(CUstream stream, double alpha, CudaVector a, double beta, CudaVector acc)
{
CUdeviceptr d_x = a.GetDevicePointer();
CUdeviceptr d_result = acc.GetDevicePointer();
int offset = 0;
cu.ParamSetp(sparseMultiply, offset, d_cellData);
offset += sizeof(long);
cu.ParamSetp(sparseMultiply, offset, d_x);
offset += sizeof(long);
cu.ParamSetp(sparseMultiply, offset, d_cellColIdx);
offset += sizeof(long);
cu.ParamSetp(sparseMultiply, offset, d_result);
offset += sizeof(long);
cu.ParamSetd(sparseMultiply, offset, alpha);
offset += sizeof(double);
cu.ParamSetd(sparseMultiply, offset, beta);
offset += sizeof(double);
cu.ParamSeti(sparseMultiply, offset, cellsize);
offset += sizeof(uint);
cu.ParamSeti(sparseMultiply, offset, cellrowsperblock);
offset += sizeof(uint);
cu.ParamSeti(sparseMultiply, offset, cellsperrow);
offset += sizeof(uint);
cu.ParamSeti(sparseMultiply, offset, stride);
offset += sizeof(uint);
cu.ParamSeti(sparseMultiply, offset, rowcount);
offset += sizeof(uint);
cu.ParamSetSize(sparseMultiply, (uint)offset);
cu.FuncSetBlockShape(sparseMultiply, blocksize, 1, 1);
cu.FuncSetSharedSize(sparseMultiply, (uint)(blocksize * sizeof(double) + 2 * cellrowsperblock * sizeof(int)));
cu.LaunchGridAsync(sparseMultiply, blockcount, 1, stream);
}
/// <summary> see CUDA doc; </summary>
public static void MemcpyDtoD(CUdeviceptr dstDevice, CUdeviceptr srcDevice, uint ByteCount)
{
testResult(my.cuMemcpyDtoD(dstDevice, srcDevice, ByteCount));
}
/// <summary> see CUDA doc; </summary>
static public void ParamSetp(CUfunction hfunc, int offset, CUdeviceptr ptr)
{
ParamSetl(hfunc, offset, (long)ptr.p);
}
/// <summary> see CUDA doc; </summary>
public static void MemcpyHtoD(CUdeviceptr dstDevice, IntPtr srcHost, uint ByteCount)
{
testResult(my.cuMemcpyHtoD(dstDevice, srcHost, ByteCount));
}
/// <summary> see CUDA doc; </summary>
static public void MemFree(CUdeviceptr dptr)
{
testResult(my.cuMemFree(dptr));
}
/// <summary> see CUDA doc; </summary>
public static void MemHostGetDevicePointer(out CUdeviceptr dptr, IntPtr p, uint flags)
{
testResult(my.cuMemHostGetDevicePointer(out dptr, p, flags));
}
/// <summary> see CUDA doc; </summary>
public static void MemsetD8(CUdeviceptr dstDevice, byte uc, uint N)
{
testResult(my.cuMemsetD8(dstDevice, uc, N));
}
/// <summary> see CUDA doc; </summary>
public static void MemAlloc(out CUdeviceptr dptr, uint bytesize)
{
testResult(my.cuMemAlloc(out dptr, bytesize));
}
