public static void Einsum(Tensor D,
Tensor A, string modeA, Tensor B, string modeB, Tensor C, string modeC,
double alpha = 1f, double beta = 0f)
{
int[] mA = new int[modeA.Length];
int[] mB = new int[modeB.Length];
int[] mC = new int[modeC.Length];
for (int i = 0; i < mA.Length; i++)
{
mA[i] = modeA[i];
}
for (int i = 0; i < mB.Length; i++)
{
mB[i] = modeB[i];
}
for (int i = 0; i < mC.Length; i++)
mC[i] = modeC[i];
fixed(int *ptrmA = mA,
ptrmB = mB,
ptrmC = mC)
fixed(long *DimensionsA = A.Shape.Dimensions, MultipliedA = A.Shape.Multiplied,
DimensionsB = B.Shape.Dimensions, MultipliedB = B.Shape.Multiplied,
DimensionsC = C.Shape.Dimensions, MultipliedC = C.Shape.Multiplied,
DimensionsD = D.Shape.Dimensions, MultipliedD = D.Shape.Multiplied)
CudaKernels.Einsum(
A.Base.Array, A.Shape.N, ptrmA, DimensionsA, MultipliedA + 1, CudaTypes.GetDataType(A),
B.Base.Array, B.Shape.N, ptrmB, DimensionsB, MultipliedB + 1, CudaTypes.GetDataType(B),
C.Base.Array, C.Shape.N, ptrmC, DimensionsC, MultipliedC + 1, CudaTypes.GetDataType(C),
D.Base.Array, D.Shape.N, ptrmC, DimensionsD, MultipliedD + 1, CudaTypes.GetDataType(D),
alpha, beta, CudaTypes.GetComputeType(C));
}
public static void MultiplyFloat32(Tensor res, Tensor a, Tensor b, float cofmul = 1, float cofadd = 0)
{
CudaManagement.SetDevice(res.Config.Device.ID);
CudaKernels.MultiplyFloat32((float *)res.Base.Array, (float *)a.Base.Array, (float *)b.Base.Array, a.Shape.TotalSize, b.Shape.TotalSize, cofmul, cofadd);
}
