Skip to content

Use native Pytorch operations for complex numbers  #19

Description

@anjali411

Hi I noticed that you have custom matmul (

def complex_matmul(A, B):
'''
Performs the matrix product between two complex matricess
'''
outp_real = torch.matmul(A.real, B.real) - torch.matmul(A.imag, B.imag)
outp_imag = torch.matmul(A.real, B.imag) + torch.matmul(A.imag, B.real)
return outp_real.type(torch.complex64) + 1j * outp_imag.type(torch.complex64)
) and tanh, neg functions defined (
def complex_tanh(input):
return tanh(input.real).type(torch.complex64)+1j*tanh(input.imag).type(torch.complex64)
def complex_opposite(input):
return -(input.real).type(torch.complex64)+1j*(-(input.imag).type(torch.complex64))
) which are actually unnecessary since these functions are supported for complex numbers in the last couple of releases of PyTorch and would be much faster too (since we call into blas operation for matmul for example).

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions