Here is a short range vs. long range solution to get you into PyTorch development. I've added some links to some relevant files you will need to change.

Short Range Solution: Add a polar() and cart() math kernels.

Learn how to register a new math kernel in PyTorch.
- aten/src/ATen/native/native_functions.yaml
Implement the non-vectorized math kernels

aten/src/ATen/native/cpu/UnaryOpsKernel.cpp
See how sinh_kernel is implemented.

Implement the vectorized math kernels

aten/src/ATen/native/cpu/UnaryOpsKernel.cpp
See how neg_kernel is implemented.
aten/src/ATen/cpu/vec256/vec256_complex_float.h
aten/src/ATen/cpu/vec256/vec256_complex_double.h

Long Range Solution: Add ComplexPolarFloat dtype

The short range solution does not provide PyTorch with the knowledge that your complex number is in polar or cartesian coordinates. Hence all operators currently assume cartesian coordinates. We have started to discuss adding a Complex Polar dtype.

Work with fb to add a new polar dtypes (ComplexPolarHalf, ComplexPolarFloat, ComplexPolarDouble).
Add math kernels to the ComplexPolar Dtype to solve this issue.
- Vectorized: Create a new vec256 file aten/src/ATen/cpu/vec256/vec256_complex_polar float.h

I would just focus on the short-range solution for now, but I think we need the long-range solution to speed up some of these computations.

Method/constructor which takes as input angle and magnitude and returns a complex tensor #35312

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions