KJlaccHoeUM9l commented on PR #13999: URL: https://github.com/apache/tvm/pull/13999#issuecomment-1450157843
Hello @masahi! Regarding the latest comments: > We should avoid allocating the imaginary part when possible. In practice, I think we only need real to complex DFT and complex to real iDFT. In both cases, we can remove one of input or output tensors. > So rather than defining one big DFT op that does everything, how about making more fine-grained definition for each possible use case? > Also, please make sure to make our definition of DFT compatible with other frameworks like PyTorch without big performance cost. Here is one for PyTorch https://pytorch.org/docs/stable/fft.html The comments are valid and should be taken into account. However, these changes require an additional investment of time. Unfortunately, the project in which I did this work ended unexpectedly. Can we merge this PR in this form (with the possible addition of `TODO`), since the direction of my work has changed a lot? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
