[Mlir-commits] [mlir] [linalg] Add quantized version of `conv_3d_ncdhw_fcdhw` (PR #113953)

[llvmlistbot at llvm.org]

MaheshRavishankar wrote:

I think we need to make a decision here.
1) We freeze any additions to Linalg named ops.
2) We allow uses of named ops of this form to come in and then build a "generalized convolution op" in Linalg that accounts for all convolution names ops.

I am not really comfortable with 1 cause we dont have a plan to fix it for downstream users. Without a solid plan and some one signed up to work on it, I would rather go with (2). I know I blocked this PR to kick start a discussion, but that isnt really fair to the contributor here (it is on us as defacto maintainers to provide a better solution).
If we are on-board with (2), we can let this one go in and add it to the list of "ops we need to deprecate".

https://github.com/llvm/llvm-project/pull/113953