Jeremy Kun
Jeremy Kun
(neglected to add a TODO) This is in the conversion pattern for ConvertTensorExtractNewLayout (which may be renamed to ConvertTensorExtract after #2199).
In https://github.com/google/heir/pull/2205, the tensor.extract kernel uses anyRangePoint to decide what to mask from the ciphertext, and then uses convert_layout to ensure the resulting scalar is packed according to the result...
With the new layout system, a linalg.reduce kernel is a bit more complicated, mainly due to the extra decisions possible with a multi-ciphertext packing. In the old setup, we only...
A continuation of https://github.com/google/heir/issues/1376 Some kernels, like ct-ct matmul, require the input dimensions have certain properties, such as being coprime. Since different kernels have different requirements, the layout-optimizer needs to...
Placeholder for TODO and discussion.
warning: first commit is raw LLM output, needs lots of cleanup.
We're starting up an effort to compare HEIR with other research compilers, and this ticket is meant to track infrastructure work related to that task, as well as be a...
We'd like to support loop structures in HEIR that contain FHE operations. Much of the auxiliary compiler work is done, as loop support for most lowerings and code generation was...
A layout change that reduces to a single rightward shift by 1 would require using all power-of-two rotation keys for a leftward shift, since it's evaluating -1 = 0b11111...1. Supporting...
Maybe `secret.conceal` should have a `trivial` attribute that denotes this, otherwise server-side uses of `secret.conceal` on static plaintexts (say, for resolving type mismatches) would cause the noise analysis to think...