Skip to content

Conversation

@Lunderberg
Copy link
Contributor

Resolve a breakage introduced in #16757. Prior to #16757, distinct TIR variables were unified if they had the same name. This commit avoids using distinct TIR variables to represent the same user input.

Resolve a breakage introduced in
apache#16757.  Prior to apache#16757, distinct
TIR variables were unified if they had the same name.  This commit
avoids using distinct TIR variables to represent the same user input.
@Lunderberg Lunderberg changed the title [SLM] Provide consistent output features in nn.Linear [SLM] Provide consistent dynamic shape when specified as python string Mar 25, 2024
If there is a trivial binding of `Var = DataflowVar`, but the
non-dataflow variable is never used outside the dataflow block in
which is is declared, then we should keep the name of the upstream
`DataflowVar`, as it is more likely to be the human-readable
name (e.g. a function parameter).
Expand handling of python strings to reproduce pre-16757 behavior

If a dynamic shape is specified with a python string, all instances of
that python string within a given function should be represented by
the same TIR variable.
@Lunderberg Lunderberg force-pushed the slm_consistent_nn_linear_out_features branch from e32aa2e to 3043702 Compare March 25, 2024 16:31
@Lunderberg
Copy link
Contributor Author

This PR is superseded by #16785.

@Lunderberg Lunderberg closed this Mar 25, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant