-
Notifications
You must be signed in to change notification settings - Fork 82
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[aievec] Canonicalize transposed B contract op. #1475
[aievec] Canonicalize transposed B contract op. #1475
Conversation
2be4a48
to
c2c73de
Compare
SmallVector<vector::IteratorType> iterators = op.getIteratorTypesArray(); | ||
auto innerMostIterators = | ||
SmallVector<vector::IteratorType>(iterators.end() - 3, iterators.end()); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It's possible that len(iterator_types)<3
?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It's possible that
len(iterator_types)<3
?
I don't think so, no. At that point I've already checked that lhs/rhs/acc have at least 2 dimensions each, which implies at least 2 parallel dimensions (acc) and there has to be at least one contracting dimension.
Edit: I've done the experiment, for the sake of it, and it looks like it does accept having 2d operands and accumulator, and only two iterators (one a reduction). I'm pretty sure that's a mistake, though. I need to check.
c2c73de
to
55e7a06
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Nice work.
@@ -411,6 +460,107 @@ struct FlattenMultDimTransferWritePattern | |||
} | |||
}; | |||
|
|||
// This pattern takes out an implicit transposition of the `rhs` operand in a |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I suggest to clarify the comment as follows:
an implicit transposition of the rhs
operand
=>
an implicit transposition of the 2 innermost dimensions of the rhs
operand
No description provided.