-
Notifications
You must be signed in to change notification settings - Fork 3.5k
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
[TensorIR][M2a] Fuse, Split #8467
Conversation
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.
Overall LGTM. Just some style comments.
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.
Thanks! @jinhongyii
@@ -881,7 +883,7 @@ PrimExpr RewriteSimplifier::Impl::VisitExpr_(const FloorModNode* op) { | |||
|
|||
TVM_TRY_REWRITE_IF(floormod(x + y * c1, c2), floormod(x, c2), | |||
c2.Eval()->value > 0 && c1.Eval()->value % c2.Eval()->value == 0); | |||
|
|||
TVM_TRY_REWRITE_IF(floormod(x * c1, x * c2), x * floormod(c1, c2), c2.Eval()->value != 0); |
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.
Please send this as separate PR with regression testcases.
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.
These lines are added to support symbolic split/fuse in some simple cases, but it still can't handle situations where there are 2 or more symbolic vars. Do you think we should support this simple situation for now by merging it in a new PR or we put it aside and support the complete symbolic usage after we improve symbolic divisible check in iter_affine_map?
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.
@tqchen @junrushao1994 what's your opinion on this?
CC @spectrometerHBH please take a look at the iter-affine-map part |
I have addressed all the comments. Please have another look. |
Comments are addressed. Please have another round of review! |
Thanks for your patience! I don't have other comments |
Sorry for bothering you. I don't know what happened and I don't know how to cancel it. |
@junrushao1994 I've addressed your new comments |
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.
Thanks for the tireless effort! LGTM
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.
LGTM
* Fuse&split (apache#408) Co-authored-by: jinhongyi <323195289@qq.com> Co-authored-by: Junru Shao <junrushao1994@gmail.com>
* Fuse&split (apache#408) Co-authored-by: jinhongyi <323195289@qq.com> Co-authored-by: Junru Shao <junrushao1994@gmail.com>
This PR is part of the TensorIR upstreaming effort (#7527) , which adds 2 schedule primitives: