Do some more simplifications specific to extents #3891
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Review updated until commit d72668e Description
Changes walkthrough 📝
PR Reviewer Guide 🔍Here are some key observations to aid the review process:
|
|
!test |
|
!test |
|
Maybe it's time for me to revive #511. With runtime info we should be able to fully simplify these expressions at concretization. |
jacobhinkle
approved these changes
Feb 17, 2025
Comment on lines
+383
to
+397
| // Simplify x + (-x) to 0 | ||
| Val* x = nullptr; | ||
| auto uop = dynamic_cast<UnaryOp*>(lhs->definition()); | ||
| if (uop != nullptr) { | ||
| // lhs may be (-x). Pick rhs as x | ||
| x = rhs; | ||
| } else { | ||
| uop = dynamic_cast<UnaryOp*>(rhs->definition()); | ||
| // rhs may be (-x). Pick lhs as x | ||
| x = lhs; | ||
| } | ||
| if (uop != nullptr && uop->getUnaryOpType() == UnaryOpType::Neg && | ||
| uop->in()->sameAs(x)) { | ||
| return lhs->fusion()->zeroVal(lhs->dtype()); | ||
| } |
There was a problem hiding this comment.
I think if you had abs(y) + (-abs(y)) this might not catch it because lhs=abs(y) is not a Neg. Instead, what about something like this?
Suggested change
| // Simplify x + (-x) to 0 | |
| Val* x = nullptr; | |
| auto uop = dynamic_cast<UnaryOp*>(lhs->definition()); | |
| if (uop != nullptr) { | |
| // lhs may be (-x). Pick rhs as x | |
| x = rhs; | |
| } else { | |
| uop = dynamic_cast<UnaryOp*>(rhs->definition()); | |
| // rhs may be (-x). Pick lhs as x | |
| x = lhs; | |
| } | |
| if (uop != nullptr && uop->getUnaryOpType() == UnaryOpType::Neg && | |
| uop->in()->sameAs(x)) { | |
| return lhs->fusion()->zeroVal(lhs->dtype()); | |
| } | |
| // Simplify x + (-x) to 0 | |
| Val* x = nullptr; | |
| Val* neg_x_in = nullptr; | |
| if (auto uop = dynamic_cast<UnaryOp*>(lhs->definition()); uop && uop->getUnaryOpType() == UnaryOpType::Neg) { | |
| // lhs is -x. Pick rhs as x | |
| neg_x_in = uop->in(); | |
| x = rhs; | |
| } else if (auto uop = dynamic_cast<UnaryOp*>(rhs->definition()); uop && uop->getUnaryOpType() == UnaryOpType::Neg) { | |
| // rhs is -x. Pick lhs as x | |
| neg_x_in = uop->in(); | |
| x = lhs; | |
| } | |
| if (x != nullptr && neg_x_in->sameAs(x)) { | |
| return lhs->fusion()->zeroVal(lhs->dtype()); | |
| } |
| const auto normalize_slice_range = [&manual_normalization]( | ||
| Slice range, Val* extent) -> Slice { | ||
| const auto get_int = [](Val* x) -> std::optional<int64_t> { | ||
| if (x != nullptr && x->isConstInt()) { |
There was a problem hiding this comment.
Minor nit: if you initialize an ExpressionEvaluator in this function I think you could use expr_eval.evaluate(x) to return a PolymorphicValue and check pv.hasValue() instead of using optional<int64_t>. Then you could avoid the isConstInt() condition here which will redundantly evaluate the val and replace it with isIntegralScalar().
Comment on lines
+812
to
+814
| if (range.start->isConstInt()) { | ||
| start_int = range.start->evaluate().as<int64_t>(); | ||
| } |
There was a problem hiding this comment.
See comment above if we have an expr_eval. Not important but will avoid re-computing these vals.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
While working on #3848, I noticed
test_unpadded_catop_issue2275_repro2took an extremely long time (> 30 min). That seems largely due to index hoisting and expression simplification. It took just a few seconds when they were disabled. That's likely due to a lot ofminandmaxdue to slicing of symbolic extents, as shown below:This PR tries to simplifies these extents a little further, which results in:
The test time is reduced to several seconds by these simplifications.
Confirmed no failure with manual_ci.sh on an H100 machine.