Skip to content

OptimizeInstruction: Optimize any boolean & (No overlap with boolean's LSB) ==> 0 #7505

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

Open
wants to merge 4 commits into
base: main
Choose a base branch
from

Conversation

xuruiyang2002
Copy link
Contributor

There are so many rules for and, but we still cannot optimize the following one:

(i32.and
 (i32.eqz
  (i32.load $0
   (i32.const 0)
  )
 )
 (i32.const 4)
)

to zero.

Adding rule for add: any boolean & (No overlap with boolean's LSB) ==> 0

Fixes: #7481

uint64_t constantValue = c->value.getInteger();
if ((constantValue & mask) == 0) {
replaceCurrent(getDroppedChildrenAndAppend(
curr, LiteralUtils::makeZero(type, *getModule())));
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

For both the replaceCurrent() calls in this function, the caller does that on the returned value, so we can just return the value here.

auto* right = curr->right;
auto leftMaxBits = Bits::getMaxBits(left, this);

if (leftMaxBits == 0) {
Copy link
Member

@kripken kripken Apr 23, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Does 0 need to be special-cased? I think it can work without it (if not, a comment would help).

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Oh, also, the case of 0 & .. should already be handled elsewhere, so we don't need to handle it here anyhow

;; CHECK-NEXT: (i32.and
;; CHECK-NEXT: (i32.const 1)
;; CHECK-NEXT: (i32.const 2)
;; CHECK-NEXT: )
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The goal of this existing test is to see the order of the operands, so this optimization prevents the test from working. To avoid that, the test can be modified to check two different numbers, perhaps (does 2, 1 - reversing the order - work perhaps?)

;; CHECK-NEXT: (i32.const 127)
;; CHECK-NEXT: (i32.const 128)
;; CHECK-NEXT: )
;; CHECK-NEXT: (i32.const 0)
;; CHECK-NEXT: )
(func $sext-24-and-127-128 (result i32)
(i32.shr_s
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The goal in this test is to see that the sign-extend can be removed. If it optimizes all the way to 0, that isn't clear. Perhaps the constant 128 a few lines below can be replaced with a local.get, so it is unknown?

;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: (i32.const 2)
;; CHECK-NEXT: )
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Perhaps the i64 2 could be an i64 param, to avoid this getting optimized away.

;; CHECK-NEXT: )
;; CHECK-NEXT: (i32.const 2)
;; CHECK-NEXT: )
;; CHECK-NEXT: (i32.const 0)
;; CHECK-NEXT: )
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i32.eqz
Copy link
Member

@kripken kripken Apr 23, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We will need new tests for the new function optimizeAndNoOverlappingBits, including the various corner cases (just one overlapping bit, no overlapping bits, etc.)

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.

O3 fails to const propagation & folding but O2 can
2 participants