JIT: Add `RangeOps::Xor` and `RangeOps::Not` by BoyBaykiller · Pull Request #126553 · dotnet/runtime

BoyBaykiller · 2026-04-04T21:43:36Z

Fix multiple regression from #126529.

One example: ParallelEnumerable.Range.
Simplified:

void TightenRangeXor(int count)
{
    if (count > 0 && count < 500)
    {
        // count range: [<1, 499>]
        // count ^ INT_MAX is equivalent to INT_MAX - x
        // new count range: [INT_MAX-499, INT_MAX - 1]
        // => can remove branch
        if ((count ^ int.MaxValue) < (count - 1000))
        {
            throw new ArgumentOutOfRangeException(nameof(count));
        }
    }
}

G_M64805_IG02:  ;; offset=0x0000
       ret      
						;; size=1 bbWeight=1 PerfScore 1.00

dotnet-policy-service · 2026-04-04T21:44:45Z

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

EgorBo · 2026-04-05T13:15:34Z

CI is down so we can't see the diffs.

Original repro: ParallelEnumerable.Range.

um.. I don't see XOR there - does it appear there after some jit transformation?

EgorBo · 2026-04-05T13:15:45Z

@MihuBot

Copilot

Pull request overview

This PR updates CoreCLR JIT range analysis to understand xor so that range tightening (and downstream optimizations like branch removal) keep working after the CONST - x → xor x, CONST transformation from #126529.

Changes:

Add RangeOps::Xor to compute ranges for a limited set of XOR patterns (constant-folding and ^ -1 / ^ INT_MAX).
Teach RangeCheck::GetRangeFromAssertions to compute XOR ranges from VN assertions (VNF_XOR).
Use RangeOps::Xor in RangeCheck::ComputeRangeForBinOp for GT_XOR (while preserving the existing Log2 special-case).

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
`src/coreclr/jit/rangecheck.h`	Adds `RangeOps::Xor` to enable tighter range inference for certain XOR cases.
`src/coreclr/jit/rangecheck.cpp`	Wires XOR into assertion-based range derivation and general binop range computation.

Copilot · 2026-04-05T13:21:10Z

src/coreclr/jit/rangecheck.h

+        // x ^ -1      is equivalent to -1 - x
+        // x ^ INT_MAX is equivalent to INT_MAX - x
+        // Example: [3..5] ^ [-1..-1] = [-6..-4]
+        if (r1IsConstVal && (r1ConstVal == -1 || r1ConstVal == INT_MAX))
+        {
+            return Subtract(r1, r2);
+        }


RangeOps::Xor maps the ^ -1 case to Subtract, but Subtract delegates through Negate which returns unknown for constant ranges that include INT_MIN. Since XOR/bitwise-NOT has no overflow semantics, this unnecessarily loses precision for ranges like [INT_MIN..k] ^ [-1..-1]. Consider handling the -1 case directly (e.g., by computing bitwise-NOT of constant limits) so it can still produce a constant range when the input range is constant.

Copilot · 2026-04-05T13:21:10Z

src/coreclr/jit/rangecheck.h

+        {
+            return Subtract(r2, r1);
+        }
+


Remove the trailing whitespace on this blank line to satisfy repo formatting/linting expectations.

Suggested change

Copilot · 2026-04-05T13:21:10Z

src/coreclr/jit/rangecheck.cpp

+        case GT_XOR:
+            r = RangeOps::Xor(op1Range, op2Range);
+            break;


This change is meant to prevent a performance regression by allowing range tightening through XOR (e.g., x ^ int.MaxValue) again. Please add a JIT regression test (likely under src/tests/JIT/opt/RangeChecks or similar) that fails without this change and verifies the relevant optimization (e.g., absence of the throw helper / simplified control flow) so future range-analysis tweaks don’t reintroduce the regression.

BoyBaykiller · 2026-04-05T13:41:51Z

um.. I don't see XOR there - does it appear there after some jit transformation?

Yes it gets produced by my other PR which I linked.
Currently this doesnt produce any diffs. But once #126529 is merged it will (fix the regression)

EgorBo · 2026-04-05T13:57:49Z

I'd say overflow doesn't exist for XOR but I saw OR also has some overflow handling

You're touching code GetRangeFromAssertions that doesn't rely on a separate path for overflow checking, so it's fine as long as it's correct.

EgorBo · 2026-04-05T14:01:12Z

um.. I don't see XOR there - does it appear there after some jit transformation?

Yes it gets produced by my other PR which I linked. Currently this doesnt produce any diffs. But once #126529 is merged it will (fix the regression)

well, then for now it's just a bunch of untested/never-triggered code without tests?

It looks correct to me, although, I think we should make Xor handling a bit smarter than focusing only on -1/maxvalue constants. E.g. the Log2 pattern involves ^ 31/^ 63 - if we can remove special handling for log2 and recognize it naturally..

BoyBaykiller · 2026-04-05T14:14:15Z

well, then for now it's just a bunch of untested/never-triggered code without tests

Well my hope is that the other PR gets merged first soon (which is why this is draft for now)

It looks correct to me, although, I think we should make Xor handling a bit smarter than focusing only on -1/maxvalue constants. E.g. the Log2 pattern involves ^ 31/^ 63 - if we can remove special handling for log2 and recognize it naturally

Yeah that'd be nice. But it seems like XOR alone is not enough to handle that. So maybe for an other PR? Although let me see if xor has a more general pattern

* add Not too

BoyBaykiller · 2026-04-10T04:27:13Z

Adding Not handling causes some diffs (even before #126529 is merged)

…ssion in System.Data.SqlTypes.SqlDecimal:AdjustScale 114664 libraries.pmi * add Not handling to ComputeRange

* add and use RangeOps::Xor

49c1b40

github-actions bot added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label Apr 4, 2026

dotnet-policy-service bot added the community-contribution Indicates that the PR has been added by a community member label Apr 4, 2026

EgorBo requested a review from Copilot April 5, 2026 13:15

MihuBot mentioned this pull request Apr 5, 2026

[JitDiff X64] [BoyBaykiller] JIT: Add and use RangeOps::Xor MihuBot/runtime-utils#1842

Open

Copilot started reviewing on behalf of EgorBo April 5, 2026 13:16 View session

Copilot AI reviewed Apr 5, 2026

View reviewed changes

BoyBaykiller mentioned this pull request Apr 5, 2026

JIT: Transform 'CONST - x' to 'xor x, CONST' when possible #126529

Open

BoyBaykiller closed this Apr 9, 2026

BoyBaykiller reopened this Apr 9, 2026

* make Xor handle all cases

dda7a50

* add Not too

* mark Xor and Neg as not overflowing as that's needed to avoid regre…

605b7e3

…ssion in System.Data.SqlTypes.SqlDecimal:AdjustScale 114664 libraries.pmi * add Not handling to ComputeRange

BoyBaykiller changed the title ~~JIT: Add and use RangeOps::Xor~~ JIT: Add RangeOps::Xor and RangeOps::Not Apr 12, 2026

build-analysis bot mentioned this pull request Apr 12, 2026

Multiple Helix work items fail on maccatalyst/tvos CoreCLR Release #126460

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

JIT: Add `RangeOps::Xor` and `RangeOps::Not`#126553

JIT: Add `RangeOps::Xor` and `RangeOps::Not`#126553
BoyBaykiller wants to merge 3 commits intodotnet:mainfrom
BoyBaykiller:add-xor-handling-to-rangecheck

BoyBaykiller commented Apr 4, 2026 •

edited

Loading

Uh oh!

dotnet-policy-service bot commented Apr 4, 2026

Uh oh!

EgorBo commented Apr 5, 2026

Uh oh!

EgorBo commented Apr 5, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Apr 5, 2026

Uh oh!

Copilot AI Apr 5, 2026

Uh oh!

Copilot AI Apr 5, 2026

Uh oh!

BoyBaykiller commented Apr 5, 2026

Uh oh!

EgorBo commented Apr 5, 2026

Uh oh!

EgorBo commented Apr 5, 2026 •

edited

Loading

Uh oh!

BoyBaykiller commented Apr 5, 2026 •

edited

Loading

Uh oh!

BoyBaykiller commented Apr 10, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

BoyBaykiller commented Apr 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dotnet-policy-service bot commented Apr 4, 2026

Uh oh!

EgorBo commented Apr 5, 2026

Uh oh!

EgorBo commented Apr 5, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Apr 5, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 5, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 5, 2026

Choose a reason for hiding this comment

Uh oh!

BoyBaykiller commented Apr 5, 2026

Uh oh!

EgorBo commented Apr 5, 2026

Uh oh!

EgorBo commented Apr 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

BoyBaykiller commented Apr 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

BoyBaykiller commented Apr 10, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

BoyBaykiller commented Apr 4, 2026 •

edited

Loading

EgorBo commented Apr 5, 2026 •

edited

Loading

BoyBaykiller commented Apr 5, 2026 •

edited

Loading