JIT: Convert multi-target switches to branchless checks by JulieLeeMSFT · Pull Request #124567 · dotnet/runtime

JulieLeeMSFT · 2026-02-18T19:25:47Z

This PR improves JIT codegen for multi-target switch-style tests by converting eligible cases into branchless checks. The goal is to produce optimal codegen that matches the intent of C# pattern matching with or (e.g., x is A or B or C).

When a switch has exactly 2 unique successors (all non-default cases target one block, default targets another), convert it from Switch to an unsigned range comparison (In fgOptimizeSwitchBranches).
When both targets are simple return blocks, fgFoldCondToReturnBlock further folds this into branchless return.

Example

        private static bool IsLetterCategory(UnicodeCategory uc)
        {
            return uc == UnicodeCategory.UppercaseLetter
                 || uc == UnicodeCategory.LowercaseLetter
                 || uc == UnicodeCategory.TitlecaseLetter
                 || uc == UnicodeCategory.ModifierLetter
                 || uc == UnicodeCategory.OtherLetter;
        }

Before

cmp      ecx, 4
ja       SHORT G_M22758_IG05
mov      eax, 1
ret
G_M22758_IG05:
xor      eax, eax
ret

After

cmp      ecx, 4
setbe    al
movzx    rax, al
ret

Details

The comparison direction (GT_LE vs GT_GT) is chosen to favor fall-through to the lexically next block.
Switches with 3 or fewer cases are skipped — the backend's native switch lowering already generates efficient comparison chains for those, and converting early can produce worse code via if-conversion (cmov).
Edge dup counts and profile weights are fixed up after conversion.
Added tests for zero-based and non-zero-based consecutive ranges.

ASMDiffs

SPMI asmdiffs show code size improvements of 83 bytes.

…etbe).

dotnet-policy-service · 2026-02-18T19:26:40Z

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

JulieLeeMSFT · 2026-02-18T19:27:23Z

@EgorBo, PTAL.
CC @dotnet/jit-contrib.

Copilot

Pull request overview

This PR optimizes multi-target switches where all non-default cases target a single block (distinct from the default target) by converting them into unsigned range comparisons. This enables branchless code generation when both targets are simple return blocks, addressing the disparity between pattern matching and explicit equality comparisons noted in issue #123858.

Changes:

Adds logic in fgOptimizeSwitchBranches to detect and optimize switches with exactly 2 unique successors
Transforms such switches into unsigned LE/GT comparisons, choosing the direction to favor fall-through
Adds comprehensive test coverage for both zero-based and non-zero-based consecutive ranges

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
src/coreclr/jit/fgopt.cpp	Implements the switch-to-range-check optimization with edge weight/dup count fixup and profile weight updates
src/tests/JIT/opt/OptSwitchRecognition/optSwitchRecognition.cs	Adds tests for zero-based (0-4) and non-zero-based (10-14) consecutive value ranges with comprehensive boundary cases

src/coreclr/jit/fgopt.cpp

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot

Pull request overview

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

src/coreclr/jit/fgopt.cpp

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot

Pull request overview

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

src/tests/JIT/opt/OptSwitchRecognition/optSwitchRecognition.cs

Copilot · 2026-02-18T19:56:25Z

src/tests/JIT/opt/OptSwitchRecognition/optSwitchRecognition.cs

+    // Test with non-zero-based consecutive values
+    [MethodImpl(MethodImplOptions.NoInlining)]
+    private static bool IsInRange10To14(int val)
+    {
+        return val == 10
+             || val == 11
+             || val == 12
+             || val == 13
+             || val == 14;
+    }
+
+    [Theory]
+    [InlineData(9, false)]
+    [InlineData(10, true)]
+    [InlineData(11, true)]
+    [InlineData(12, true)]
+    [InlineData(13, true)]
+    [InlineData(14, true)]
+    [InlineData(15, false)]
+    [InlineData(-1, false)]
+    public static void TestSwitchToRangeCheckNonZeroBased(int arg1, bool expected) => Assert.Equal(expected, IsInRange10To14(arg1));


This “non-zero-based consecutive values” test uses 10..14, but SwitchRecognition forces minValue=0 when maxValue <= SWITCH_MAX_DISTANCE, so the recognized switch will likely cover 0..14 with many entries targeting the false block. That shape is not eligible for the new switch→range-check simplification (which requires all non-default cases to share one target). If the intent is to exercise the new optimization, consider using a range where maxValue > SWITCH_MAX_DISTANCE (e.g., 100..104) so recognition keeps a non-zero minValue and emits a dense 0..4 switch after subtracting.

Suggested change

// Test with non-zero-based consecutive values

[MethodImpl(MethodImplOptions.NoInlining)]

private static bool IsInRange10To14(int val)

{

return val == 10

|| val == 11

|| val == 12

|| val == 13

|| val == 14;

}

[Theory]

[InlineData(9, false)]

[InlineData(10, true)]

[InlineData(11, true)]

[InlineData(12, true)]

[InlineData(13, true)]

[InlineData(14, true)]

[InlineData(15, false)]

[InlineData(-1, false)]

public static void TestSwitchToRangeCheckNonZeroBased(int arg1, bool expected) => Assert.Equal(expected, IsInRange10To14(arg1));

// Test with non-zero-based consecutive values (range 100 to 104)

[MethodImpl(MethodImplOptions.NoInlining)]

private static bool IsInRange100To104(int val)

{

return val == 100

|| val == 101

|| val == 102

|| val == 103

|| val == 104;

}

[Theory]

[InlineData(99, false)]

[InlineData(100, true)]

[InlineData(101, true)]

[InlineData(102, true)]

[InlineData(103, true)]

[InlineData(104, true)]

[InlineData(105, false)]

[InlineData(-1, false)]

public static void TestSwitchToRangeCheckNonZeroBased(int arg1, bool expected) => Assert.Equal(expected, IsInRange100To104(arg1));

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated no new comments.

EgorBo · 2026-02-19T13:35:22Z

src/coreclr/jit/fgopt.cpp

+        // For small switches with 2 or fewer non-default cases (caseCount <= 3 including the default),
+        // the backend's switch lowering already generates efficient comparison chains. Converting
+        // early can enable if-conversion (cmov) that produces worse code for these cases.
+        if (caseCount <= 3)


Is it the case or just an assumption?

EgorBo · 2026-02-19T13:36:24Z

src/coreclr/jit/fgopt.cpp

        return true;
    }
+    else if (block->GetSwitchTargets()->GetSuccCount() == 2 && block->GetSwitchTargets()->HasDefaultCase() &&
+        (block->IsLIR() || fgNodeThreading == NodeThreading::AllTrees))


Can we only do this for !block->IsLIR() I don't think it's beneficial to support LIR (Lower will handle such switches anyway). So we can simplify code by removing special cases for IsLIR()

EgorBo · 2026-02-19T13:38:45Z

src/coreclr/jit/fgopt.cpp

+                FlowEdge* trueEdge;
+                FlowEdge* falseEdge;
+
+                if (block->NextIs(defaultDest))


We generally avoid relying on lexical order, you can delete this if and unconditionally use the GT_LE path. We used to do it, but it's no longer required.

EgorBo · 2026-02-19T13:42:24Z

src/coreclr/jit/fgopt.cpp

+        if (caseDest != defaultDest)
+        {
+            bool allCasesSameTarget = true;
+            for (unsigned i = 1; i < caseCount - 1; i++)


I think you only need to check defaultEdge->getDupCount() being 1 here

EgorBo · 2026-02-19T13:46:15Z

src/coreclr/jit/fgopt.cpp

+                // The highest case index is caseCount - 2 (caseCount includes the default case).
+                // Using an unsigned GT comparison handles negative values correctly because they
+                // wrap to large unsigned values, making them greater than maxCaseIndex as expected.
+                const unsigned maxCaseIndex = caseCount - 2;


I think you can avoid this math and the explanation by using maxCaseIndex = firstCaseEdge->getDupCount() and then GT_LT instead of GT_LE

EgorBo · 2026-02-19T13:54:07Z

src/coreclr/jit/fgopt.cpp

+                const unsigned trueDupCount = trueEdge->getDupCount();
+                const unsigned falseDupCount = falseEdge->getDupCount();
+
+                if (trueDupCount > 1)


I don't follow this logic. We basically only need to adjust the firstCaseEdge's dupCount - set it to 1. and adjust bbRefs of firstCaseEdge's destionation block.
The defaultEdge and its destionation block is unchanged.

JIT: Convert multi-target switches to branchless range checks (cmp; s…

50f76bf

…etbe).

JulieLeeMSFT added this to the 11.0.0 milestone Feb 18, 2026

Copilot AI review requested due to automatic review settings February 18, 2026 19:25

JulieLeeMSFT added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label Feb 18, 2026

dotnet-policy-service bot assigned JulieLeeMSFT Feb 18, 2026

Copilot started reviewing on behalf of JulieLeeMSFT February 18, 2026 19:26 View session

Copilot AI reviewed Feb 18, 2026

View reviewed changes

src/coreclr/jit/fgopt.cpp Outdated Show resolved Hide resolved

Update src/coreclr/jit/fgopt.cpp

e26174f

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings February 18, 2026 19:41

Copilot started reviewing on behalf of JulieLeeMSFT February 18, 2026 19:42 View session

Copilot AI reviewed Feb 18, 2026

View reviewed changes

src/coreclr/jit/fgopt.cpp Show resolved Hide resolved

src/coreclr/jit/fgopt.cpp Outdated Show resolved Hide resolved

Update src/coreclr/jit/fgopt.cpp

b94a3f6

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings February 18, 2026 19:48

Copilot started reviewing on behalf of JulieLeeMSFT February 18, 2026 19:49 View session

Copilot AI reviewed Feb 18, 2026

View reviewed changes

Update src/tests/JIT/opt/OptSwitchRecognition/optSwitchRecognition.cs

2b0a347

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings February 18, 2026 23:01

Copilot started reviewing on behalf of JulieLeeMSFT February 18, 2026 23:02 View session

Copilot AI reviewed Feb 18, 2026

View reviewed changes

EgorBo reviewed Feb 19, 2026

View reviewed changes

Conversation

JulieLeeMSFT commented Feb 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Example

Before

After

Details

ASMDiffs

Uh oh!

dotnet-policy-service bot commented Feb 18, 2026

Uh oh!

JulieLeeMSFT commented Feb 18, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Copilot AI Feb 18, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

EgorBo Feb 19, 2026

Choose a reason for hiding this comment

Uh oh!

EgorBo Feb 19, 2026

Choose a reason for hiding this comment

Uh oh!

EgorBo Feb 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

EgorBo Feb 19, 2026

Choose a reason for hiding this comment

Uh oh!

EgorBo Feb 19, 2026

Choose a reason for hiding this comment

Uh oh!

EgorBo Feb 19, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Comments

JulieLeeMSFT commented Feb 18, 2026 •

edited

Loading

EgorBo Feb 19, 2026 •

edited

Loading