Update GetRangeFromAssertions to handle some basic TYP_LONG scenarios where it FitsIn<int32_t> by tannergooding · Pull Request #128906 · dotnet/runtime

tannergooding · 2026-06-02T16:45:32Z

This is an alternative to #128676. It needs confirmation that the diffs/TP is acceptable and may require a few iterations or pulling back prior to it being ready for review.

dotnet-policy-service · 2026-06-02T16:46:58Z

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

Copilot

Pull request overview

This PR broadens assertion-based range derivation in the JIT so it can reason about some TYP_LONG value numbers when the resulting values are known to fit in int32, and wires the updated API through rangecheck and assertion propagation to enable additional folding / bounds-check reasoning.

Changes:

Update ValueNumStore::IsVNIntegralConstant to coerce constants as int64_t, allowing TYP_LONG constants that fit to be recognized as int32 constants.
Extend RangeCheck::GetRangeFromAssertions/worker to accept an explicit var_types and add limited handling for TYP_LONG scenarios (notably RSZ/RSH shift cases and other VN ops).
Update assertion propagation and range analysis callsites to pass the expression type and tolerate unknown ranges where TYP_LONG can’t be represented as an int32-based Range.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
src/coreclr/jit/valuenum.h	Enables integral-constant extraction from `TYP_LONG` VNs via `int64_t` coercion.
src/coreclr/jit/rangecheck.h	Updates `GetRangeFromAssertions` signature and adds `Range::IsUnknown()` helper.
src/coreclr/jit/rangecheck.cpp	Implements the new typed assertion-range logic and extends range computation to consult it in more cases.
src/coreclr/jit/assertionprop.cpp	Adapts assertion-prop folding to the new API and to possibly-unknown ranges for wider types.

Copilot

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

Comments suppressed due to low confidence (1)

src/coreclr/jit/rangecheck.cpp:796

In VNF_Cast handling, when result is non-constant (e.g., casting to/from types that Range can’t represent) the code unconditionally propagates castOpRange if it’s a constant range. This is unsound for sign-changing casts (e.g., int -> uint, uint -> long) when the operand range can include negatives: the cast changes negative values to large positives, but castOpRange would still contain negatives and exclude the large values.

This can cause incorrect tightening and downstream folding/removal based on a range that doesn’t describe the cast result.

                // Now see if we can do better by looking at the cast source.
                // if its range is within the castTo range, we can use that (and the cast is basically a no-op).
                if (varTypeIsIntegral(arg0Typ))
                {
                    Range castOpRange =
                        GetRangeFromAssertionsWorker(comp, arg0Typ, arg0VN, assertions, --budget, visited);

                    if (castOpRange.IsConstantRange())
                    {
                        if (!result.IsConstantRange())
                        {
                            result = castOpRange;
                        }
                        else if ((castOpRange.LowerLimit().GetConstant() >= result.LowerLimit().GetConstant()) &&
                                 (castOpRange.UpperLimit().GetConstant() <= result.UpperLimit().GetConstant()))
                        {
                            result = castOpRange;
                        }
                    }

tannergooding · 2026-06-02T23:32:53Z

~~Extracted two parts out of this into #128922 and #128923 to get better TP and diff metrics to decide how much to preserve or not.~~ Most of this change requires things working together to fully lightup, otherwise we only get relatively small diffs for any singular portion.

…ent/symbolic cases (#128922) This is a smaller change from #128906 that doesn't involve more complex handling around `TYP_LONG`

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

… where it FitsIn<int32_t>

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

…ll GetRangeFromAssertions

tannergooding · 2026-06-04T10:25:29Z

/azp run fuzzlyn, runtime-coreclr jitstress, runtime-coreclr jitstressregs

azure-pipelines · 2026-06-04T10:25:46Z

Azure Pipelines successfully started running 3 pipeline(s).

tannergooding · 2026-06-04T10:30:43Z

Diffs are here.

Linux Arm64

Overall (-476,080 bytes)
FullOpts (-476,080 bytes)

Linux x64

Overall (-432,892 bytes)
FullOpts (-432,892 bytes)

Windows Arm64

Overall (-393,596 bytes)
FullOpts (-393,596 bytes)

Windows x64

Overall (-260,911 bytes)
FullOpts (-260,911 bytes)

Linux arm

Overall (-83,344 bytes)
FullOpts (-83,344 bytes)

Windows x86

Overall (-46,131 bytes)
FullOpts (-46,131 bytes)

Linux x64

Overall (+0.05% to +0.18%)
FullOpts (+0.05% to +0.20%)

Windows arm64

Overall (+0.08% to +0.25%)
FullOpts (+0.08% to +0.26%)

Windows x64

Overall (+0.07% to +0.25%)
FullOpts (+0.07% to +0.25%)

tannergooding · 2026-06-04T10:42:52Z

Overall the diffs are teh standard set you'd expect. We have places that change from sign-extension to zero-extension because we know its never negative and we have removal of code that is now provably dead, unreachable, or unnecessary.

This lights up for places where we're explicitly using long or nint on 64-bit platforms, including places like TensorPrimitives, BigInteger, and the various SpanHelpers where we extend the length up to nint to do the rest of the algorithm.

EgorBo · 2026-06-04T11:10:57Z

                    }
+                    else if ((elementCount == 32) && varTypeIsLong(rangeType))
+                    {
+                        return {SymbolicIntegerValue::Zero, UpperBoundForType(TYP_UINT)};


why not just rangeType = TYP_UINT like above?

EgorBo · 2026-06-04T11:11:34Z

-            *isKnownNonNegative = true;
-        }
-        if ((rng.LowerLimit().GetConstant() > 0) || (rng.UpperLimit().GetConstant() < 0))
+        Range rng = RangeCheck::GetRangeFromAssertions(this, tree->TypeGet(), treeVN, assertions);


I really don't like the fact we need to pass type. It should be evaluated from VN, shouldn't it?

EgorBo · 2026-06-04T11:13:26Z

-        if ((rng.LowerLimit().GetConstant() > 0) || (rng.UpperLimit().GetConstant() < 0))
+        Range rng = RangeCheck::GetRangeFromAssertions(this, tree->TypeGet(), treeVN, assertions);
+
+        if (rng.IsConstantRange())


it's unfortunate we broke the contract for GetRangeFromAssertions to always return a constant range. My opinion we shouldn't do it and instead either upgrade Range to TYP_LONG or introduce a new GetRangeFromAssertions for 64-bit ranges

EgorBo · 2026-06-04T11:22:48Z

                }
+                else
+                {
+                    // TODO: We could return `0, keUnknown` for `elementCount == 32` if the result is TYP_LONG


I still don't understand what 0, keUnknown is supposed to mean, keUnknown implies it can be something that overflows making the range invalid

EgorBo · 2026-06-04T11:24:28Z


+    bool IsUnknown() const
+    {
+        return lLimit.IsUnknown() && uLimit.IsUnknown();


seems unused

Copilot AI review requested due to automatic review settings June 2, 2026 16:45

Copilot started reviewing on behalf of tannergooding June 2, 2026 16:45 View session

dotnet-policy-service Bot assigned tannergooding Jun 2, 2026

github-actions Bot added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label Jun 2, 2026

Copilot AI reviewed Jun 2, 2026

View reviewed changes

Comment thread src/coreclr/jit/rangecheck.cpp

Copilot AI review requested due to automatic review settings June 2, 2026 19:19

Copilot started reviewing on behalf of tannergooding June 2, 2026 19:20 View session

Copilot AI reviewed Jun 2, 2026

View reviewed changes

Comment thread src/coreclr/jit/rangecheck.h

This was referenced Jun 2, 2026

Have ComputeRange call into GetRangeFromAssertions for non dependent/symbolic cases #128922

Merged

Improve handling of VNF_Cast in GetRangeFromAssertionsWorker #128923

Closed

tannergooding added a commit that referenced this pull request Jun 3, 2026

Have ComputeRange call into GetRangeFromAssertions for non depend…

3752108

…ent/symbolic cases (#128922) This is a smaller change from #128906 that doesn't involve more complex handling around `TYP_LONG`

Copilot AI review requested due to automatic review settings June 4, 2026 00:34

tannergooding force-pushed the better-rngchk2 branch from 64aae0c to 2e4ba16 Compare June 4, 2026 00:34

Copilot started reviewing on behalf of tannergooding June 4, 2026 00:34 View session

Copilot AI reviewed Jun 4, 2026

View reviewed changes

Comment thread src/coreclr/jit/rangecheck.cpp

Update GetRangeFromAssertions to handle some basic TYP_LONG scenarios…

f72c56d

… where it FitsIn<int32_t>

tannergooding force-pushed the better-rngchk2 branch from 2e4ba16 to f72c56d Compare June 4, 2026 02:38

This was referenced Jun 4, 2026

slow macOS - "##[error]The job running on agent Azure Pipelines 9 ran longer than the maximum time of 60 minutes." dotnet/dnceng#1883

Open

The Operation will be canceled. The next steps may not contain expected logs. dotnet/dnceng#3008

Open

tannergooding marked this pull request as ready for review June 4, 2026 10:07

Copilot AI review requested due to automatic review settings June 4, 2026 10:07

Copilot started reviewing on behalf of tannergooding June 4, 2026 10:08 View session

Copilot AI reviewed Jun 4, 2026

View reviewed changes

Comment thread src/coreclr/jit/rangecheck.cpp

Don't change to the more expensive budget when having ComputeRange ca…

492a936

…ll GetRangeFromAssertions

EgorBo reviewed Jun 4, 2026

View reviewed changes

Conversation

tannergooding commented Jun 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dotnet-policy-service Bot commented Jun 2, 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!

tannergooding commented Jun 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

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 left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

tannergooding commented Jun 4, 2026

Uh oh!

azure-pipelines Bot commented Jun 4, 2026

Uh oh!

tannergooding commented Jun 4, 2026

Linux Arm64

Linux x64

Windows Arm64

Windows x64

Linux arm

Windows x86

Linux x64

Windows arm64

Windows x64

Uh oh!

tannergooding commented Jun 4, 2026

Uh oh!

EgorBo Jun 4, 2026

Choose a reason for hiding this comment

Uh oh!

EgorBo Jun 4, 2026

Choose a reason for hiding this comment

Uh oh!

EgorBo Jun 4, 2026

Choose a reason for hiding this comment

Uh oh!

EgorBo Jun 4, 2026

Choose a reason for hiding this comment

Uh oh!

EgorBo Jun 4, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

tannergooding commented Jun 2, 2026 •

edited

Loading

tannergooding commented Jun 2, 2026 •

edited

Loading