Preserve mask element granularity in SIMD transformations by tannergooding · Pull Request #133708 · dotnet/runtime

tannergooding · 2026-09-11T17:07:34Z

Summary

Preserve mask producer/consumer element granularity during SIMD folding and lowering:

Prove blend masks are uniform at the consumer's element width before replacing sign-bit selection with AND/AND_NOT.
Recompute mask lane count when comparison fallback changes element type. Restrict direct constant-mask shortcuts to the zero/all-bits-set cases they implement.
Widen embedded broadcasts only when their scalar is constant. Preserve runtime broadcast load widths, including during PTESTM containment. Correct total mask lane accounting for wider vectors and duplicate constant bits with unsigned arithmetic.
Preserve SVE mask-to-vector conversions when producer and consumer element widths differ.

Add six focused xarch regression cases and a compact SVE regression for the shared folding guard. Existing valid same-width optimizations remain enabled.

Fixes #133548
Fixes #133552
Fixes #133555

Validation

Built Checked/Release x64 JITs, Checked cross-JITs, the merged regression runner, and the ARM64 SVE regression project.
Each of the six final xarch cases fails against the preserved original JIT and passes with the fixes on Checked and Release, with actual AVX512BW/VL execution. The final cases also pass with AVX disabled (SSE4.1-only). Broader AVX2, MinOpts, and portable software-fallback coverage passed before trimming redundant test variants.
SVE before/after cross-codegen confirms vector operations are retained for reinterpreted masks and the same-width predicate optimization remains. SVE runtime execution was unavailable on the x64 host.
Local BenchmarkDotNet checks retained identical instructions and code sizes for homogeneous blend (28 bytes), constant-broadcast select (44 bytes), and same-width comparison (33 bytes). Timings were noisy; no speedup claim. These measurements predate the rebase onto current main.
JIT formatting passed before publication.

Note

This PR description was drafted by GitHub Copilot.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

azure-pipelines · 2026-09-11T17:08:23Z

Azure Pipelines: Successfully started running 5 pipeline(s). 11 pipeline(s) were filtered out due to trigger conditions. There may be pipelines that require an authorized user to comment /azp run to run.

dotnet-policy-service · 2026-09-11T17:08:56Z

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

Copilot

Copilot review overview

🟡 Changes recommended

A runtime broadcast can still be widened unsafely through the blend rewrite path.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Review tier: Lite
Findings: 1

New issues introduced by this change (1)

Severity	Finding
	`src/coreclr/jit/gentree.cpp` — Do not bypass broadcast validation for the rationalizer caller View comment

What changed in this PR

This PR preserves SIMD mask element granularity across xarch lowering and SVE folding.

Changes:

Tightens mask and comparison optimizations.
Preserves runtime broadcast widths.
Adds xarch and SVE regression coverage.

File	Description
`src/tests/JIT/Regression_ro_2/Runtime_133555.cs`	Tests reinterpreted comparison masks.
`src/tests/JIT/Regression_ro_2/Runtime_133552.cs`	Tests runtime and constant broadcast behavior.
`src/tests/JIT/Regression_ro_2/Runtime_133548.cs`	Tests blend mask granularity.
`src/tests/JIT/opt/SVE/PredicateInstructions.cs`	Tests SVE mask granularity preservation.
`src/coreclr/jit/lowerxarch.cpp`	Updates xarch mask and broadcast lowering.
`src/coreclr/jit/gentree.cpp`	Guards broadcast widening and SVE mask folding.

Copilot

Copilot review overview

🔵 Needs a closer look

An unresolved moderate issue remains in vector-constant broadcast widening in gentree.cpp.

Review tier: Lite
Findings: None

Resolved findings (1)

Do not bypass broadcast validation for the rationalizer caller

tannergooding · 2026-09-15T01:39:42Z

/ba-g unrelated and known issues

) ## Summary Preserve mask producer/consumer element granularity during SIMD folding and lowering: - Prove blend masks are uniform at the consumer's element width before replacing sign-bit selection with AND/AND_NOT. - Recompute mask lane count when comparison fallback changes element type. Restrict direct constant-mask shortcuts to the zero/all-bits-set cases they implement. - Widen embedded broadcasts only when their scalar is constant. Preserve runtime broadcast load widths, including during PTESTM containment. Correct total mask lane accounting for wider vectors and duplicate constant bits with unsigned arithmetic. - Preserve SVE mask-to-vector conversions when producer and consumer element widths differ. Add six focused xarch regression cases and a compact SVE regression for the shared folding guard. Existing valid same-width optimizations remain enabled. Fixes dotnet#133548 Fixes dotnet#133552 Fixes dotnet#133555 ## Validation - Built Checked/Release x64 JITs, Checked cross-JITs, the merged regression runner, and the ARM64 SVE regression project. - Each of the six final xarch cases fails against the preserved original JIT and passes with the fixes on Checked and Release, with actual AVX512BW/VL execution. The final cases also pass with AVX disabled (SSE4.1-only). Broader AVX2, MinOpts, and portable software-fallback coverage passed before trimming redundant test variants. - SVE before/after cross-codegen confirms vector operations are retained for reinterpreted masks and the same-width predicate optimization remains. SVE runtime execution was unavailable on the x64 host. - Local BenchmarkDotNet checks retained identical instructions and code sizes for homogeneous blend (28 bytes), constant-broadcast select (44 bytes), and same-width comparison (33 bytes). Timings were noisy; no speedup claim. These measurements predate the rebase onto current main. - JIT formatting passed before publication. > [!NOTE] > This PR description was drafted by GitHub Copilot. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

tannergooding · 2026-09-18T19:15:13Z

/backport to release/11.0

Note

This backport request was posted by GitHub Copilot at the user's direction.

github-actions · 2026-09-18T19:15:21Z

Started backporting to release/11.0 (link to workflow run)

@tannergooding

…ons (#134213) Backport of #133708 to release/11.0 /cc @tannergooding ## Customer Impact - [ ] Customer reported - [x] Found internally This fixes incorrect SIMD results and JIT assertion failures in #133548 and #133552. Both issues reproduce on .NET 11 RC2 but not .NET 8, 9, or 10, so .NET 11 would otherwise ship with regressions in SIMD mask folding and lowering. Because the fixes share the same mask-granularity invariants, this backport also includes #133555, which affects .NET 9 through .NET 11. ## Regression - [x] Yes - [ ] No The #133548 and #133552 failures were introduced during the .NET 11 development cycle; the exact introducing change was not identified. #133555 predates .NET 11 and is included because it is fixed by the same change. ## Testing The source PR added focused regressions for all three xarch failures and SVE coverage for the shared folding guard. The xarch cases failed with the original JIT and passed after the fix in Checked and Release builds, including native AVX-512 execution and reduced-ISA configurations. SVE before/after cross-codegen preserved the required vector operations. Representative valid same-width optimizations retained identical instruction counts and code sizes. ## Risk Low. The changes are narrowly scoped element-granularity guards and invariant corrections that preserve existing valid SIMD transformations. Focused regression coverage exercises the affected vector widths and ISA configurations, SVE cross-codegen validates the shared guard, and representative unaffected optimizations retained identical instruction counts and code sizes. **IMPORTANT**: If this backport is for a servicing release, please verify that: - For .NET 8 and .NET 9: The PR target branch is `release/X.0-staging`, not `release/X.0`. - For .NET 10+: The PR target branch is `release/X.0` (no `-staging` suffix). ## Package authoring no longer needed in .NET 9 **IMPORTANT**: Starting with .NET 9, you no longer need to edit a NuGet package's csproj to enable building and bump the version. Keep in mind that we still need package authoring in .NET 8 and older versions. > [!NOTE] > This backport template was completed by GitHub Copilot at the user's direction. Co-authored-by: Tanner Gooding <tagoo@outlook.com> Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Preserve mask element granularity in SIMD transformations

af7a07d

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>

Copilot AI lite review requested due to automatic review settings September 11, 2026 17:07

dotnet-policy-service Bot assigned tannergooding Sep 11, 2026

github-actions Bot added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label Sep 11, 2026

Copilot started reviewing on behalf of tannergooding September 11, 2026 17:08 View session

Copilot AI reviewed Sep 11, 2026

View reviewed changes

Comment thread src/coreclr/jit/gentree.cpp

This was referenced Sep 11, 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 requested a review from EgorBo September 12, 2026 02:03

Merge branch 'main' into tannergooding-simd-mask-granularity-fixes

f03167f

Copilot AI review requested due to automatic review settings September 13, 2026 17:35

Copilot started reviewing on behalf of tannergooding September 13, 2026 17:36 View session

Copilot AI reviewed Sep 13, 2026

View reviewed changes

build-analysis Bot mentioned this pull request Sep 13, 2026

github60486 failed with NRE at Program.ValidateRecursiveCurrentMethod(Int32 depth) #133798

Closed

EgorBo approved these changes Sep 14, 2026

View reviewed changes

tannergooding merged commit ae47da8 into dotnet:main Sep 15, 2026
140 of 143 checks passed

tannergooding deleted the tannergooding-simd-mask-granularity-fixes branch September 15, 2026 01:42

dotnet-maestro Bot mentioned this pull request Sep 16, 2026

[main] Source code updates from dotnet/runtime dotnet/dotnet#9402

Merged

dotnet-milestone-bot Bot added this to the 12.0-preview1 milestone Sep 16, 2026

This was referenced Sep 18, 2026

JIT: (bug) BlendVariable on float ignores the sign-bit rule when the mask is byte-granular #133548

Closed

JIT: (bug) an embedded broadcast of a runtime value hits a JIT assert when it is widened for masking #133552

Closed

github-actions Bot mentioned this pull request Sep 18, 2026

[release/11.0] Preserve mask element granularity in SIMD transformations #134213

Merged

4 tasks

Navigation Menu

Preserve mask element granularity in SIMD transformations - #133708

Preserve mask element granularity in SIMD transformations#133708
tannergooding merged 2 commits into
dotnet:mainfrom
tannergooding:tannergooding-simd-mask-granularity-fixes

tannergooding commented Sep 11, 2026

Uh oh!

azure-pipelines Bot commented Sep 11, 2026

Uh oh!

dotnet-policy-service Bot commented Sep 11, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

tannergooding commented Sep 15, 2026

Uh oh!

Uh oh!

tannergooding commented Sep 18, 2026

Uh oh!

github-actions Bot commented Sep 18, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

tannergooding commented Sep 11, 2026

Summary

Validation

Uh oh!

azure-pipelines Bot commented Sep 11, 2026

Uh oh!

dotnet-policy-service Bot commented Sep 11, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot review overview

🟡 Changes recommended

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot review overview

🔵 Needs a closer look

Uh oh!

tannergooding commented Sep 15, 2026

Uh oh!

Uh oh!

tannergooding commented Sep 18, 2026

Uh oh!

github-actions Bot commented Sep 18, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants