JIT: don't claim rotates set ZF on xarch by AndyAyersMS · Pull Request #129348 · dotnet/runtime

AndyAyersMS · 2026-06-12T19:45:58Z

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixes#129288.

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixesdotnet#129288. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-12T19:46:54Z

#129346 tracks a more comprehensive approach to a fix. Just doing the surgical thing here.

@tannergooding PTAL
fyi @dotnet/jit-contrib

dotnet-policy-service · 2026-06-12T19:47:22Z

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

Copilot

Pull request overview

Updates the JIT’s xarch flag-setting classification so GT_ROL / GT_ROR are no longer treated as operations that can provide a valid ZF result for zero-comparison lowering, preventing branches from consuming stale ZF after a rotate.

Changes:

Remove GT_ROL / GT_ROR from the xarch “shift group” in GenTree::SupportsSettingZeroFlag().
Add an explanatory comment documenting xarch rotate flag semantics (CF, and OF for 1-bit rotates; not ZF/SF/AF/PF).

AndyAyersMS · 2026-06-13T02:14:47Z

Test failure on xarch may be related. Investigating.

AndyAyersMS · 2026-06-13T14:57:39Z

CI triage for the 4 failing legs — all pre-existing infra, none touch xarch JIT:

Leg	Known issue
Libraries Test Run checked coreclr linux x64 Release	#117138 — checked-build assert in `handletable.cpp:306` (`HndCreateHandle`), crash before any test ran (`Starting: System.Text.Json.Tests` then SIGABRT). `blocking-clean-ci` label.
Libraries Test Run checked coreclr linux_musl arm / arm64 Release	#129339 — `System.Security.Cryptography.Tests` SIGSEGV on Alpine 3.23. arm-only; cannot be caused by an xarch-gated change.
Build browser-wasm windows checked CoreCLR_ReleaseLibraries	#129229 — `wasm-ld : error : undefined symbol: SystemInteropJS_GetManagedStackTrace` in JS interop.

Sibling JIT-codegen-heavy pipelines on the same SHA are green:

runtime-coreclr hardware-intrinsics (1462172) ✅
runtime-coreclr superpmi-diffs (1462171) — partiallySucceeded baseline, no diffs flagged
runtime-dev-innerloop (1462168) ✅
dotnet-linker-tests (1462170) ✅

No retry needed; safe to merge.

AndyAyersMS · 2026-06-13T15:12:13Z

Added a regression test in src/tests/JIT/Regression/JitBlue/Runtime_129288/ (commit 07be5fc).

The test places a bounds check immediately before each of six rotate-then-branch-on-zero patterns (ROR/ROL × {ulong const, uint const}, plus < 1UL and != 0 polarity variants). A .csproj sets DOTNET_TieredCompilation=0 so the buggy FullOpts lowering is exercised on the first call.

Verified locally against windows.x64.Checked:

pre-fix JIT (no rotate split): exit 107 — all 6 subtests get the wrong branch
post-fix JIT (this PR): exit 100 — pass
MinOpts (DOTNET_JitMinOpts=1): exit 100 — confirms the bug is FullOpts-only as documented

Verified to fail (exit 107 = 6 subtests wrong) under the pre-fix JIT and pass (exit 100) under the fix. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-13T15:17:42Z

Revised: dropped the .csproj (and its RequiresProcessIsolation + CLRTestEnvironmentVariable). CI runs JIT regression tests with tiering disabled globally, so the bare .cs file is sufficient — the test infrastructure auto-generates the project. Test source updated to note that.

PR is now back to a clean fix-commit + test-commit pair (force-pushed d629cae).

Copilot

Pull request overview

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

Reword to match the IR: GT_ROL/GT_ROR cover only ROL/ROR (not RCL/RCR), and CF/OF on those are only updated when the rotate count is non-zero. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-14T14:24:56Z

/ba-g unrelated build timeouts

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixes#129288. --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-12T19:45:58Z

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixes#129288.

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixesdotnet#129288. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-12T19:46:54Z

#129346 tracks a more comprehensive approach to a fix. Just doing the surgical thing here.

@tannergooding PTAL
fyi @dotnet/jit-contrib

dotnet-policy-service · 2026-06-12T19:47:22Z

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

Copilot

Pull request overview

Updates the JIT’s xarch flag-setting classification so GT_ROL / GT_ROR are no longer treated as operations that can provide a valid ZF result for zero-comparison lowering, preventing branches from consuming stale ZF after a rotate.

Changes:

Remove GT_ROL / GT_ROR from the xarch “shift group” in GenTree::SupportsSettingZeroFlag().
Add an explanatory comment documenting xarch rotate flag semantics (CF, and OF for 1-bit rotates; not ZF/SF/AF/PF).

AndyAyersMS · 2026-06-13T02:14:47Z

Test failure on xarch may be related. Investigating.

AndyAyersMS · 2026-06-13T14:57:39Z

CI triage for the 4 failing legs — all pre-existing infra, none touch xarch JIT:

Leg	Known issue
Libraries Test Run checked coreclr linux x64 Release	#117138 — checked-build assert in `handletable.cpp:306` (`HndCreateHandle`), crash before any test ran (`Starting: System.Text.Json.Tests` then SIGABRT). `blocking-clean-ci` label.
Libraries Test Run checked coreclr linux_musl arm / arm64 Release	#129339 — `System.Security.Cryptography.Tests` SIGSEGV on Alpine 3.23. arm-only; cannot be caused by an xarch-gated change.
Build browser-wasm windows checked CoreCLR_ReleaseLibraries	#129229 — `wasm-ld : error : undefined symbol: SystemInteropJS_GetManagedStackTrace` in JS interop.

Sibling JIT-codegen-heavy pipelines on the same SHA are green:

runtime-coreclr hardware-intrinsics (1462172) ✅
runtime-coreclr superpmi-diffs (1462171) — partiallySucceeded baseline, no diffs flagged
runtime-dev-innerloop (1462168) ✅
dotnet-linker-tests (1462170) ✅

No retry needed; safe to merge.

AndyAyersMS · 2026-06-13T15:12:13Z

Added a regression test in src/tests/JIT/Regression/JitBlue/Runtime_129288/ (commit 07be5fc).

The test places a bounds check immediately before each of six rotate-then-branch-on-zero patterns (ROR/ROL × {ulong const, uint const}, plus < 1UL and != 0 polarity variants). A .csproj sets DOTNET_TieredCompilation=0 so the buggy FullOpts lowering is exercised on the first call.

Verified locally against windows.x64.Checked:

pre-fix JIT (no rotate split): exit 107 — all 6 subtests get the wrong branch
post-fix JIT (this PR): exit 100 — pass
MinOpts (DOTNET_JitMinOpts=1): exit 100 — confirms the bug is FullOpts-only as documented

Verified to fail (exit 107 = 6 subtests wrong) under the pre-fix JIT and pass (exit 100) under the fix. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-13T15:17:42Z

Revised: dropped the .csproj (and its RequiresProcessIsolation + CLRTestEnvironmentVariable). CI runs JIT regression tests with tiering disabled globally, so the bare .cs file is sufficient — the test infrastructure auto-generates the project. Test source updated to note that.

PR is now back to a clean fix-commit + test-commit pair (force-pushed d629cae).

Copilot

Pull request overview

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

Reword to match the IR: GT_ROL/GT_ROR cover only ROL/ROR (not RCL/RCR), and CF/OF on those are only updated when the rotate count is non-zero. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-14T14:24:56Z

/ba-g unrelated build timeouts

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixes#129288. --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-12T19:45:58Z

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixes#129288.

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixesdotnet#129288. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-12T19:46:54Z

#129346 tracks a more comprehensive approach to a fix. Just doing the surgical thing here.

@tannergooding PTAL
fyi @dotnet/jit-contrib

dotnet-policy-service · 2026-06-12T19:47:22Z

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

Copilot

Pull request overview

Updates the JIT’s xarch flag-setting classification so GT_ROL / GT_ROR are no longer treated as operations that can provide a valid ZF result for zero-comparison lowering, preventing branches from consuming stale ZF after a rotate.

Changes:

Remove GT_ROL / GT_ROR from the xarch “shift group” in GenTree::SupportsSettingZeroFlag().
Add an explanatory comment documenting xarch rotate flag semantics (CF, and OF for 1-bit rotates; not ZF/SF/AF/PF).

AndyAyersMS · 2026-06-13T02:14:47Z

Test failure on xarch may be related. Investigating.

AndyAyersMS · 2026-06-13T14:57:39Z

CI triage for the 4 failing legs — all pre-existing infra, none touch xarch JIT:

Leg	Known issue
Libraries Test Run checked coreclr linux x64 Release	#117138 — checked-build assert in `handletable.cpp:306` (`HndCreateHandle`), crash before any test ran (`Starting: System.Text.Json.Tests` then SIGABRT). `blocking-clean-ci` label.
Libraries Test Run checked coreclr linux_musl arm / arm64 Release	#129339 — `System.Security.Cryptography.Tests` SIGSEGV on Alpine 3.23. arm-only; cannot be caused by an xarch-gated change.
Build browser-wasm windows checked CoreCLR_ReleaseLibraries	#129229 — `wasm-ld : error : undefined symbol: SystemInteropJS_GetManagedStackTrace` in JS interop.

Sibling JIT-codegen-heavy pipelines on the same SHA are green:

runtime-coreclr hardware-intrinsics (1462172) ✅
runtime-coreclr superpmi-diffs (1462171) — partiallySucceeded baseline, no diffs flagged
runtime-dev-innerloop (1462168) ✅
dotnet-linker-tests (1462170) ✅

No retry needed; safe to merge.

AndyAyersMS · 2026-06-13T15:12:13Z

Added a regression test in src/tests/JIT/Regression/JitBlue/Runtime_129288/ (commit 07be5fc).

The test places a bounds check immediately before each of six rotate-then-branch-on-zero patterns (ROR/ROL × {ulong const, uint const}, plus < 1UL and != 0 polarity variants). A .csproj sets DOTNET_TieredCompilation=0 so the buggy FullOpts lowering is exercised on the first call.

Verified locally against windows.x64.Checked:

pre-fix JIT (no rotate split): exit 107 — all 6 subtests get the wrong branch
post-fix JIT (this PR): exit 100 — pass
MinOpts (DOTNET_JitMinOpts=1): exit 100 — confirms the bug is FullOpts-only as documented

Verified to fail (exit 107 = 6 subtests wrong) under the pre-fix JIT and pass (exit 100) under the fix. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-13T15:17:42Z

Revised: dropped the .csproj (and its RequiresProcessIsolation + CLRTestEnvironmentVariable). CI runs JIT regression tests with tiering disabled globally, so the bare .cs file is sufficient — the test infrastructure auto-generates the project. Test source updated to note that.

PR is now back to a clean fix-commit + test-commit pair (force-pushed d629cae).

Copilot

Pull request overview

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

Reword to match the IR: GT_ROL/GT_ROR cover only ROL/ROR (not RCL/RCR), and CF/OF on those are only updated when the rotate count is non-zero. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-14T14:24:56Z

/ba-g unrelated build timeouts

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixes#129288. --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-12T19:45:58Z

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixes#129288.

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixesdotnet#129288. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-12T19:46:54Z

#129346 tracks a more comprehensive approach to a fix. Just doing the surgical thing here.

@tannergooding PTAL
fyi @dotnet/jit-contrib

dotnet-policy-service · 2026-06-12T19:47:22Z

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

Copilot

Pull request overview

Updates the JIT’s xarch flag-setting classification so GT_ROL / GT_ROR are no longer treated as operations that can provide a valid ZF result for zero-comparison lowering, preventing branches from consuming stale ZF after a rotate.

Changes:

Remove GT_ROL / GT_ROR from the xarch “shift group” in GenTree::SupportsSettingZeroFlag().
Add an explanatory comment documenting xarch rotate flag semantics (CF, and OF for 1-bit rotates; not ZF/SF/AF/PF).

AndyAyersMS · 2026-06-13T02:14:47Z

Test failure on xarch may be related. Investigating.

AndyAyersMS · 2026-06-13T14:57:39Z

CI triage for the 4 failing legs — all pre-existing infra, none touch xarch JIT:

Leg	Known issue
Libraries Test Run checked coreclr linux x64 Release	#117138 — checked-build assert in `handletable.cpp:306` (`HndCreateHandle`), crash before any test ran (`Starting: System.Text.Json.Tests` then SIGABRT). `blocking-clean-ci` label.
Libraries Test Run checked coreclr linux_musl arm / arm64 Release	#129339 — `System.Security.Cryptography.Tests` SIGSEGV on Alpine 3.23. arm-only; cannot be caused by an xarch-gated change.
Build browser-wasm windows checked CoreCLR_ReleaseLibraries	#129229 — `wasm-ld : error : undefined symbol: SystemInteropJS_GetManagedStackTrace` in JS interop.

Sibling JIT-codegen-heavy pipelines on the same SHA are green:

runtime-coreclr hardware-intrinsics (1462172) ✅
runtime-coreclr superpmi-diffs (1462171) — partiallySucceeded baseline, no diffs flagged
runtime-dev-innerloop (1462168) ✅
dotnet-linker-tests (1462170) ✅

No retry needed; safe to merge.

AndyAyersMS · 2026-06-13T15:12:13Z

Added a regression test in src/tests/JIT/Regression/JitBlue/Runtime_129288/ (commit 07be5fc).

The test places a bounds check immediately before each of six rotate-then-branch-on-zero patterns (ROR/ROL × {ulong const, uint const}, plus < 1UL and != 0 polarity variants). A .csproj sets DOTNET_TieredCompilation=0 so the buggy FullOpts lowering is exercised on the first call.

Verified locally against windows.x64.Checked:

pre-fix JIT (no rotate split): exit 107 — all 6 subtests get the wrong branch
post-fix JIT (this PR): exit 100 — pass
MinOpts (DOTNET_JitMinOpts=1): exit 100 — confirms the bug is FullOpts-only as documented

Verified to fail (exit 107 = 6 subtests wrong) under the pre-fix JIT and pass (exit 100) under the fix. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-13T15:17:42Z

Revised: dropped the .csproj (and its RequiresProcessIsolation + CLRTestEnvironmentVariable). CI runs JIT regression tests with tiering disabled globally, so the bare .cs file is sufficient — the test infrastructure auto-generates the project. Test source updated to note that.

PR is now back to a clean fix-commit + test-commit pair (force-pushed d629cae).

Copilot

Pull request overview

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

Reword to match the IR: GT_ROL/GT_ROR cover only ROL/ROR (not RCL/RCR), and CF/OF on those are only updated when the rotate count is non-zero. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-14T14:24:56Z

/ba-g unrelated build timeouts

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixes#129288. --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-12T19:45:58Z

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixes#129288.

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixesdotnet#129288. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-12T19:46:54Z

#129346 tracks a more comprehensive approach to a fix. Just doing the surgical thing here.

@tannergooding PTAL
fyi @dotnet/jit-contrib

dotnet-policy-service · 2026-06-12T19:47:22Z

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

Copilot

Pull request overview

Updates the JIT’s xarch flag-setting classification so GT_ROL / GT_ROR are no longer treated as operations that can provide a valid ZF result for zero-comparison lowering, preventing branches from consuming stale ZF after a rotate.

Changes:

Remove GT_ROL / GT_ROR from the xarch “shift group” in GenTree::SupportsSettingZeroFlag().
Add an explanatory comment documenting xarch rotate flag semantics (CF, and OF for 1-bit rotates; not ZF/SF/AF/PF).

AndyAyersMS · 2026-06-13T02:14:47Z

Test failure on xarch may be related. Investigating.

AndyAyersMS · 2026-06-13T14:57:39Z

CI triage for the 4 failing legs — all pre-existing infra, none touch xarch JIT:

Leg	Known issue
Libraries Test Run checked coreclr linux x64 Release	#117138 — checked-build assert in `handletable.cpp:306` (`HndCreateHandle`), crash before any test ran (`Starting: System.Text.Json.Tests` then SIGABRT). `blocking-clean-ci` label.
Libraries Test Run checked coreclr linux_musl arm / arm64 Release	#129339 — `System.Security.Cryptography.Tests` SIGSEGV on Alpine 3.23. arm-only; cannot be caused by an xarch-gated change.
Build browser-wasm windows checked CoreCLR_ReleaseLibraries	#129229 — `wasm-ld : error : undefined symbol: SystemInteropJS_GetManagedStackTrace` in JS interop.

Sibling JIT-codegen-heavy pipelines on the same SHA are green:

runtime-coreclr hardware-intrinsics (1462172) ✅
runtime-coreclr superpmi-diffs (1462171) — partiallySucceeded baseline, no diffs flagged
runtime-dev-innerloop (1462168) ✅
dotnet-linker-tests (1462170) ✅

No retry needed; safe to merge.

AndyAyersMS · 2026-06-13T15:12:13Z

Added a regression test in src/tests/JIT/Regression/JitBlue/Runtime_129288/ (commit 07be5fc).

The test places a bounds check immediately before each of six rotate-then-branch-on-zero patterns (ROR/ROL × {ulong const, uint const}, plus < 1UL and != 0 polarity variants). A .csproj sets DOTNET_TieredCompilation=0 so the buggy FullOpts lowering is exercised on the first call.

Verified locally against windows.x64.Checked:

pre-fix JIT (no rotate split): exit 107 — all 6 subtests get the wrong branch
post-fix JIT (this PR): exit 100 — pass
MinOpts (DOTNET_JitMinOpts=1): exit 100 — confirms the bug is FullOpts-only as documented

Verified to fail (exit 107 = 6 subtests wrong) under the pre-fix JIT and pass (exit 100) under the fix. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-13T15:17:42Z

Revised: dropped the .csproj (and its RequiresProcessIsolation + CLRTestEnvironmentVariable). CI runs JIT regression tests with tiering disabled globally, so the bare .cs file is sufficient — the test infrastructure auto-generates the project. Test source updated to note that.

PR is now back to a clean fix-commit + test-commit pair (force-pushed d629cae).

Copilot

Pull request overview

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

Reword to match the IR: GT_ROL/GT_ROR cover only ROL/ROR (not RCL/RCR), and CF/OF on those are only updated when the rotate count is non-zero. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-14T14:24:56Z

/ba-g unrelated build timeouts

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixes#129288. --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-12T19:45:58Z

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixes#129288.

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixesdotnet#129288. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-12T19:46:54Z

#129346 tracks a more comprehensive approach to a fix. Just doing the surgical thing here.

@tannergooding PTAL
fyi @dotnet/jit-contrib

dotnet-policy-service · 2026-06-12T19:47:22Z

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

Copilot

Pull request overview

Updates the JIT’s xarch flag-setting classification so GT_ROL / GT_ROR are no longer treated as operations that can provide a valid ZF result for zero-comparison lowering, preventing branches from consuming stale ZF after a rotate.

Changes:

Remove GT_ROL / GT_ROR from the xarch “shift group” in GenTree::SupportsSettingZeroFlag().
Add an explanatory comment documenting xarch rotate flag semantics (CF, and OF for 1-bit rotates; not ZF/SF/AF/PF).

AndyAyersMS · 2026-06-13T02:14:47Z

Test failure on xarch may be related. Investigating.

AndyAyersMS · 2026-06-13T14:57:39Z

CI triage for the 4 failing legs — all pre-existing infra, none touch xarch JIT:

Leg	Known issue
Libraries Test Run checked coreclr linux x64 Release	#117138 — checked-build assert in `handletable.cpp:306` (`HndCreateHandle`), crash before any test ran (`Starting: System.Text.Json.Tests` then SIGABRT). `blocking-clean-ci` label.
Libraries Test Run checked coreclr linux_musl arm / arm64 Release	#129339 — `System.Security.Cryptography.Tests` SIGSEGV on Alpine 3.23. arm-only; cannot be caused by an xarch-gated change.
Build browser-wasm windows checked CoreCLR_ReleaseLibraries	#129229 — `wasm-ld : error : undefined symbol: SystemInteropJS_GetManagedStackTrace` in JS interop.

Sibling JIT-codegen-heavy pipelines on the same SHA are green:

runtime-coreclr hardware-intrinsics (1462172) ✅
runtime-coreclr superpmi-diffs (1462171) — partiallySucceeded baseline, no diffs flagged
runtime-dev-innerloop (1462168) ✅
dotnet-linker-tests (1462170) ✅

No retry needed; safe to merge.

AndyAyersMS · 2026-06-13T15:12:13Z

Added a regression test in src/tests/JIT/Regression/JitBlue/Runtime_129288/ (commit 07be5fc).

The test places a bounds check immediately before each of six rotate-then-branch-on-zero patterns (ROR/ROL × {ulong const, uint const}, plus < 1UL and != 0 polarity variants). A .csproj sets DOTNET_TieredCompilation=0 so the buggy FullOpts lowering is exercised on the first call.

Verified locally against windows.x64.Checked:

pre-fix JIT (no rotate split): exit 107 — all 6 subtests get the wrong branch
post-fix JIT (this PR): exit 100 — pass
MinOpts (DOTNET_JitMinOpts=1): exit 100 — confirms the bug is FullOpts-only as documented

Verified to fail (exit 107 = 6 subtests wrong) under the pre-fix JIT and pass (exit 100) under the fix. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-13T15:17:42Z

Revised: dropped the .csproj (and its RequiresProcessIsolation + CLRTestEnvironmentVariable). CI runs JIT regression tests with tiering disabled globally, so the bare .cs file is sufficient — the test infrastructure auto-generates the project. Test source updated to note that.

PR is now back to a clean fix-commit + test-commit pair (force-pushed d629cae).

Copilot

Pull request overview

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

Reword to match the IR: GT_ROL/GT_ROR cover only ROL/ROR (not RCL/RCR), and CF/OF on those are only updated when the rotate count is non-zero. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-14T14:24:56Z

/ba-g unrelated build timeouts

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixes#129288. --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-12T19:45:58Z

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixes#129288.

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixesdotnet#129288. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-12T19:46:54Z

#129346 tracks a more comprehensive approach to a fix. Just doing the surgical thing here.

@tannergooding PTAL
fyi @dotnet/jit-contrib

dotnet-policy-service · 2026-06-12T19:47:22Z

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

Copilot

Pull request overview

Updates the JIT’s xarch flag-setting classification so GT_ROL / GT_ROR are no longer treated as operations that can provide a valid ZF result for zero-comparison lowering, preventing branches from consuming stale ZF after a rotate.

Changes:

Remove GT_ROL / GT_ROR from the xarch “shift group” in GenTree::SupportsSettingZeroFlag().
Add an explanatory comment documenting xarch rotate flag semantics (CF, and OF for 1-bit rotates; not ZF/SF/AF/PF).

AndyAyersMS · 2026-06-13T02:14:47Z

Test failure on xarch may be related. Investigating.

AndyAyersMS · 2026-06-13T14:57:39Z

CI triage for the 4 failing legs — all pre-existing infra, none touch xarch JIT:

Leg	Known issue
Libraries Test Run checked coreclr linux x64 Release	#117138 — checked-build assert in `handletable.cpp:306` (`HndCreateHandle`), crash before any test ran (`Starting: System.Text.Json.Tests` then SIGABRT). `blocking-clean-ci` label.
Libraries Test Run checked coreclr linux_musl arm / arm64 Release	#129339 — `System.Security.Cryptography.Tests` SIGSEGV on Alpine 3.23. arm-only; cannot be caused by an xarch-gated change.
Build browser-wasm windows checked CoreCLR_ReleaseLibraries	#129229 — `wasm-ld : error : undefined symbol: SystemInteropJS_GetManagedStackTrace` in JS interop.

Sibling JIT-codegen-heavy pipelines on the same SHA are green:

runtime-coreclr hardware-intrinsics (1462172) ✅
runtime-coreclr superpmi-diffs (1462171) — partiallySucceeded baseline, no diffs flagged
runtime-dev-innerloop (1462168) ✅
dotnet-linker-tests (1462170) ✅

No retry needed; safe to merge.

AndyAyersMS · 2026-06-13T15:12:13Z

Added a regression test in src/tests/JIT/Regression/JitBlue/Runtime_129288/ (commit 07be5fc).

The test places a bounds check immediately before each of six rotate-then-branch-on-zero patterns (ROR/ROL × {ulong const, uint const}, plus < 1UL and != 0 polarity variants). A .csproj sets DOTNET_TieredCompilation=0 so the buggy FullOpts lowering is exercised on the first call.

Verified locally against windows.x64.Checked:

pre-fix JIT (no rotate split): exit 107 — all 6 subtests get the wrong branch
post-fix JIT (this PR): exit 100 — pass
MinOpts (DOTNET_JitMinOpts=1): exit 100 — confirms the bug is FullOpts-only as documented

Verified to fail (exit 107 = 6 subtests wrong) under the pre-fix JIT and pass (exit 100) under the fix. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-13T15:17:42Z

Revised: dropped the .csproj (and its RequiresProcessIsolation + CLRTestEnvironmentVariable). CI runs JIT regression tests with tiering disabled globally, so the bare .cs file is sufficient — the test infrastructure auto-generates the project. Test source updated to note that.

PR is now back to a clean fix-commit + test-commit pair (force-pushed d629cae).

Copilot

Pull request overview

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

Reword to match the IR: GT_ROL/GT_ROR cover only ROL/ROR (not RCL/RCR), and CF/OF on those are only updated when the rotate count is non-zero. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-14T14:24:56Z

/ba-g unrelated build timeouts

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixes#129288. --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-12T19:45:58Z

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixes#129288.

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixesdotnet#129288. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-12T19:46:54Z

#129346 tracks a more comprehensive approach to a fix. Just doing the surgical thing here.

@tannergooding PTAL
fyi @dotnet/jit-contrib

dotnet-policy-service · 2026-06-12T19:47:22Z

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

Copilot

Pull request overview

Updates the JIT’s xarch flag-setting classification so GT_ROL / GT_ROR are no longer treated as operations that can provide a valid ZF result for zero-comparison lowering, preventing branches from consuming stale ZF after a rotate.

Changes:

Remove GT_ROL / GT_ROR from the xarch “shift group” in GenTree::SupportsSettingZeroFlag().
Add an explanatory comment documenting xarch rotate flag semantics (CF, and OF for 1-bit rotates; not ZF/SF/AF/PF).

AndyAyersMS · 2026-06-13T02:14:47Z

Test failure on xarch may be related. Investigating.

AndyAyersMS · 2026-06-13T14:57:39Z

CI triage for the 4 failing legs — all pre-existing infra, none touch xarch JIT:

Leg	Known issue
Libraries Test Run checked coreclr linux x64 Release	#117138 — checked-build assert in `handletable.cpp:306` (`HndCreateHandle`), crash before any test ran (`Starting: System.Text.Json.Tests` then SIGABRT). `blocking-clean-ci` label.
Libraries Test Run checked coreclr linux_musl arm / arm64 Release	#129339 — `System.Security.Cryptography.Tests` SIGSEGV on Alpine 3.23. arm-only; cannot be caused by an xarch-gated change.
Build browser-wasm windows checked CoreCLR_ReleaseLibraries	#129229 — `wasm-ld : error : undefined symbol: SystemInteropJS_GetManagedStackTrace` in JS interop.

Sibling JIT-codegen-heavy pipelines on the same SHA are green:

runtime-coreclr hardware-intrinsics (1462172) ✅
runtime-coreclr superpmi-diffs (1462171) — partiallySucceeded baseline, no diffs flagged
runtime-dev-innerloop (1462168) ✅
dotnet-linker-tests (1462170) ✅

No retry needed; safe to merge.

AndyAyersMS · 2026-06-13T15:12:13Z

Added a regression test in src/tests/JIT/Regression/JitBlue/Runtime_129288/ (commit 07be5fc).

The test places a bounds check immediately before each of six rotate-then-branch-on-zero patterns (ROR/ROL × {ulong const, uint const}, plus < 1UL and != 0 polarity variants). A .csproj sets DOTNET_TieredCompilation=0 so the buggy FullOpts lowering is exercised on the first call.

Verified locally against windows.x64.Checked:

pre-fix JIT (no rotate split): exit 107 — all 6 subtests get the wrong branch
post-fix JIT (this PR): exit 100 — pass
MinOpts (DOTNET_JitMinOpts=1): exit 100 — confirms the bug is FullOpts-only as documented

Verified to fail (exit 107 = 6 subtests wrong) under the pre-fix JIT and pass (exit 100) under the fix. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-13T15:17:42Z

Revised: dropped the .csproj (and its RequiresProcessIsolation + CLRTestEnvironmentVariable). CI runs JIT regression tests with tiering disabled globally, so the bare .cs file is sufficient — the test infrastructure auto-generates the project. Test source updated to note that.

PR is now back to a clean fix-commit + test-commit pair (force-pushed d629cae).

Copilot

Pull request overview

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

Reword to match the IR: GT_ROL/GT_ROR cover only ROL/ROR (not RCL/RCR), and CF/OF on those are only updated when the rotate count is non-zero. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-06-14T14:24:56Z

/ba-g unrelated build timeouts

ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixes#129288. --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

JIT: don't claim rotates set ZF on xarch
84efbfe
ROL/ROR only update CF (and OF in the 1-bit form); SF/ZF/AF/PF are not affected for any count, so they don't belong in SupportsSettingZeroFlag's shift group. Fixesdotnet#129288. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

CopilotAI review requested due to automatic review settings June 12, 2026 19:45

github-actionsBot added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label Jun 12, 2026

dotnet-policy-serviceBot assigned AndyAyersMS Jun 12, 2026

Copilotstarted reviewing on behalf of AndyAyersMS June 12, 2026 19:46 View session

CopilotAI reviewed Jun 12, 2026
View reviewed changes

Comment threadsrc/coreclr/jit/gentree.cpp Outdated

JulieLeeMSFT approved these changes Jun 12, 2026
View reviewed changes

tannergooding approved these changes Jun 12, 2026
View reviewed changes

JIT: add regression test fordotnet#129288
d629cae
Verified to fail (exit 107 = 6 subtests wrong) under the pre-fix JIT and pass (exit 100) under the fix. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

CopilotAI review requested due to automatic review settings June 13, 2026 15:17

AndyAyersMS force-pushed the fix-129288-rotate-zf branch from 07be5fc to d629caeCompare June 13, 2026 15:17

Copilotstarted reviewing on behalf of AndyAyersMS June 13, 2026 15:17 View session

CopilotAI reviewed Jun 13, 2026
View reviewed changes

Comment threadsrc/coreclr/jit/gentree.cpp Outdated

JIT: clarify rotate-flags comment per review feedback
31b9e40
Reword to match the IR: GT_ROL/GT_ROR cover only ROL/ROR (not RCL/RCR), and CF/OF on those are only updated when the rotate count is non-zero. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

tannergooding approved these changes Jun 13, 2026
View reviewed changes

AndyAyersMS mentioned this pull request Jun 13, 2026
JIT: fix FitsIn<int32_t> assert in BitOperations.Rotate{Left,Right} const-fold #129136
Closed

AndyAyersMS enabled auto-merge (squash) June 13, 2026 18:43

build-analysisBot mentioned this pull request Jun 13, 2026
tracing/runtimeeventsource/nativeruntimeeventsource/nativeruntimeeventsource failing in CI #90605
Open

AndyAyersMS merged commit 3a87981 into dotnet:mainJun 14, 2026
137 of 145 checks passed

AndyAyersMS mentioned this pull request Jun 14, 2026
JIT: STRESS_OPT_REPEAT silent miscompile produces wrong-direction branch into divide-by-zero #129386
Closed

dotnet-maestroBot mentioned this pull request Jun 15, 2026
[main] Source code updates from dotnet/runtime dotnet/dotnet#7196
Merged

CopilotAI mentioned this pull request Jun 16, 2026
JIT: fix RBO constant relop truth evaluation under opt-repeat #129487
Closed

dotnet-milestone-botBot added this to the 11.0-preview6 milestone Jun 17, 2026

tannergooding mentioned this pull request Jul 16, 2026
Wire up orphaned Runtime_129288 regression test #130836
Merged

github-actionsBot locked and limited conversation to collaborators Jul 18, 2026