Fix ICE in const eval of packed SIMD types with non-power-of-two element counts by enthropy7 · Pull Request #151628 · rust-lang/rust

enthropy7 · 2026-01-25T03:10:59Z

const evaluation of packed SIMD types with non-power-of-two element counts (like Simd<_, 3>) was hitting an ICE. the issue was in check_simd_ptr_alignment - it asserted that backend_repr must be BackendRepr::SimdVector, but for packed SIMD types with non-power-of-two counts the compiler uses BackendRepr::Memory instead (as mentioned in rustc_abi/src/layout.rs:1511).

was fixed by making check_simd_ptr_alignment accept both BackendRepr::SimdVector and BackendRepr::Memory for SIMD types. added a check to ensure we're dealing with a SIMD type, and the alignment logic works the same for both representations.

also i added a test that reproduces the original ICE.

rustbot · 2026-01-25T03:11:03Z

Some changes occurred to the CTFE machinery

cc @RalfJung, @oli-obk, @lcnr

Some changes occurred to the CTFE / Miri interpreter

cc @rust-lang/miri

rustbot · 2026-01-25T03:11:05Z

r? @nnethercote

rustbot has assigned @nnethercote.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Kivooeo · 2026-01-25T12:07:35Z

cc @folkertdev

folkertdev · 2026-01-25T12:12:19Z

I discussed this with @workingjubilee, and the true underlying issue is more complicated. Really we should not be using BackendRepr::Memory for SIMD types, that is a historical accident.

However we might want this fix in the meantime (though with a FIXME: remove the BackendRepr::Memory case when SIMD vectors are always passed as BackendRepr::SimdVector or something like that).

enthropy7 · 2026-01-25T12:17:37Z

I discussed this with @workingjubilee, and the true underlying issue is more complicated. Really we should not be using BackendRepr::Memory for SIMD types, that is a historical accident.

However we might want this fix in the meantime (though with a FIXME: remove the BackendRepr::Memory case when SIMD vectors are always passed as BackendRepr::SimdVector or something like that).

so, what should we do now? implement this fix and add a comment?

folkertdev · 2026-01-25T12:36:26Z

Let's see if Jubilee can give an indication on how quickly this can be fixed.

You can add the comment already. If nothing else it'll re-run CI and hopefully pass this time.

enthropy7 · 2026-01-25T12:37:52Z

Let's see if Jubilee can give an indication on how quickly this can be fixed.

You can add the comment already. If nothing else it'll re-run CI and hopefully pass this time.

got it, thanks. also i would like to work on fix in future, if needed :)

rustbot · 2026-01-25T12:40:58Z

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

RalfJung · 2026-01-25T19:02:50Z

r? me

RalfJung

This LGTM. Let's see what @workingjubilee says.

View changes since this review

RalfJung · 2026-01-25T19:46:29Z

tests/ui/consts/const-eval/simd/simd-packed-non-pow2.rs

Please add a link to the issue that is fixed by this PR as a comment inside this test.

…ngth Packed SIMD types with non-power-of-two element counts use BackendRepr::Memory instead of BackendRepr::SimdVector. The check_simd_ptr_alignment function now handles both cases to prevent ICEs when evaluating SIMD intrinsics in const contexts.

RalfJung · 2026-02-21T16:28:59Z

I'll approve now; @workingjubilee please chime in asynchronously if you think that's a mistake.
@bors r+

rust-bors · 2026-02-21T16:29:01Z

📌 Commit c45a513 has been approved by RalfJung

It is now in the queue for this repository.

… r=RalfJung Fix ICE in const eval of packed SIMD types with non-power-of-two element counts fixes rust-lang#151537 const evaluation of packed SIMD types with non-power-of-two element counts (like `Simd<_, 3>`) was hitting an ICE. the issue was in `check_simd_ptr_alignment` - it asserted that `backend_repr` must be `BackendRepr::SimdVector`, but for packed SIMD types with non-power-of-two counts the compiler uses `BackendRepr::Memory` instead (as mentioned in `rustc_abi/src/layout.rs:1511`). was fixed by making `check_simd_ptr_alignment` accept both `BackendRepr::SimdVector` and `BackendRepr::Memory` for SIMD types. added a check to ensure we're dealing with a SIMD type, and the alignment logic works the same for both representations. also i added a test that reproduces the original ICE.

Rollup of 14 pull requests Successful merges: - #150468 (rustc_target: callconv: powerpc64: Use the ABI set in target options instead of guessing) - #151628 (Fix ICE in const eval of packed SIMD types with non-power-of-two element counts) - #151740 (Fix short backtraces from stripped executables) - #151871 (don't use env with infer vars) - #152591 (Simplify internals of `{Rc,Arc}::default`) - #152865 (Fixed ByteStr not padding within its Display trait when no specific alignment is mentioned) - #152908 (Enable rust.remap-debuginfo in the dist profile) - #152705 (Test(lib/win/proc): Skip `raw_attributes` doctest under Win7) - #152767 (fix typo in `carryless_mul` macro invocation) - #152837 (fix(codegen): Use `body_codegen_attrs` For Caller In `adjust_target_feature_sig`) - #152871 (Fix warnings in rs{begin,end}.rs files) - #152921 (Add build.rustdoc option to bootstrap config) - #152933 (Start migration for `LintDiagnostic` items by adding API and migrating `LinkerOutput` lint) - #152937 (remove unneeded reboxing)

… r=RalfJung Fix ICE in const eval of packed SIMD types with non-power-of-two element counts fixes rust-lang#151537 const evaluation of packed SIMD types with non-power-of-two element counts (like `Simd<_, 3>`) was hitting an ICE. the issue was in `check_simd_ptr_alignment` - it asserted that `backend_repr` must be `BackendRepr::SimdVector`, but for packed SIMD types with non-power-of-two counts the compiler uses `BackendRepr::Memory` instead (as mentioned in `rustc_abi/src/layout.rs:1511`). was fixed by making `check_simd_ptr_alignment` accept both `BackendRepr::SimdVector` and `BackendRepr::Memory` for SIMD types. added a check to ensure we're dealing with a SIMD type, and the alignment logic works the same for both representations. also i added a test that reproduces the original ICE.

Rollup of 15 pull requests Successful merges: - #149366 (GVN: consider constants of primitive types as deterministic) - #150468 (rustc_target: callconv: powerpc64: Use the ABI set in target options instead of guessing) - #151628 (Fix ICE in const eval of packed SIMD types with non-power-of-two element counts) - #151871 (don't use env with infer vars) - #152591 (Simplify internals of `{Rc,Arc}::default`) - #152657 (std: move `exit` out of PAL) - #152865 (Fixed ByteStr not padding within its Display trait when no specific alignment is mentioned) - #152908 (Enable rust.remap-debuginfo in the dist profile) - #152705 (Test(lib/win/proc): Skip `raw_attributes` doctest under Win7) - #152767 (fix typo in `carryless_mul` macro invocation) - #152837 (fix(codegen): Use `body_codegen_attrs` For Caller In `adjust_target_feature_sig`) - #152871 (Fix warnings in rs{begin,end}.rs files) - #152921 (Add build.rustdoc option to bootstrap config) - #152933 (Start migration for `LintDiagnostic` items by adding API and migrating `LinkerOutput` lint) - #152937 (remove unneeded reboxing)

… r=RalfJung Fix ICE in const eval of packed SIMD types with non-power-of-two element counts fixes rust-lang#151537 const evaluation of packed SIMD types with non-power-of-two element counts (like `Simd<_, 3>`) was hitting an ICE. the issue was in `check_simd_ptr_alignment` - it asserted that `backend_repr` must be `BackendRepr::SimdVector`, but for packed SIMD types with non-power-of-two counts the compiler uses `BackendRepr::Memory` instead (as mentioned in `rustc_abi/src/layout.rs:1511`). was fixed by making `check_simd_ptr_alignment` accept both `BackendRepr::SimdVector` and `BackendRepr::Memory` for SIMD types. added a check to ensure we're dealing with a SIMD type, and the alignment logic works the same for both representations. also i added a test that reproduces the original ICE.

Rollup of 14 pull requests Successful merges: - #150468 (rustc_target: callconv: powerpc64: Use the ABI set in target options instead of guessing) - #151628 (Fix ICE in const eval of packed SIMD types with non-power-of-two element counts) - #151871 (don't use env with infer vars) - #152591 (Simplify internals of `{Rc,Arc}::default`) - #152657 (std: move `exit` out of PAL) - #152865 (Fixed ByteStr not padding within its Display trait when no specific alignment is mentioned) - #152908 (Enable rust.remap-debuginfo in the dist profile) - #152705 (Test(lib/win/proc): Skip `raw_attributes` doctest under Win7) - #152767 (fix typo in `carryless_mul` macro invocation) - #152837 (fix(codegen): Use `body_codegen_attrs` For Caller In `adjust_target_feature_sig`) - #152871 (Fix warnings in rs{begin,end}.rs files) - #152921 (Add build.rustdoc option to bootstrap config) - #152933 (Start migration for `LintDiagnostic` items by adding API and migrating `LinkerOutput` lint) - #152937 (remove unneeded reboxing)

… r=RalfJung Fix ICE in const eval of packed SIMD types with non-power-of-two element counts fixes rust-lang#151537 const evaluation of packed SIMD types with non-power-of-two element counts (like `Simd<_, 3>`) was hitting an ICE. the issue was in `check_simd_ptr_alignment` - it asserted that `backend_repr` must be `BackendRepr::SimdVector`, but for packed SIMD types with non-power-of-two counts the compiler uses `BackendRepr::Memory` instead (as mentioned in `rustc_abi/src/layout.rs:1511`). was fixed by making `check_simd_ptr_alignment` accept both `BackendRepr::SimdVector` and `BackendRepr::Memory` for SIMD types. added a check to ensure we're dealing with a SIMD type, and the alignment logic works the same for both representations. also i added a test that reproduces the original ICE.

Rollup of 13 pull requests Successful merges: - #150468 (rustc_target: callconv: powerpc64: Use the ABI set in target options instead of guessing) - #151628 (Fix ICE in const eval of packed SIMD types with non-power-of-two element counts) - #151871 (don't use env with infer vars) - #152591 (Simplify internals of `{Rc,Arc}::default`) - #152865 (Fixed ByteStr not padding within its Display trait when no specific alignment is mentioned) - #152908 (Enable rust.remap-debuginfo in the dist profile) - #152705 (Test(lib/win/proc): Skip `raw_attributes` doctest under Win7) - #152767 (fix typo in `carryless_mul` macro invocation) - #152837 (fix(codegen): Use `body_codegen_attrs` For Caller In `adjust_target_feature_sig`) - #152871 (Fix warnings in rs{begin,end}.rs files) - #152921 (Add build.rustdoc option to bootstrap config) - #152933 (Start migration for `LintDiagnostic` items by adding API and migrating `LinkerOutput` lint) - #152937 (remove unneeded reboxing)

… r=RalfJung Fix ICE in const eval of packed SIMD types with non-power-of-two element counts fixes rust-lang#151537 const evaluation of packed SIMD types with non-power-of-two element counts (like `Simd<_, 3>`) was hitting an ICE. the issue was in `check_simd_ptr_alignment` - it asserted that `backend_repr` must be `BackendRepr::SimdVector`, but for packed SIMD types with non-power-of-two counts the compiler uses `BackendRepr::Memory` instead (as mentioned in `rustc_abi/src/layout.rs:1511`). was fixed by making `check_simd_ptr_alignment` accept both `BackendRepr::SimdVector` and `BackendRepr::Memory` for SIMD types. added a check to ensure we're dealing with a SIMD type, and the alignment logic works the same for both representations. also i added a test that reproduces the original ICE.

Rollup of 15 pull requests Successful merges: - #150468 (rustc_target: callconv: powerpc64: Use the ABI set in target options instead of guessing) - #151628 (Fix ICE in const eval of packed SIMD types with non-power-of-two element counts) - #151871 (don't use env with infer vars) - #152591 (Simplify internals of `{Rc,Arc}::default`) - #152865 (Fixed ByteStr not padding within its Display trait when no specific alignment is mentioned) - #152908 (Enable rust.remap-debuginfo in the dist profile) - #147859 (reduce the amount of panics in `{TokenStream, Literal}::from_str` calls) - #152705 (Test(lib/win/proc): Skip `raw_attributes` doctest under Win7) - #152767 (fix typo in `carryless_mul` macro invocation) - #152837 (fix(codegen): Use `body_codegen_attrs` For Caller In `adjust_target_feature_sig`) - #152871 (Fix warnings in rs{begin,end}.rs files) - #152879 (Remove `impl IntoQueryParam<P> for &'a P`.) - #152933 (Start migration for `LintDiagnostic` items by adding API and migrating `LinkerOutput` lint) - #152937 (remove unneeded reboxing) - #152953 (Fix typo in armv7a-vex-v5.md)

… r=RalfJung Fix ICE in const eval of packed SIMD types with non-power-of-two element counts fixes rust-lang#151537 const evaluation of packed SIMD types with non-power-of-two element counts (like `Simd<_, 3>`) was hitting an ICE. the issue was in `check_simd_ptr_alignment` - it asserted that `backend_repr` must be `BackendRepr::SimdVector`, but for packed SIMD types with non-power-of-two counts the compiler uses `BackendRepr::Memory` instead (as mentioned in `rustc_abi/src/layout.rs:1511`). was fixed by making `check_simd_ptr_alignment` accept both `BackendRepr::SimdVector` and `BackendRepr::Memory` for SIMD types. added a check to ensure we're dealing with a SIMD type, and the alignment logic works the same for both representations. also i added a test that reproduces the original ICE.

…uwer Rollup of 14 pull requests Successful merges: - #150468 (rustc_target: callconv: powerpc64: Use the ABI set in target options instead of guessing) - #151628 (Fix ICE in const eval of packed SIMD types with non-power-of-two element counts) - #151871 (don't use env with infer vars) - #152591 (Simplify internals of `{Rc,Arc}::default`) - #152865 (Fixed ByteStr not padding within its Display trait when no specific alignment is mentioned) - #147859 (reduce the amount of panics in `{TokenStream, Literal}::from_str` calls) - #152705 (Test(lib/win/proc): Skip `raw_attributes` doctest under Win7) - #152767 (fix typo in `carryless_mul` macro invocation) - #152837 (fix(codegen): Use `body_codegen_attrs` For Caller In `adjust_target_feature_sig`) - #152871 (Fix warnings in rs{begin,end}.rs files) - #152879 (Remove `impl IntoQueryParam<P> for &'a P`.) - #152933 (Start migration for `LintDiagnostic` items by adding API and migrating `LinkerOutput` lint) - #152937 (remove unneeded reboxing) - #152953 (Fix typo in armv7a-vex-v5.md)

Rollup merge of #151628 - enthropy7:fix-simd-const-eval-ice, r=RalfJung Fix ICE in const eval of packed SIMD types with non-power-of-two element counts fixes #151537 const evaluation of packed SIMD types with non-power-of-two element counts (like `Simd<_, 3>`) was hitting an ICE. the issue was in `check_simd_ptr_alignment` - it asserted that `backend_repr` must be `BackendRepr::SimdVector`, but for packed SIMD types with non-power-of-two counts the compiler uses `BackendRepr::Memory` instead (as mentioned in `rustc_abi/src/layout.rs:1511`). was fixed by making `check_simd_ptr_alignment` accept both `BackendRepr::SimdVector` and `BackendRepr::Memory` for SIMD types. added a check to ensure we're dealing with a SIMD type, and the alignment logic works the same for both representations. also i added a test that reproduces the original ICE.

…uwer Rollup of 14 pull requests Successful merges: - rust-lang/rust#150468 (rustc_target: callconv: powerpc64: Use the ABI set in target options instead of guessing) - rust-lang/rust#151628 (Fix ICE in const eval of packed SIMD types with non-power-of-two element counts) - rust-lang/rust#151871 (don't use env with infer vars) - rust-lang/rust#152591 (Simplify internals of `{Rc,Arc}::default`) - rust-lang/rust#152865 (Fixed ByteStr not padding within its Display trait when no specific alignment is mentioned) - rust-lang/rust#147859 (reduce the amount of panics in `{TokenStream, Literal}::from_str` calls) - rust-lang/rust#152705 (Test(lib/win/proc): Skip `raw_attributes` doctest under Win7) - rust-lang/rust#152767 (fix typo in `carryless_mul` macro invocation) - rust-lang/rust#152837 (fix(codegen): Use `body_codegen_attrs` For Caller In `adjust_target_feature_sig`) - rust-lang/rust#152871 (Fix warnings in rs{begin,end}.rs files) - rust-lang/rust#152879 (Remove `impl IntoQueryParam<P> for &'a P`.) - rust-lang/rust#152933 (Start migration for `LintDiagnostic` items by adding API and migrating `LinkerOutput` lint) - rust-lang/rust#152937 (remove unneeded reboxing) - rust-lang/rust#152953 (Fix typo in armv7a-vex-v5.md)

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. labels Jan 25, 2026

rustbot assigned nnethercote Jan 25, 2026

This comment has been minimized.

Sign in to view

RalfJung mentioned this pull request Jan 25, 2026

rustdoc-gui tests sporadically failing #93784

Open

enthropy7 force-pushed the fix-simd-const-eval-ice branch from 805a3cf to 5baf4e6 Compare January 25, 2026 12:40

rustbot assigned RalfJung and unassigned nnethercote Jan 25, 2026

RalfJung approved these changes Jan 25, 2026

View reviewed changes

enthropy7 force-pushed the fix-simd-const-eval-ice branch from 5baf4e6 to c45a513 Compare January 25, 2026 20:11

rust-bors bot added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Feb 21, 2026

jhpratt mentioned this pull request Feb 21, 2026

Rollup of 14 pull requests #152945

Closed

jhpratt mentioned this pull request Feb 22, 2026

Rollup of 15 pull requests #152950

Closed

jhpratt mentioned this pull request Feb 22, 2026

Rollup of 14 pull requests #152951

Closed

jhpratt mentioned this pull request Feb 22, 2026

Rollup of 13 pull requests #152957

Closed

jhpratt mentioned this pull request Feb 22, 2026

Rollup of 15 pull requests #152960

Closed

This was referenced Feb 22, 2026

Rollup of 13 pull requests #152964

Closed

Rollup of 14 pull requests #152965

Merged

rust-bors bot merged commit 8a3d112 into rust-lang:main Feb 22, 2026
11 checks passed

rustbot added this to the 1.95.0 milestone Feb 22, 2026

Uh oh!

Conversation

enthropy7 commented Jan 25, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rustbot commented Jan 25, 2026

Uh oh!

rustbot commented Jan 25, 2026

Uh oh!

This comment has been minimized.

Kivooeo commented Jan 25, 2026

Uh oh!

folkertdev commented Jan 25, 2026

Uh oh!

enthropy7 commented Jan 25, 2026

Uh oh!

folkertdev commented Jan 25, 2026

Uh oh!

enthropy7 commented Jan 25, 2026

Uh oh!

rustbot commented Jan 25, 2026

Uh oh!

RalfJung commented Jan 25, 2026

Uh oh!

RalfJung left a comment • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

RalfJung Jan 25, 2026

Choose a reason for hiding this comment

Uh oh!

RalfJung commented Feb 21, 2026

Uh oh!

rust-bors bot commented Feb 21, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

enthropy7 commented Jan 25, 2026 •

edited by rustbot

Loading

RalfJung left a comment •

edited by rustbot

Loading