Issues

[LoopStrengthReduce] Use-list order picks between two equal-cost solutions, and one of them costs an extra instruction (`d - 1` in a counted loop)

llvm/llvm-project#235399 · opened

on Oct 25, 2026

CI code formatter steps inconsistency in `libcxx`

llvm/llvm-project#235395 · opened

on Oct 25, 2026

[PowerPC] `The register $carry needs to be live in to %bb.1, but is missing from the live-in list.` with `crbits` and wide integers on 32-bit

Bug

llvm/llvm-project#235393 · opened

on Oct 25, 2026

1

[ExecutionEngine] Interpreter has no support for the LLVM-23 opaque byte types (b8/b16/b32/b64/b128): 6 abort sites plus a StoreValueToMemory that silently writes nothing for byte-element vectors (stable wrong values and uninitialized-memory reads; ORC and MCJIT are correct on all of them)

llvm/llvm-project#235388 · opened

on Oct 25, 2026

[InstCombine] Replacing an entry-block freeze(poison) with a loop-varying select arm changes the result

Bug

llvm/llvm-project#235385 · opened

on Oct 25, 2026

[InstCombine] Narrowing an aggregate load through extractvalue incorrectly increases alignment

Bug

llvm/llvm-project#235384 · opened

on Oct 25, 2026

[MLIR][Affine] `--canonicalize` can create an `affine.for` with an empty bound map

llvm/llvm-project#235382 · opened

on Oct 25, 2026

2

[MemCpyOpt] Hoists an unreachable aggregate store above a noreturn call

Bug

llvm/llvm-project#235381 · opened

on Oct 25, 2026

[InstCombine] Negator reuses a cached nsw negation for a non-nsw request

Bug

llvm/llvm-project#235380 · opened

on Oct 25, 2026

[DAGCombiner] `X - (X + Y) -> -Y` fold checks `reassoc` only on the fsub, not on the inner fadd

Bug

llvm/llvm-project#235379 · opened

on Oct 25, 2026

[FixIrreducible] `reconnectPhis` replaces a guard PHI of moved `undef` values with `poison`

Bug

llvm/llvm-project#235378 · opened

on Oct 25, 2026

[ExpandIRInsts] frem expansion branches on `fabs(x) > fabs(y)` without freezing the operands

Bug

llvm/llvm-project#235377 · opened

on Oct 25, 2026

[Clang][CodeGen] VFE miscompiles virtual member-function pointer calls with relative vtables

llvm/llvm-project#235376 · opened

on Oct 25, 2026

1

[AArch64] Instruction order at the end of a divide-bound loop costs ~14% on Apple M2 (shows up as dispatch stalls)

llvm/llvm-project#235371 · opened

on Oct 25, 2026

3

[InstCombine] PHI CSE replaces uses of an unflagged PHI with a fast-math-flagged duplicate without intersecting flags. miscompile on phi ninf/phi nnan duplicates (GVN twin fixed by #232149)

Bug

llvm/llvm-project#235370 · opened

on Oct 24, 2026

[NVPTX] "Cannot select: v2bf16 = scalar_to_vector" for a byte-wise copy of an unaligned bfloat through a local array (regression in LLVM 23)

Bug

llvm/llvm-project#235368 · opened

on Oct 24, 2026

[NVPTX] "Cannot select: v2i32 = and" when storing `trunc <2 x i32> to <2 x i1>` on sm_100 and newer

Bug

llvm/llvm-project#235365 · opened

on Oct 24, 2026

1

Clang rejects calling the function pointer conversion operator of a consteval lambda in a consteval context

llvm/llvm-project#235353 · opened

on Oct 24, 2026

3

[clang] Partial ordering for the address of a function template skips an `auto` return type on one side only

llvm/llvm-project#235330 · opened

on Oct 24, 2026

3

Request Commit Access For sriramshastry

llvm/llvm-project#235326 · opened

on Oct 24, 2026

4

[clang] SIGSEGV emitting implicit destructor when a module holds (transitively) a class defined with a qualified-id at global scope in another module interface

Bug

llvm/llvm-project#235320 · opened

on Oct 24, 2026

1

[SimplifyCFG] Switch lookup table is one entry short for a full-range index, turning the default result into poison

Bug

llvm/llvm-project#235318 · opened

on Oct 24, 2026

1

[GlobalISel] Constant folding llvm.sqrt.f80 through double loses extended precision

Bug

llvm/llvm-project#235317 · opened

on Oct 24, 2026

[ComplexDeinterleaving] Single-reduction path replaces a non-add-reduction user of the accumulator with llvm.vector.reduce.add

Bug

llvm/llvm-project#235316 · opened

on Oct 24, 2026

[SelectionDAG] Store merging fuses `p[sext(x + 1)]` and `p[sext(x + 2)]` into one i16 store although the i8 add wraps

Bug

llvm/llvm-project#235315 · opened

on Oct 24, 2026

Navigation Menu

Repository navigation

Issues

[LoopStrengthReduce] Use-list order picks between two equal-cost solutions, and one of them costs an extra instruction (`d - 1` in a counted loop)

CI code formatter steps inconsistency in `libcxx`

[PowerPC] `The register $carry needs to be live in to %bb.1, but is missing from the live-in list.` with `crbits` and wide integers on 32-bit

[ExecutionEngine] Interpreter has no support for the LLVM-23 opaque byte types (b8/b16/b32/b64/b128): 6 abort sites plus a StoreValueToMemory that silently writes nothing for byte-element vectors (stable wrong values and uninitialized-memory reads; ORC and MCJIT are correct on all of them)

[InstCombine] Replacing an entry-block freeze(poison) with a loop-varying select arm changes the result

[InstCombine] Narrowing an aggregate load through extractvalue incorrectly increases alignment

[MLIR][Affine] `--canonicalize` can create an `affine.for` with an empty bound map

[MemCpyOpt] Hoists an unreachable aggregate store above a noreturn call

[InstCombine] Negator reuses a cached nsw negation for a non-nsw request

[DAGCombiner] `X - (X + Y) -> -Y` fold checks `reassoc` only on the fsub, not on the inner fadd

[FixIrreducible] `reconnectPhis` replaces a guard PHI of moved `undef` values with `poison`

[ExpandIRInsts] frem expansion branches on `fabs(x) > fabs(y)` without freezing the operands

[Clang][CodeGen] VFE miscompiles virtual member-function pointer calls with relative vtables

[AArch64] Instruction order at the end of a divide-bound loop costs ~14% on Apple M2 (shows up as dispatch stalls)

[InstCombine] PHI CSE replaces uses of an unflagged PHI with a fast-math-flagged duplicate without intersecting flags. miscompile on phi ninf/phi nnan duplicates (GVN twin fixed by #232149)

[NVPTX] "Cannot select: v2bf16 = scalar_to_vector" for a byte-wise copy of an unaligned bfloat through a local array (regression in LLVM 23)

[NVPTX] "Cannot select: v2i32 = and" when storing `trunc <2 x i32> to <2 x i1>` on sm_100 and newer

Clang rejects calling the function pointer conversion operator of a consteval lambda in a consteval context

[clang] Partial ordering for the address of a function template skips an `auto` return type on one side only

Request Commit Access For sriramshastry

[clang] SIGSEGV emitting implicit destructor when a module holds (transitively) a class defined with a qualified-id at global scope in another module interface

[SimplifyCFG] Switch lookup table is one entry short for a full-range index, turning the default result into poison

[GlobalISel] Constant folding llvm.sqrt.f80 through double loses extended precision

[ComplexDeinterleaving] Single-reduction path replaces a non-add-reduction user of the accumulator with llvm.vector.reduce.add

[SelectionDAG] Store merging fuses `p[sext(x + 1)]` and `p[sext(x + 2)]` into one i16 store although the i8 add wraps