Issues

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

[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

2

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

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

[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

[GlobalISel] IRTranslator returns the destination when inserting a one-element vector into a one-element vector

Bug

llvm/llvm-project#235314 · opened

on Oct 24, 2026

[MLIR][Affine] `affine-expand-index-ops` emits invalid IR for a nested `affine.if` use of `affine.delinearize_index`

llvm/llvm-project#235302 · opened

on Oct 24, 2026

1

[NewGVN] EH funclet token replaced with poison: "cleanupret from %cl" becomes "cleanupret from poison" on unreachable cleanup blocks -> verifier abort (Broken module)

Bug

llvm/llvm-project#235301 · opened

on Oct 24, 2026

[lli] Interpreter asserts (APInt.h:127 isUIntN) when main returns a negative constant; JIT returns the wrapped value correctly

Bug

llvm/llvm-project#235299 · opened

on Oct 24, 2026

2

[InstCombine] foldSelectCttzCtlz flips cttz/ctlz isZeroPoison false→true without guarding undef true-value — miscompile on select(icmp x,0,K,cttz(x,false))

llvm/llvm-project#235297 · opened

on Oct 24, 2026

1

[CodeGen] Assertion `getLangOpts().OpenMPIsTargetDevice && "OpenMP AMDGPU/NVPTX/SPIRV is only prepared to deal with device code."' failed.

Bug

llvm/llvm-project#235290 · opened

on Oct 24, 2026

2

[NVPTX][DAGCombiner] `__builtin_clzll(x | 1) >> 5` is folded to 0: visitSRL's `srl (ctlz x), log2(BW)` fold uses the i32 result width of NVPTX's ctlz of an i64 value

Bug

llvm/llvm-project#235271 · opened

on Oct 24, 2026

[clang][RISCV] canonicalizeMaskForVLPredicate accepts a wrapping (mod 2^EltBits) step sequence, so lanes of `(splat(b) | <..., 4,5,6,7>) < n` are computed wrong

Bug

llvm/llvm-project#235270 · opened

on Oct 24, 2026

1

[BPF] Assertion `Expr->getKind() == MCExpr::SymbolRef` in BPFMCCodeEmitter::getMachineOpValue for `ld_imm64` with `sym + const`

Bug

llvm/llvm-project#235269 · opened

on Oct 24, 2026

1

[libc] Implement getresuid and getresgid in unistd.h

llvm/llvm-project#235268 · opened

on Oct 24, 2026

1

[clang][ExpandIRInsts] `[su]itofp iN -> x86_fp80` (N > 128) converts via fp128 and then truncates, so it rounds twice: `(long double)(unsigned _BitInt(129))(2^127 + 2^63 + 1)` gives 2^127 instead of 2^127 + 2^64

Bug

llvm/llvm-project#235266 · opened

on Oct 24, 2026

1

[clang][AMDGPU][R600] PerformDAGCombine folds `(f32 fp_round (f64 uint_to_fp i64 a))` into `(f32 uint_to_fp a)`: `(float)(double)a` loses its double rounding

Bug

llvm/llvm-project#235264 · opened

on Oct 24, 2026

1

[DAG] foldSelectToABD negates an INT_MIN constant lane: `|a + INT_MIN|` becomes `a + 2^31`

Bug

llvm/llvm-project#235263 · opened

on Oct 24, 2026

· LLVM 23.x Release

2

[DAG] visitVSELECT folds `x > C-1 ? x + -C : 0` to `usubsat x, C` also for lanes with C = 0, where the compare is the never-true `x > UINT_MAX`: those lanes return x instead of 0

Bug

llvm/llvm-project#235262 · opened

on Oct 24, 2026

2

[clang] [Mips] lowerShiftRightParts/lowerShiftLeftParts select on `and shamt, 32` (0 or 32, not a 0/1 boolean): `0x1FFFFFFFF >> (b & 32)` gives 0x2100000001 on MIPS32

Bug

llvm/llvm-project#235261 · opened

on Oct 24, 2026

2

[clang] crashes with SIGSEGV when evaluating __builtin_bit_cast in a static_assert involving a self-referential struct

Bug

llvm/llvm-project#235252 · opened

on Oct 24, 2026

4

[lld][WebAssembly] wasm-ld crashes on an object whose Function section declares functions but which has no Code section

Bug

llvm/llvm-project#235251 · opened

on Oct 24, 2026

1

Navigation Menu

Repository navigation

Issues

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

[GlobalISel] IRTranslator returns the destination when inserting a one-element vector into a one-element vector

[MLIR][Affine] `affine-expand-index-ops` emits invalid IR for a nested `affine.if` use of `affine.delinearize_index`

[NewGVN] EH funclet token replaced with poison: "cleanupret from %cl" becomes "cleanupret from poison" on unreachable cleanup blocks -> verifier abort (Broken module)

[lli] Interpreter asserts (APInt.h:127 isUIntN) when main returns a negative constant; JIT returns the wrapped value correctly

[InstCombine] foldSelectCttzCtlz flips cttz/ctlz isZeroPoison false→true without guarding undef true-value — miscompile on select(icmp x,0,K,cttz(x,false))

[CodeGen] Assertion `getLangOpts().OpenMPIsTargetDevice && "OpenMP AMDGPU/NVPTX/SPIRV is only prepared to deal with device code."' failed.

[NVPTX][DAGCombiner] `__builtin_clzll(x | 1) >> 5` is folded to 0: visitSRL's `srl (ctlz x), log2(BW)` fold uses the i32 result width of NVPTX's ctlz of an i64 value

[clang][RISCV] canonicalizeMaskForVLPredicate accepts a wrapping (mod 2^EltBits) step sequence, so lanes of `(splat(b) | <..., 4,5,6,7>) < n` are computed wrong

[BPF] Assertion `Expr->getKind() == MCExpr::SymbolRef` in BPFMCCodeEmitter::getMachineOpValue for `ld_imm64` with `sym + const`

[libc] Implement getresuid and getresgid in unistd.h

[clang][ExpandIRInsts] `[su]itofp iN -> x86_fp80` (N > 128) converts via fp128 and then truncates, so it rounds twice: `(long double)(unsigned _BitInt(129))(2^127 + 2^63 + 1)` gives 2^127 instead of 2^127 + 2^64

[clang][AMDGPU][R600] PerformDAGCombine folds `(f32 fp_round (f64 uint_to_fp i64 a))` into `(f32 uint_to_fp a)`: `(float)(double)a` loses its double rounding

[DAG] foldSelectToABD negates an INT_MIN constant lane: `|a + INT_MIN|` becomes `a + 2^31`

[DAG] visitVSELECT folds `x > C-1 ? x + -C : 0` to `usubsat x, C` also for lanes with C = 0, where the compare is the never-true `x > UINT_MAX`: those lanes return x instead of 0

[clang] [Mips] lowerShiftRightParts/lowerShiftLeftParts select on `and shamt, 32` (0 or 32, not a 0/1 boolean): `0x1FFFFFFFF >> (b & 32)` gives 0x2100000001 on MIPS32

[clang] crashes with SIGSEGV when evaluating __builtin_bit_cast in a static_assert involving a self-referential struct

[lld][WebAssembly] wasm-ld crashes on an object whose Function section declares functions but which has no Code section