LLVM Weekly - #652, June 29th 2026

Welcome to the six hundred and fifty-second issue of LLVM Weekly, a weekly newsletter (published every Monday) covering developments in LLVM, Clang, and related projects. LLVM Weekly is brought to you by Alex Bradbury. Subscribe to future issues at https://llvmweekly.org and pass it on to anyone else you think may be interested. Please send any tips or feedback via email: asb@asbradbury.org, or Mastodon: @llvmweekly@fosstodon.org / @asb@fosstodon.org, or Bluesky: @llvmweekly.org / @asbradbury.org.

Recordings from EuroLLVM 2026 are now available on YouTube.

According to the LLVM Calendar in the coming week there will be the following:

Office hours with the following hosts: Quentin Colombet, Renato Golin.
Online sync-ups on the following topics: formal specification, pointer authentication, OpenMP, Clang C and C++ language working group, Flang, RISC-V, HLSL.
For more details see the LLVM calendar, getting involved documentation on online sync ups and office hours.

Fangrui Song started a new thread to further the discussion on revisiting commit-access criteria, suggesting a revised policy would consider signals such as single-employer vouching and that a “substantial contribution” standard would be better.
Alexander Yermolovich started an RFC discussion on multi-level line table support in LLVM, noting that profiling tools need to map machine instructions not just to high-level source code, but to specific intermediate representations as well.
KotoriAster proposed the addition of ObjCConstantIVarOffsetPass which would constify Objective-C ivar offset globals.
Tom Stellard posted a new RFC on enabling immutable releases (i.e. using GitHub’s Immutable Releases feature for llvm releases).

WebAssembly reference types are now represented in LLVM IR as target extension types rather than pointers in specific address spaces. 765f0d1.
FileCheck failures are now annotated with search ranges which aims to make it easier to understand the pattern’s behaviour. 138e217.
-force-vector-width=... now accepts scalable vectorisation factors, e.g. vscale x 4. 5951daf.
LLVM’s bump pointer allocator gained a MinAlign template argument which keeps the pointer at a minimum alignment. c91c9fc.
BasicAA was modified to allow some additional recursion across getelementptr/phis. 95decf7.
Profile-aware branch condition merging was enabled for AArch64. 2668971.
utls/docs was created to contain a shared sphinx configuration. 42ff45b.
The noipa function attribute was added, which disables interprocedural analysis that inspects the definition of the function. c07bc29.
ShadowCallStack support was added to the Hexagon backend. 2919ad7.
A compact enum table was added to LLVM ADT collection. If all uses of EnumEntry are ported to this representation, it will save 4700 relocations on libLLVM.so. e2e7612.

The high-level documentation for the Clang bytecode interpreter was updated. b0bf2c7.
OpenMP target region support was added to ClangIR. 213c7b7.
The ThreadSanitizer runtime flags and suppressions were documented. 0179379.
A UserAfterLifetimeEnd checker was added for detecting lifetime safety violations involving the [[clang::lifetimebound]] annotation. f3a6975.

It’s now possible to build flang in a way that it is optimised with PGO and BOLT. 44409a0.
LLVM libc gained documentation on which builtins the codebase references today and the minimum GCC version that ships each one. d4148a8.
Optimized ARM assembly single precision FP add/sub were added to compiler-rt. affc89f.
An acc-emit-remarks-loop pass was added to MLIR, which will emit optimisation remarks or loops in acc.compute_region and describe their mapping to OpenACC parallel levels and GPU dimensions. 65a7ccf.
RaiseWasmMLIRPass was added to MLIR, which will convert WasmSSA MLIR to core dialects. 5314be5.