LLVM Weekly - #491, May 29th 2023

Welcome to the four hundred and ninety-first 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 http://llvmweekly.org and pass it on to anyone else you think may be interested. Please send any tips or feedback to asb@asbradbury.org, or @llvmweekly or @asbradbury on Twitter, or @llvmweekly@fosstodon.org or @asb@fosstodon.org.

News and articles from around the web and events

Save the date for the 2023 US LLVM Developers Meeting! It will take place on October 10-12 in Santa Clara, CA.

Min-Yih “Min” Hsu blogged about what’s new in M68k LLVM.

According to the LLVM calendar in the coming week there will be:

• Office hours with the following hosts: Johannes Doerfert.
• Online sync-ups on the following topics: Flang, OpenMP, classic Flang, loop optimisations, OpenMP in flang, HLSL, SPIR-V, MLGO.
• For more details see the LLVM calendar, getting involved documentation on online sync ups and office hours.

On the forums

• Yeoul Na posted an RFC on enforcing bounds safety in C with a new -fbounds-safety extension.

• Jeroen Dobbelaere shared a write-up of the EuroLLVM round table on alias analysis and full restrict.

• The slides and recording are now available from the last open MLIR meeting, an RFC discussion about ‘distinct attribute’ in MLIR.

• Chris Lattner shared thoughts on the topic of renaming flang-new to flang, ahead of a planned meeting of review managers.

• “isuckatcs” made a proposal for improving RegionStore in clang’s static analyzer.

• James Henderson started an RFC discussion on adding the preferred casting style (C++ vs C) to LLVM’s coding standards.

• Matt Arsenault proposed adding llvm.ldexp and llvm.frexp intrinsics.

LLVM commits

• A skeleton PPC64 JITLink backend was added, which also aims to serve as a good example of how to introduce a JITLink backend for a new architecture. 8313507.

• The X86FixupVectorConstantsPass was introduced, which will re-fold AVX512 vector load folds as broadcast folds. 0b91de5.

• Small (less than 32/64-bit width) vector copies and vector stores are now scalarized on RISC-V. 28b21c4, a8a36ee.

• PostShrinkWrap was enabled by default. c4a60c9.

• A GCNRewritePartialRegUses pass was added to AMDGPU to rewrite partially used virtual super-registers with registers of minimal size. 8d0412c.

• The .option arch directive is now supported in RISC-V assembly. 9e8ed34.

• The load-store-vectorizer was rewritten using a new algorithm, aiming to prefer scalar loads if faster. 2be0abb7.

Clang commits

• __bf16 was made an arithmetic (rather than storage-only) type. e621757.

• RISC-V vector tuple types are now implemented. 5e92298, 0019226, and more.

• Python files in clang and clang-tools-extra were reformatted with black. dd3c26a.

• Initial interactive CUDA support for clang-repl was committed. ddeab07.

Other project commits

• libc++ <format> is now considered stable. dff62f5.

• Python files in the non-LLVM and Clang directories were reformatted with black. f98ee40, 2238dcc, f9008e6.

• PowerPC vector types are now supported in Flang. ef93417.

• The LLVM libc math function status page was updated to show more targets (including linux-riscv64). 0bda541.

• lld gained a new --print-memory-usage option to print details of the target’s memory usage. 811cbfc.

• A generic scalar replacement of aggregates pass was added to MLIR. 4ed502e.