LLVM Weekly - #537, April 15th 2024

Welcome to the five hundred and thirty-seventh 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 [email protected], @llvmweekly or @asbradbury on Twitter, or @[email protected] or @[email protected].

News and articles from around the web and events

No articles to highlight this week - tips always welcome.

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

• Office hours with the following hosts: Phoebe Wang, Johannes Doerfert, Aaron Ballman.
• Online sync-ups on the following topics: Flang, pointer authentication, SYCL, libc++, LLVM security group, new contributors, LLVM/offload, classic flang, Clang C/C++ language working group, loop optimisations, floating point, OpenMP in Flang, MLIR open meeting.
• For more details see the LLVM calendar, getting involved documentation on online sync ups and office hours.

On the forums

• Renato Golin is curious about interest in organising LLVM conferences in other regions, e.g. Asia, South America, Africa.

• Vy Nguyen shared an updated design proposal for LLDB telemetry/metrics.

• Tom Stellard shared a PSA on a reported suspicious login for llvmbot and the actions taken.

• Nick Desaulniers shared that he’s drafted a policy for hand-written assembly in libc

• Matthias Gehre posted an RFC on adding arithmetic, logical and comparison expressions in MLIR’s PDLL.

• Vitaly Buka is seeking to introduce a new clang builtin __builtin_allow_runtime_check which is intended to be used to guard expensive runtime checks.

• Kristof Beyls shared the slides of his keynote on a BOLT-based binary analysis tool.

• The 64th edition of MLIR News is now available.

• Guillaume Chatelet posted notes from the LLVM libc round table at EuroLLVM.

• Justin Bogner kicked off an RFC thread on adding a bunch of additional LLVM intrinsics for math operations present in C++ or HLSL.

• Andy Kaylor proposed modifications to the LLVM language reference to describe the allowed impact on fast math flags of transformations.

LLVM commits

• Initial support was added to the superword-level parallelism vectorizer for vectorization of non–power-of-2 ops. 6d66db3.

• uitofp now supports the nneg flag. 9170e38.

• An ExecutionSession state verified was added to ORC, enabled under EXPENSIVE_CHECKS builds. 649523f.

• ExpandLargeFpConvert gained support for bfloat types. 110c22f.

• Statepoint and patchpoint support was added for RISC-V. 53003e3.

Clang commits

• The _cpp_concepts macro is now set to 202002L which enables <expected> from libstdc++ to work correctly with Clang. 2875e24.

• Work started on implementing vector types in the clang interpreter. b7a93bc.

• The syntax difference between HIP and CUDA is now documented. 2bf4889.

• The CLANG_ENABLE_CIR CMake variable was added in preparation for allowing the building of Clang with ClangIR. 44de2bb.

• The ‘if’ clause was implemented for OpenACC compute constructs. daa8836.

Other project commits

• A design document for debug info generation in Flang was committed. 357f6c7.

• The scudo secure memory allocator now has an EnableContiguousRegions mode. bab0507.

• The libclc build system was refactored to allow in-tree builds. 72f9881.

• libcxx format was updated to Unicode 15.1. 59e66c5.

• The AMDGPU shared memory optimisation pass was removed from MLIR due to a range of issues. 4471831.