LLVM Weekly - #283, June 3rd 2019

Welcome to the two hundred and eighty-third 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 to asb@asbradbury.org, or @llvmweekly or @asbradbury on Twitter.

News and articles from around the web

I didn't spot anything in particular. If you think I missed something, feel free to drop me tips for next week.

On the mailing lists

• Francesco Petrogalli shared an RFC on exposing user-provided vector functions for auto-vectorisation, which generated lots of discussion.

• Michael Kruse, Bardia Mahjour, Hal Finkel, and Johannes Doerfert are forming a loop optimisation working group, and share a doodle poll for anyone interested in taking part in weekly or bi-weekly calls.

• Kit Barton proposes making loops guards part of the canonical loop structure.

• Oliver Giroux proposes an approach for std::atomic_wait for libcxx.

• On an architecture with a status register, how to optimise compare instructions where the status register already has the right condition? Eli Friedman explains two options.

• Melanie Blower is seeking feedback on Intel's plans to submit patches to support -fp-model= and -fp-speculation=.

• The recent thread on removing magic numbers assuming 8-bit bytes generated substantial discussion. Many thanks to Quentin Neill for sending in a list of key posts:

  • The standards vs. language vs. frontend vs. backend interplay: http://lists.llvm.org/pipermail/llvm-dev/2019-May/132088.html
  • Give and take of upstreaming new stuff vs. maintenance vs.acceptance: http://lists.llvm.org/pipermail/llvm-dev/2019-May/132221.html
  • Out of tree maintenance: http://lists.llvm.org/pipermail/llvm-dev/2019-May/132293.html
  • The community "discussion model" http://lists.llvm.org/pipermail/llvm-dev/2019-May/132565.html
  • Community objections: http://lists.llvm.org/pipermail/llvm-dev/2019-May/132331.html
  • Resolving objections: http://lists.llvm.org/pipermail/llvm-dev/2019-May/132333.html

LLVM commits

• It's been in discussion for a long time, but a scalable vector IR type has now been added! r361953.

• A (for now) optional type was added to byval function parameters in order to prepare the way for opaque pointer types. r362128.

• The ARM fp-only-sp and d16 options were replaced with fp64 and d32. Each subtargetfeature for an FP architecture now comes with different variants for combinations of these options.r361845.

• New ISD::LRINT and ISD::LLRINT SelectionDAG nodes were added along with new intrinsics. r361875.

• An llvm-lipo tool was started, aiming to be a drop-in replacement for the lipo tool for manipulating universal binaries. r361896.

• MC layer support for more AArch64 SVE2 instructions is being added. r361933, and many more.

• MC layer support for the new x86 'ENQCMD' instructions was added. r362053.

• The AsmMatcher now allows repeated input operands if there's a custom converter. r362066.

Clang commits

• SSE intrinsics have been ported to PowerPC using Altivec. r362190.

• clangd gained support for outputting a "structured response for Hover", which can be rendered by the client as it needs. r361803.

Other project commits

• Documentation was added for LLD's planned partitioning feature. r361923.

• libcxx was updated to support the recently added ELF dependent libraries feature. r362048.