LLVM Weekly - #206, Dec 11th 2017

Welcome to the two hundred and sixth 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 [email protected], or @llvmweekly or @asbradbury on Twitter.

News and articles from around the web

LLVM 5.0.1-rc3 has been tagged. The release should take place this week.

Madhur Amilkanthwar has written a simple LLVM metadata visualizer using dot.

On the mailing lists

Hans Wennborg has started the discussion on the scheduling of the 6.0.0 release. Following the regular schedule, it would be branched on Wednesday 17th January. One large consumer of the branch has requested it takes place earlier to better align with their internal process, and Hans is seeking feedback on this.
Hideki Saito has shared a status update on VPlan, the effort to improve the Loop Vectorizer. "We are currently at a point where we can slow down the refactoring effort for the purpose of expediting the big functionality boost: outer loop vectorization -and by doing so encourage more participation from the wider LLVM community in the refactoring effort to expedite the overall transition to the VPlan framework." Hideki's follow-up RFC outlines the plan for outer loop vectorisation.
Julie Hockett has proposed a clang-doc tool, a documentation generator for C/C++ code supporting Doxygen-style markup.
Joel E. Denny has shared an RFC on clacc, a project that extends clang with OpenACC support. It works by translating OpenACC to OpenMP.
Mikhail Zolotukhin kicked off a discussion on reducing LLVM build times by minimising includes.
Rahul Chaudhry has shared an update on efforts to reduce the size of the dynamic relocations section in position independent executables.
Dean Michael Berris has shared an XRay-generated trace of a Clang invocation which can be opened in the Chrome trace viewer. Dean hopes such instrumentation will be useful for Clang and LLVM development.

LLVM commits

Loop predication is now enabled for reverse loops. r319659.
Initial support for a hardware-assisted AddressSanitizer has been added. As described in the design doc, this is based around embedding metadata in pointers as supported on AArch64. r320217.
All outstanding MC-layer patches for RV{32,64}IMAFD support in the RISC-V backend have now been committed, as well as initial MC layer support for the compressed (C) instruction set extension. r320020, r320023, r320037, and more.
The Nios2 backend continues to develop, and now has enough infrastructure to compile a simple empty function. r320035.
The machine basic block reference format has been changed to unify output between MIR and debug output. A line previously printed with the comment "BB#0" will now be annotated with "%bb.0".The commit messages contains a series of shell commands that can be used to update tests and the like. r319665.
The register allocator can now generate and store multiple regalloc hints per virtual register. This results in codegen improvements but must be enabled on a per-target basis (due to the churn in tests). r319754.
guessInstructionProperties=0 is now set for the SystemZ backend. As noted in the commit message, bugs were exposed in the process. See here for more info on guessInstructionProperties and consider enabling it for your backend! r319756.
"Full coverage" of MC encoding for the X87 and FMA instructions for X86 has been added. r320015, r320016.
It is now possible to ask the tablegenerated register parsing function to tolerate duplicate register names. This is useful for targets such as RISC-V which use the same register names for 32 and 64-bit floating point. r320018.
The "debugify" pass to test debug info preservation has been added. r320202.

Clang commits

Clang's default C++ dialect is now gnu++14 rather than gnu++98. r320250.
A design document for an AddressSanitizer inspired hardware-assisted memory safety tool has been committed. r319684.

Other project commits

Initial libFuzzer support for Fuchsia OS has been added. r320210.
LLD gained a detector for Cortex-A53 erratum 843419. r319780, r319781.

Review corner

The LLVM Weekly review corner serves to highlight patches that are stuck waiting awaiting review, or work from first-time contributors. See here for more information and how to submit you work for inclusion. Of course the hope is that highlighting these patches will enable LLVM Weekly readers will step up and help to get them merged. I'll be reporting back each week on any activity generated on these patches, as well as sharing a new batch. If you want your patch included you must submit it via the linked form.

No new patch submissions this week, but thanks to "Cameron", Hal Finkel, and Richard Smith for contributing review comments to last week's patches. Nikolai Kosjar is still looking for a code review of D39903.