LLVM Weekly - #200, Oct 30th 2017

Welcome to the two hundredth 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

The next Zurich Compiler and Code Generation Social will take place on November 9th at 7pm and feature the talk "The RISC-V effort and the PULP project at ETH Zurich".

The Arm HPC workshop will take place 12th-13th Dec in Tokyo, and the call for papers/participation is now open.

On the mailing lists

• "escha" has shared a patch and RFC on modifying reassociate for improved common subexpression elimination. On private graphics shader benchmarks, this has helped to reduce instruction count. The patch is now up for review.

• Dan Gohman has kicked off a new discussion on handling infinite loops with no side effects. Such loops are undefined behaviour in C++, but may have defined behaviour for other source languages.

• Sam McCall has posted an RFC on adding a newly written JSON library to LLVM Support.

• Alessandro Di Federico has shared work on using XPath as a query language for LLVM IR.

• Chandler Carruth has started an RFC thread on reserving function names recognised by TargetLibraryInfo. David Chisnall suggests that the frontend should be responsible for adding metadata to indicate when a given function is "special".

• Geoff Berry has shared an RFC on adding a bit to register MachineOperands. This bit would be used to track whether a register definition was a virtual register before register allocation.

• Rui Ueyama posted a proposal on making .eh_frame more linker-friendly.

• Bingfeng Mei asked about differences in bitfield handling between Clang/LLVM and GCC. Joseph Myers provided some insight on the GCC list.

• Vedant Kumar shared notes and updates from the code coverage birds of a feather session at the LLVM Dev Meeting.

• Andrew Trick has provided insight into the MicroOpBufferSize parameter.

LLVM commits

• Infrastructure to support padding code with nop instructions in the MC layer has been committed. This will be used for performance improvements on Intel targets (see the RFC). r316413.

• GlobalISel can now import store SelectionDAG rules. r316360.

• Runtime preemption specifiers can now be attached to global variables, functions and aliases. dso_preemptable indicates the function/variable could be replaced by a symbol from a different linkage unit at runtime, while dso_local means this is not possible. r316668.

• The X86 backend gained a "domain reassignment" pass. This will look for cases where it is profitable to e.g. convert from the GPR domain to the Mask domain (vector mask registers). r316288.

• The greedy register allocator was improved to avoid bad eviction chains (see the commit message for an example). r316295.

• Brief documentation has been added on SelectionDAG rule imports for GlobalISel. r316350.

• Instruction scheduling information as been added for the Intel Broadwell microarchitecture. r316492.

• The CalledValuePropagation pass has been introduced, which will attach !callees metadata to indirect call sites. r316576.

Clang commits

• The basics of clang-refactor "extract function" functionality has been implemented. r316465.

• A new clang::tooling:ToolExecutor interface was introduced, which is expected to replace the ClangTool interface for executing frontend actions on a set of translation units. r316653.

Other project commits

• LLD can now emit relocations in Android's relocation packing format. As described in the commit message, the algorithm used here is more advanced than the one in the gold linker, producing even smaller relocation sections. r316775.

• LLD gained support for range extension thunks on ARM. See Peter Smith's previous mailing list message for an introduction to range extension thunks. r316752, r316754.

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.

There were two patch submissions last week, both of which saw fresh review activity. Thank you to Quentin Colombet and Simon Pilgrim for contributing review comments.

I haven't received any submissions to the review corner this week.