[llvm-dev] LLVM Weekly - #245, September 10th 2018

Mon Sep 10 12:09:31 PDT 2018

LLVM Weekly - #245, September 10th 2018
=====================================

If you prefer, you can read a HTML version of this email at
<http://llvmweekly.org/issue/245>.

Welcome to the two hundred and forty-fifth 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](https://www.linkedin.com/in/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 at asbradbury.org>, or
@llvmweekly or @asbradbury on Twitter.

## News and articles from around the web

Applications for LLVM Foundation Travel Grants for full time undergraduate and
graduate students are [open until September
12th](http://lists.llvm.org/pipermail/llvm-dev/2018-September/125872.html).

The next Zurich LLVM Compiler Social will be [held on Thursday 13th
September](http://lists.llvm.org/pipermail/llvm-dev/2018-September/125858.html).
It features talks on the Lean and Isabelle theorem provers.

The 4th LLVM/GCC social in Shenzhen will [take place on September
16th](http://lists.llvm.org/pipermail/llvm-dev/2018-September/125933.html).

Bekket McClane's series on writing an LLVM pass continues with guidance on
[integrating the pass into
Clang](https://medium.com/@mshockwave/writing-llvm-pass-in-2018-part-iv-d69dac57171d).

Nuno Lopes and John Regehr have published a fascinating paper [Future
Directions for Optimizing Compilers](https://arxiv.org/pdf/1809.02161.pdf) on
arXiv. "This paper
examines the problem of making optimizing compilers faster, less buggy, and
more capable of generating high-quality output."

## On the mailing lists

* Paul Kirth has shared an RFC on [upstreaming Syringe, a dynamic behaviour
injection
framework](http://lists.llvm.org/pipermail/cfe-dev/2018-September/059260.html).
Developed as part of a Google internship, this framework allows users to
specify alternate implementations of functions and dynamically switch
behaviour at runtime. Gábor Márton provided
[feedback](http://lists.llvm.org/pipermail/cfe-dev/2018-September/059269.html)
based on his work on `fci_mock` which was presented at EuroLLVM this year and
suggests joining forces. David Chisnall
[suggests](http://lists.llvm.org/pipermail/cfe-dev/2018-September/059282.html)
looking at the LOOM framework which provides similar functionality.

* Jakub Kuderski has posted an RFC on [per-callsite inline
intrinsics](http://lists.llvm.org/pipermail/cfe-dev/2018-September/059232.html),
making it possible to manually force desired inlining decisions with a fine
granularity.

* Jonas Paulsson kicked off a discussion on [improving sequences of
shufflevector instructions produced by the loop
vectorizer](http://lists.llvm.org/pipermail/llvm-dev/2018-September/125826.html).

* The 7.0.0 release is [almost
ready](http://lists.llvm.org/pipermail/llvm-dev/2018-September/125850.html),
but there will be a final release candidate (rc3) before the final tag.
Hans Wennborg is requesting further additions to the release notes.

* Kristina Brooks
[proposes](http://lists.llvm.org/pipermail/cfe-dev/2018-September/059222.html)
a new `__builtin_type_name` Clang extension to retrieve a type name from a
pointer. Full justification is given in the thread, which generated a
reasonable amount of discussion.

## LLVM commits

* A new "hot cold" splitting pass was introduced, which can use profile
information to outline cold regions to a separate function.
[r341669](https://reviews.llvm.org/rL341669).

* A new control height reduction pass was merged. This uses information from
profiling to merge conditional blocks of code, merging the number of
conditional branches in the hot path.
[r341386](https://reviews.llvm.org/rL341386).

* There were a number of changes related to the SpeculativeLoadHardening
'Spectre' mitigation pass. It learned to harden indirect branches and switches
without retpolines, gained a mitigation against the ret2spec attack, and an
LLVM IR attribute was added indicating a function body has speculative load
hardening enabled.
[r341356](https://reviews.llvm.org/rL341356),
[r341358](https://reviews.llvm.org/rL341358),
[r341363](https://reviews.llvm.org/rL341363).

* `TargetInstrInfo::has{StoreTo,LoadFrom}StackSlot` was extended to return an
array of accesses. [r341301](https://reviews.llvm.org/rL341301).

* `-Rpass-analysis=size-info` will now output per-function size-info remarks
rather than per-module summaries.
[r341588](https://reviews.llvm.org/rL341588).

* llvm-objcopy gained support for compressing DWARF .debug sections.
[r341635](https://reviews.llvm.org/rL341635).

* MemorySanitizer was extended with KMSAN, for kernel instrumentation.
[r341637](https://reviews.llvm.org/rL341637).

## Clang commits

* A new compact binary serialization format was defined for clangd's symbol
index. This is 10x more compact than YAML and 40% more compact than gzipped
AML. [r341375](https://reviews.llvm.org/rL341375).

* DWARF accelerator tables are now enabled by default when tuning for LLDB.
[r341564](https://reviews.llvm.org/rL341564).

* The deprecated getLocStart, getBeginLoc and getLocEnd APIs were removed.
[r341573](https://reviews.llvm.org/rL341573).

## Other project commits

* LLVM's OpenMP library gained an initial implementation of the OMP 5.0 memory
management routines. [r341687](https://reviews.llvm.org/rL341687).

* libFuzzer tests were enabled on Windows.
[r341385](https://reviews.llvm.org/rL341385).