[llvm-dev] LLVM Weekly - #188, Aug 7th 2017

[Alex Bradbury via llvm-dev]

LLVM Weekly - #188, Aug 7th 2017
================================

If you prefer, you can read a HTML version of this email at
<http://llvmweekly.org/issue/188>.

Welcome to the one hundred and eighty-eighth 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](http://asbradbury.org). 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

Jonathan Boccara has [summarised the recent proposal for 'metaclasses' in
C++](http://www.fluentcpp.com/2017/08/04/metaclasses-cpp-summary/).

Version 3.1 of the [SLEEF Vectorized Math Library](http://sleef.org/) has been
released, adding AArch64 support as well as completing the set of supported
C99 math functions. SLEEF exploits SIMD to provide high performance real
floating point math functions.

[generator-llvm](https://github.com/mshockwave/generator-llvm) is a small tool
for generating project templates for out-of-tree LLVM passes.


## On the mailing lists

* Christian Dietrich has written an [RFC on 'cHash' AST
hashing](http://lists.llvm.org/pipermail/cfe-dev/2017-August/054911.html).
This can be used to [avoid redundant
recompilation](https://www.usenix.org/conference/atc17/technical-sessions/presentation/dietrich).

* Jakub Kaderski has shared an RFC on [adding a batch update API for dominator
trees](http://lists.llvm.org/pipermail/llvm-dev/2017-August/116070.html).

* Francis Visoiu Mistrih has been working on improvements to the current LLVM
shrink wrapping optimisations and has [written an RFC on his
changes](http://lists.llvm.org/pipermail/llvm-dev/2017-August/116131.html).

* Pete Couperus is [proposing the upstreaming of the Synopsys ARC
backend](http://lists.llvm.org/pipermail/llvm-dev/2017-August/116192.html).

* Chad Rosier is seeking feedback on [modeling JumpThreading in the inline
cost model](http://lists.llvm.org/pipermail/llvm-dev/2017-August/116183.html).

* Matthias Braun [explains the existence of the SelectionDAG
schedulers](http://lists.llvm.org/pipermail/llvm-dev/2017-July/116005.html).


## LLVM commits

* The XRay docs have been expanded with a full description of the Flight Data
Recorder binary format. [r309891](http://reviews.llvm.org/rL309891).

* The stack colouring pass now correctly maintains AliasAnalysis information.
[r309651](http://reviews.llvm.org/rL309651),
[r309849](http://reviews.llvm.org/rL309849).

* Stack frame handling for AArch64 win64 vararg functions has been rewritten.
[r309744](http://reviews.llvm.org/rL309744).

* PowerPC, ARM, and X86 tests have been cleaned up to not use `llc -march`.
The `-march` switch only changes the target architecture, but the OS target is
unchanged, which can lead to unexpected test output. If you have out-of-tree
targets, you might want to check your tests for this potential issue.
[r309574](http://reviews.llvm.org/rL309574),
[r309575](http://reviews.llvm.org/rL309575).

* GlobalISel is now always built as part of LLVM.
[r309990](http://reviews.llvm.org/rL309990).

* The libFuzzer tests have been ported to LIT.
[r310075](http://reviews.llvm.org/rL310075).

* After a series of patches over the past few weeks and months, the ARM
backend is now MachineVerifier clean. This means the machine verifier can be
enabled (e.g. for bots with expensive checks turned on) without resulting in
spurious errors. [r310180](https://reviews.llvm.org/rL310180).


## Clang commits

* The `-Wpragma-pack` warning will complain about missing `#pragma pac (pop)`
directives. [r309559](http://reviews.llvm.org/rL309559).

* OpenCL 2.0 atomic builtin functions have been added as Clang builtins, in
order to support the generation of synchronization scope operands.
[r310082](http://reviews.llvm.org/rL310082).


## Other project commits

* LLD's `--gdb-index` is now 40% faster, as it now avoids a linear search.
[r309652](http://reviews.llvm.org/rL309652).

* LLDB now allows an auto-continue flag to be set on breakpoints.
[r309969](http://reviews.llvm.org/rL309969).