[LLVMdev] LLVM Weekly - #20, May 19th 2014

[Alex Bradbury]

LLVM Weekly - #20, May 19th 2014
================================

If you prefer, you can read a HTML version of this email at
<http://llvmweekly.org/issue/20>

Welcome to the twentieth 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.

This week's issue is perhaps a little less thorough than normal. I've been in
San Francisco most of the week for Maker Faire this weekend, where I was at
the Raspberry Pi booth with some other Foundation members. As this issue goes
out, I'll be enjoying my last day in SF before heading to the airport for the
long flight home and the ensuing jetlag.

## News and articles from around the web

The WebKit blog features an [excellent and detailed article about the new
Fourth Tier LLVM
JIT](https://www.webkit.org/blog/3362/introducing-the-webkit-ftl-jit/) which
sheds light on the how and why.

The [Neu framework](http://neu.andrometa.net/) has recently been announced. It
is a C++11 framework, collection of programming languages and software system
designed for artificial intelligence applications and technical computing in
general. It makes use of the LLVM MC JIT for its NPL language as well as
generating high performance neural networks.


## On the mailing lists

* Tim Northover has posted an [RFC on removing the AArch64 backend and
renaming ARM64 to
AArch64](http://article.gmane.org/gmane.comp.compilers.llvm.devel/73034). This
would represent the culmination of the work to merge the two backends.

* In case you're wondering about when we'll see LLVM/Clang 3.5, Bill Wending
has [posted to inform us the current plan is to start testing in July and
target a release in
August](http://article.gmane.org/gmane.comp.compilers.clang.devel/36750).

* I always enjoy pointing out excellent answers to questions on the mailing
lists. The large and helpful community is one of the greatest assets of LLVM.
This week, I draw special attention to Jordan Rose's [answer to a question
about tracking a function parameter with the static
analyzer](http://article.gmane.org/gmane.comp.compilers.clang.devel/36755).

* The testing phase for LLVM 3.4.2 [has
started](http://article.gmane.org/gmane.comp.compilers.llvm.devel/73069).
According to [the
schedule](http://article.gmane.org/gmane.comp.compilers.llvm.devel/72919) it
should be released on May 25 provided the testing phase didn't raise issues.
Tom Stellard is also looking for backup release managers.


## LLVM commits

* The inliner has been taught how to preserve musttail invariants.
[r208910](http://reviews.llvm.org/rL208910).

* A new C API has been added for a thread yielding callback.
[r208945](http://reviews.llvm.org/rL208945).

* Another patch in the series to improve MergeFunctions performance has been
committed. A total ordering has now been implemented among operations.
[r208973](http://reviews.llvm.org/rL208973),
[r208976](http://reviews.llvm.org/rL208976).

* The ARM load/store optimisation pass has been fixed to work with Thumb1.
[r208992](http://reviews.llvm.org/rL208992).

* GlobalValue has been split into GlobalValue and GlobalObject, which allows a
code to statically accept a Function or a GlobalVariable but not an alias.
[r208716](http://reviews.llvm.org/rL208716).

* Integral reciprocal was optimised to not use division. This optimisation was
influenced by John Regehr's [Souper](http://www.cs.utah.edu/~regehr/souper/).
[r208750](http://reviews.llvm.org/rL208750). Another optimisation opportunity
uncovered by Souper was signed icmp of -(zext V).
[r208809](http://reviews.llvm.org/rL208809).

* I rather like that these transforms for single bit tests were [verified with
Z3](http://rise4fun.com/Z3/Pmsh). [r208848](http://reviews.llvm.org/rL208848).

* PowerPC gained global named register support, for r1, r2 and r13 (depending
on the subtarget). [r208509](http://reviews.llvm.org/rL208509).

* Documentation was added for the ARM64 BigEndian NEON implementation.
[r208577](http://reviews.llvm.org/rL208577).

* The constant folder is now better at looking through bitcast constant
expressions. This is a first step towards fixing this [poor performance of
these range
comprehensions](http://ericniebler.com/2014/04/27/range-comprehensions/.
[r208856](http://reviews.llvm.org/rL208856).


## Clang commits

* Initial support for MS ABI compliant RTTI mangling has been committed.
[r208661](http://reviews.llvm.org/rL208661),
[r208668](http://reviews.llvm.org/rL208509).

* Clang will no longer copy objects with trivial, deleted copy constructors.
This fixes bugs and improves ABI compatibility with GCC and MSVC.
[r208786](http://reviews.llvm.org/rL208786). Though the itanium C++ ABI part
was reverted for now. [r208836](http://reviews.llvm.org/rL208836).


## Other project commits

* The LLDB Machine Interface has been committed. This is an implementation of
the [GDB Machine
Interface](https://sourceware.org/gdb/onlinedocs/gdb/GDB_002fMI.html),
useful
for implementing your own frontend to LLDB.
[r208972](http://reviews.llvm.org/rL208972).

* AddressSanitizer started to gain some windows tests.
[r208554](http://reviews.llvm.org/rL208554),
[r208859](http://reviews.llvm.org/rL208859),
[r208873](http://reviews.llvm.org/rL208873) and more.

* The instrumented profiling library API was fixed to work with shared
objects, and profiling is now supported for dlopened shared libraries..
[r208940](http://reviews.llvm.org/rL208940),
[r209053](http://reviews.llvm.org/rL209053).