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).