LLVM Weekly - #39, Sep 29th 2014
Welcome to the thirty-ninth 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 http://llvmweekly.org and pass it on to anyone else you think may be interested. Please send any tips or feedback to firstname.lastname@example.org, or @llvmweekly or @asbradbury on Twitter.
The canonical home for this issue can be found here at llvmweekly.org.
News and articles from around the web
An implementation of Common Lisp with an LLVM backend, Clasp, has been announced. There's a lot of work to be done on performance, but development is very active on Github.
A backend for the educational 'y86' instruction set architecture has been started. The source is on Github.
A new binary snopshot of the ELLCC cross compilation toolchain is now available. Pre-compiled binaries are available for ARM, MIPS, PPC, and x86. All tarballs contain header files and runtime libraries for all targets to allow you to build for any supported target.
On the mailing lists
Bill Schmidt started off an interesting discussion about whether sqrt of a negative number should be optimised to 0 or NaN. The xalanc benchmark in SPEC CPU2006 relies on it returning a NaN apparently.
Chandler Carruth has posted an RFC proposing that LLVM should require a working C++11 thread, mutex, and atomic with the intention being that support for MinGW and win32threads is dropped. Switching to MinGW-win64 seems like a feasible path forwards.
Wondering how to use noalias and alias.scope metadata notations? Hal Finkel has the answer for you.
Should the LLVM project standardise on a commit message policy? Renato Golin suggests trying to keep the first line short followed by some number of 80 character paragraphs. It seems there's massive agreement on this sort of guidance.
Richard Pennington has posted a size comparison of bzip2 compiled for different architectures. All executables are statically linked with musl and compiler-rt.
Segmented stacks support for the x32 ABI has been fixed. r218247.
Robin Morisset's work on optimisation of atomics continues. AtomicExpandPass now inserts fences itself rather than SelectionDAGBuilder. r218329.
LLVM's libSupport gained a type-safe alternative to
llvm-vtabledump learned how to dump RTTI structures for the MS ABI. r218498.
assume_alignedfunction attribute is now supported. r218500.
The thread safety analysis documentation has seen a hefty update. r218420.
MS compatibility is further improved with support for the
__superscope specifier. r218484.