LLVM Weekly - #31, Aug 4th 2014
Welcome to the thirty-first 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 asb@asbradbury.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
Renato Golin has posted a summary of feedback from his talk on LLVM and GCC collaboration at the GNU Tools Cauldron. This both summarises the main areas he's looking for collaboration, and the feedback that people gave at the event or afterwards.
This blog post describes how to use Obfuscator-LLVM to to obfuscate Android NDK binaries.
On the mailing lists
Johannes Kapfhammer, a Google Summer of Code student has posted an RFC on adding a fuzzy parser for highlighting C++. The Clang parser can't be used for this as it may be highlighting incomplete snippets where not all types or functions are included. It was pointed out in responses that this is similar to clang-format's parser, but apparently this parser is not easily reusable and very tied to the clang-format implementation.
In response to a query about whether there is interest in a performance tracking BOF at the LLVM Dev meeting, Renato Golin has posted a summary of related issues discussed at the last meeting and progress made on them.
John Kåre Alsaker posted a question about adding a stack probe function attribute. This would be used for languages like Rust that attempt to detect stack overflow with the use of a guard page. Philip Reames responded with a number of thoughts on this feature.
Daniel Stewart has implemented ELF/AArch64 support for lld and is looking for how to upstream it. People are happy for a single patch to be posted for review.
In response to a question about documentation for adding builders to the LLVM buildbot service, Dan Liew has posted a summary of how he has done it. He's looking for feedback on whether this is the best way to do things.
John Kåre Alsaker has asked for feedback on enhancing BasicAliasAnalysis for Rust. Rust's restrictions on aliasing are not currenly fully taken advantage of. Both Hal Finkel and Cameron Zwarich respond with thoughts on how to proceed.
Tom Stellard proposes renaming the R600 target to AMDGPU. The motivation is that the backend has the name since the R600 was the first AMD GPU targeted, but it has added support for all AMD GPUs since then. There seems to be agreement this would be a sensible renaming.
LLVM commits
FastISel for AArch64 saw a number of improvements, including support for shift-immediate, arithmetic with overflow intrinsics. r214345, r214348, and more.
The SLPVectorizer has seen a largeish commit that implements an "improved scheduling algorithm". Sadly the commit message offers no further details. r214494.
TargetInstrInfo gained
isAsCheapAsMovewhich takes a MachineInstruction and returns true if that instruction is as cheap as a move instruction. r214158.LLVM libraries can now be exported as importable CMake targets, making it easier for those building LLVM-based projects. This is now documented. r214077.
Release notes for PowerPC changes during 3.5 development have been committed. r214403.
Initial work towards supporting debug locations for fragmented variables (e.g. by-value struct arguments passed in registers) has been committed. r214576.
Clang commits
Work on support for the MSVC ABI continues. Clang will now consider required alignment constraints on fields. r214274.
AddressSanitizer now passes source-level information from Clang to ASan using metadata rather than by creating global variables. r214604.
The PowerPC backend now support selection of the ELFv1/ELFv2 ABI via the
-mabi=option. r214074.