LLVM Weekly - #94, Oct 19th 2015
Welcome to the ninety-fourth 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.
A good time was had by all at ORConf last week at CERN. We had over 100 open source hardware enthusiasts join us in Geneva. You can find my slides updating on lowRISC here. Videos should appear on youtube in the next week or so.
The canonical home for this issue can be found here at llvmweekly.org.
News and articles from around the web
A six month retrospective of LLILC, the project to produce an open source LLVM-based compiler for .NET, has been posted. It describes work still to be done for garbage collection and exception handling, code size and code quality, and JIT throughput.
The schedule for the 2015 LLVM Developers' Meeting is now available.
The new ELF linker in LLD is looking pretty fast. Right now it can link Clang in about half the time of binutils gold. However, the resulting binary is larger. It will be interesting to see how the performance compares when both are at feature parity, but this is looking promising.
On the mailing lists
Chandler Carruth, on behalf of the board of the LLVM Foundation has posted an RFC on introducing an LLVM Community Code of Conduct. The proposal is based on the Django code of conduct and generated masses of discussion. A couple of days later, Chandler posted a second draft incorporated feedback and answering many of the questions raised. The response appears to be good so far. I'll just highlight one of the questions and answers: "Q: Is this trying to change how the community behaves?" "A: I think the resounding answer is no, this is very much meant to formalize the existing extremely polite and respectful behavior that the LLVM community has had for many years."
There is going to be a birds of a feather section about the future of LLVM's C APIs at the upcoming LLVM developers' meeting, and Justin Bogner has helpfully shared some notes in preparation for this.
Philip Reames has shared some suggested topics for the managed languages birds of a feather meeting at the upcoming LLVM devmeeting. Joe Ranieri suggests some additional topics.
Chris Matthews has shared an RFC on adding background workers to LNT.
Diego Novillo is going to be hosting a birds of a feather on profile-guided optimisations at the upcoming dev meeting and has shared a preliminary list of topics for discussion.
Sanjoy Das has updated us on his work with operand bundles and gc transition arguments, and is seeking input and opinions on his suggested ways forward.
Zachary Turner has written about his efforts to support Python 3 with LLDB.
Evgenii Stepanov has posted an RFC on adding an internal linkage attribute. The message explains why setting
always_inline
and hidden symbol visibility is not enough.Arch Robison initiated a discussion on extending the SLP vectorizer to work with tuples in Julia.
LLVM commits
Hexagon gained a new pass to merge adjacent stores. r250542.
Hexagon gained skeleton support for the 'HVX' extension instructions. r250600.
The loop vectorizer will now shrink integer operations into the smallest type possible. r250032.
Documentation has been added for binary sample profile encoding. r250309.
RewriteStatpointsForGC is starting to make use of operand bundles. r250489.
Clang commits
Clang gained support for the
-fdebug-prefix-map=
option as in GCC. r250094.The PS4 toolchain definition has been added to Clang. r250262.
Clang now understands
-flto=thin
. r250398.