LLVM Weekly - #18, May 5th 2014
Welcome to the eighteenth 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.
I'm going to be in the San Francisco area May 13th-20th with some other Raspberry Pi people. We'll be at Maker Faire Bay Area on the 17th and 18th. Let me know if there's anything else I should check out while over there.
The canonical home for this issue can be found here at llvmweekly.org.
News and articles from around the web
Andrew Ruef has written a blog post about using static analysis and Clang to find the SSL heartbleed bug. The code for the checker described in the blog post is available on Github.
Eli Bendersky has written an article about how to use libTooling to implement source to source transformations.
The next Paris LLVM Social will take place on May 5th (i.e. this evening).
The LLVM Bay Area social will take place on May 8th. Please RSVP if you are interested.
On the mailing lists
Discussion is still going on regarding last week's proposal to add intrinsics for safe division. Philip Reames has shared his reasoning againt the proposal. This triggered a lot of discussion which is unfortunately more than I have time to summarise.
Kevin Modzelewski is working on exceptions for Pyston, the LLVM-based Python JIT and asks for advice/experiences on how others have implemented them in projects like VMKit. There are a number of useful responses.
Yi Kong has suggested improvements to LNT for benchmarking. One of the concerns raised in followup responses is that benchmark results may not be stable and may take too long to be practical on e.g. ARM hardware. Chris Matthews suggests performing a geometric mean similar to the SPEC benchmarks as an alternative to a plain total execution time metric.
Dario Domizioli has posted an RFC for an optnone pragma. This proposal triggered robust discussion about whether there is any advantage to supporting this in addition to the optnone function attribute, though it seems as though there is enough support for the feature for it to make its way in.
The patch to perform common subexpression elimination for a group of getelementptrs that was discussed a couple of weeks ago has been merged. It is currently only enabled for the NVPTX backend. r207783.
X86 code generation has been implemented for the musttail function attribute. r207598.
Pass run listeners were added to the pass manager. This adds C/C++ APIs to enable fine-grain progress report and safe suspension points. See the commit message for more info r207430.
The SLPVectorizer gained the ability to recognize and vectorize intrinsic math functions. r207901.
NRVO (named return value optimisation) determination was rewritten. According to the commit message, "a variable now has NRVO applied if and only if every return statement in that scope returns that variable." Also, NRVO is performed roughly 7% more often in a bootstrap clang build. r207890.
libclang's documentation comment API has been split in to a separate header. r207392.
The libclang API now supports attributes 'pure', 'const', and 'noduplicate'. r207767.
The comment parser no longer attempts to validate HTML attributes (the previous solution was insufficient). r207712.