LLVM Weekly - #28, Jul 14th 2014
Welcome to the twenty-eighth 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 email@example.com, or @llvmweekly or @asbradbury on Twitter.
I'll be at the GNU Tools Cauldron 2014 next weekend, being held at the University of Cambridge Computer Laboratory (which handily is also where I work). If you're there, do say hi.
The canonical home for this issue can be found here at llvmweekly.org.
News and articles from around the web
An update on Clang/LLVM on Windows has been posted on the LLVM blog. Impressive progress has been made, and as I mentioned last week the MSVC compatibility page has been updated.
There is (somewhat amazingly) now a Pascal-86 frontend for LLVM. The compiler frontend is written entirely in Python. More information is available in the author's Master's thesis (via Phoronix).
On the mailing lists
Edward Diener has kicked off a discussion on whether clang-cl should reproduce buggy VC++ preprocessor behaviour. A concern is that this might hold back people wanting to use preprocessor metaprogramming on the Windows platform. As is pointed out in one of the responses, bug-for-bug compatibility mode could always be disabled via a new flag.
Corey Richardson has a query about using clang-c for a bindings generator. No responses yet, but it's certainly of interest to me.
Peter Collingbourne has proposed offering a native object on-disk module format storing LLVM bitcode in an ELF/COFF/MACH-O section. Dan Liew points out that unrecognised sections are included in the output by the linker, meaning if you linked these .o files the bitcode would be included in the produced binary. Peter says the current plan wouldn't have these object files used directly by the linker.
DominanceFrontier has been templatified, so in theory it can now be used for MachineBasicBlocks (where previously it was only usable with BasicBlocks). r212885.
The quality of results for CallSite vs CallSite BasicAA queries has been improved by making use of knowledge about certain intrinsics such as memcpy and memset. r212572.
Work on overhauling x86 vector lowering continues. Chandler now reports that with the new codepath enabled, LLVM is now at performance pairty with GCC for the core C loops of the x264 code when compiling for SSE2/SSE3. r212610.
ASM instrumentation for AddressSanitizer is now generated entirely in MachineCode, without relying on runtime helper functions. r212455.
Generation of the new mips.abiflags section was added to the MIPS backend. r212519.
isDereferenceablePointer will now look through some bitcasts. r212686.
A new checker was added, to flag code that tests a variable for 0 after using it as a denominator (implying a potential division by zero). r212731.
On the ARM target, LLVM's atomicrmw instructions will be used when ldrex/strex are available. r212598.
Support was adding for mips-img-linux-gnu toolchains. r212719.