LLVM Weekly - #122, May 2nd 2016
Welcome to the one hundred and twenty-second 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.
The canonical home for this issue can be found here at llvmweekly.org.
News and articles from around the web
GCC 6.1 has been released. Perhaps the most apparent user-visible change is that the C++ frontend now defaults to C++14.
The Rust compiler has introduced a new intermediate representation, MIR, used for optimisations prior to lowering to LLVM IR.
Tanya Lattner has written about the LLVM Foundation's plans for 2016. The LLVM Foundation has established 3 main programs: Educational Outreach, Grants and Scholarships, and Women in Compilers and Tools.
On the mailing lists
Dean Michael Berris has shared an RFC on upstraming Google's 'XRay' function call tracing system. For more information, you can read the XRay whitepaper.
Sanjoy Das has suggested generalising the AssumptionCache to AxiomAche. He proposes maintaining separate lists of guards and assumptions within the AxiomCache.
There's been some more activity in response to Phil Tomson's question about instruction scheduling. Christof Douma followed up with some advice.
Chris Bieneman has suggested raising the CMake minimum version to 3.4.3. Renato Golin flagged up some concerns, with Chandler Carruth offering a counterpoint.
Discussion has continued on the proposal to introduce a new LLVM sub-project for parallelism runtime and support libraries. This is probably best summarised by reading Hal Finkel's thoughts on the way forward, C Bergström's concerns, and Jason Henline's proposed charter for the subproject.
Peter Collingbourne has shared an RFC on redesigning the LLD symbol table in order to improve memory locality.
LLVM now supports indirect call promotion based on value-profile information. This will promote indirect calls to a direct call guarded by a precondition. r267815.
The LLVM documentation has been extended with a CMake primer covering the basics of the CMake scripting language. r268096.
The PDB dumper has been refactored into a library. r267431.
The MinLatency attributed has been removed from SchedMachineModel. r267502.
CodeGenPrepare will now use branch weight metadata to decide if a select should be turned into a branch. r267572.
llvm.loop.distribute.enablemetadata was added. This indicates a loop should be split in to multiple loops. r267672.
The SystemZ backend now supports the Swift calling convention. r267823.
libFuzzer's documentation has been expanded and improved. r267892.
clang-tidy gained a new checker for redundant expressions on both sides of a binary operator. r267574.
A new clang-tidy check will warn for use of functions like
atolthat don't report conversion errors. r268100.
nodebugattribute on a global or static variable will now suppress all debug info for that variable. r267746.
A number of OpenMP features gained codegen support, such as the map clause and target data directive. r267808, r267811.
Other project commits
LLD now supports an
-O0option to produce output as quickly as possible. Currently this disables section merging at the cost of a potentially much larger output. r268056.
The symbol table in LLD's ELF linker has been redesigned with the intent of improving memory locality. The new design produces measurable speedups for the binaries tested in the commit message. r268178.
LLD's linkerscript support expanded to encompass comparison operators. r267832.
LLD performance on large executables has been improved by skipping scanRelocs on sections that are never mapped to memory at runtime (e.g. debug sections). r267917.