LLVM Weekly - #49, Dec 8th 2014

Welcome to the forty-ninth 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 https://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.

News and articles from around the web

Most of the 2014 LLVM Developers' Meeting videos and slides are now online. Sadly, there are no videos from the talks by Apple employees yet. Hopefully they'll be appearing later.

QuarksLab has a rather nice write-up of deobfuscating an OLLVM-protected program.

The LLVM-based ELLCC has been making progress on ELK, a bare-metal POSIX-like environment.

Support for statepoints landed in LLVM this week, and Philip Reames has a blog post detailing some notes and caveats. See also the mailing list discussion linked to below about future plans for GC in LLVM.

On the mailing lists

• Sami Liedes shares his workflow for using afl-fuzz with the Clang test suite. In 11 hours of testing he managed to find 34 distinct assertion failures and at least one segmentation fault.

• Duncan P.N. Exon Smith has shared an updated on his work on the metadata-value split, detailing the new semantic restrictions this will entail.

• Philip Reames has a post detailing his future plans for GC in LLVM. Comments are invited. The aim is to eventually delete the existing gcroot lowering code. If you are actively using this, please do speak up.

• John Yates, who worked on the compiler for the Apollo Computer's DN10K has shared a description of how that compiler would have handled one of the recent examples from the Souper work.

• How can you reproduce Clang's -O3 using opt? The answer, thanks for Tobias Grosser, is clang -O3 -mllvm -disable-llvm-optzns followed by opt -O3.

• Tobias Grosser is seeking community feedback on where in the pipeline the Polly loop optimiser should run. The post is well worth a read for the discussion of expected trade-offs.

• Rafael Espíndola has been working on type merging during LTO and ultimately proposes moving to a single pointer type in LLVM IR. There seems to be positive feedback on the idea, given that pointer types don't convey useful information to the optimizer and don't really provide safety.

LLVM commits

• The statepoint infrastructure for garbage collection has landed. See the final patch in the series for documentation. r223078, r223085, r223137, r223143.

• The LLVM assembler gained support for ARM's funky modified-immediate assembly syntax. r223113.

• The OCaml bindings now has a CMake buildsystem. r223071.

• The PowerPC backend gained support for readcyclecounter on PPC32. r223161.

• Support for 'prologue' metadata on functions has been added. This can be used for inserting arbitrary code at a function entrypoint. This was previously known as prefix data, and that term has been recycled to be used for inserting data just before the function entrypoint. r223189.

• PowerPC gained a Power8 instruction schedule definition r223257.

Clang commits

• LLVM IR for vtable addresses now uses the type of the field being pointed to, to enable more optimisations. r223267.

• New attributes have been added to specify AMDGPU register limits. This is a performance hint that can be used to attempt to limit the number of used registers. r223384.

• Clang gained the __has_declspec_attribute preprocessor macro. r223467.

• __has_attribute now only looks for GNU-style attributes. You should be able to use __has_cpp_atribute or __has_declspec_attribute instead. r223468.

Other project commits

• DataFlowSanitizer is now supported for MIPS64. r223517.

• libcxx now supported std::random_device on (P)NaCl. r223068.

• An effort has started in lld to reduce abstraction around InputGraph, which has been found to get in the way of new features due to excessive information hiding. r223330. The commit has been temporarily reverted due to breakage on Darwin and ELF.

• A large chunk of necessary code for Clang module support has been added to LLDB. r223433.

• LLDB now has documented coding conventions. r223543.