LLVM Weekly - #252, October 29th 2018
Welcome to the two hundred and fifty-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 https://llvmweekly.org and pass it on to anyone else you think may be interested. Please send any tips or feedback to [email protected], or @llvmweekly or @asbradbury on Twitter.
News and articles from around the web
Stephen Kelly has published blog posts on extending clang-tidy and examining the Clang AST with clang-query.
On the mailing lists
-
Alexander Ivchenko posted an RFC on implementing asm-goto support in Clang and LLVM. One of the main motivations is to support the GCC inline assembly extension used by the Linux kernel. But the mechanism is more general and may support control flow tracking for setjmp/longjmp, or intrinsics with multiple successors (such as an atomic cmpxchg primitive lowered to ll/s with two successor blocks).
-
Graham Hunter reports that he has posted a proposal for scalable vector types to Phabricator.
-
Kristof Beyls summarised the discussion from the LLVM bug cycle Birds of a Feather session at the LLVM Dev Meeting.
-
Reid Kleckner proposes moving llvm.dbg.value out of the instruction stream.
-
James Y Knight continues to work on the new git canonical LLVM git repository. Although the intent is to mark it "final" as soon as possible, I haven't seen that declaration yet. Therefore, now is the time to jump in and check for any issues.
-
Anastasia Stulova summarised the LLVM Dev Meeting round table discussion on SPIR-V. The preference from the developers attending that meeting was to continue development of the SPIR-V translation tool.
LLVM commits
-
The AArch64 backend gained a new pass to compress jump-table entries when possible. r345188.
-
Documentation and a script were added for building clang+LLVM with PGO. r345427.
-
The bisect-skip-count utility gained documentation. r344903
-
New
FMINNUM_IEEE/FMAXNUM_IEEESelectionDAG nodes were introduced to perform floating-point min/max following the IEEE-754 2008 definition. r344914. -
FixedLenDecoderEmitter can now use APInt-like objects as InsnType, useful for targets where
uint64_tis not sufficient for encoded instructions. r345056. -
A list of potential benchmarks for the LLVM test-suite has been committed. r345074.
Clang commits
-
Clang command-line options were added for AArch64 branch protection and return address signing. r345273.
-
Function multiversioning is now supported non-ELF systems. r345298.