LLVM Weekly - #494, June 19th 2023
If you prefer, you can read a HTML version of this email at http://llvmweekly.org/issue/494.
Welcome to the four hundred and ninety-fourth 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 asb@asbradbury.org, or @llvmweekly or @asbradbury on Twitter, or @llvmweekly@fosstodon.org or @asb@fosstodon.org.
News and articles from around the web and events
The call for speakers is now open for the 2023 LLVM Dev Meeting, with submissions due by July 31st. Additionally, there’s a thread asking about what tutorials people would like to see and a call for workshops (with several linked threads already created to gauge interest in various topics).
The toolchains track at the Linux Plumbers Conf'23 has an open call for participation, with submissions due by September 25th.
The next LLVM Social in Berlin is taking place this Wednesday June 21st and will feature a presentation from Lawrence Benson on SIMD vector abstractions.
LLVM 16.0.6 was released. This is intended to be the last 16.0.x release.
Herb Sutter’s trip report from the Summer ISO C++ standards meeting is now up.
According to the LLVM calendar in the coming week there will be:
- Office hours with the following hosts: Phoebe Wang, Johannes Doerfert.
- Online sync-ups on the following topics: LLVM Security Group, OpenMP, Clang C/C++ language working group, Flang, RISC-V, MLIR, LLVM embedded toolchains, SPIR-V.
- For more details see the LLVM calendar, getting involved documentation on online sync ups and office hours.
On the forums
Sjoerd Meijer kicked off an RFC discussion on an assembly super optimiser. The current proof of concept focuses on instruction scheduling and uses the Z3 SAT solver.
Chris Lattner summarised the outcome of the discussion on the proposal to rename flang-new to flang including a summary of the proposal that all parties agreed with.
Matt Arsenault shared some early thoughts on improving strictfp support and changing our current approach with constrained FP intrinsics.
Prem Chintalapudi suggests introducing memfill and memtransfer intrinsics.
Joshua Cranmer proposes new metadata for loads and stores to help solve a semantics gap around the @llvm.memcpy intrinsic.
Johannes Doerfert initiated a discussion on conformance and timing CI for GPUs, in the context of the efforts to build a libm for the GPU.
Philip Reames shared a heads up on planned changes to the RISC-V vector pseudos.
Stephen Tozer posted a draft implementation of the two-level line tables proposal in LLVM.
In response to a user question, Artem Dergachev gave some pointers on how to implement a checker
Slides and a recording from part 2 of the Numba presentation at the MLIR open meeting are now available.
LLVM commits
The DAG combiner now adds nodes back to the worklist in topological order rather than reverse topological order. a70d5e2.
A scheduling model was added for the Neoverse V2. 887362d.
SLP is now enabled by default for the RISC-V backend when the V extension is enabled. 7f26c27.
Clang commits
The undefined behaviour sanitizer documentation was improved. ee0367e.
Multilib layering is now supported in Clang, and a design doc was committed. a5aeba7, ab2c80b, edc1130.
The Clang interpreter gained support for virtual function calls. 976d8b4.
TargetInfo.cpp was broken up into separate files per target. 992cb98.
The
__builtin_isfpclassfunction was added. 7dd387d.