LLVM Weekly - #543, May 27th 2024
Welcome to the five hundred and forty-third 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 protected], @llvmweekly or @asbradbury on Twitter, or @[email protected] or @[email protected].
News and articles from around the web and events
Min-Yih Hsu blogged about legalizations in LLVM backends.
According to the LLVM calendar in the coming week there will be the following:
- Office hours with the following hosts: Johannes Doerfert.
- Online sync-ups on the following topics: Flang, new contributors, pointer authentication, LLVM/Offload, classic flang, loop optimisations, OpenMP, MLIR.
- For more details see the LLVM calendar, getting involved documentation on online sync ups and office hours.
On the forums
Tanya Lattner shared a save the date for the US LLVM Developers' Meeting, which will take place October 22-24th in Santa Clara. Applications are open to volunteer for the program and travel grant committees.
“pablo” posted an MLIR RFC on adding
transform.loop.tile_indirection_using_forall.Corentin Ferry proposed removing signless type support in MLIR’s EmitC dialect.
There was quite a lot of further discussion about the idea of introducing new ‘one-short’ dialect conversion driver in MLIR. See in particular, Alex Zinenko’s discussion of the history behind the current design and assessment of the current state.
Eric Fiselier posted an RFC aiming to end a point of disagreement between libc++ contributors concerning support for “in-tree headers” during development.
LLVM commits
Conditional indirect call promotion using vtable-based comparison was implemented. 5d3f296.
constexpr getelementptrs are now canonicalised to i8 element types. 8e8d259.
Contributions to LLVM as of June 1st will no longer need to be licensed under both the current and legacy license. ee76f1e.
The
-call-latencycommand-line option to llvm-mca can be used to control the assumed latency of a call instruction. 848bef5.AMDGPU-specific module splitting was implemented, allowing
--lto-partitionsto work more consistently. d7c3713.Inline assembly is now supported for SPIR-V. 214e6b4.
For RISC-V vector, the vsetvl insertion pass now runs after RVV register allocation. 675e7bd.
Clang commits
Work started to move away from Clang assuming the default address space is AS 0. 10edb49.
Sized deallocation was enabled by default in C++14 onwards. 130e93c.
The security.SetgidSetuidOrdr checker was implemented in Clang’s static analyzer. 11b97da.