LLVM Weekly - #521, December 25th 2023

Welcome to the five hundred and twenty-first 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, @llvmweekly or @asbradbury on Twitter, or @llvmweekly@fosstodon.org or @asb@fosstodon.org.

For those who celebrate it, Merry Christmas!

News and articles from around the web and events

The next LLVM bay area meetup will take place on Jan 29th.

A few more LLVM Dev Meeting videos have been posted on YouTube.

I didn’t spot any more LLVM related news articles over the last week, but if you’ll allow me a little bit of cross-promotion, you might be interested in my recent posts on a somewhat overlooked reference counting paper and my appreciation for terminal bells.

According to the LLVM calendar in the coming week there will be (note: I’d double check if these are still going ahead - I’d imagine with the holidays many of these are events that just weren’t cancelled on the calendar):

• Office hours with the following hosts: Johannes Doerfert, Amara Emerson.
• Online sync-ups on the following topics: Flang, libc++, OpenMP, classic flang, loop optimisations.
• For more details see the LLVM calendar, getting involved documentation on online sync ups and office hours.

On the forums

• Konstantin Varlamov posted an RFC on adding __builtin_verbose_trap. Unlike __builtin_trap, this provides a message about what went wrong.

• Ariel Burton suggested adding support for IBM’s hexademical floating-point format.

• I think I missed this thread first time round. Johannes Doerfert started a thread about unifying clang flags / language options for offloading languages.

• The 59th edition of MLIR news is now available.

• Oskar Wirga proposed extending Clang’s CFI to support dynamically resolved function pointers.

• Anton Lydike provided a summary of the discussion on an MPI MLIR dialect so far and proposed a path forwards.

• Ramkumar Ramachandra suggested a Mastodon instance for the LLVM community (e.g. something like social.llvm.org, similar to social.kernel.org). Though based on feedback received, noted that the moderation cost may be too high.

LLVM commits

• Usage instructions were added for FatLTO. 8190e09.

• The dominance algorithm used in RegAllocFast was made O(1) rather than O(n). 40ec791.

• Release branches will now start numbered as N.1.0. 4532617.

• LoopAccessAnalysis will now hoist runtime checks by default. 49b0e6d.

• The X86 assembly parser learned to error for an out of range displacement. 83680f8.

• The RISC-V vector crypto extensions are no longer marked as experimental, however the C intrinsics are not yet finalised. 0905865.

• The AArch64LoadStoreOptimizer pass can now merge an index address with a large offset into the base address. 32878c2.

• MC layer and codegen support was introduced for the Armv9.5-A pointer authentication code enhancement instructions. 192f720, 7bd1721.

Clang commits

• The multilib reuse logic for RISC-V bare-metal toolchains was improved. 379d32d.

• Support for the counted_by attribute was reverted. cca4d6c.

Other project commits

• The libcxx codebase was reformatted. 9783f28.

• ranges::contains and ranges::fold_left[_with_iter] was implemented in libcxx. fdd089b, 3903438.

• Non-interprocedural dataflow analysis is now supported in MLIR, rather than just interprocedural analysis. 32a4e3f.

• The emitc.expression operation was introduced for the MLIR emitc dialect. d980384.