LLVM Weekly - #557, September 2nd 2024

Welcome to the five hundred and fifty-seventh 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.

News and articles from around the web and events

The next Cambridge Compiler Social will take place on 3rd September 2024, featuring not one but two talks: Quidditch: An end-to-end deep learning compiler for highly-concurrent accelerators with software-managed caches (Markus Böck) and Mojo’s Wishlist for MLIR 2.0 (Jeff Niu).

The call for proposals for the MLIR workshop at the LLVM Developer Meeting is open.

According to the LLVM calendar in the coming week there will be the following:

• Office hours with the following hosts: Renato Golin, Anastasia Stulova, Quentin Colombet, Johannes Doerfert,
• Online sync-ups on the following topics: Flang, pointer authentication, AArc64, libc++, new contributors, LLVM/Offload, C/C++ language working group, OpenMP for Flang, MLIR, HLSL, MLGO.
• For more details see the LLVM calendar, getting involved documentation on online sync ups and office hours.

On the forums

• “H. Vetinari” proposed distributing patch releases over the lifecycle of an LLVM version in order to avoid a situation where there are known issues to be addressed, but no more planned point releases and a long wait to the next major release.

• David Spickett requested feedback on the idea of authoring an LLVM blog post for the flang-new to flang renaming.

• Miguel A. Arroyo started an RFC thread on adding support for basic block sections on Windows (COFF).

• Tulio Magno Quites Machado Filho suggested that LLVM/Clang builds should change to avoid exposing information on downstream git repositories that may be sensitive (e.g. URL).

• Michael Maitland posted an RFC on support for vectorisation of conditional scalar assignment.

• Chris Bieneman provided an update on the LLVM governance proposal discussion.

LLVM commits

• The semantics of the experimental.get.vector.length intrinsic were made more strict in order to enable more optimisations. e806370.

• The benchmark apps included in the LLVM test-suite were documented. c4a5381.

• Optimisation remarks can now be emitted for ArgumentPromotion and DeadArgumentElimination. 470f55f.

• The amdgpu-sw-lower-lds pass was introduced, lowering local data store (LDS) uses to use dynamically allocated global memory. 7bc9d95.

• The DXIL pretty printer now handles target extension type resources. 87157ab.

• The stepvector intrinsic was moved out of the experimental namespace. 95d2d1c.

• SimplifyCFG gained a new option to hoist load/stores with conditional faulting. 87c86aa.

• The Windows release packaging now only builds the AArch64, Arm, and X86 targets. 2a28df6.

• The llvm.fake.use intrinsic was introduced to extend variable lifetimes. 3d08ade.

Clang commits

• The clang-tidy contributing guide was updated. 362d37a.

• RISC-V vector crypto intrinsics are no longer considered experimental. 051054e6f.

• HLSL multi-argument resolution in Clang vs DXC was documented. 02654f7.

• RISC-V vector tuple type support was added. 239127d.

Other project commits

• Range extension thunks are now supported in LLD for ARM64EC targets. efad561.

• libcxx’s lit test suite now runs against a version of the library installed in a dummy location in order to more closely approximate the library as shipped to users. 0e8208e.

• lldb-dap gained support for instruction breakpoints. 89c27d6.