LLVM Weekly - #487, May 1st 2023

Welcome to the four hundred and eighty-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, or @llvmweekly or @asbradbury on Twitter, or @llvmweekly@fosstodon.org or @asb@fosstodon.org.

News and articles from around the web and events

nlvm, an Nim language frontend for LLVM has been released. The author provided a summary.

GCC 13.1 was released.

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

• Office hours with the following hosts: Tobias Grosser, Kristof Beyls, Quentin Colombet, Johannes Doerfert.
• Online sync-ups on the following topics: Flang, OpenMP, Clang C/C++ language working group, loop optimisation, OpenMP for flang, MLIR, HLSL, SPIR-V, MLGO, GPU.
• For more details see the LLVM calendar, getting involved documentation on online sync ups and office hours.

On the forums

• Wang Pengcheng proposed introducing functions and lambdas into TableGen, pointing to RISC-V vector scheduling as a motivating use case.

• The dates for the US LLVM Developers' Meeting haven’t been announced yet, but Tanya Lattner indicates the current goal is to hold it in October.

• Daniel Thornburgh started a discussion about supporting linker section packing in LLD. tannewt gave some insight on how this could be used in the context of CircuitPython.

• Aiden Grossman initiated a discussion on platform support required for llvm-exegesis.

• William Junda Huang proposed deprecating the compact binary sample profile format.

• In response to a question, Kiran Chandramohan summarised the status of OpenMP in MLIR.

LLVM commits

• SamplePGO gained a new algorithm to perform better when input profiles are stale. a98d6a1, 892daed.

• The !range operator was added to TableGen. ab2187d.

• Utility functions were introduced for converting between EBCDIC and UTF-8. b42718d.

• The “adding a builder” documentation was updated. 73f4f56.

• The “compiler launcher” (e.g. ccache) is now disabled on external projects and for the later stages of a multi-stage clang build. As these build steps use a fresh from-source toolchain build, they pollute the build cache with no chance of speeding things up. e38cdc5.

• MachineValueType.h is now fully generated and IIT_Info was moved to Intrinsics.td. ddaf085, c49f850.

• llvm.is.fpclass is now custom-lowered for RISC-V. 5b869f3.

• powi, log, and frem are now handed in computeKnownFPClass. 8e72219, 2ff52ea, de7bfec.

• Vector strict rounding operations are now supported on RISC-V. 1855c0a.

• The preserve_all calling convention was added for AArch64. d75e70d.

• denormal-fp-math now has a “dynamic” option. bc37be1.

• -fsanitizer=kernel=memory was enabled for SystemZ. a3e56a8.

Clang commits

• LoongArch multiarch triples are now supported in the Clang driver. 61fee67.

• Stale projects were removed from the Clang open projects page, and some new projects added. 5e10cd7, f0630a3.

• Gather intrinsics descriptions were added to avx2intrin.h. 039ae62.

• __attribute__((interrupt("user"))) is no longer supported for RISC-V. This was part of the ‘N’ extension that didn’t make it into the ratified privileged specification. 05d0cae.

• Flag output inline asm constraints on AArch64 are now supported. 4ab76b7.

• The riscv_rvv_vector_bits(N) attribute was added, based on AArch64 arm_sve_vector_bits. 42e79d9.

Other project commits

• An integer bitwidth narrowing pass was added to MLIR, as well as a generic mem2reg implementation. da0730b, f88f8fd.

• A GDB/LLDB interface was added for interactive debugging of MLIR actions. 1020150.

• A CI job was added for pstl in libcxx. 2445603.

• Short-range thunk support was added for AArch64 in LLD. d0cdc5d.