LLVM Weekly - #335, June 1st 2020

Welcome to the three hundred and thirty-fifth 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 https://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.

News and articles from around the web

PLDI 2020 (now of course an online conference) will feature a series of "Ask Me Anything" live Q+A sessions, including with Chris Lattner.

On the mailing lists

• Chris Lattner provided a batch of responses to comments on his thread about LLVM decision making and created a new draft of the decision making proposal, this time hosted on Google Docs for ease of commenting.

• Tanya Lattner is seeking volunteers for the LLVM Developers' Meeting Program Committee.

• The 8th MLIR newsletter is now available, containing the usual mix of updates on MLIR commits, discussions, and ecosystem projects.

• David Blaikie started a thread to discuss issues arond DWARF-awareness in the linker, specifically focusing on zero length GCed/duplicated functions.

• Christudasan Devadasan is seeking feedback on allowing CostperUse for registers to be set more dynamically.

• Bill Wendling posted an RFC on loading bitfields with the smallest needed types.

• Chris Tetreault followed up to the thread on refactoring the class hierarchy of VectorType with a summary of a discussion on the SVE call about this.

• Jeremy Morse provided an update on an alternate approach to specifying variable locations post-isel, providing a report on his initial prototype.

• Dean Sutherland provided a summary of how debug locations were handled in a previous commercial compiler. Chris Lattner commented that this is similar to the model MLIR ended up at, and suggested it may be interesting to port this design to LLVM IR.

LLVM commits

• A BFloat MachineValueType was added, and basic support for bfloat was added to AArch64. 0508fb4, 42eba9b, 78bd0c0.

• The hardware loop intrinsics are now documented in LLVM's language reference. 880c35a.

• The llvm.roundeven intrinsic was introduced, implementing the roundToIntegralTiesToEven IEEE-754 operation. 4d20e31.

• FileCheck now allows parenthesized expressions. 3be5e53.

• The VE backend gained a minimal machine code layer, including AsmParser support. 0e0907f.

Clang commits

• An explicit -fno-semantic-interposition will now infer dso_local. 9d55e4e.

• The release notes were updated with a porting guide for AST matchers. 63f927b.

• A new webkit.WebKitNoUncountedMemberChecker was added to the static analyzer. 660cda57.

• A new generate_formatted_state.py tool was added to generate a summary of the clang-format status for the whole LLVM project. 6f56a58.

Other project commits

• Support for constexpr in std::array was overhauled. 77b9abf.

• Basic OpenMP parallel operations were added to the OpenMP MLIR dialect. 5ba874e.