[llvm-dev] LLVM Weekly - #343, July 27th 2020 (original) (raw)
Alex Bradbury via llvm-dev llvm-dev at lists.llvm.org
Mon Jul 27 12:15:20 PDT 2020
- Previous message: [llvm-dev] Discourse category for the AMDGPU target
- Next message: [llvm-dev] Difficulties with arc land
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
LLVM Weekly - #343, July 27th 2020
If you prefer, you can read a HTML version of this email at <http://llvmweekly.org/issue/343>.
Welcome to the three 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 <asb at asbradbury.org>, or @llvmweekly or @asbradbury on Twitter.
News and articles from around the web
LLVM 10.0.1-final has been tagged.
Hassam Uddin wrote up a quick introduction to register allocation. Note this is a high-level introduction, and isn't tied to any particular compiler framework (sample code is given in OCaml).
On the mailing lists
Arthur Eubanks started a thread about the new pass manager status, which spawned discussion bout flipping to the new pass manager by default. Sjoerd Meijer shared concerns related to observed code size regressions.
Varun Gandhi suggested that LLVM should explain the lack of C++ ABI stability in the developer policy, prompting a reasonable amount of discussion about the policy and the possible use of deprecation notices.
Alexander Richardson posted an RFC on requiring explicit address space arguments for PointerType, motivated by reducing the difficulty of updating out-of-tree targets like CHERI as well as avoiding bugs for in-tree targest like AVR that use non-zero pointer address spaces.
David Blaikie started a thread to discuss a potential misoptimisation with zero-length functions.
Tanya Lattner sent a reminder that applications for the LLVM Foundation Board of Directors are due by July 30th.
LLVM commits
The CMake minimum version has been bumped to 3.13.4. afa1afd.
Target-specific InstCombine code has now been moved into the targets. 2a6c871.
An "extract" test utility was added, to support multiple documents / test cases in a single test file. d054c7e.
update_test_checksandupdate_cc_test_checksgained a new option to allow function attributes to be checked. 937bad3.A new
byrefargument attribute was introduced. Unlike byval, it doesn't imply that a copy is made anywhere or that the argument is passed on the stack. 5e999cb.AMDGPU GlobalISel gained initial support for calls. 61f1f2a.
A "development-mode InlineAdvisor" was added, supporting loading models with a command-line parameter and collecting training logs. 70f8d0a.
Loop peeling support was added to the loop fusion pass. 38a8217.
The abs/umin/umax/smin/smax intrinsics were added. fef0cf0, deb4bb2.
As part of the work to integrate lowering of GC live operands into the register allocator, GC relocation can now be lowered to virtual registers. 3da1a96.
Clang commits
A new
--ld-pathcommand line flag was added, and should be used in favour of-fuse-ldfor specifying the path to the linker. 1bc5c84.clang-format gained a new BitFieldColonSpacing option 52ab7aa.
Other project commits
LLVM's libc gained implementations of strstr, fmin, fminf, fminl, strnlen, fmax, fmaxf, fmaxl and memrchr. d080635, 7ce32f8, f773d37, 4096088, 5a9630b.
The
[[clang::trivial_abi]]annotation has now been added tostd::unique_ptrin libcxx, allowing further optimisations. 76887bc.MLIR's SPIR-V to LLVM conversion is now documented. b74ab49.
lld-macho gained partial support for weak definitions. 31d5885.
LLDB now has support for AArch64 SVE register information and core file register access. 7e017de.
- Previous message: [llvm-dev] Discourse category for the AMDGPU target
- Next message: [llvm-dev] Difficulties with arc land
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]