LLVM 19 will soon be released. This post provides a summary of mycontributions in this release cycle to record my learning progress.

• LLVM MC: Integratedassembler improvements in LLVM 19
• lld/ELF: lld 19 ELFchanges
• AArch64
  • [AArch64]Implement -fno-plt for SelectionDAG/GlobalISel
  • [AArch64]Support optional constant offset for constraint "S"
  • [AArch64AsmParser]Allow branch target symbol to have a shift/extend modifier name
  • [MC,AArch64]Create mapping symbols with non-unique names
• ARM
  • [ARM,MC]Support FDPIC relocations. See MMU-less systems andFDPIC
• RISC-V
  • [RISCV]Support constraint "s"
  • [MC,RISCV]Create mapping symbols with non-unique names
• sanitizers
  • [ubsan]Support static linking with standalone runtime
  • [sanitizer]Reject unsupported -static at link time
  • __asan_register_elf_globals:properly check the "no instrumented global variable" case
  • [asan,test]Disable _FORTIFY_SOURCE test incompatible with glibc 2.40

LLVM binary utilities

• [llvm-readobj,ELF]Support --decompress/-z
• [llvm-objcopy]Improve help messages
• [llvm-readelf]Print a blank line for the first hex/string dump
• [llvm-objcopy]Add --compress-sections
• [llvm-readelf]Print more information for RELR

Hashing

I optimized the bit mixer used byllvm::DenseMap<std::pair<X, Y>> andllvm::DenseMap<std::tuple<X...>>.llvm/ADT/Hashing.h, used by StringRef hashingand DenseMap, was supposed to be non-deterministic. Despitethis, a lot of code relied on a specific iteration order. I mademultiple fixes across the code base and landed [Hashing] Use anon-deterministic seed if LLVM_ENABLE_ABI_BREAKING_CHECKS to improvetest coverage (e.g. assertion builds) and ensure future flexibility toreplace the algorithm.

The change has a noticeable code size reduction

1
2
3
4
5
6
7
8
9

# old
movq    _ZN4llvm7hashing6detail19fixed_seed_overrideE@GOTPCREL(%rip), %rax
movq    (%rax), %rax
testq   %rax, %rax
movabsq $-49064778989728563, %rcx       # imm = 0xFF51AFD7ED558CCD
cmoveq  %rcx, %rax

# new
movabsq $-49064778989728563, %rcx

... and significantcompile time improvement.

I optimizedDenseMap::{find,erase}, yielding compile timeimprovement.

The significant code size decrease from replacing Hashing.h bit mixerand simplifying/optimizing DenseMap code highlightspotential implementation challenges. While boost::unordered_flat_map,Abseil's Swiss Table, and Folly's F14 are likely faster, they couldincrease the code size significantly.

NumericalStabilitySanitizer

NumericalStabilitySanitizer is a new feature for the 19.x releases. Ihave made many changes on the compiler-rt part.

Clang

Driver maintenance

• [Driver]Ensure ToolChain::LibraryPaths is not empty for non-Darwin
• [Driver] Support -Wa,--defsym similar to -Wa,-defsym.
• [Driver]Don't default to -mrelax-all for non-RISCV -O0. See Clang's-O0 output: branch displacement and size increase

Options used by the LLVM integrated assembler are currently handledin an ad-hoc way. There is deduplication with and without LTO.Eventually we might want to adopt TableGen for these -Wa,options.

Others:

• [Sema]-Wpointer-bool-conversion: suppress lambda function pointer conversiondiagnostic during instantiation
• [Sema]Mark alias/ifunc targets used and consider mangled names. This patchreceived a lot of scrutiny because it fixed an issue reported the XZbackdoor author(s).
• Makediagnostic pragma override -Werror=foo and DefaultErrorwarnings
• Many patches migrated code off from potentially non-deterministicllvm/ADT/Hashing.h
• [CodeGen]Set attributes on resolvers emitted after ifuncs

Code review

I reviewed a wide range of patches, including areas like ADT/Support,binary utilities, MC, lld, clangDriver, LTO, sanitizers, LoongArch,RISC-V, and new features like NumericalStabilitySanitizer andRealTimeSanitizer.

To quantify my involvement, a search for patches I commented on(repo:llvm/llvm-project is:pr -author:MaskRay commenter:MaskRay created:>2024-01-23)yields 780 results.