lld 16 ELF changes

llvm-project 16 was just released. I added some lld/ELF notes to https://github.com/llvm/llvm-project/blob/release/16.x/lld/docs/ReleaseNotes.rst. Here I will elaborate on some changes.

Link speed improved greatly compared with lld 15.0. Notably input section initialization and relocation scanning are now parallel. (D130810) (D133003)
ELFCOMPRESS_ZSTD compressed input sections are now supported. (D129406)
--compress-debug-sections=zstd is now available to compress debug sections with zstd (ELFCOMPRESS_ZSTD). (D133548)
--no-warnings/-w is now available to suppress warnings. (D136569)
DT_RISCV_VARIANT_CC is now produced if at least one R_RISCV_JUMP_SLOT relocation references a symbol with the STO_RISCV_VARIANT_CC bit. (D107951)
DT_STATIC_TLS is now set for AArch64/PPC32/PPC64 initial-exec TLS models when producing a shared object.
--no-undefined-version is now the default; symbols named in version scripts that have no matching symbol in the output will be reported. Use --undefined-version to revert to the old behavior. (D135402)
-V is now an alias for -v to support gcc -fuse-ld=lld -v on many targets.
-r no longer defines __global_pointer$ or _TLS_MODULE_BASE_.
A corner case of mixed GCC and Clang object files (STB_WEAK and STB_GNU_UNIQUE in different COMDATs) is now supported. (D136381)
The output SHT_RISCV_ATTRIBUTES section now merges all input components instead of picking the first input component. (D138550)
For x86-32, -fno-plt GD/LD TLS models call *[email protected](%reg) are now supported. Previous output might have runtime crash.
Armv4(T) thunks are now supported. (D139888) (D141272)

Speed

Link speed has greatly improved compared to lld 15.0.0.

In this release cycle, I made input section initialization and relocation scanning parallel. (D130810 D133003)

% hyperfine --warmup 1 --min-runs 20 "numactl -C 20-27 "{/tmp/out/custom15,/tmp/out/custom16}"/bin/ld.lld @response.txt --threads=8"
Benchmark 1: numactl -C 20-27 /tmp/out/custom15/bin/ld.lld @response.txt --threads=8
  Time (mean ± σ):     676.3 ms ±   4.4 ms    [User: 751.5 ms, System: 472.3 ms]
  Range (min … max):   667.9 ms … 686.1 ms    20 runs

Benchmark 2: numactl -C 20-27 /tmp/out/custom16/bin/ld.lld @response.txt --threads=8
  Time (mean ± σ):     453.5 ms ±   5.3 ms    [User: 760.5 ms, System: 448.9 ms]
  Range (min … max):   445.4 ms … 462.8 ms    20 runs

Summary
  'numactl -C 20-27 /tmp/out/custom16/bin/ld.lld @response.txt --threads=8' ran
    1.49 ± 0.02 times faster than 'numactl -C 20-27 /tmp/out/custom15/bin/ld.lld @response.txt --threads=8'

(--threads=4 => 1.38x (0.7009s => 0.5096s))

Linking a -DCMAKE_BUILD_TYPE=Debug build of clang:

% hyperfine --warmup 2 --min-runs 25 "numactl -C 20-27 /tmp/out/custom"{15,16}"/bin/ld.lld @response.txt --threads=8"
Benchmark 1: numactl -C 20-27 /tmp/out/custom15/bin/ld.lld @response.txt --threads=8
  Time (mean ± σ):      3.664 s ±  0.039 s    [User: 6.657 s, System: 3.434 s]
  Range (min … max):    3.605 s …  3.781 s    25 runs

Benchmark 2: numactl -C 20-27 /tmp/out/custom16/bin/ld.lld @response.txt --threads=8
  Time (mean ± σ):      3.141 s ±  0.028 s    [User: 7.071 s, System: 3.150 s]
  Range (min … max):    3.086 s …  3.212 s    25 runs

Summary
  'numactl -C 20-27 /tmp/out/custom16/bin/ld.lld @response.txt --threads=8' ran
    1.17 ± 0.02 times faster than 'numactl -C 20-27 /tmp/out/custom15/bin/ld.lld @response.txt --threads=8'

(--threads=1 => 1.06x (7.620s => 7.202s), --threads=4 => 1.11x (4.138s => 3.727s))

Linking a default build of chrome:

% hyperfine --warmup 2 --min-runs 25 "numactl -C 20-27 /tmp/out/custom"{15,16}"/bin/ld.lld @response.txt --threads=8"
Benchmark 1: numactl -C 20-27 /tmp/out/custom15/bin/ld.lld @response.txt --threads=8
  Time (mean ± σ):      4.159 s ±  0.018 s    [User: 6.703 s, System: 2.705 s]
  Range (min … max):    4.136 s …  4.207 s    25 runs

Benchmark 2: numactl -C 20-27 /tmp/out/custom16/bin/ld.lld @response.txt --threads=8
  Time (mean ± σ):      3.700 s ±  0.027 s    [User: 7.417 s, System: 2.556 s]
  Range (min … max):    3.660 s …  3.771 s    25 runs

Summary
  'numactl -C 20-27 /tmp/out/custom16/bin/ld.lld @response.txt --threads=8' ran
    1.12 ± 0.01 times faster than 'numactl -C 20-27 /tmp/out/custom15/bin/ld.lld @response.txt --threads=8'

(--threads=4 => 1.11x (4.387s => 3.940s))