Skip to content

Latest commit

 

History

History
77 lines (52 loc) · 1.7 KB

benchmarks.md

File metadata and controls

77 lines (52 loc) · 1.7 KB
Raw

Benchmarks

LLVM Compilation

For this benchmark, LLVM (2021-01-17, Git commit: cfec6cd50c36f3db2fcd4084a8ef4df834a4eb24) was built from source as follows:

mkdir build && cd build
cmake -DCMAKE_C_COMPILER_LAUNCHER=buildcache -DCMAKE_CXX_COMPILER_LAUNCHER=buildcache -G Ninja -DCMAKE_BUILD_TYPE=Release ../llvm
time ninja

The system used for the benchmark was:

  • CPU: AMD Ryzen 7 1800x (8-core x86-64, underclocked to 3.0 GHz).
  • Disk: 1TB NVMe (960 EVO)
  • RAM: 32GiB DDR4 @ 3000 MHz
  • OS: Linux (Ubuntu 20.04)
  • Compiler: GCC 9.3.0
  • BuildCache: 0.24.0

TL;DR graph

LLVM build time speedup

No cache

Time Speed
19m50.9s 1.0x

Local cache, preprocessor mode, no compression

Time Speed
Cold cache 20m55.7s 0.95x
Warm cache 1m38.7s 12.1x

Cache size: 240.8 MiB

Local cache, direct mode, no compression

Time Speed
Cold cache 20m56.2s 0.95x
Warm cache 0m32.0s 37.2x

Cache size: 326.7 MiB

Local cache, direct mode, LZ4 compression

Time Speed
Cold cache 20m59.0s 0.95x
Warm cache 0m31.8s 37.4x

Cache size: 134.6 MiB

Local cache, direct mode, ZSTD compression

Time Speed
Cold cache 20m56.7s 0.95x
Warm cache 0m31.9s 37.3x

Cache size: 86.9 MiB

Ccache

For reference, the same compilation was performed with Ccache version 3.7.7 (using the default configuration) on the same system, with the following results:

Time Speed
Cold cache 20m55.7s 0.95x
Warm cache 0m36.0s 33.1x

Cache size: 354.8 MiB