Skip to content

Instantly share code, notes, and snippets.

@michaeljclark
Last active May 11, 2017 09:11
Show Gist options
  • Save michaeljclark/c4fded96d8c341317f0fc67e700143f3 to your computer and use it in GitHub Desktop.
Save michaeljclark/c4fded96d8c341317f0fc67e700143f3 to your computer and use it in GitHub Desktop.
RISC-V to x86 binary translation benchmarks
Runtime (seconds)
RV32 x86-32 RV64 x86-64
meta meta qemu meta meta qemu
interp JIT user native interp JIT user native
primes 11.57 0.59 0.52 0.34 5.07 0.17 0.27 0.11
miniz 32.55 1.58 2.17 0.81 41.52 2.09 2.20 0.76
SHA-512 55.27 2.47 2.74 0.73 23.69 0.70 1.12 0.24
AES 32.75 0.96 1.52 0.37 38.39 1.00 1.61 0.27
qsort 3.86 0.25 1.01 0.13 3.96 0.28 0.94 0.13
dhrystone 23.34 0.49 2.18 0.18 22.30 0.49 2.21 0.10
DMIPS (v1.1) 243 11579 2610 31386 254 11547 2576 55132
Performance Ratio (smaller is better)
RV32 x86-32 RV64 x86-64
meta meta qemu meta meta qemu
interp JIT user native interp JIT user native
primes 34.33 1.75 1.55 1.00 44.43 1.52 2.36 1.00
miniz 40.19 1.95 2.68 1.00 55.00 2.77 2.92 1.00
SHA-512 76.23 3.40 3.78 1.00 100.39 2.97 4.76 1.00
AES 88.02 2.57 4.08 1.00 140.12 3.66 5.88 1.00
qsort 28.82 1.86 7.50 1.00 29.52 2.05 7.04 1.00
dhrystone 128.97 2.72 12.04 1.00 214.45 4.75 21.23 1.00
Ratio (average) 66:1 2.4:1 5.3:1 1:1 97:1 3:1 7.4:1 1:1
Notes:
- riscv32/riscv64 (gcc -Os) 7.0.1 20170321 (experimental)
- x86-32/x86-64 (gcc -O3) (Debian 6.3.0-6) 6.3.0 20170205
- native Intel(R) Core(TM) i7-5557U CPU @ 3.10GHz
- all RISC-V code is compiled with RVC compression enabled
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment