Benchmarks

From RevaWiki

Jump to: navigation, search

These benchmarks are against the "bench.f" program in the "bench" subdirectory of Reva. The gcc versions are agains the C version of the benchmarks, also there.

The machines:

  • A (Athlon XP 2600+ 2GHz [6.10.0]) Windows XPsp2, 1G RAM
  • B (Athlon XP 2400+ 2GHz [6.8.1]) Linux 2.6.11 1G RAM, 3948 Bmips
  • C (Celeron M 1.4Ghz [6.13.6]) Linux 2.6.13 512M Ram, 2802 Bmips
  • D (Xeon 3.06Ghz [15.2.0]) Windows XPsp2, 1G RAM

The timings:

version A B C D
reva-6.0.3 -- 5.999 6.891
reva-6.0 6.406 6.072 9.794 6.812
reva-5.1 9.750 8.944 13.523 9.094
gforth 0.62 18.478 16.874 31.757
gforth-fast 13.707 15.486 23.881
Win32Forth 30.609
VFX/Win32 8.766
SwiftForth 6.735
IForth 2.1 7.230 7.634
gcc 5.828 6.130 10.173
gcc -O24.249 4.876 5.259

NOTE: Several of the other forths do agressive optimization, so the 'noop' gets removed. In order to restore parity I modified the bench so that each compiler does the same 'call noop...' (or equivalent). VFX actually optimizes away a call to the empty word but for some reason doesn't optimize away the loop.

Retrieved from "http://ronware.org/reva/wiki/Benchmarks"
Personal tools