forthvsc (and Pascal)

While Forth advocates routinely describe Forth in the most beautiful colors, I could not find any actual benchmarking comparing Forth to C. There's a reference implementation of RC4 on Wikipedia which I took to perform benchmarking with. I also took the RC4 description on Wikipedia and implemented it naively in C with no optimizations whatsoever, just the algorithm as it was described in pseudocode there.

Also there's at least one charming person claiming that Pascal has a performance penalty of roughly 5% and who dared me to test it while themselves being too lazy, scared or incompetent to try it themselves. So I did.

Test Conditions

Measured on an Intel Core i7-5930K CPU @ 3.50GHz using gcc 8.3.0 and gforth 0.7.3 on Linux 5.0.0. Since the speed difference is so vast, the number of generated schedules/keystream varies from C to Forth, consult the code for details. Pascal code was compiled with Free Pascal Compiler v3.0.4+dfsg-22 [2019/01/24] for x86_64.

Forth: Key Schedules

$ time gforth key_schedules.for

real	0m21,616s
user	0m21,616s
sys	0m0,000s

$ time gforth key_schedules.for

real	0m21,573s
user	0m21,568s
sys	0m0,004s

$ time gforth key_schedules.for
real	0m21,682s
user	0m21,682s
sys	0m0,000s

Best of three: 21.573 seconds.

Forth: RC4 Data Stream

$ time gforth key_stream.for

real	0m19,560s
user	0m19,559s
sys	0m0,001s

$ time gforth key_stream.for
real	0m19,595s
user	0m19,595s
sys	0m0,001s

$ time gforth key_stream.for
real	0m19,601s
user	0m19,592s
sys	0m0,009s

Best of three: 19.560 seconds.

C: Key Schedules

$ gcc -Wall -O3 -o rc4_c_impl rc4_c_impl.c
$ time ./rc4_c_impl schedule
real	0m6,822s
user	0m6,822s
sys	0m0,000s

$ time ./rc4_c_impl schedule
real	0m6,843s
user	0m6,843s
sys	0m0,000s

$ time ./rc4_c_impl schedule
real	0m6,946s
user	0m6,946s
sys	0m0,000s

Best of three: 6.822 seconds.

C: RC4 Data Stream

$ gcc -Wall -O3 -o rc4_c_impl rc4_c_impl.c
$ time ./rc4_c_impl stream
real	0m5,990s
user	0m5,986s
sys	0m0,004s

$ time ./rc4_c_impl stream
real	0m6,001s
user	0m6,001s
sys	0m0,000s

$ time ./rc4_c_impl stream
real	0m5,976s
user	0m5,975s
sys	0m0,001s

Best of three: 5.976 seconds.

Pascal: Key Schedules

$ fpc -S2 -O3 rc4test.pas
$ time ./rc4test
real	0m23,604s
user	0m23,603s
sys	0m0,000s

$ time ./rc4test
real	0m23,601s
user	0m23,600s
sys	0m0,000s


$ time ./rc4test
real	0m23,651s
user	0m23,648s
sys	0m0,000s

Best of three: 23.601 seconds.

Pascal: RC4 Data Stream

$ fpc -S2 -O3 rc4test.pas
$ time ./rc4test
real	0m8,913s
user	0m8,913s
sys	0m0,000s

$ time ./rc4test
real	0m8,844s
user	0m8,843s
sys	0m0,001s

$ time ./rc4test
real	0m8,847s
user	0m8,847s
sys	0m0,001s

Best of three: 8.844 seconds.

Comparison

Test	Language	Iterations	Time/sec	Iterations/Time	Factor	Factor
Key Schedules	Forth	100000	21.573	4635	1	1 / 91
	C	10000000	6.822	1465846	316	3.46
	Pascal	10000000	23.601	423711	91	1
Key Stream	Forth	10000000	19.560	511247	1	1 / 442
	C	2000000000	5.976	334672021	655	1.48
	Pascal	2000000000	8.844	226142017	442	1

I.e., in Forth on this particular test you get about 4.64 kSchedules/sec while in C you get 1.47 MSchedules/sec. In Forth, you get a data stream with about 511 kB/s while in C you get 335 MB/s. For Pascal, it's about 424 kSchedules/sec and a data stream of 226 MB/s.

License

All my code (C, Pascal) is CC-0. Forth code was taken from Wikipedia and is under the respective license.

elrzn/forthvsc