(Useless benchmark) Dhrystone on Sas C/Gcc/Vbcc

[alkis]

Standard benchmark disclaimers. You shouldn't draw any conclusions based on this. Benchmarks are evil. Etc. :-)

Anyways, I came across some really old source on the net of the dhrystone benchmark. I cross-compiled it with three compilers and run it on fs-uae (emulated A1200 +8mb fast)

So, tada, here it is:

 *  LOOPS 500000
 *  A1200 (+8mb Fast Ram) emulated on linux (kubuntu 64bit) with fs-uae 2.2.3
 *	Comp	Version	DHRY/SEC	TIME
 *  gcc		3.4.0	14285   	(35 secs)
 *  sc 		6.58	13513   	(37 secs)
 *  vbcc 	V0.9b	13157		(38 secs)
 *
 *  sc link dhry.c nostkchk opt optcomp=100 optdep=100 optinline optloop
 *  	optrdep=100 optschd cpu=68020
 *
 *  gcc -O3 -Wall -c -fmessage-length=0 -fomit-frame-pointer -m68020
 *  	-funroll-loops -finline-functions
 *
 *  vc +aos68k -O4 -cpu=68020

Source code attached.

[Jope]

If you want to run it under emulation, you should use a500 emulation + everything cycle exact turned on. Even then it's a bit pointless. :-)

[alkis]

It's pointless as you say.
Well, as absolute values I am pretty sure these are useless.
As relative values though, they *might* not be so useless. I suppose if you keep everything else as a constant and the only thing you change is the running code, it could relatively show the most efficient code? (I can see it going the other way....i.e. measuring some emulator sweetspot that a certain code managed by accident to trigger)

So, yeah, pretty much useless :-)

[phx]

It's interesting to see that newer gcc versions have become much better. Many years ago I made a benchmark which showed gcc V2.95 far behind vbcc:
http://sun.hasenbraten.de/vbcc/index...iew=benchmarks