pete@ohm.york.ac.uk (-Pete French.) (03/14/91)
I have just compiled c68 version II patchlevel 1 on one of our SUNs here in the department and was interested in comparing the speed of the code produced. To do this I grabbed a copy of the nearest benchmark program (dhrystone) and ran a few tests - using c68gas and then the gnu assembler/linker to produce a SUN executable. I then discovered an interesing fact... The code runs FASTER if it isnt optimised. Timings on this SUN are : c68 -opt -reg -short : 1562 ds/s c68 -noopt -short : 2173 ds/s c68 -opt -reg -noshort : 1315 ds/s c68 -noopt -noshort : 1851 ds/s The optimised code seems to run at about 71% of the speed of the unoptimised code ! This seems very perculiar to me - have I done something very stupid or is there something funny about the optimisation of the code - the optimised code does come out to be 80% of the size of the unoptimised so something is being done. any comments ? (I have yet to try it under MINIX) -- -Pete French. / "Two wrongs don't make a right, Adaptive Systems Engineering / - but three lefts do !"
HBO043%DJUKFA11.BITNET@cunyvm.cuny.edu (Christoph van Wuellen) (03/19/91)
I think it is just a typo in cmain.c, function options() replace .. opt_option = ! value by ... opt_option = value I never use -opt, -noopt, this was meant for compiler debugging. It is somehow striking that the non-optimized code is shorter since upon optimization, useless code is eliminated, arithmetic expressions are combined etc. C.v.W.