I use the micro second function now. I test all file in a romdisk to prevent CD access to affect timing measures.
First i design a 100 time slice array space to make 100 timing measures.
Then i launch the main routine without calling the decode library i obtain 100 values in micro seconds :
73 values at 5 + 20 values at 6 + 6 values at 7 + one huge value i suspect to be KOS INIT routine that i dropped from my test.
So for an average time slice without decoding, i obtain (73*5 + 20*6 +6*7) / 99 = 5.32 micro seconds !!
Then i launch the main and call the decoder routs with two different sound all stereo and 44100 Hz:
Sound A (i do not know where it is from)
I obtain (50*6 + 44*7 +5*8) / 99 = 6.54 micro seconds. So to have only the average decode time slice : 6.54 - 5.32 = 1.22 micro seconds.
Sound B (part of space adventure cobra Sound track)
I obtain (43*7 + 46*6 + 10*8) / 99 = 6.63 micro seconds. So to have only the average decode time slice : 6.63-5.32 = 1.31 micro seconds.
I use my routs on real DC in 60 Hz mode : 60 Hz means 1/60 fps time slice so 1/60 = 0.016 = 1.6 * 10 exp -2.
An average time slice decode take a % of ( (1.31*10exp-6) / (1.6 * 10 exp -2) ) *100 = 0.81*10exp -4 = 0.0081 % of a frame ...
If i made no mistake i think it's fair
I use your first version but with modifications to suit to my needs ...