I'm slightly annoyed to have Álmos (seemingly) mis-quoted me on his blog (and out of context to boot), without even having replied or messaged me once first:
Quote:
Many of the users have irrational expectations for how much the JIT compiling will speed up the emulator. (According to somebody: it supposed to be "ten times faster than the interpretive"... Err... Not likely. How did they come up with any number anyway?)
That doesn't appear to be an actual quote (despite the use of quotes), since Google fails to find an exact match, but I guess he was paraphrasing what I wrote ages ago in this thread:
Quote:
2. Does he plan any further optimisations, or changes which might give big speed-ups? I was a bit disappointed that the last speed test he made (in May!) only showed a 4x speed-up (which is significant, but not nearly enough for many of the reasons people want JIT in the first place). I would have hoped for perhaps a 10x speed-up...
I'm not sure how he translated "a bit disappointed", "would have hoped" & "perhaps 10x" into "irrational expectations"... But if he *had* chosen to ask me where I got that "10x" figure from, I would have answered something like this:
Interpreted programming languages are (very roughly speaking) 10x slower than natively compiled languages (could be 5x or 20x slower, or worse if badly done). In an ideal case JIT could approach natively compiled speeds (if super optimised), hence a 10x speed-up might be possible.
To prove I'm not entirely talking out my arse, a quick search
found this:
Quote:
Comparing interpreted code with JIT, a JIT is usually about ten times as fast. Certainly in the Kaffe Java virtual machine, the JIT is ten times as fast as the interpreter.
But maybe 68k emulation is especially difficult to JIT, so you can't speed it up as much? I
found this benchmark of a 68k JIT:
Quote:
Manipulating a large .JPG graphic
...
Basilisk II JIT R5: 3.1 seconds
...
Basilisk II .9: 18.8 seconds
(i.e. a speed-up of 6.0 times)
Quote:
Stuffit Test Two
...
Basilisk II JIT R3: 2.3 seconds
Basilisk II .9: 12.6 seconds
(i.e. a speed-up of 5.5 times)
A couple of other tests showed speed-ups of 3.2 & 3.5 times, but I am on-purposely trying to find the best-possible JIT vs Interpretive performance (since that is what is being disputed).
Finally, I quickly ran my own tests using
BogoMIPS:
AmigaOS4.1 interpreted emulation gave 36.5 BogoMIPS
AmigaOS4.1 Petunia (JIT) gave 189.0 BogoMIPS
(And just for fun, AmigaOS4.1 natively compiled gave 1138.0 BogoMIPS)
That makes Petunia JIT 5.2 times faster than interpreted emulation. Álmos should be slightly familiar with Petunia
So I think given all the above, it's not unreasonable that a CPU-bound test of E-UAE's new JIT could realistically aim for 5-6 times faster. And my pick-a-figure-our-of-the-air guess of 10 times faster was certainly not an "irrationally expectation", even if it was slightly higher than appears to be possible.