| From: | Oliver Roberts |
| Date: | 9 Aug 2000 at 19:05:31 |
| Subject: | Re: OGR/RC5 and MorphOS (A bit long) |
On 09-Aug-00 16:08:49 BST, Michael.J.Every wrote:
> Appologies to anyone who does not want to see this...
> I think this is of interest to most.
> If I am wrong (get flames) then I'll move over to CU-List so you dont get
> anymore of this crap!
Better still, move it to the RC5 list :)
>> Core #0 or #1 may be faster under 68k emulation.
> Here we have the results of the dnetc v2.8010-463-CPR-00071321 RC5 client
> for 68k > running under M0S68k emulation, using the different RC5 cores
> available.
Well, that blows my theory - I guess the emulation is more like the 060
than I thought.
> Suprisingly poor performance from the 020/030 core.
Not really - this core is much slower on a real 060, than the 060 core.
The 68000 core is in fact faster on an 060 than the 020/030 core. The
relative differences in your benchmarks tend to match that of real 060
benchmarks.
> If the 68k Emulation is anything like Oxypatcher in so far as it
> optimises instructions as it needs them then this could explain the poor
> performance when compared to the 000/010(I hope that makes sense, if not
Nope. The fact is that the 020/030 core is simply just slower than the
other 2 cores when run on an 060. As I said, the cores are highly optimized
for the intended cpu, making optimum use of caches and such like. For
example, the 020/030 core is based on code that uses lots of loops, which
produces more compact and faster code on an 020. In the 040/060 core, the
loops have been unrolled, which produces larger and more repetitive code,
but that suits the 040 and 060 much better than the loopy core code. Other
than that, I don't know all the techie details, as optimizing assembler
code for optimum speed is not a speciality of mine :)
> Note inconsistancy with time & keys/sec the 040/25 took (16.87 &
> 21,851.12 keys/sec)
> & time the 68060 emulation (040/060 core) on M0S took (16.21 & 16,170.46
> keys/sec)
Don't pay much attention to that - the 16.xx sec values just say for how
long the benchmark ran for. The amount of keys cracked (which will usually
be slightly different each time you -benchmark) is divided by this value
to produce the benchmark speed in keys/sec.
Quote carefully and read all ADMIN:README mails