Re: [Help-smalltalk] Bench.st .. surprising results

[Paolo Bonzini]

> I just ran Bench.st for run, and got the following figures, on my
> Pentium IV 2.8Ghz 512Kb cache machine:
>
>      72,644,721 bytecodes/sec; 2,460,329 sends/sec
>      66,832,006 bytecodes/sec; 2,447,278 sends/sec
>      71,438,537 bytecodes/sec; 2,187,121 sends/sec
>
> These are 3 times the reported values for a Pentium II 300Mhz machine.
> It makes me wonder if there is some machine tick (which is constant)
> related performance constraint in the vm.

The Pentium 4 is absolutely not suited to bytecode interpretation. 
Indeed the faster VM is almost 50% faster on the Pentium 4 because it 
executes fewer bytecodes (longer traces in the P4's trace cache and 
fewer stalls due to indirect jumps).

Most of the mprotect's are outside the inner loop (GC's happen outside it).

> Am I wrong? should I not expect a more significant improvement from a
> machine an order of magnitude faster mhz + bus etc, + paolo's 20% faster
> VM ?
>
> I would have tried with --enable-jit, but:

Thanks, I'll look into this.

Paolo