Re: [Jython-dev] A few micro-benchmarks for Jython 2 and two other ideas

[Jeff A. <ja...@fa...>]

On 07/01/2021 11:50, Andrew Haley wrote:
> On 1/6/21 4:28 PM, Jeff Allen wrote:
>>> And to see if intrinsic methods are being utilized or not and where in
>>> compiled code, you can add:
>>>      -XX:+PrintCompilation -XX:+UnlockDiagnosticVMOptions
>>> -XX:+PrintInlining
>>>
>> Unlock has to come first, it seems. I've experimented with those options
>> and found what they produced was pretty incomprehensible. I never made
>> the disassembly option work.
> If you're interested in persisting a bit more, I can help with this.

I did persist thanks, and found a pre-built hsdis which produced the 
disassembly from which I got the few lines in the e-mail. (Also, I  
confess I am working on Windows.) So now I just have the comprehension 
problem.

I will return to this, but for now it is enough to have a sense where it 
falls short. Some of the things I hoped for happen, some don't, and I 
know roughly what I will try.

>> Going back and trying a little harder, thanks to your suggestion, I got
>> further this morning. The output remains too complex for me to follow
>> (so many jumps!), but a superficial inspection supports the conjectures
>> I made based only on timing.
> WRT intrinsics: if there's something important, we can do it.

Thanks! I don't think I know yet what would be critical, but will bank 
the offer.

Thad is clearly knowledgeable in this area, but I didn't see the mixed 
results from my first use of invokedynamic as a failure of the JVM to 
use intrinsics. I've probably misunderstood. The method handle adaptors 
mostly represent stack manipulations (or simply register selections) 
around the ultimate arithmetic operations, and I think the JVM will get 
the idea if I can decrease the nesting to the point where inlining 
completes. However, if it doesn't do so in critical cases, it's good to 
have somewhere to go.

Jeff Allen