About JEP 338: Vector API

[forax at univ-mlv.fr]

----- Mail original -----
> De: "Paul Sandoz" <paul.sandoz at oracle.com>
> À: "Remi Forax" <forax at univ-mlv.fr>
> Cc: "b lacombe" <b.lacombe at lug.com>, "panama-dev at openjdk.java.net'" <panama-dev at openjdk.java.net>
> Envoyé: Jeudi 11 Février 2021 17:36:28
> Objet: Re: About JEP 338: Vector API

> Nicely done!
> 
> Challenge: support control flow, translating to the use of masks.
> 
> Not bad, far from good :-) Not a criticism of what you did, but a criticism of
> how you had to do it. The Java platform can do a much better job making easier
> for you to do this kind of thing.

I agree if it is not at the cost of making the Java AST part of the API.
Using the Java bytecode as an AST is better because the Java bytecode don't change frequently, sometime, some fanatics add a new instruction but it's one instruction every ten years.

> 
> Paul.

Rémi

> 
>> On Feb 10, 2021, at 2:22 PM, forax at univ-mlv.fr wrote:
>> 
>> 
>> De: "Paul Sandoz" <paul.sandoz at oracle.com>
>> À: "Remi Forax" <forax at univ-mlv.fr>
>> Cc: "b lacombe" <b.lacombe at lug.com>, "panama-dev at openjdk.java.net'"
>> <panama-dev at openjdk.java.net>
>> Envoyé: Mercredi 10 Février 2021 19:11:47
>> Objet: Re: About JEP 338: Vector API
>> Hi Remi,
>> 
>> Thanks for pointing out it can be done using serialization, and people do that,
>> since that is what's currently possible. I think it's problematic and we can do
>> much better.
>> 
>> Paul.
>> 
>> It's not that awful :)
>> Enjoy https://github.com/forax/vector-handle
>> 
>> It should perform well but i've not done any perf tests.
>> And obviously, i've not implemented all the operations, only +, -, * and /, feel
>> free to do some pull requests.
>> 
> > Rémi