RFR: 8259925: [Vector API] Unreasonable IndexOutOfBoundsException message when length < vlen

Jie Fu jiefu at openjdk.java.net
Thu Jan 21 09:37:53 UTC 2021


On Wed, 20 Jan 2021 19:30:41 GMT, Paul Sandoz <psandoz at openjdk.org> wrote:

> Unfortunately it is still problematic because you have replaced the intrinsic check (that performed by `Object.checksIndex`, or more specifically [here](https://github.com/openjdk/jdk/blob/master/src/java.base/share/classes/jdk/internal/util/Preconditions.java#L261)).
> 
> Further, you have replaced the bounds check options, in place for experimentation. We are not ready yet to collapse our options, further analysis is required as bounds checks can be very sensitive in C2.
> 
> I would be open to you adding a third case, so that you can analyze the performance without perturbing the default behavior. I suspect the correct fix is to make intrinsic `Objects.checkFromIndexSize` in a similar manner to `Object.checksIndex`.

Hi @PaulSandoz ,

Thanks for your kind review and comments.

To be honest, I didn't get your first point.
As for the example above, the intrinsified Objects.checkIndex seems to generate more code than inlined if-statements.
So I'm confused why it's a problem to replace the intrinsic check.
Did you mean an intrinsic is always (or for most cases) better than non-intrinc or inlined if-statements?
And why?

Could you please make it more clearer to us?
Thanks.

-------------

PR: https://git.openjdk.java.net/jdk/pull/2127


More information about the core-libs-dev mailing list