RFR: 8271820: Implementation of JEP 416: Reimplement Core Reflection with Method Handle [v13]

[Peter Levart]

On Thu, 14 Oct 2021 16:05:19 GMT, Mandy Chung <mchung at openjdk.org> wrote:

>> src/java.base/share/classes/jdk/internal/reflect/DirectMethodHandleAccessor.java line 86:
>> 
>>> 84:     }
>>> 85: 
>>> 86:     private static final int PARAM_COUNT_MASK = 0x00FF;
>> 
>> Is this packing logic required? I get it that it saves footprint - but then you have to always unmask bits to get the argument count (see invoke and other places). If you keep this, it might be worth documenting what the bit layout is supposed to be?
>
> It's not driven by performance data.   It's part of Peter's contribution.   I also prefer it without the packing.   I propose to keep the parameter count as a separate field and examine it when there is footprint issue.

The reason for this packing is (was) ORing the value with a non-zero value so that field never held zero value. When for example an individual value (say paramCount) is used in a separate @Stable field, its value set may include the default value (zero) which is then not optimized by JIT as a constant. This is from @Stable docs:

 * By extension, any variable (either array or field) which has annotated
 * as stable is called a stable variable, and its non-null or non-zero
 * value is called a stable value.

...and:

 * The HotSpot VM relies on this annotation to promote a non-null (resp.,
 * non-zero) component value to a constant, thereby enabling superior
 * optimizations of code depending on such a value (such as constant folding).
 * More specifically, the HotSpot VM will process non-null stable fields (final
 * or otherwise) in a similar manner to static final fields with respect to
 * promoting the field's value to a constant.  Thus, placing aside the
 * differences for null/non-null values and arrays, a final stable field is
 * treated as if it is really final from both the Java language and the HotSpot

So now that some of these fields are final and not annotated with @Stable any more but are treated as "trusted final" fields, the question is whether JIT is treating the default (zero, null) values differently or not. If they are treated as static final fields where default value makes no difference, then it's ok to split this multi-valued field into individual fields.

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

PR: https://git.openjdk.java.net/jdk/pull/5027