RFR: 8347009: Speed ​​up parseInt and parseLong [v15]

[Shaojin Wen]

On Wed, 5 Feb 2025 16:48:06 GMT, Raffaello Giulietti <rgiulietti at openjdk.org> wrote:

>> src/java.base/share/classes/jdk/internal/util/DecimalDigits.java line 143:
>> 
>>> 141:      * @return If both characters are numbers, return d0 * 10 + d1, otherwise return -1
>>> 142:      */
>>> 143:     @ForceInline
>> 
>> Can we leave this out? From the doc: "This annotation must be used sparingly."
>> I tried without and on M1 performance does not seem to be affected.
>
> Again, can we avoid `@ForceInline`?

The codeSize of the digit2 method is 56, greater than C1 MaxInline 35. I was worried about the performance in the early runtime, so I added @ForceInline

>> src/java.base/share/classes/jdk/internal/util/DecimalDigits.java line 175:
>> 
>>> 173:          */
>>> 174:         int d;
>>> 175:         short x = UNSAFE.getShortUnaligned(str, Unsafe.ARRAY_BYTE_BASE_OFFSET + offset, false);
>> 
>> @wenshao I'm a bit worried about the use of Unsafe here.
>> 
>> This method is `public` (although in an internal package), and while it is used correctly in this PR, there's no warning in the doc that `str` and `offset` must come from a trusted caller that ensures that they are safe to use with, well..., Unsafe.
>> 
>> Did you consider safer alternatives, like usage of `VarHandle`, even if that might mean a slight performance degradation?
>
> Never mind, `VarHandle` cannot be used in this case.

Using VarHandle in this scenario may affect JVM startup performance

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

PR Review Comment: https://git.openjdk.org/jdk/pull/22919#discussion_r1943325107
PR Review Comment: https://git.openjdk.org/jdk/pull/22919#discussion_r1943328441