RFR: 8341566: Add Reader.of(CharSequence) [v10]
Markus KARG
duke at openjdk.org
Fri Oct 11 06:18:15 UTC 2024
On Thu, 10 Oct 2024 22:42:19 GMT, Bernd <duke at openjdk.org> wrote:
> > I have slightly adapted the copy loop to allow CPUs to use solely single-operand INC within the loop instead of dual-operand ADD.
>
> If that makes a difference we might need to consider fixing the Compiler :)
Hm... really? In fact I am not sure if the compiler is actually able to optimize replacement from ADD to INC *in this particular* scenario: The operand of the ADD is *neither* a constant and in particular *not* `1`, the upper limit of the loop is *not* a constant, and is typically far beyond any reasonable unrolling limit (readers rather often are asked for multiple KB here to optimize I/O transmissions). This is not just `for (int i = 0; i < 5; i++;)` (always same limit, reasonably low limit, using INC only).
Anyways, if you can proof me wrong I am willing to learn (and to undo this commit)! :-)
-------------
PR Comment: https://git.openjdk.org/jdk/pull/21371#issuecomment-2406625958
More information about the core-libs-dev
mailing list