RFR: 8332265: RISC-V: Materialize pointers faster by using a temp register [v7]

[Robbin Ehn]

On Fri, 24 May 2024 11:27:07 GMT, Robbin Ehn <rehn at openjdk.org> wrote:

> Updated change looks good. It would be nice to see how much this will benefit performance.

Here are 'some' number, it still unclear if these actually are significant:

BASELINE                       | movptr2
fop                  3120 msec |  2811 msec =0.900962
h2                  19156 msec | 17600 msec =0.918772
jython              24060 msec | 23343 msec =0.9702
luindex              3222 msec |  3226 msec =1.00124
lusearch             4383 msec |  4380 msec =0.999316
lusearch-fix         4096 msec |  4359 msec =1.06421
pmd                  7417 msec |  7342 msec =0.989888
jython              24060 msec | 23343 msec =0.9702
fop(Xcomp)           3060 msec |  3058 msec =0.999346
h2(Xcomp)           38724 msec | 38717 msec =0.999819
jython(Xcomp)       29999 msec | 29694 msec =0.989833
luindex(Xcomp)       5259 msec |  5195 msec =0.98783
lusearch(Xcomp)      6364 msec |  6269 msec =0.985072
lusearch-fix(Xcomp)  6430 msec |  6534 msec =1.01617
pmd(Xcomp)           7360 msec |  6999 msec =0.950951
jython(Xcomp)       29999 msec | 29694 msec =0.989833
Avg:0.983353


Integrating later today!

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

PR Comment: https://git.openjdk.org/jdk/pull/19246#issuecomment-2134472452