Hi Erik,

We are in the process of completing the legal work. Hopefully we can
settle that soon…

As for the code, I will have to inform it is not a small patch. Our
current works include RA and Binary Operation Optimization.
Fortunately, we can report that it already passes the testsuite right
now.

We have a few numbers comparing the two RA implementations, on a
Marvell ARMv7 platform:

SVN revision 4380
Richards:          163  171  169
DeltaBlue:         143  143  146
Crypto:            162  162  161
RayTrace:          171  171  169
EarleyBoyer:       559  578  578
RegExp:            142  139  136
----
Score (version 5): 193  195  194

Our work base on SVN revision 4367
Richards:          177  177  174
DeltaBlue:         161  162  161
Crypto:            227  227  228
RayTrace:          179  172  176
EarleyBoyer:       617  614  612
RegExp:            126  126  124
----
Score (version 5): 212  210  210

As we can see, the “heavier-weight” RA route seems to be profitable
currently.
Of course, the current bleeding_edge RA has just been committed, so
more tuning may be possible there.
Also, raw performance of different ARM processors may also tip the
balance, so more variations might be seen on other ARM CPUs.

--
Neil Hsueh
Marvell Taiwan

-- 
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev

To unsubscribe, reply using "remove me" as the subject.

Reply via email to