[aarch64-port-dev ] Review, please

[Johnson, E Andrew]

I have discovered a number of places where both the aarch64-port/jdk7 and aarch64-port/jdk8 sources have conditionals which include "defined(AARCH64)" that are not in the jdk9/dev sources.  One of these, in hotspot/src/share/vm/runtime/advancedThresholdPolicy.cpp, causes a regression in the performance of the SPECjvm2008 benchmark scimark.monte_carlo, since a method which had been been expanded inline in both jdk7 and jdk8 is not inlined in jdk9.  Because of the number of calls made to this method, the calling overhead reduces the score to 1/6 of the score reported when the method is expanded inline.  I am sure that other benchmarks, as well as other applications, are impacted by this.


Several of the places where I discovered missing conditionals are in c1 sources.  While the default build is for the server compiler, I see JTREG results posted routinely on the Linaro OpenJDK site for both the client and server builds, so I believe that the client compiler should be kept up-to-date as much as possible until it is officially dropped from all 64-bit ports.


I will be submitting a bug report and a webrev which includes these patches.  However, since the principal developers of the aarch64-port are on this email list, I figured it would be worthwhile to get a "sanity check" on these changes.  If there are good reasons why any of these should not be propagated into jdk9/dev, I will be happy to remove them from the proposed patches.  If indeed, these are all "errors of omission", then I will submit the entire set of patches.


See the attached file for the proposed patches.


-AndyJ

-------------- next part --------------
A non-text attachment was scrubbed...
Name: aarch64.patch
Type: text/x-diff
Size: 5166 bytes
Desc: aarch64.patch
URL: <http://mail.openjdk.java.net/pipermail/aarch64-port-dev/attachments/20150413/41a121ea/aarch64.patch>