RFR: Fix Shenandoah build on s390

[Aleksey Shipilev]

On 08/23/2018 05:37 AM, Andrew Hughes wrote:
> Bug: https://icedtea.classpath.org/bugzilla/show_bug.cgi?id=3619
> Webrev: http://cr.openjdk.java.net/~andrew/shenandoah-8/pr3619/

Thanks!

> A couple of build issues were found when trying to build the
> Shenandoah source code base on s390. As the build failures in the
> linked bug show, issues arise on s390 that are not visible on other
> platforms, because size_t is not equal to uintx on that platform. On
> s390, uintx is a unsigned int, while size_t is a long unsigned int.
> 
> The attached patch fixes the build issues by switching two format
> specifiers in the heuristics code from UINTX_FORMAT to SIZE_FORMAT,

The block in shenandoahAdaptiveHeuristics is gone in our current development builds, and it will
sift down to 8u as we next round of backporting next week. The block in shenandoahStaticHeuristics
is not needed and can be removed:
  http://hg.openjdk.java.net/shenandoah/jdk/rev/8cb652cfde2c

> and casting the region size from uintx to size_t in
> shenandoahHeapRegion.cpp. The latter is necessary, as otherwise the
> calls to MAX2 have mismatched types (uintx, size_t). A possible
> alternative may be to use a size_t for the region size instead, but I
> went with the solution that seemed the lowest risk on all platforms.

Yes, that is a better option:
  http://hg.openjdk.java.net/shenandoah/jdk/rev/1af5bd0d84eb

> The changes are applied against aarch64/shenandoah-jdk8u, as that's
> where s390 builds were tested,

So it would be easier for us to propagate the changes to shenandoah/jdk8, and then make a bulk
integration with all these bugfixes into aarch64-port/shenandoah-jdk8u. We can put your change in
aarch64-port/shenandoah-jdk8u as is, but that would show up as merge conflict later. So if there is
no time pressure, I'd prefer to wait for bulk integration.

-Aleksey