RFR: JDK-8201536 configure fails compiler check due to bad -m32 flag

[Magnus Ihse Bursie]

This is intended to be a proper fix for the -m32/-m64 compiler flag on 
unsupported platforms regression due to my previous flag refactoring.

Here's my analysis of the old code, from the bug report:

The old code set the -m<bits> flag in these cases:
if OS is solaris or aix,
or if OS_TYPE is unix and build_type is reduced

While this captures incorrect aspects of the situation, it has worked 
without complaints. I think what it *really* tries to do is:
* We can only run reduced builds on platforms where we can set the 
-m<bits> flag.
* xlc and sunstudio compilers needs to have -m<bits> (or -q<bits> in the 
case of xlc) to function properly. I assume this also means it's always 
supported for those compilers on the platforms we support there.

For clang/gcc, the flag applies to x86, sparc and ppc. Maybe we should 
just set it on x86, but then again, it won't hurt to set it on all these 
three architectures.

In theory, this means that reduced builds should be able to be supported 
on all platforms that support sunstudio and xlc. However, these are, 
afaik: solaris-sparcv9, solaris-x86_64, aix-ppc64, aix-s390x, where we 
do not support a 32-bit variant anyway. I do not know if there is an 
aix-x86 and aix-x86_64? If so, we should be able to do a reduced build 
for aix-x86_64.

So in practice, reduced builds are supported only for clang/gcc on 
x86_64. (And Windows, which uses a different mechanism.)


Bug: https://bugs.openjdk.java.net/browse/JDK-8201536
WebRev: http://cr.openjdk.java.net/~ihse/JDK-8201536-fix-m64-cflag/webrev.01

/Magnus