hardcoding -m32/-m64 is more harmful than using the toolchain defaults

[Martin Buchholz]

On Wed, Oct 28, 2015 at 2:32 AM, Andrew Haley <aph at redhat.com> wrote:

> On 27/10/15 20:36, Martin Buchholz wrote:
> > The world would be a better place if the current x86 32-bit ABI was
> > replaced by "x32"
> > https://en.wikipedia.org/wiki/X32_ABI
> > but it's looking unlikely that we will get there.  For starters, all
> > compilers must be rewritten to target x32, and that includes the jdk
> > jits.
>
> No they wouldn't.  The jits don't have to be changed at all because we
> already can handle compressed OOPs with no shift or base offset,
> i.e. native 32-bit addresses.  The Java calling convention would
> therefore stay as it is.  All that would have to be changed is the
> native <-> Java interfaces and accesses to native pointer-sized fields
> in C++ structures.  It's not trivial, but neither is it a rewrite.
>

I'll buy that - it's an "easy" job, but still a job for a compiler writer.

Perhaps x32 has less benefit for java programs because jits are already
free to generate code that uses 32-bit pointers but takes advantage of the
x86_64 instruction set?