Request for reviews (XL): 6829187: compiler optimizations required for JSR 292

[Christian Thalinger]

On Mon, 2009-11-30 at 17:16 -0800, Vladimir Kozlov wrote:
> src/cpu/x86/vm/frame_x86.inline.hpp
> Did you miss the next check or it was intentional?
>   && last_sp >= sp()

This is intentional.  The SP can be shrunk my some adapters so that the
assert does not hold anymore.

> src/cpu/x86/vm/x86_{32|64}.ad
> Instead of using "/MethodHandle" in format %{%}
> I would fix MachCallStaticJavaNode::dump_spec().

Hmm, I tried that but it is not printed with PrintOptoAssembly.  The
printed output seems to be generated by output_h.cpp.  Am I missing
something here?

> Why you can't add effect KILL ebp to CallStaticJavaHandle()?
> If you do it do you still need idealreg2mhdebugmask[] masks?

No.  Tom, John and I talked a lot about this one.  The problem is that
when EBP is killed the compiler uses it for deopt information and that
kills the saved value.

> src/share/vm/ci/ciCPCache.hpp.html
>    27 // The class represents and entry in the constant pool cache.
>                               ^ an ???
> Should the class be called ciCPCacheEntry?

No, the comment is misleading.  I change it.

Here is the updated webrev:

http://cr.openjdk.java.net/~twisti/6829187/webrev.02/

-- Christian