Request for review: 8004076: Move _max_locals and _size_of_parameters to ConstMethod for better sharing

[Jiangli Zhou]

Hi John,

Thanks a lot for the review and suggestion. I'll add Address constants 
where it's feasible.

Jiangli

On 12/07/2012 02:40 PM, John Rose wrote:
> On Dec 7, 2012, at 1:45 PM, Jiangli Zhou wrote:
>
>> Could I have another reviewer for the change? Either official or 
>> non-official reviewer is okay.
>
> I have a comment on this one.  (And you can use me as a reviewer.)
>
> In the sparc code, I would prefer to see the Address constants kept 
> around, because they make the code easier to read.
>
> -  const Address size_of_parameters(G5_method, 
> Method::size_of_parameters_offset());
> -  const Address size_of_locals    (G5_method, 
> Method::size_of_locals_offset());
> +  const Register G3_constMethod = Glocals_size;
> +  const Address size_of_parameters(G3_ConstMethod, 
> ConstMethod::size_of_parameters_offset());
> +  const Address size_of_locals   
>  (G3_ConstMethod, ConstMethod::size_of_locals_offset());
>    const Address constMethod       (G5_method, Method::const_offset());
>    int rounded_vm_local_words = round_to( 
> frame::interpreter_frame_vm_local_words, WordsPerLong );
> @@ -511,7 +509,8 @@
>    // Lscratch can't be used as a temporary because the call_stub uses
>    // it to assert that the stack frame was setup correctly.
> +  __ ld_ptr( constMethod, G3_constMethod );
> -  __ lduh( size_of_parameters, Glocals_size);
>
> Maybe the details make it difficult, but I think it's possible in most 
> places.
>
> As a side effect, using a fixed register (like G3) for the ConstMethod 
> will make debugging easier.
>
> Same comment for x86/64.
>
> This may be more difficult to do on x86/32, since there are fewer temp 
> registers.  But that's an oddball case.
>
> Thanks,
> — John

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/hotspot-runtime-dev/attachments/20121207/e631479e/attachment.html