Review request: 8026391: The Metachunk header wastes memory - 8026392: Metachunks and Metablocks are using a too large alignment

[Coleen Phillimore]

This is good, so we align metachunks to 8 byte alignment rather than 16 
in 64 bit?
#define ARENA_AMALLOC_ALIGNMENT (2*BytesPerWord)

Coleen

On 10/15/2013 02:11 PM, Stefan Karlsson wrote:
> From an offline discussion with Jon, I decided to change the following:
>
> $ hg diff
> diff -r 8e2b57a0132c src/share/vm/memory/metachunk.cpp
> --- a/src/share/vm/memory/metachunk.cpp    Tue Oct 15 14:32:20 2013 +0200
> +++ b/src/share/vm/memory/metachunk.cpp    Tue Oct 15 20:07:51 2013 +0200
> @@ -37,8 +37,8 @@
>    // so that 64 bit types get correctly aligned.
>    const size_t alignment = 8;
>
> +  // Make sure that the Klass alignment also agree.
>    STATIC_ASSERT(alignment == (size_t)KlassAlignmentInBytes);
> -  STATIC_ASSERT(alignment == (size_t)BytesPerLong);
>
>    return alignment;
>  }
>
> StefanK
>
> On 10/14/13 1:58 PM, Stefan Karlsson wrote:
>> Please, review these two patches to remove some of the wastages 
>> introduced in the metaspace.
>>
>>
>> 8026391: The Metachunk header wastes memory
>>  http://cr.openjdk.java.net/~stefank/8026391/webrev.00/
>>  https://bugs.openjdk.java.net/browse/JDK-8026391
>>
>> There are a couple of fields in Metachunk that store already known 
>> information, like the beginning and the end of the Metachunk. I 
>> replaced those fields with code instead.
>>
>> To make this easier, I extracted some duplicated code in Metablock 
>> and Metachunk into a super class named Metamem. The extracted code is 
>> purely needed to allow these objects to be placed on the FreeList and 
>> in the BinaryTreeDictionary.
>>
>>
>> 8026392: Metachunks and Metablocks are using a too large alignment
>>  http://cr.openjdk.java.net/~stefank/8026392/webrev.00/
>>  https://bugs.openjdk.java.net/browse/JDK-8026392
>>
>> The metaspace code was using ARENA_ALIGN to align the Metablocks and 
>> Metachunk header. This forced all allocations of metadata to be 16 
>> bytes aligned. This patch patch changes the restriction to be 8 bytes 
>> instead, since that's the restriction needed for 64 bit types like 
>> long, double, pointers and klass pointers,
>>
>>
>> Testing: Added unit test for Metachunk. JPRT (before splitting of the 
>> patches)
>>
>> thanks,
>> StefanK
>