RFR (S) JDK-8014912 Restore PrintSharedSpaces after NPG

[Coleen Phillimore]

I think you can inherit from HistogramElement and add fields but I'm not 
going to make further suggestions at this time, I just wanted you to 
consider reusing this code.   Your code is good.

Coleen


On 5/23/2013 12:51 PM, Ioi Lam wrote:
> Coleen,
>
> I looked at histogram.hpp, but it has only one _count field:
>
>    class HistogramElement : public CHeapObj<mtInternal> {
>      protected:
>       jint _count;
>       const char* _name;
>
>      public:
>       HistogramElement();
>       virtual int count();
>       virtual const char* name();
>       virtual void increment_count();
>       void print_on(outputStream* st) const;
>       virtual int compare(HistogramElement* e1,HistogramElement* e2);
>    };
>
> For my purpose, I would need a count and size for each subtype of 
> metadata, so it doesn't look like this class is immediately usable for 
> me.
>
> I am actually planning (with a different bug ID) to dump a detailed 
> "map file" of every object in CDS. Now that I record all the 
> allocations under Metaspace::_alloc_record_head, I can iterate over 
> them and call, for example, Method::print_codes_on() on all the 
> methods. This will allow you to do all kinds of analysis, like
>
>     How many unquickened getfield bytecodes are inside CDS
>     How many methods contain unquickened bytecodes
>
> ... etc. When we were doing Monty's romizer, a mapfile like this was 
> very helpful in identifying possible areas for improvements.
>
> While I do that, I can also dump a table of the size distribution of 
> each type objects in a histogram format, similar to 
> PrintSymbolTableSizeHistogram.
>
> Thanks
> - Ioi
>
>
> On 05/22/2013 06:47 PM, Coleen Phillimore wrote:
>>
>> Ioi,
>>
>> This looks really good.  The only thing is that we have code to 
>> create histograms in the VM (see utilities/histogram.hpp). Couldn't 
>> you reuse this class instead of writing your own histogram creating 
>> function?   It would sort your data for you! I don't know if it would 
>> have the data you are looking for but lots of other jvm code uses 
>> this class for a similar purpose.
>>
>> Thanks,
>> Coleen
>>
>> On 5/21/2013 6:29 PM, Ioi Lam wrote:
>>> Please review:
>>>
>>> http://cr.openjdk.java.net/~iklam/8014912/print_sharedspc_001/
>>>
>>> Bug: Restore PrintSharedSpaces functionality after NPG
>>>
>>> http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=8014912
>>>     https://jbs.oracle.com/bugs/browse/JDK-8014912
>>>
>>> Summary of fix:
>>>
>>> I added a new enum, MetaspaceObj::Type, which is passed to
>>>
>>>     Metaspace::allocate(ClassLoaderData* loader_data, size_t word_size,
>>>                         bool read_only, MetaspaceObj::Type type, 
>>> TRAPS);
>>>
>>> During CDS dumping, all allocations are logged in a linked list to
>>> be processed later by metaspaceShared.cpp to generate statistics for
>>> objects in the CDS archive. See sample output below.
>>>
>>> * No such logging is doneat normal VM execution time so there
>>>   is no performance impact.
>>>
>>> * I verified that the patch can be applied cleanly on Jon's
>>>   recentpatch for JDK-8014862 "Add fast Metasapce capacity and
>>>   used per MetadataType"
>>>
>>> http://hg.openjdk.java.net/hsx/hotspot-gc/hotspot/rev/2b1a9d972fc2
>>>
>>> Testing:
>>>
>>>     JPRT, UTE (vm.quick.testlist)
>>>
>>> Sample output:
>>>
>>>                         ro_cnt   ro_bytes     % | rw_cnt 
>>> rw_bytes     % |  all_cnt  all_bytes     %
>>> --------------------+---------------------------+---------------------------+-------------------------- 
>>>
>>> Unknown             :        1         64   0.0 | 1         64 0.0 
>>> |        2        128   0.0
>>> Class               :        0          0   0.0 | 2364 2006080 14.7 
>>> |     2364    2006080   9.2
>>> Symbol              :    49858    2507168  30.9 | 0          0 0.0 
>>> |    49858    2507168  11.5
>>> TypeArrayU1         :    13264     608336   7.5 | 2365 428784 3.1 
>>> |    15629    1037120   4.8
>>> TypeArrayU2         :     4796     341952   4.2 | 0          0 0.0 
>>> |     4796     341952   1.6
>>> TypeArrayU4         :     2371     178528   2.2 | 0          0 0.0 
>>> |     2371     178528   0.8
>>> TypeArrayU8         :     3708     365920   4.5 | 0          0 0.0 
>>> |     3708     365920   1.7
>>> TypeArrayOther      :        0          0   0.0 | 0          0 0.0 
>>> |        0          0   0.0
>>> Method              :        0          0   0.0 | 32552 3659888 26.8 
>>> |    32552    3659888  16.8
>>> ConstMethod         :    32552    4099008  50.6 | 0          0 0.0 
>>> |    32552    4099008  18.8
>>> MethodData          :        0          0   0.0 | 0          0 0.0 
>>> |        0          0   0.0
>>> ConstantPool        :        0          0   0.0 | 2365 3225600 23.6 
>>> |     2365    3225600  14.8
>>> ConstantPoolCache   :        0          0   0.0 | 2350 2212320 16.2 
>>> |     2350    2212320  10.2
>>> Annotation          :       26       1248   0.0 | 0          0 0.0 
>>> |       26       1248   0.0
>>> MethodCounters      :        0          0   0.0 | 0          0 0.0 
>>> |        0          0   0.0
>>> SymbolHashentry     :        0          0   0.0 | 49858 1595456 11.7 
>>> |    49858    1595456   7.3
>>> SymbolBuckets       :        0          0   0.0 | 20011 320176 2.3 
>>> |    20011     320176   1.5
>>> Other               :        0          0   0.0 | 0     203893 1.5 
>>> |        0     203893   0.9
>>> --------------------+---------------------------+---------------------------+-------------------------- 
>>>
>>> Total               :   106576    8102224 100.0 | 111866 13652261 
>>> 100.0 |   218442   21754485 100.0
>>>
>>>
>>> Thanks,
>>> Ioi
>>>
>>>
>>
>