review for 7088955: add C2 IR support to the SA
Vladimir Kozlov
vladimir.kozlov at oracle.com
Fri Sep 9 17:10:27 PDT 2011
Than you, Tom.
This looks good.
Vladimir
Tom Rodriguez wrote:
> On Sep 9, 2011, at 4:30 PM, Vladimir Kozlov wrote:
>
>> Tom Rodriguez wrote:
>>> On Sep 9, 2011, at 4:16 PM, Vladimir Kozlov wrote:
>>>> Push it :)
>>>>
>>>> What criteria you used to map only some C2 node classes?
>>> I mapped the ones that seemed important and didn't require extensive work. We can always add more. Any in particular you want?
>> No, for initial push I think it is enough. Do I understand correctly that we don't need them for Replay? It is only for debugging compiler threads.
>
> The post mortem replay stuff uses the SA's support for the CI to build replay but these changes are all just targeted at debugging the compiler.
>
> tom
>
>>>> I don't see next fields in Node.java (may be not needed)
>>>>
>>>> jushort _class_id;
>>>> jushort _flags;
>>> The information in the class_id should match the inheritance so it didn't seem that necessary. I can at least add the declaration to vmStructs.cpp.
>> OK.
>>
>>>> and no C2 types.
>>> Do you mean type.hpp? I just ran out of stream at that point I think.
>> Yes, and it is fine to not have it now.
>>
>> Thanks,
>> Vladimir
>>
>>> tom
>>>> Thanks,
>>>> Vladimir
>>>>
>>>> Tom Rodriguez wrote:
>>>>> http://cr.openjdk.java.net/~never/7088955
>>>>> 21066 lines changed: 7246 ins; 13673 del; 147 mod; 29134 unchg
>>>>> 7088955: add C2 IR support to the SA
>>>>> Reviewed-by:
>>>>> These are bunch of SA improvements I collected as part of the replay
>>>>> support. It includes support for C2 types and dumping of the graph,
>>>>> the PhaseCFG, the InlineTree and MDOs. There are a bunch of new
>>>>> classes to support this but for the most part they are simple mirrors
>>>>> for their corresponding C++ classes and could be generated directly
>>>>> from the vmStructs declarations. About half of the new lines are
>>>>> copyright notices and Java boilerplate.
>>>>> The C++ changes consist only of friend declarations and moving
>>>>> nmethodBucket to the header so it can be described by vmStructs.
>>>>> This also includes support in the SA for augmenting the type database
>>>>> of a JVM during reading of a core file and dumping the type database
>>>>> with a new vmstructdump command. -Dsun.jvm.hotspot.typedb=<file> will
>>>>> read <file> after parsing the vmStructs from the child and add any new
>>>>> definitions to the type database. The saenv scripts recognize the
>>>>> environment variable SA_TYPEDB and pass the value in the property to
>>>>> the invoked VM. I also augmented the type database logic so that it
>>>>> can create const and pointer variants of types on the fly so they no
>>>>> longer need to be declared in vmStructs.cpp. Additionally I added
>>>>> support for GrowableArray templating to support reading various data
>>>>> structures.
>>>>> I also deleted the win32 and dbx debugger backends since those were
>>>>> supplanted by the windbg and proc backends.
>>>>> The webrev itself is quite large but most of the newly added classes
>>>>> were generated from the vmStructs.cpp declarations or by simple
>>>>> transliteration of C++ code. The meaningful bits are all at the
>>>>> beginning of the webrev.
>>>>> The replay stuff will be laid on top of these changes.
>>>>> Tested with sajdi tests.
>
More information about the hotspot-compiler-dev
mailing list