RFR: Make Zero build and run with JDK8

[Christian Thalinger]

On Oct 17, 2012, at 3:34 PM, Christian Thalinger <christian.thalinger at oracle.com> wrote:

> 
> On Oct 17, 2012, at 3:05 PM, Roman Kennke <rkennke at redhat.com> wrote:
> 
>> Am Mittwoch, den 17.10.2012, 14:40 -0700 schrieb Christian Thalinger:
>>> On Oct 16, 2012, at 8:01 AM, Roman Kennke <rkennke at redhat.com> wrote:
>>> 
>>>> Hi Christian,
>>>> 
>>>>>>>>> In the recent weeks I worked on the Zero interpreter, to get it to build
>>>>>>>>> and run with JDK8, and in particular with the latest changes that came
>>>>>>>>> from mlvm (meth-lazy). The following webrev applies to hsx/hotspot-main:
>>>>>>>>> 
>>>>>>>>> http://cr.openjdk.java.net/~rkennke/zerojdk8/webrev.00/
>>>>>>> 
>>>>>>> src/share/vm/asm/codeBuffer.cpp:
>>>>>>> 
>>>>>>> -  if (dest->blob() == NULL) {
>>>>>>> +  if (dest->blob() == NULL && dest_filled != 0x0) {
>>>>>>> 
>>>>>>> Do we really need this when you have this:
>>>>>> 
>>>>>> The above is needed, because the loop above it that initializes
>>>>>> dest_filled is never executed. However, I will change the test to
>>>>>> dest_filled != NULL :-)
>>>>>> 
>>>>>>> static void pd_fill_to_bytes(void* to, size_t count, jubyte value) {
>>>>>>> 
>>>>>>> -  memset(to, value, count);
>>>>>>> +  if ( count > 0 ) memset(to, value, count);
>>>>>>> 
>>>>>>> }
>>>>>> 
>>>>>> Turns out that this is 1. not related to the other change above and 2.
>>>>>> not needed. I'll remove it.
>>>>>> 
>>>>>>> src/share/vm/interpreter/interpreter.cpp:
>>>>>>> 
>>>>>>> -  assert(_entry_table[kind] == _entry_table[abstract], "previous value must be AME entry");
>>>>>>> +  assert(_entry_table[kind] == NULL || _entry_table[kind] == _entry_table[abstract], "previous value must be AME entry or NULL");
>>>>>>> 
>>>>>>> Why did you need that change?
>>>>>> 
>>>>>> Apparently, before the whole table was initialized, and the methodhandle
>>>>>> related entries left as abstract. Now the methodhandle entries are
>>>>>> simply left to NULL. I suppose we could change the assert to 
>>>>>> 
>>>>>> assert(_entry_table[kind] == NULL, "previous value must be NULL");
>>>>>> 
>>>>>> Alternatively, we could fix the code that puts the other entries to also
>>>>>> set the methodhandle entries to AME and go back to the original assert.
>>>>>> What do you think?
>>>>> 
>>>>> TemplateInterpreterGenerator::generate_all sets all MH entries to AME:
>>>>> 
>>>>> // method handle entry kinds are generated later in MethodHandlesAdapterGenerator::generate:                                                                  
>>>>> for (int i = Interpreter::method_handle_invoke_FIRST; i <= Interpreter::method_handle_invoke_LAST; i++) {                                                      
>>>>>  Interpreter::MethodKind kind = (Interpreter::MethodKind) i;                                                                                                  
>>>>>  Interpreter::_entry_table[kind] = Interpreter::_entry_table[Interpreter::abstract];                                                                          
>>>>> }                                                                                                                                                              
>>>>> 
>>>>> You need similar code in Zero.
>>>> 
>>>> Alright, I extracted this piece of code into a separate protected method
>>>> void AbstractInterpreterGenerator::initializeMethodHandleEntries() and
>>>> call it both from templateInterpreter and cppInterpreter to initialize
>>>> the method handle entries to AME.
>>>> 
>>>> This new webrev also reverts this:
>>>> 
>>>>>> static void pd_fill_to_bytes(void* to, size_t count, jubyte value) {
>>>>>>> 
>>>>>>> -  memset(to, value, count);
>>>>>>> +  if ( count > 0 ) memset(to, value, count);
>>>>>>> 
>>>>>>> }
>>>>> 
>>>> 
>>>> .. and changes the 0x0 to NULL here:
>>>> 
>>>> 
>>>>>> src/share/vm/asm/codeBuffer.cpp:
>>>>>>> 
>>>>>>> -  if (dest->blob() == NULL) {
>>>>>>> +  if (dest->blob() == NULL && dest_filled != 0x0) {
>>>>>>> 
>>>>> 
>>>> 
>>>> I tried JRuby a little more and verified that it's actually using +indy,
>>>> and it works all well.
>>>> 
>>>> http://cr.openjdk.java.net/~rkennke/zerojdk8/webrev.01/
>>> 
>>> It seems this webrev contains the old changes.
>> 
>> Arg! I did the changes in my hotspot-comp tree then made the webrev in
>> my hotspot-main :-) Here's the correct one:
>> 
>> http://cr.openjdk.java.net/~rkennke/zerojdk8/webrev.02/
> 
> That looks better.  The only thing is we don't use camel-case for methods:
> 
> +  void initializeMethodHandleEntries();
> 
> Could you change it to:
> 
> +  void initialize_method_handle_entries();
> 
> I would do it myself but I cannot verify that I didn't break Zero.  I really should set up a build environment for Zero...

I'm in the game:

$ java -version
java version "1.8.0-ea"
Java(TM) SE Runtime Environment (build 1.8.0-ea-b60)
OpenJDK 64-Bit Zero VM (build 25.0-b06-internal-jvmg, interpreted mode)

-- Chris

> 
> -- Chris
> 
>> 
>> Cheers,
>> Roman
>> 
>> 
>