[jdk7u]: What is or should be the difference between BytecodeInterpreter::run and BytecodeInterpreter::runWithChecks

[Volker Simonis]

On Wed, Oct 23, 2013 at 7:43 AM, Francis ANDRE
<francis.andre.kampbell at orange.fr> wrote:
> Hi Volker
>
>
> Le 21/10/2013 09:22, Volker Simonis a écrit :
>
>> Hi Francis,
>>
>> if you build with CC_INTERP the build will create a file called
>> bytecodeInterpreterWithChecks.cpp under generated/jvmtifiles in the
>> output directory which looks as follows:
>>
>> #define VM_JVMTI
>> #include "interpreter/bytecodeInterpreter.cpp"
>>
>> So yes, you will end up with both versions in the VM.
>
> Since bytecodeInterpreterWithCheck differs only from bytecodeInterpreter by
> the VM_JVMTI code conditionned by the #define VM_JVMTI, why not using a
> single run method with an additional boolean parameter as
>
> BytecodeInterpreter::run(interpreterState istate, bool runWithCheck);
>
> This would avoid the messy xslt step producing the
> bytecodeInterpreterWithCheck
>

It's for performance reasons. You would have to check 'runWithCheck'
for every single bytecode.

>
>>
>> If bytecodeInterpreterWithChecks.cpp doesn't get created during your
>> build it may be that the Windows makefiles are rotten with regard to
>> the C++ interpreter.
>
> Ok I found the problem with the windows make.
>
>>
>> You may have a look at the Linux Makefiles in general or at our
>> ppc-aix port (http://hg.openjdk.java.net/ppc-aix-port/jdk7u/) where we
>> support the C++ interpreter (but only on Linux/PPC64 and AIX/PPC64.
>> You can download build logs of our Linux/PPC64 builds from
>> http://cr.openjdk.java.net/~simonis/ppc-aix-port/index.html to get an
>> idea how a C++ interpreter build should look like.
>>
>> Regards,
>> Volker
>>
>>
>> On Sat, Oct 19, 2013 at 3:04 PM, Francis ANDRE
>> <francis.andre.kampbell at orange.fr> wrote:
>>>
>>> Hi
>>>
>>> ON a WXP/Cygwin/jdk7u platform with CC_INTERP as true, the hotspot does
>>> not
>>> compile or link because of the alternate definition of the method
>>>   BytecodeInterpreter::run and BytecodeInterpreter::runWIthCheck
>>>
>>> #if defined(VM_JVMTI)
>>> void
>>> BytecodeInterpreter::runWithChecks(interpreterState istate) {
>>> #else
>>> void
>>> BytecodeInterpreter::run(interpreterState istate) {
>>> #endif
>>>
>>>    // In order to simplify some tests based on switches set at runtime
>>>    // we invoke the interpreter a single time after switches are enabled
>>>    // and set simpler to to test variables rather than method calls or
>>> complex
>>>    // boolean expressions.
>>>
>>>    static int initialized = 0;
>>>    static int checkit = 0;
>>> ...
>>> ...
>>>
>>> while elsewhere, both run and runWIthChecks are referenced as for example
>>>
>>>    // Allow c++ interpreter to do one initialization now that switches
>>> are
>>> set, etc.
>>>    BytecodeInterpreter start_msg(BytecodeInterpreter::initialize);
>>>    if (JvmtiExport::can_post_interpreter_events())
>>>      BytecodeInterpreter::runWithChecks(&start_msg);
>>>    else
>>>      BytecodeInterpreter::run(&start_msg);
>>>
>>>
>>> So, should both run and runWithChecks exist at the same time or should
>>> they
>>> be conditionned by VM_JVMTI?
>>>
>>> Francis
>
>