RFR: bootstrap might exit early in debug JVM

[Christian Thalinger]

No problem.  Are you pushing it or should I do it?

> On Mar 5, 2015, at 1:32 AM, Doug Simon <doug.simon at oracle.com> wrote:
> 
> Looks good to me. However, getting it integrated may take a while given the current gate instability.
> 
>> On Mar 5, 2015, at 12:53 AM, Christian Thalinger <christian.thalinger at oracle.com> wrote:
>> 
>> Sometimes a debug JVM is too slow to schedule compilations in time during a bootstrap and the result is an early exit:
>> 
>> Bootstrapping Graal in 1000 ms (compiled 0 methods)
>> 
>> Here is a fix that waits until there are some methods in the queue the first time:
>> 
>> diff --git a/src/share/vm/graal/graalCompiler.cpp b/src/share/vm/graal/graalCompiler.cpp
>> --- a/src/share/vm/graal/graalCompiler.cpp
>> +++ b/src/share/vm/graal/graalCompiler.cpp
>> @@ -85,12 +85,15 @@
>>  }
>> 
>>  int qsize;
>> -  jlong sleep_time = 1000;
>> +  bool first_round = true;
>>  int z = 0;
>>  do {
>> -    os::sleep(THREAD, sleep_time, true);
>> -    sleep_time = 100;
>> -    qsize = CompileBroker::queue_size(CompLevel_full_optimization);
>> +    // Loop until there is something in the queue.
>> +    do {
>> +      os::sleep(THREAD, 100, true);
>> +      qsize = CompileBroker::queue_size(CompLevel_full_optimization);
>> +    } while (first_round && qsize == 0);
>> +    first_round = false;
>>    if (PrintBootstrap) {
>>      while (z < (_methodsCompiled / 100)) {
>>        ++z;
>> 
>