RFR 8012371: Adjust Tiered compile threshold according to available space in codecache

[Albert Noll]

Hi,

are we ready to commit the changes?

Best,
Albert

On 16.05.2013 19:14, Vladimir Kozlov wrote:
> On 5/16/13 8:41 AM, Albert Noll wrote:
>> Hi,
>>
>> Thanks again for looking at the code. I made the proposed changes.
>>
>> http://cr.openjdk.java.net/~anoll/8012371/webrev.02/
>
> This is good.
>
>> I also thought about adding:
>>
>> if (CodeCache::needs_flushing()) {
>>    return Double.MAX_VALUE;
>> }
>
> As Igor pointed it could be temporary state so we should not do that.
>
> Also compileBroker does that check before creating compile task:
>
>   if (!should_compile_new_jobs() || (UseCodeCacheFlushing && 
> CodeCache::needs_flushing())) {
>     CompilationPolicy::policy()->delay_compilation(method());
>     return NULL;
>   }
>
> Thanks,
> Vladimir
>
>>
>> Before:
>>
>> // Increase C1 compile threshold when the code cache is filled more
>> // than specified by IncreaseFirstTierCompileThresholdAt percentage.
>> // The main intention is to keep enough free space for C2 compiled code
>> // to achieve peak performance if the code cache is under stress.
>> if ((TieredStopAtLevel == CompLevel_full_optimization) && (level != 
>> CompLevel_full_optimization))  {
>>    double current_reverse_free_ratio = CodeCache::reverse_free_ratio();
>>    if (current_reverse_free_ratio > _increase_threshold_at_ratio) {
>>      k *= exp(current_reverse_free_ratio - 
>> _increase_threshold_at_ratio);
>>    }
>> }
>>
>>
>> What do you think?
>>
>> Thanks for looking again :-)
>>
>> Albert
>>
>> On 16.05.2013 04:44, Igor Veresov wrote:
>>> Sure :)
>>>
>>> igor
>>>
>>> On May 15, 2013, at 2:34 PM, Vladimir Kozlov 
>>> <vladimir.kozlov at oracle.com> wrote:
>>>
>>>> On 5/15/13 1:32 PM, Igor Veresov wrote:
>>>>> Well, there is a Tier23InlineeNotifyFreqLog option there already.. :)
>>>> I did not know about that flag but I still would like to use 
>>>> FirstTier.
>>>>
>>>> Thanks,
>>>> Vladimir
>>>>
>>>>> But I'm not insisting. Hopefully not too many people will have to 
>>>>> tweak any of that.
>>>>>
>>>>> igor
>>>>>
>>>>> On May 15, 2013, at 11:02 AM, Vladimir Kozlov 
>>>>> <vladimir.kozlov at oracle.com> wrote:
>>>>>
>>>>>> Tier23 could be confusing - we don't have 23 levels of tier :)
>>>>>> That is why I thought FirstTier is more general name for Tier1, 
>>>>>> Tier2, Tier3.
>>>>>>
>>>>>> Vladimir
>>>>>>
>>>>>> On 5/15/13 9:53 AM, Igor Veresov wrote:
>>>>>>> May I also comment about that option name?
>>>>>>> Most tier-specific options in are in the form of TierXFooBar, 
>>>>>>> where X is a compilation level(s). So, may be
>>>>>>> IncreaseTier23CompileThresholdAt?
>>>>>>>
>>>>>>> igor
>>>>>>>
>>>>>>> On May 15, 2013, at 9:04 AM, Vladimir Kozlov 
>>>>>>> <vladimir.kozlov at oracle.com> wrote:
>>>>>>>
>>>>>>>> IncreaseFirstTierCompileThresholdAt
>>