RFR(M): 6921087: G1: remove per-GC-thread expansion tables from the fine-grain remembered sets

[John Cuthbertson]

Hi Thomas,

I would rather not remove the HRRS_VERBOSE code. It has proven to be 
invaluable in the past when debugging the dreaded "missing entry in 
remembered set" failure. So as potential candidate for investigating an 
error like this in the future - I would like to keep the code in place 
rather than recreate it. I'm OK with not using conditional compilation 
and using a 'develop' flag as suggested by Bengt.

Anyway a new webrev based upon your latest patch can be found at: 
http://cr.openjdk.java.net/~johnc/6921087/webrev.1/

Thanks,

JohnC

On 06/21/12 09:04, Thomas Schatzl wrote:
> Hi,
>
> On Thu, 2012-06-21 at 12:27 +0200, Bengt Rutisson wrote:
>   
>> John and Thomas,
>>
>> I think we should try to remove dead code and I consider code that has 
>> been #if-deffed out for several years dead. So, I'm all for removing 
>> both PRT_COUNT_OCCUPIED and SAMPLE_FOR_EVICTION. All our testing and 
>> performance measurements have been based on these being enabled. Let's 
>> remove the unused code for having those flags disabled.
>>
>> A bit more tricky question is HRRS_VERBOSE. I am not sure why this is a 
>> #define and not a develop flag. In my opinion we can remove that too 
>> along with the code it guards since HRRS_VERBOSE is disabled. On the 
>> other hand this is just logging so it would also be fine to turn 
>> HRRS_VERBOSE into a develop flag. Personally I prefer to remove the code 
>> and add it in again later if we see the need for it.
>>
>>     
>
>   I totally agree. Attached is a diff that removes the ifdef's and
> defines using the current values, and fixes the copyright year in the
> header.
> Tested with a few specjbb runs only.
>
> Thomas
>
>   

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/hotspot-gc-dev/attachments/20120622/77cfb69a/attachment.htm>