libstdc++/41975

Paolo Carlini paolo.carlini@oracle.com
Fri Oct 14 11:28:00 GMT 2011


Hi,
> On 10/12/2011 10:42 PM, Paolo Carlini wrote:
>> Hi,
>>> Here is an other step forward to fix this issue. In this step nodes 
>>> are now doubly linked to each other. It avoids some loops on buckets 
>>> but performance are still not ok. I have created a performance test, 
>>> see attached 41975.cc.
>> I think we can as well commit it. I can do it.
>     You are right, to see the evolution the performance test should be 
> pushed first before we start changing hashtable data structure.
So I did it with the below. I increased a bit the numbers to make the 
test more meaningful on modern machines. We should probably also iterate 
over different sizes, etc. Anyway, it's a start.

Later today, I'll also add the overdue count correctness test.

Thanks,
Paolo.

////////////////////
-------------- next part --------------
2011-10-14  François Dumont  <francois.cppdevs@free.fr>

	* testsuite/performance/23_containers/insert_erase/41975.cc: New.
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: patchlet
URL: <http://gcc.gnu.org/pipermail/libstdc++/attachments/20111014/aa0e8130/attachment.ksh>


More information about the Libstdc++ mailing list