|
From: Josef W. <Jos...@gm...> - 2012-12-21 10:57:34
|
Am 21.12.2012 11:34, schrieb sv...@va...: > + /* Adding guarded memory actions and merging them with the existing > + queue is too complex. Simply flush the queue and add this > + action immediately. Since guarded loads and stores are pretty > + rare, this is not thought likely to cause any noticeable > + performance loss as a result of the loss of event-merging > + opportunities. */ On x86, for sure they are rare. I am not so sure for ARM. Apart from merging events, another benefit of the queue is to do the dirty helper calls in chunks, reducing register save/restore overhead. I really need to check out Valgrind on my recently bought Nexus 7 ;-) The changes for Callgrind should be exactly the same. Thanks, Josef |