Reply to thread

Message

<blockquote data-quote="vpupkin" data-source="post: 842379" data-attributes="member: 112198">...and what do you know, I was right on the money. It turns out that ffdshow that is included in SAF 6.0.5 is leaking like there is no tomorrow with high quality conversion from YV12-&gt;RGB enabled. Took me a while to identify it, too. It's not very visible if you just eyeball realtime taskman chart (increase in consumed memory is rather subtle), but number of handles is rapidly increasing even in realtime, and ultimately amount of free physical memory drops, too. After ~1 hour of playback free physical memory decreased from ~900Mb down to less than 100Mb, and number of open handles increased from ~50K to more than 800K!The funny part is, one of the first things that I tried was to change audio and video decoders in MP, and I was certain that the problem was still reproduceable, so I ruled out decoders. That was before I updated dshowhelper though, and I'm not going back to re-check.In any case, I can definitely see in ffdshow changelogs that there were some memory fixes in colorspace convertors, and after updating it to a new version (from SAF 6.0.6) and with the v92 dshowhelper everything seems to be rock solid in my limited testing (~2.5 hours) yesterday. Will keep monitoring, but looks like the problem is solved.</blockquote>

[QUOTE="vpupkin, post: 842379, member: 112198"] ...and what do you know, I was right on the money. It turns out that ffdshow that is included in SAF 6.0.5 is leaking like there is no tomorrow with high quality conversion from YV12->RGB enabled. Took me a while to identify it, too. It's not very visible if you just eyeball realtime taskman chart (increase in consumed memory is rather subtle), but number of handles is rapidly increasing even in realtime, and ultimately amount of free physical memory drops, too. After ~1 hour of playback free physical memory decreased from ~900Mb down to less than 100Mb, and number of open handles increased from ~50K to more than 800K! The funny part is, one of the first things that I tried was to change audio and video decoders in MP, and I was certain that the problem was still reproduceable, so I ruled out decoders. That was before I updated dshowhelper though, and I'm not going back to re-check. In any case, I can definitely see in ffdshow changelogs that there were some memory fixes in colorspace convertors, and after updating it to a new version (from SAF 6.0.6) and with the v92 dshowhelper everything seems to be rock solid in my limited testing (~2.5 hours) yesterday. Will keep monitoring, but looks like the problem is solved. [/QUOTE]

Verification