Reply to thread

Message

<blockquote data-quote="tourettes" data-source="post: 508308" data-attributes="member: 10858">I'm still wondering why it seems to plague only 24fps files. About the clock inaccuracy, we could wake up (say 5 ms early) and then use some busy looping to sleep the rest of the time to reach the target. This is actually used in many game engines to get the exact position. Also with Vista we can use the Multimedia Scheduler to boost the priority (it will quarantee pretty good timing, much better than with XP).update: The timer code is at least a little bit buggy on current SVN version (not sure if it has any big enough inaccuraties that would trigger any side effects, but here is something that I have been using while implementing the improved syncing. Maybe it would be worth to test if it reduces the dropped frames you are seeing (I see zero late frames on my own testing with PAL material).[code]static CCritSec lock; // lock for timer initialization (multiple threads are using the timer during startup)static LONGLONG GetCurrentTimestamp(){&nbsp; LONGLONG ticks100ns;&nbsp; {&nbsp; &nbsp; CAutoLock lock(&amp;lock);&nbsp; &nbsp; if( !g_bTimerInitializer ) &nbsp; &nbsp; {&nbsp; &nbsp; &nbsp; g_bQPCAvail = QueryPerformanceFrequency( &amp;g_liQPCFreq );&nbsp; &nbsp; &nbsp; Log(&quot;GetCurrentTimestamp(): Performance timer available: %d&quot;, g_bQPCAvail);&nbsp; &nbsp; &nbsp; g_bTimerInitializer = true;&nbsp; &nbsp; }&nbsp; }&nbsp; if( g_bQPCAvail ) &nbsp; {&nbsp; &nbsp; // [url]http://msdn.microsoft.com/en-us/library/ms644904(VS.85).aspx[/url]&nbsp; &nbsp; // Use always the same CPU core (no clue if it actually makes any difference...)&nbsp; &nbsp; DWORD_PTR oldmask = SetThreadAffinityMask(::GetCurrentThread(), 0);&nbsp; &nbsp; &nbsp; QueryPerformanceCounter ((LARGE_INTEGER*)&amp;ticks100ns);&nbsp; &nbsp; &nbsp; ticks100ns&nbsp; &nbsp; = LONGLONG((double(ticks100ns) * 10000000) / double(g_liQPCFreq.QuadPart) + 0.5);&nbsp; &nbsp; &nbsp; &nbsp; SetThreadAffinityMask(::GetCurrentThread(), oldmask);&nbsp; }&nbsp; else &nbsp; {&nbsp; &nbsp; // to milliseconds to microseconds&nbsp; &nbsp; ticks100ns = GetTickCount() * 10000; &nbsp; }&nbsp; return ticks100ns;}[/code]</blockquote>

[QUOTE="tourettes, post: 508308, member: 10858"] I'm still wondering why it seems to plague only 24fps files. About the clock inaccuracy, we could wake up (say 5 ms early) and then use some busy looping to sleep the rest of the time to reach the target. This is actually used in many game engines to get the exact position. Also with Vista we can use the Multimedia Scheduler to boost the priority (it will quarantee pretty good timing, much better than with XP). [b]update:[/b] The timer code is at least a little bit buggy on current SVN version (not sure if it has any big enough inaccuraties that would trigger any side effects, but here is something that I have been using while implementing the improved syncing. Maybe it would be worth to test if it reduces the dropped frames you are seeing (I see zero late frames on my own testing with PAL material). [code] static CCritSec lock; // lock for timer initialization (multiple threads are using the timer during startup) static LONGLONG GetCurrentTimestamp() { LONGLONG ticks100ns; { CAutoLock lock(&lock); if( !g_bTimerInitializer ) { g_bQPCAvail = QueryPerformanceFrequency( &g_liQPCFreq ); Log("GetCurrentTimestamp(): Performance timer available: %d", g_bQPCAvail); g_bTimerInitializer = true; } } if( g_bQPCAvail ) { // [url]http://msdn.microsoft.com/en-us/library/ms644904(VS.85).aspx[/url] // Use always the same CPU core (no clue if it actually makes any difference...) DWORD_PTR oldmask = SetThreadAffinityMask(::GetCurrentThread(), 0); QueryPerformanceCounter ((LARGE_INTEGER*)&ticks100ns); ticks100ns = LONGLONG((double(ticks100ns) * 10000000) / double(g_liQPCFreq.QuadPart) + 0.5); SetThreadAffinityMask(::GetCurrentThread(), oldmask); } else { // to milliseconds to microseconds ticks100ns = GetTickCount() * 10000; } return ticks100ns; } [/code] [/QUOTE]

Verification