PDA

View Full Version : Help me get really fast MPEG-2 decoding into AviSynth


Blue_MiSfit
9th October 2008, 02:18
Hi all,

I have a lot of extremely high bitrate (50mbps CBR 4:2:2) MPEG-2 sources that I need to encode as quickly as possible. DGIndex / DGDecode has been my MPEG-2 decoding method of choice, but it's simply too slow (barely 30fps, which makes for _slow_ encodes)

So, I've been experimenting with DirectShowSource. I can't seem to get video into AviSynth through this method, even though I can easily play the video in Media Player Classic.

When I replicate the filter graph in GraphEdit (File Source (async) -> Nero Splitter -> ffdshow video decoder -> Haali renderer) I can get playback, but when I delete Haali renderer, save the GRF, and open in DirectShowSource, I get the "Class Not Registered" error.

Ideas???

My only real goal is to get very fast MPEG-2 decoding. Fast enough (hopefully) to saturate a quad core running fast x264 first-pass settings, with no AviSynth filtering.

~MiSfit

neuron2
9th October 2008, 02:45
The upcoming GPU version of DGMPGDec will give you a massive performance improvement.

Blue_MiSfit
9th October 2008, 02:54
But only on a system with a fast GPU, right?

I have to work on Dell 1950 servers, which only have basic integrated graphics (and 2x quad core 2.66 GHz Xeons ;))

neuron2
9th October 2008, 03:09
Try using a different DirectShow decoder, such as one that comes with a software DVD player.

Blue_MiSfit
9th October 2008, 04:25
Good idea. I tried Arcsoft, but I couldn't get it working. GraphEdit can be such a PITA..

Will do..

Other possibilities - Cyberlink, Nero, others?

-MiSfit

neuron2
9th October 2008, 04:36
I've had success with Sonic Cinemaster and the Elecard decoder that came with XMuxer Pro.

If you'd like to post a sample stream, I can try to give you a working GRF.

squid_80
9th October 2008, 07:08
The upcoming GPU version of DGMPGDec will give you a massive performance improvement.
Are you sure about that? When I tested CUDA mpeg2 decoding the speed was well below what dgmpgdec can do now. I assumed this was due to VP2 not having full mpeg2 acceleration.