home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!cs.utexas.edu!uwm.edu!spool.mu.edu!agate!doc.ic.ac.uk!uknet!acorn!rwilson
- From: RWilson@acorn.co.uk
- Newsgroups: comp.sys.acorn
- Subject: Misc. Replay Answers
- Message-ID: <20111@acorn.co.uk>
- Date: 13 Nov 92 10:27:31 GMT
- Sender: rwilson@acorn.co.uk
- Lines: 53
-
- The Harrier file hasn't got any 'key frames" and so doesn't start well from
- arbitary positions!
-
- The Replay compressor generates a 'forwards stream' of information - if the
- decompressor leaps into the middle of it it doesn't have frame (N-1) and so
- can't decompress frame N correctly (though it can decompress it without
- crashing!). With a long enough sequence (or with some sequence cuts) frames
- greater than (N-1) will eventually be decompressed correctly.
-
- Obviously this is a disaster for random access to an arbitary movie, so the
- decompressor periodically (actually once per chunk) writes out the state of
- a frame WITHOUT any other frame dependency. To start at any frame, the
- decompressor starts at the beginning of a chunk, loads the key frame and
- quickly (the replay decompressor is QUICK) decompresses from the 'forward
- stream' to the right frame. There is no 'palette' in a Replay file (as Gary
- Palmer theorised): all pixel values refer to 15bpp RGB or YUV.
-
- For compactness the key frame information can be omitted except for that in
- the first chunk - and then you get a movie like Harrier2 which can't be
- started anywhere except the beginning. The key frame information is not read
- during normal play, so the bandwidth of data transfer doesn't need to take
- it into account: its around 20KBytes extra per second of the movie.
-
- Playing the movie at -speed 2 (or even -speed 2.4 if you want to try 30fps)
- isn't an accurate test of Replay: objects in the 12.5fps world are moving
- 'unexpectedly far' compared to those in the 25fps or 30fps world: this
- causes less efficient coding at 12.5fps compared with the same scene at
- 25fps: as a consequence the decompressor ends up doing more work to
- reconstruct a frame. So I usually tell the compressor to go for a higher
- quality (spend more bytes per frame) at 12.5fps compared with 25fps, since
- the supporting media has an easy job to provide it. 5800 bytes per frame at
- 12.5fps (average) is much easier than 5000 bytes per frame at 25fps! This
- greater number of bytes is hard to transfer at the higher speed...
-
- Playing Replay movies from a RAM disc is not particularly advantageous: the
- decompressor might have been able to use the extra memory in a better way
- (such as greater buffering of decompressed frames allowing the decompressor
- to decompress more frames 'at once'). An IDE or SCSI hard disc provides
- >300KBytes per second data transfer at not much more CPU overhead (which is
- what's important) than the RAM disc: Replay is designed to cope with the
- additional latency of a real drive (it can cope with the 0.4 second latency
- of CD-ROM, after all!) and does sequential access to the file so there's
- little actuator overhead.
-
- As for the ARM3 (or better) comment: I wrote that expecting the
- documentation (which, after all, gets burned into various CD-ROMs) to be
- eventually over taken by newer chips - we didn't want people with ARM800
- unearthing these old CD-ROMs, reading the blurb ("works on ARM2 or ARM3")
- and ringing up Customer Services to ask if that meant "works on ARM800".
- Obviously ARM600 is not a rumour. Presently ARM700 and ARM800 are the
- material for rumours ("which I cannot confirm or deny").
-
- --Roger
-