home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.atari.st.tech
- Path: sparky!uunet!mdisea!uw-coco!uw-beaver!bgilbert
- From: bgilbert@cs.washington.edu (Ben Gilbert)
- Subject: Re: Co-operating processors on the Falcon
- Message-ID: <1992Nov13.212206.15649@beaver.cs.washington.edu>
- Sender: bgilbert@cs.washington.edu (Ben Gilbert)
- Organization: Computer Science & Engineering, U. of Washington, Seattle
- References: <721574179.11857@minster.york.ac.uk>
- Date: Fri, 13 Nov 92 21:22:06 GMT
- Lines: 67
-
- In article <721574179.11857@minster.york.ac.uk> andrew@minster.york.ac.uk writes:
- >If the Falcon is multitasking how will the operating system
- >ensure that memory access by the DSP are safe? Will programs running in the
- >DSP get swapped in and out or is it used like a 'dumb' peripheral which
- >processes have to contend for? I wonder how machines like the NeXT do it?
-
- No, programs running on the DSP run in their own private memory. The DSP
- has its own bus and 32 KWords of memory, in which it can run many processes
- concurrently (well, give them shared time anyway). The DSP cannot do memory
- accesses to main main memory, so there is no conflict with the memory
- management of the 68030.
-
- >
- >Will this also mean that two assemblers are needed too? One which
- >programs for the 68030 CPU are written/compiled and one which is used for
- >the 56001 DSP. Writing programs for the Falcon will be complicated. First
- >I'd write a program in C for the CPU and then one for the DSP(?) and somehow
- >link the two so that the DSP program is downloaded into the DSP where it is
- >initiated by the main program?
-
- Yes, there are separate assemblers/compilers for writing DSP code and normal
- 68030 code. I believe the developers have been sent the DSP compilers and
- docs quite a while ago. Your linking idea is on the right track, but I
- imagine it to be more like using assembly code in your C programs : you
- write your normal interface and all in C, for example, then write the
- super-duper high-speed compression algorithm in DSP code, then your normal
- code sets up the DSP with the appropriate data and sets it loose. Kind
- of like you call the SDMA sound engine to play a sound (hands off as far
- as your program is concerned), you'll call the DSP with a bunch of parameters
- and let it go to work.
-
- >
- >I'm sorry if this all sounds a little naive but I've never explored co-
- >processor systems but when I get my Falcon I'll have to.
- >
- >-Andrew.
- >
- >P.S.
- >Will it be possible to perform JPEG conversions on the Falcon using the DSP
- >so that they are 'super' fast? The conversion programm will obviously have
- >to read and write from main memory which is where my question about security
- >on memory access comes in for multiprocessing systems.
- >
-
- They have mentioned this, and it probably will be able to do these in real
- time. Like I said, the DSP can't access main memory, I think it has to
- request a DMA transfer or something to get to memory. A previous poster
- said that the DSP-68030 bus interface was capable of 1 MB/sec or so, but
- this spec could be off. At any rate, it will be fast enough to allow the
- things you were mentioning. (JPEG, virtual FAX, etc)
-
- >---------------
- >
- > Andrew Hague - Researching Computer Aided Learning & On-Line Lectures.
- > Department of Computer Science.
- > University of York, "To get that all over tan stand
- > YORK. still at the north pole during
- > Y01 5DD the summer." A. Hague 6-Nov-92
- > Tel: (+44/0)904 432762
- >
- > E-MAIL: andrew@minster.york.ac.uk
-
- Cheers,
- --
- Ben Gilbert e-mail : bgilbert@cs.washington.edu
- Department of Computer Science
- University of Washington
-