Path: chonsp.franklin.ch!not-for-mail From: Neil Franklin Newsgroups: alt.sys.pdp8,alt.sys.pdp10,alt.folklore.computers Subject: Byte access (was Re: Emulation,simulation and NO-Operation.) Date: 21 Mar 2000 20:48:01 +0100 Organization: My own Private Self Lines: 38 Message-ID: <6uwvmwm7f2.fsf_-_@chonsp.franklin.ch> References: <38d5f55d$0$20818@nntp1.ba.best.com> <38D7ADDA.AC2560EB@thinkage.ca> NNTP-Posting-Host: chonsp.franklin.ch X-Trace: chonsp.franklin.ch 953668085 421 10.0.3.2 (21 Mar 2000 19:48:05 GMT) X-Complaints-To: news@chonsp.franklin.ch NNTP-Posting-Date: 21 Mar 2000 19:48:05 GMT X-Newsreader: Gnus v5.7/Emacs 20.4 Xref: chonsp.franklin.ch alt.folklore.computers:52469 "Alan T. Bowler" writes: > Joe Smith wrote: > > > The down side of the PDP-10 instruction set is that memory was word > > addressable. Therefore, to modify just one byte in a word meant reading > > in a 36-bit word, diddling with the bits, then writing the full word > > back to memory. > > Which is of course what is actually happening "under the hood" > with modern single chip system. With _some_ modern systems, not all. This is not just manufacturer, but even chip generation dependant. > Setting a single byte causes > a memory read of the cache chunk (usually something between > 1 to 4, 32 or 64 bit words), stuffing the 8 bits, and writing > the whole chunk back. A few examples: 80286 transfers only 16bits. 80386/486 transfers 8/16/32bits (uses 4 signals ByteEnable0..3). 68000 transfers 8/16bits (uses 2 signals Upper/LowerDataStrobe(?)). Sorry, I have no pinouts from newer chips. Damn magazines don't cater for self-builders any more. -- Neil Franklin, neil@franklin.ch.remove http://neil.franklin.ch/ Nerd, Geek, Hacker, Unix Guru, Sysadmin, Roleplayer, LARPer, Mystic Use a WIMP (Windows Icons Mouse Pulldowns) interface - or get one with a CLUE (Command Linue User Environment)?