Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!usc!wuarchive!sdd.hp.com!samsung!gca!hinton From: hinton@gca.UUCP (Edward Hinton) Newsgroups: comp.arch Subject: Re: Compilers and efficiency Message-ID: <627@gca.UUCP> Date: 14 May 91 11:11:14 GMT References: <12216@mentor.cc.purdue.edu> Organization: GCA Corporation, Andover MA Lines: 48 In article rwa@cs.athabascau.ca (Ross Alexander) writes: >hrubin@pop.stat.purdue.edu (Herman Rubin) writes: > >>To clarify, I meant the specific ability to place the transmission in low gears. > >Why are you so d*mned eager to second guess >this mechanism, and divert your attention from the thing that, >compared to the machinery, you do well (guidance)? BTW, I don't >accept the `snow hypothesis'; with proper tires, second gear starts to >prevent slippage are redundant (and besides, they're hard on the >engine - low speed/high torque (lugging) strains the crankshaft and >its bearings). > I'm new to this thread, but I have a bone to pick with this reasoning, as well as whether Ross has ever lived through a winter driving an automatic transmission car in New England or a similar snow area. Snow here is WET. It slides. It becomes something akin to sawdust and motor oil mixed together during the early portions of a storm, which happens to be when MOST accidents due to skidding occur up here. I much prefer the control of a manual (translate: RISC) transmission, although when driving an automatic (translate: CISC), I have many times found that lower gears are the ONLY way to go up hill unless you care to invest in chains which contribute to road damage unless the road is covered with ice. Low gears are also necessary when you have certain engine problems miles from the nearest service station. CISC might be claimed to do things better under the conditions designed for, but I recently had a problem which required me to use the two low gears to get home in order to prevent stalling every few feet. In computer architecture terms, the principles are the same. CISC can do wonderful things that the designers thought of. But those are usually the easy cases to predict. 'When things go wrong' and 'under adverse conditions' are the ones where no amount of hardware forsight will suffice, and exceptions are where software folks really make their money. RISC makes the hard cases easier. Of course, one could claim that just as automatic transmissions also allow manual use of lower gears, CISC machines don't need to forbid simple instructions also. But we're dealing with extra baggage then, and just as the lower gears in my truck don't work nearly as well as my standard transmission car, so too, extra baggage in a CISC is unlikely to be as efficient as the regular RISC instructions. Please don't simply flame my next remark, but my philosophy is simple: "Hardware designers shouldn't keep trying to guess what unexpected ways I will want to use their hardware." I am just as creative as a hardware guy, just in a different line of work. I want fast, simple, general tools that let me be creative. RISC gives me that.