Talk:Multiply–accumulate operation

From Wikipedia, the free encyclopedia
Jump to: navigation, search
WikiProject Computing (Rated Start-class)
WikiProject icon This article is within the scope of WikiProject Computing, a collaborative effort to improve the coverage of computers, computing, and information technology on Wikipedia. If you would like to participate, please visit the project page, where you can join the discussion and see a list of open tasks.
Start-Class article Start  This article has been rated as Start-Class on the project's quality scale.
 ???  This article has not yet received a rating on the project's importance scale.

VAX EMUL?[edit]

Is the VAX's EMUL instruction a multiply-add? Its description says: "The multiplicand operand is multiplied by the multiplier operand, giving a double-length result. The addend operand is sign extended to double length and added to the result. The product operand is replaced by the final result." That sounds like a description of a multiply-add. 13:28, 16 September 2007 (UTC)

Yes, a multiply-add. No, not a multiply-accumulate, because the addend is too short to be useful as an accumulator. Dicklyon (talk) 21:26, 6 June 2011 (UTC)

Division and square root[edit]

Because of this instruction there is no need for a hardware divide or square root unit, since they can both be implemented efficiently in software using the FMA.

How is this done? --Abdull (talk) 19:56, 19 June 2008 (UTC)

I guess this is alluding to division (digital)#Fast division methods, the methods of computing square roots#Babylonian method, and Methods of computing square roots#Iterative methods for reciprocal square roots, all of which generate their results in "a few" iterations of multiply-and-add.
What's a good way of pointing this out without diverging too far from the topic of this article? -- (talk) 15:44, 17 September 2008 (UTC)
Yes check.svg Done. This article now directly links to division algorithm and methods of computing square roots, which give details of implementing these things with only multiply and add, which I hope is obvious enough can be combined into a fused multiply-add. --DavidCary (talk) 08:35, 18 November 2013 (UTC)

Dry explanation[edit]

One should try to redefine the first definition of fused add, now it's useless... —Preceding unsigned comment added by (talk) 13:49, 6 January 2011 (UTC)

Nounize it?[edit]

Instead of Multiply–accumulate, should we move it to a noun? Multiplier–accumulator? Or Multiply–accumulate instruction? What think? Dicklyon (talk) 03:05, 13 June 2011 (UTC)

Multiply-add would be more common. Certainly not accumulator or adder - that would refer to an actual bit of hardware. No need to say instruction or operation. Dmcq (talk) 05:53, 13 June 2011 (UTC)
In the hardware and DSP biz, multiplier–accumulator is a common topic (e.g. in contexts where there are no op codes or such); it presently redirects to this verb-named article. Should it have its own article then? I understand that this verb is commonly used to stand for the operation or instruction, but the article title guidelines strongly suggest noun forms for titles. Couldn't we supply the implied noun, an "operation" as the lead says? I should have suggested Multiply–accumulate operation among the alternative noun forms. Dicklyon (talk) 14:57, 13 June 2011 (UTC)
Even sticking with multiply-accumulate if you look at the first page of things google returns you get algorithm, unit, architecture, operation, four uses as a noun, plus multiply accumulator. So just on its own would seem to be the most common. Dmcq (talk) 16:07, 13 June 2011 (UTC)
But on its own isn't a noun (uses as a noun just omit the noun, whether operation, unit, etc.); titles are supposed to be topics, expressed as nouns, not common subphrases of other sorts. See WP:NOUN. Dicklyon (talk) 01:45, 14 June 2011 (UTC)
If people use it as a noun it's a noun. Why send people through redirects or decide between the various possible nouns that can follow? The IEEE 2008 standard says 'Addition, subtraction, multiplication, division, fused multiply add, square root, compare, and other operations' for instance. If it is good enough for a standard it is good enough for us I think. Dmcq (talk) 09:30, 14 June 2011 (UTC)
That sentence doesn't strike you as oddly non-parallel with a verb form in the list of nouns? Why refer to an "operation" by the verb form of what it does, just because some others sometimes do? What's the objection to multiply–accumulate operation? How is sending people through redirects a disadvantage? Dicklyon (talk) 15:28, 14 June 2011 (UTC)
It sounds perfectly okay to me. I don't think people normally stick operation onto it. In fact sticking operation at the end sounds clunky to me and I'd only do it to specially emphasise the operation part as opposed something like using a multiply-add to do something or a multiply-accumulate unit. Dmcq (talk) 17:14, 14 June 2011 (UTC)
I understand how one accustomed to a term can interpret a verb as a noun, or whatever. But we're writing for an audience who are often not that familiar, so it's a good idea to use standard grammar and punctuation to help clarify what we mean. Perhaps I'll do an RM and see if there are other opinions. Dicklyon (talk) 20:44, 14 June 2011 (UTC)

Requested move[edit]

The following discussion is an archived discussion of a requested move. Please do not modify it. Subsequent comments should be made in a new section on the talk page. No further edits should be made to this section.

The result of the move request was: no page move. I'm closing this discussion, as it seems to have run its course, and I note that the page has already been moved to the suggested target. Cheers. - GTBacchus(talk) 15:29, 28 June 2011 (UTC)

Multiply-accumulateMultiply–accumulate operation – I propose we change the verb title to the implied noun, per WP:NOUN (and change the hyphen to an en dash while we're at it, per WP:DASH). I'm listing this as a possibly controversial move, since I got pushback from an editor in my suggestion above. The "operation" seems to be the most common noun that goes with this article, and it's what the lead says the article is about. It's also general enough to support a section on a "multiplier–accumulator" or "multiply–accumulate unit". Dicklyon (talk) 20:56, 14 June 2011 (UTC)

  • Support as nominator. Dicklyon (talk) 21:21, 14 June 2011 (UTC)
  • Support but I imagine somebody is going to bring up the ArbCom injunction here... –CWenger (^@) 23:09, 14 June 2011 (UTC)
Since you brought it up, why not link it and remind us what it says? Dicklyon (talk) 23:38, 14 June 2011 (UTC)
Here it is. It says "There is to be a moratorium on article title changes that are due to hyphen/endash exchange." I suppose you could argue it is more than a hyphen/en dash exchange, but it may be better to play it safe and wait on this. –CWenger (^@) 23:44, 14 June 2011 (UTC)
It's certainly not "due to"; if someone wants to argue to keep it with a hyphen, in spite of the recent broad agreement with the provisions of WP:DASH, we could consider that as an option. Dicklyon (talk) 23:51, 14 June 2011 (UTC)
  • Oppose It is more common to refer to it without operation or instruction or function or unit or whatever. We don't have the square root operation article and a multiply or an add are perfectly okay as nouns without putting in arbitrary suffixes. Dmcq (talk) 23:13, 14 June 2011 (UTC)
    • Square root can be used as a noun though. Doubtful that multiply–accumulate could. –CWenger (^@) 23:30, 14 June 2011 (UTC)
    • A square root is certainly a thing, but we say "square root operation" (as this article currently does) when we mean the operation; a multiply or an add is not so, usually; those would be called multiplications or additions, normally. Indeed, multiply redirects to multiplication. And what is an accumulate? And odd usages are not a good reason to avoid satisfying WP:NOUN, esp. when it's so easy to fix. Dicklyon (talk) 23:43, 14 June 2011 (UTC)
  • Comment Also "multiply-add" seems to be distinctly more common than "multiply-accumulate" and the the fused form in the IEEE specification is always referred to that way. Of the ones which have something following the terms in books 'instruction' seems to be more common for multiply-add and 'operation' or 'unit' is more common for multiply-accumulate, presumably because of its digital signalling background. Dmcq (talk) 13:15, 15 June 2011 (UTC)
    • Multiply–accumulate instruction is fine with me too. I think it describes it better than multiply–add instruction. –CWenger (^@) 15:17, 15 June 2011 (UTC)
      • If we do that, the scope of the article will be significantly narrowed to instructions, and we'd need another article on the general M–A operation, as used in DSP and often in the context of systems that don't have instructions. Could do that. And multiply–add is a more generic operation, which does not always support accumulation, like the VAX instruction discussed above. Dicklyon (talk) 17:34, 15 June 2011 (UTC)
  • Oppose on two grounds:
    • This is more than a ternary operation, which would evaluate (A * B) + C; it also stores the result in C, and will not function without that aspect. Glancing at the literature suggests several other nouns. Although all of them seem to have been back-formed to the acronym MAC, Multiply accumulate calculation seems best, if any move is necessary. (The reasons to have a title be a noun are to make it the subject of the lead sentence and to make linking easy; both are accomplished now.)
    • I deeply regret Dicklyon's insistence on unidiomatic dashes; in this case, a space, a slash (/), and a hyphen are all more common, in that order; of these, multiply/accumulate is clearest; multiply accumulate is usage. Septentrionalis PMAnderson 21:08, 15 June 2011 (UTC)
      • Regret noted. In some styles, the grammatical relationship of "multiply-and-accumulate" is signaled by a slash; in some by a hyphen, and in some by a space; I agree that the latter two are uninformative, and the slash is clearest. But in wikipedia style, this is what we use en dashes for; this style is also widely recommended in style guides, as you know, some of which particularly recommend it as better than slash. And the "calculation" form is a new one on me, relatively rare; better to stick to what it's called in good sources, but style it per WP:MOS. Dicklyon (talk) 21:42, 15 June 2011 (UTC)
        • You mean the style a handful of editors wish to force on Wikipedia, to the detriment of the encyclopedia and the English language. If this dash is not dropped, this conversation will continue at WP:AE. Septentrionalis PMAnderson 21:58, 15 June 2011 (UTC)
          • If someone closes this debate as move, he will need to consider whether the en dash would run afoul of the moratorium quoted above. That's a distraction from the main point here. Dicklyon (talk) 00:06, 16 June 2011 (UTC)
  • Support. This is an especially difficult case, and one that I would not have brought in the current climate. OED associates "multiply" and "accumulate" in only one instance: "1588 T. Kyd tr. Tasso Housholders Philos. f. 25, Exchaunge that doth multiply or accumulat infinite and excessiue profits." I fear that here we are looking rather at infinite and excessive losses, not profits. But since the move only incidentally involves hyphen-versus-dash, I will comment on it. (Indeed, it does not infringe the ArbCom injunction: it is not "due to" an exchange of hyphen and dash, but only involves one incidentally to the matter of wording.) As it stands, the title is um, multiply ambiguous. If fails against at least two of the five principles given at WP:TITLE: precision (it doesn't tell the reader what the article is about, and in fact I genuinely thought "multiply" was the adverb from "multiple", when I came to investigate this RM!); and consistency (similar articles typically incorporate a noun where they can: Analog-to-digital redirects to Analog-to-digital converter). So as for the words themselves, yes: I would want {Multiply+accumulate+operation}; and since the question of punctuation inevitably then arises, I would want the final form to be "Multiply–accumulate operation". Actual sources typically use a hyphen, an unspaced slash, a spaced slash, or a mere space between the components "multiply" and "accumulate"; but sources do not determine punctuation on Wikipedia, any more than they do for other publishers. Wikipedia uses its own standards, as all publishers do. Currently, WP:MOS would call for a dash in this case, and the relevant guideline is shown to be pretty well supported by the community in polling (as it is also supported by several major style guides). NoeticaTea? 22:29, 15 June 2011 (UTC)
    • And there are quite a few sources that do use the en dash, too. E.g. [1], [2], [3], [4]. And "operation" does seem to be the most common and most relevant noun form. It is not out of character for our friend to oppose styling per the MOS, but I don't see why he has thrown in the other distractors, like backronyms for MAC. Dicklyon (talk) 23:43, 15 June 2011 (UTC)
      • Yes, your evidence shows that the en dash is used. The rationale: "Multiply" does not modify "accumulate". These two elements jointly modify "operation" (though what they are, or do, in the present title "Multiply-accumulate" is obscure). Furthermore, without the en dash "multiply" is ambiguous: adverb, as I had thought, or verb? With the en dash the meaning is immediately clear, even if most readers do not notice this clarifying effect. That's quite normal: we all read serif fonts more efficiently than sans-serif fonts, but few of us know that we do, or why this is so. NoeticaTea? 00:08, 16 June 2011 (UTC)
Interesting points to ponder, after the injunction ends. I suggest this be left as it is until that time. Tony (talk) 00:21, 16 June 2011 (UTC)
Interesting observation about the adverb. As a computer guy, it hadn't occurred to me that one might read multiply as an adverb, as in multiply-connected. Maybe for now the slash is the safe bet: Multiply/accumulate operation. Dicklyon (talk) 00:23, 16 June 2011 (UTC)
In fact I agree with Tony. No rush to settle this one. But in answer to Dick: The en dash successfully removes the unwanted adverbial reading, but it is justified on firmer grounds also. A slash would not be so apt: it would suggest alternation or (disjunctive) selection, rather than coordination. In other words, the overall operation includes the component operations multiply and accumulate (it doesn't count against this that the results of the multiply operation are subjected to the accumulate operation); so the overall operation is most aptly named as a composite characterised by the terms "multiply" and "accumulate" equally (though only partially) by each. Each is essential, and neither is selected in the overall operation (contrast "on/off switch", which can of course be justified). The fact that some sources choose slash is immaterial; that is an unprincipled choice – or at least, it does not accord with the principles (at least equally sound) that Wikipedia has adopted. NoeticaTea? 03:44, 16 June 2011 (UTC)
Good point. And we may be admonished to wait anyway; we'll see. As I said, keeping the hyphen a while is OK; the main point is to nounize the title. Dicklyon (talk) 04:04, 16 June 2011 (UTC)
  • Comment Could I suggest that people look at what is actually done rather than thinking about adverbs and suchlike. If what people actually do reads badly to an editor is no justification for going and sticking in one's own ideas. The MOS about dashes or hyphens is practically an internal thing as they look so similar. However deciding between slash or blank or dash based on your own feelings about grammar is WP:Original research or something like it. I did not sign up to Wikipedia to invent stuff but to summarize what is already there. Dmcq (talk) 07:35, 16 June 2011 (UTC)
Yes Dmcq, you could and will suggest what you like. So will others; so will I. Your assertion that dashes and hyphens "look so similar" is original research, is it not? It is at odds with the majority of style guides and published books on punctuation (which are the reliable sources in this domain). They find a significant difference. "What is already there" amounts to 1) choice of words (that is, what one would say to identify the thing discussed in the article); and then 2) spelling, which is subject to some regional variations; but then we sometimes impose 3) particular punctuation, for which every publisher has its own take on what is best practice, and on what serves best to convey the meaning. A publisher's exploration of the options, and its setting of standards, cannot be regarded as "original research"; Wikipedia's practice is no exception. The Wikipedia policy on original research concerns the content of articles, not the establishment of guidelines for punctuation, or any other style issue. Haven't I shown, in the case of my own misreading of the present title, how informative wording and imposition of rational punctuation serve to convey sense reliably?
More could be said; but our time could be better spent, yes? NoeticaTea? 08:09, 16 June 2011 (UTC)
Firstly I was not talking about the distinction between hyphens and dashes. Secondly whether or not you found the title confusing is not relevant. What is relevant for a title is what is used in the real world. Slashes are not used in anything near as many cases as blanks or hyphens/dashes. Dmcq (talk) 08:26, 16 June 2011 (UTC)
You want to talk a little longer? All right. You wrote: "The MOS about dashes or hyphens is practically an internal thing as they look so similar." Now, I don't know what you meant by that. What is an "internal thing"? But your sentence does include the assertion that "they [dashes and hyphens] look so similar." So whatever else you were doing, you did address dashes and hyphens; and your assertion that they "look so similar" provides an excellent example of "original research", which you condemn. Now, it is highly relevant that I found the title confusing – as a sophisticated reader and a competent native speaker of English, but unfamiliar with the topic. It is excellent evidence that something needs to be fixed. I have shown that adding the noun at the end is only a first step: "Multiply-accumulate operation" is still indeterminate in meaning, in more ways than one. In fact, it is positively misleading. (Such a hyphen would normally signal that multiply, however it is parsed, modifies accumulate.) "Multiply/accumulate operation" is more definite, but still leans toward a wrong reading, by which multiply and accumulate somehow alternate, or are selected between in the operation. The only option that does not steer the newcomer to a false reading is "Multiply–accumulate operation". Uncertainty is still possible; but it is avoided to the extent that punctuation can avoid it. That makes it the best form of title – the one that many major style guides would advocate, and the one that WP:MOS would advocate. "Reliable sources" for content are not always reliable sources for optimal (or even informative) punctuation. Content experts are often unable to communicate clearly because they already know what the term means without effort, however it is punctuated. Hence the need for style guides, editors, and encyclopedias for communication with the world at large, not just within a clique of initiates. NoeticaTea? 09:06, 16 June 2011 (UTC)
If you don't understand what somebody means it is better to ask than to start writing something irrelevant based on it. By 'The MOS about dashes or hyphens is practically an internal thing as they look so similar.' I meant that I was not going to dispute the MOS about dashes or hyphens on this matter as most people wouldn't even spot the difference, I wouldn't know or care if it was a dash or a hyphen in a book. As far as I'm concerned it is just some stylistic thing Wikipedia goes in for. And yes I reiterate it does not matter one whit whether you found it confusing or not if it is the main name as used. The title is for finding the article. The title is not the article. The article lead is the thing that says what it is about.There is absolutely no requirement to go around changing article titles so they correspond with your preconceptions about what things should be named like. The article is there to tell you about it, not for you to go and change to correspond with how you would name it. If you want things named your own way you're perfectly entitled to write a book and try and get it published and then perhaps the world will change to your way and the articles will need to be renamed. Dmcq (talk) 09:39, 16 June 2011 (UTC)
Dmcq, the last thing I want to do is gratuitously give offence; and I will assume, in good faith, that you share that disposition. We clearly have different views concerning the purpose of titles of Wikipedia. Your view, though perfectly understandable if titles are to be mere conventional labels – akin to serial numbers, or numbered sections of statutes – is at odds with WP:TITLE, the Project's policy on the matter. Look again at the five principles, near the top in Deciding on an article title. Two of the five concern avoidance of ambiguity and consistency with similar articles, as I have pointed out. What we do in RMs is discuss how to cover those five requirements optimally. I have shown a way to do that in the present case, and have argued at length for it. You don't like it? That's your choice; but your not liking it is no argument. You speak of my "preconceptions", but ignore the arguments I adduce that suggest that I do not work from preconceptions but from published sources concerning punctuation and style generally (including WP:MOS), and from the practice of publishers everywhere. Your refusal to look at the evidence I exhibit, or work in terms of cogent argument (dismissing it instead as "original research"), says nothing about what I offer in this dialogue, but more about your own contribution. I don't think we're going to break through this impasse; so again I suggest that we stop. Good idea? NoeticaTea? 10:10, 16 June 2011 (UTC)
MOS of publishers elsewhere is not anything to do with avoiding ambiguity and ensuring consistency with similar articles. The similar articles are articles on computing. Dmcq (talk) 10:34, 16 June 2011 (UTC)
Actually technical fields use en dashes more than most, and currently most CS and engineering articles do use the en dash where appropriate. A few from my watch list:

Bessel–Clifford function, Bezold–Brücke shift, Blinn–Phong shading model, Bohlen–Pierce scale, Brain–computer interface, Choi–Williams distribution function, Current–voltage characteristic, Diode–transistor logic, Eckert–Mauchly Award, Electrolyte–insulator–semiconductor sensor, Gain–bandwidth product, Gram–Schmidt process, Gummel–Poon model, Hartree–Fock method, Horn–Schunck method, Itakura–Saito distance, Johnson–Nyquist noise, Kanade–Lucas–Tomasi feature tracker, Karhunen–Loève theorem, Lead–lag effect, Lemelson–MIT Prize, Metal–oxide–semiconductor structure, Michelson–Morley experiment, Nyquist–Shannon sampling theorem, Pole–zero plot, Power–delay product, Push–pull output, p–n junction, Rate–distortion optimization, Rate–distortion theory, Reed–Solomon error correction, Shannon–Fano coding, Shannon–Fano–Elias coding, Shannon–Hartley theorem, Shannon–Weaver model, Source–measurement unit, Stimulus–response model, Time–frequency analysis, Time–frequency representation, Transistor–transistor logic, Vertical–horizontal illusion, Wavenumber–frequency diagram, Wave–particle duality, Weber–Fechner law, Wiener–Khinchin theorem, Zipf–Mandelbrot law

The ones I know of that don't use en dash, but should, include this one, and Adder-subtractor and Time-frequency analysis for music signal; there are probably others, but probably not many, that should be fixed. There are many more articles with wrong hyphenation and wrong capitalization; I do work on moving these toward toward consistency; the MOS is the guide, since trying to form a local consistency that's area specific is not an idea that has had much support in WP. Dicklyon (talk) 18:00, 16 June 2011 (UTC)
For the record: They should not be fixed; they are fixed. Standard English is what it has been for two centuries: we use hyphens, spaces, or nothing for compounds; dashes are punctuation, marking abrupt changes of sense or construction between the words of a sentence. Septentrionalis PMAnderson 21:28, 16 June 2011 (UTC)
You're referring to the normal dash, or em dash. The en dash fills the role of what the Cambridge grammar calls a "long hyphen". Perhaps you're right that it was not in use yet in 1811. Dicklyon (talk) 02:17, 17 June 2011 (UTC)
Yes, it's true. In 1817, "love–hate relationship" was styled with an em dash; I'm glad that didn't stick. Dicklyon (talk) 02:35, 17 June 2011 (UTC)
Update: oops, no, that book is scan is messed up; looks like a 1967 thing got appended. Anyway, odd to see the em dash styling for what would conventionally be an en dash. Dicklyon (talk) 07:17, 17 June 2011 (UTC)

Leave the hyphen[edit]

Let's defer the hyphen/dash fix until the politics around that blows over. Back to the main point: which noun form should we use? Are there serious alternatives at this point to Multiply-accumulate operation? Or any opposition in light of the fact that this seems to follow everyone's willingness to use what's most common in sources? Dicklyon (talk) 14:10, 16 June 2011 (UTC)

  • Done – Seeing no opposition, and being admonished at AE to proceed to implement my suggestion, I'll go ahead and move the page. If someone wants to close this RM with a different result, I won't object. Dicklyon (talk) 15:21, 17 June 2011 (UTC)
I just had a look at AE. Colour of the bikeshed comes to mind. Perhaps that should be bike-shed. Or perhaps bike–shed. Dmcq (talk) 20:44, 17 June 2011 (UTC)
But you didn't list the one that is correct: "bike shed". And we could argue about the spelling of color. Dicklyon (talk) 23:49, 17 June 2011 (UTC)
The above discussion is preserved as an archive of a requested move. Please do not modify it. Subsequent comments should be made in a new section on this talk page. No further edits should be made to this section.