OMNIBAC Symbolic assembler(ID:6033/)
First symbolic machine code
Advanced internal project on computing at GE, featuring machine-organised memory.
Katz is the recounter
References:
We had full intentions of using the IT language for our input to the Univac, as Mr. Bemer reported here last year, then changed our minds and decided that we needed something with "name" variables similar to Fortran or AT-3 (Mathmatic).
We also desire to use a language that will be usable on both the Univac and the 650 without too much changing-around of the pseudo-code. This desire is motivated largely by both the presence of the two machines as well as by the very nature of our main activity, namely, educational work. The advantages of a common language (or, more properly, a compatible language) are quite clear but not easily attained on two machines of such different capacities.
At the present time we are planning to use a multistage process in translating from the pseudo-code (problem-oriented language) to the machine language.
It is likely that we will process the original pseudo-code into a simpler pseudo-code, then translate that code, etc., according to the following list:
POL
POL1
SML
Preprocessor
Unisap
C-10
POL will be an algebraic problem-oriented language which will be quite rich in its abilities (possibly including such things as general summation notation, matrix manipulation, "find maximum," "find minimum," differentiation, integration, etc.). The POLl will be another algebraic problem-oriented language with about the same capabilities as presently found in AT-3 (Math-matic) or Fortran. SML is Simple Machine Language and is a language which has the characteristics of a computer but no reference at all to any particular computer. We have already completed the specifications of SML and how it is to work. Following SML will come a "preprocessor," so named because it was invented for the very purpose of preceding Unisap before its utility in this particular connection was evident.
The preprocessor is currently being debugged. The coding for SML to preprocessor is now being written. Unisap has already been mentioned as an excellent example of a symbolic assembly program and has been widely distributed. C-10 neither needs any explanation (it is the Univac machine language) nor does it particularly deserve explanation.
Extract: SOAP, CASE SOAP
I shall now descend from my blue-sky position and describe what we have already finished doing in the automatic programming business. We, as have several others, started our first efforts at automatic programming after taking a long close look at Dr. Perlis' IT compiler for the IBM 650. As usual, when one examines something created elsewhere, we thought that we could make some improvements in the existing scheme without giving up any of the advantages that were already present. Our first effort was then to write a program named "Compiler II-SOAP II." This program worked very well, turned out nice "tight" coding, and included IT as a subset language. The bulk of the flow charting, planning, and coding for Compiler II-SOAP II was done by two of my undergraduate students, Mr. Lynch, a sophomore, and Mr. Knuth, a freshman.
0ne of the more unpleasant experiences encountered in the work on the compiler was the discovery that SOAP II was unable to assemble the entire compiler owing to the symbol table becoming filled up at an early stage of the assembly process. The solution to the problem was obvious but not very satisfactory.
As a matter of fact we did modify SOAP II to dump the symbol table and then reload it again in modified form, but we abandoned this philosophy as not being a worthwhile solution to the problem. Therefore, Mr. Knuth suggested that he write a new symbolic assembly program with some new features incorporated in it. Accordingly, SOAP III (later renamed CASE-SOAP III due to some rather peculiar complaints from a large corporation) was written. CASE-SOAP-III solved the symbol-table difficulty by introducing a fairly new idea--the program point. Program points are addresses which the programmer needs to introduce in order to cause the machine to function properly but which have no mnemonic value to the functioning of the program. For example, one may frequently encounter the use of addresses such as NEXT and NXT, etc., which are included solely to link the program together but which have no significance at all in the logical structure of the problem. The program point was introduced just to solve the problem of naming these "random" addresses. The program point uses no room at all in the symbol table and is continuously redefinable by the simple expedient of using the same point over again. As an example consider the following segment of coding:
RAU ROO02
BMI 1F
STL PO005
RSL ONE6 1F
1 ALO PCON
In both cases of the use of "1F" in the instruction address position, the meaning is to use the address of program point ''1" forward. Now look at the next example (which, by the way, does nothing) :
2 RAU ROO01 1F
1 SUP ONES
NZE 1F 2B
1 RAL 6 1B
Note that the use of "1F" still refers to the forward program point, while "1B" refers to the correspondingly numbered backward program point. The use of "6" in the data address of the last instruction refers to the address of that particular line of coding (i.e., the line "1 RAL 6 1B"). Using CASE-SOAP III to write our next compiler revealed that there was an economy of approximately 50 per cent in the number of symbols used in CASE-SOAP III over the equivalent program written in SOAP II. Since we are looking forward to acquiring some day some of the optional attachments for our 650, Mr. Knuth incorporated additional pseudo-operations into CASE-SOAP III which allow one to effectively write programs for the augmented machine.
The new compiler referred to in the previous paragraph has been named RUNCIBLE. I shall now stick out my neck and state quite flatly that I do not believe that any tighter program will ever be devised for the 650. RUNCIBLE is again an algebraic problem-oriented compiler for the 650. The input language includes both IT and Compiler II-SOAP II as proper subset languages. The compiler is in a single deck of cards but exists logically in twenty-four different versions. The various versions are selected at read-in time by the storage-entry switch settings. The various possibilities are given from examining the following:
(Multipass OR One-pass) AND (650 OR 653) AND (Minimum Clocking OR
Full Clocking OR No Clocking) AND (Normal Output OR Error Search)
= (Operating Mode)
We now examine in some detail the options appearing in the foregoing equation. The multipass mode of operation is the one usually associated with 650 compilers (i.e., an intermediate language step is used). In the case at hand, the intermediate language is CASE-SOAP III. The one-pass mode turns out a machine-language program directly in a five instruction per card format. In the second option the term 650 refers to the basic 2,000-word 650 as being used as the object machine (i.e., the machine on which the compiled problem is to be run), the term 653 means that the compiled program is to be run on a machine equipped with floating point, index registers, and (at the programmer's option) core storage. It is also possible to cause the output machine language to be tailored to fit a machine equipped with core storage only and neither index registers nor floating point. The various "clocking" options mentioned refer to the several "tracing" modes that can be called upon to help debug the object program. It is possible to ascertain (while actually operating the object program) upon which statement in the pseudolanguage the machine is working. This one feature is proving to be one of the most helpful debugging aids that we have ever encountered.
Extract: IT, Runcible
The input language to RUNCIBLE is an augmented version of Compiler II-SOAP II which includes names of subroutines and some other English-language options in the control words. A sample program which appears on page 20 of the manual follows :
The program reads in a non-negative value (and tests to see that it is non-negative) of I1 and then calculates I1 factorial and punches both the input number and the factorial value.
We are now naturally looking forward to the day when we shall have to consider a 650 compiler for the machine with tapes and RAMAC unit.
In conclusion let me state that in our opinion the most important thing to worry about, as far as application of machines to scientific and engineering problems is concerned, is the time consumed from the statement of the problem in English to the presentation of the numerical answer to the problem proposer.
We see no reason why the time consumed on the computer should be the only time interval which comes under close scrutiny. It seems much more reasonable to conserve the time of the scientist and engineer rather than the machine-after all, one may obtain machines by the simple expedient of ordering them from the manufacturer, but a foolproof, economic method of obtaining competent personnel is as yet an almost unsolved problem.
Extract: MATH-MATIC vs. RUNCIBLE
M. O. LOCKS (Remington Rand Univac) :
Could you give me your interpretation of the difference between an assembly system and a compiler system? Also, could you explain how something which is as flexible as a compiler can be gotten onto a 2,000-word drum with no further erasable storage?
MR. WAY:
Our view of an assembly system is a facility whereby one writes a program in some language-usually similar to machine language-and gets out an essentially one-to-one translation of that language. In our case we write in a symbolic language, using pseudo-codes and symbolic addresses, and get out a machine-language code along with the actual locations-in the case of the Univac I, for instance, C-10 operation codes and addresses-when we assemble the tape which will run the problem. A compiler, on the other hand, is usually a one-to-many translator. You put in, usually, algrebraic statements of your problem and get out a machine code to do the problem; the latter, of course, in appearance bears no resemblance whatsoever to the original problem statement.
Now your question about how we do this on a 2,000-word drum is of interest. A number of people said it could not be done, but Perlis proved that this was wrong. I notice that you are with Remington Rand Univac, so I see why the question comes up! If anybody ever writes a compiler that will fit inside the 1,000- word memory of the Univac I, I would ce~tainly like to see it. The whole philosophy of compiling is dictated by the machine available to do it. In the case of the Univac I, with ten tape units, one can spin wheels and do nearly anything if he has time. As examples, witness Math-matic and Flow-matic, which do spin wheels and take time, but do compile. With the 650 one does not use this type of philosophy, mainly because he cannot. So, one uses a different method of generating the machine language. I realize I haven't told you in detai! how to write a compiler; I didn't intend to. Does that answer your question?
MR. LOCKS:
Yes, that does basically answer the question, but I would like to inquire how much your 650 compiler can do as compared to a compiler which does spin wheels, Math-matic, for instance.
MR. WAY:
I am sure we can run rings around it with this 650 compiler, the basic reason being that the 650 storage of 2,000 words allows us to get more in the main Current Developments in Computer Pvogremming Techniques 131 storage of the machine at once. In our view the Math-matic system at present is rather handicapped by the input-output of the Univac I. The man who writes pseudo-code in Math-matic must know a lot about the Univac I, or he will not be able successfully to segment a problem. At least this has been our experience. Extract: FORTRANSIT, IT, RUNCIBLE
R. B. WISE (Armour Research Foundation) :
You mentioned that you were going to bypass the Perlis language compiler for the Univac I and write something more ambitious. Will this allow for IT compiler input?
MR. WAY:
We haven't decided as yet, but we would like to allow it if possible.
MR. WISE:
I, for one, would want to register a small protest if you do not. The IT language is about the closest thing we have today to the universal language among computers, and it seems to me that it would be very much worthwhile to have something-even something which some people may classify as mediocrewhich will allow communication among computers. This has been written for the 650; through the use of For Transit, you have access to some programs for the 704; it has been written for the 1103A; and, in somewhat modified form, it also fits in the Datatron 205. I think there is a big argument for allowing that type of input.
MR. WAY:
YOU opened a loophole when you said "in somewhat modified form." When I said before that our compiler would accept previous programs, I meant without any alteration. Extract: OMNIBAC Assembler
J. I,. KATZ (General Electric Co.) :
Compilers are designed to simplify the complexity of the instruction code itself and of the clerical problem of keeping track of storage. Now, has anybody given any serious thought to building a machine which will work directly in a symbolic code-in a three-letter mnemonic code, say, or directly in a Fortran language? Or has anybody given any thought to building a machine with fewer operations in its vocabulary, so that one person could remember them all?
MR. WAY:
I do not know of anybody considering such a machine, and, personally, I frankly wonder why you would want one. If you have a machine which will manage its own storage through the use of a compiler and will allow you to solve problems, then what do you care what goes on inside? I think that if you had a machine which would allow you to use a language such as Fortran as direct input, * before long you would not be satisfied with it, and you would be writing compilers and interpretive systems so that you would be able to use some other language.
W. F. BAUER (Space Technology Laboratories):
I would like to agree with Mr.Katz. I think there is a good possibility that machines of the future will operate in symbolic language. To back that up, look at history, and you will see that in 1951, 1952, and 1953 everyone was happy to do floating point in an interpretive way. Everyone said, "Why build the hardware into the machines?" But the hardware did eventually get built intc the machines. There are other examples like this where the hardware is getting built into the machines. I think everybody in the country today is using symbolic addresses. I think we will have computers of the future which will accept symbolic addresses. I think the user won't know and he won't care in which circuits his information is residing.
MR. KATZ:
There are several reasons for my making the comment. Very few people have heard of the first machine I coded because it was built within my own organization on research and development funds and the description was never published. It was the machine OMIBAC (Ordinal Memory Inspecting Binary Automatic Computer). OMIBAC was built during the years 1949 through 1951. I t was an operating compater in 1951. I t had a vocabulary of eleven different operations which could accomplish almost anything. At the time there was considerable discussion as to whether or not to build in the hardware for floating point. The final decision was in the affirmative. The conclusion was reached that it was simpler to let the maintenance men do the sweating and take the pressure off the programmer, since he has enough problems setting up the logic. Whether his scaling is going to hold up as the problem progresses is just too much to add to the burden. Let him code directly, add A and B, without worrying where the decimal points are. Let the machine do the worrying. This, at that time, seemed to be the answer, and I'm glad to see that this judgment is being verified
by the trend in computers.
in Proceedings of the 1958 Computer Applications Symposium, Armour Research Foundation, Illinois Institute of Technology, Chicago, Illinois view details