Abstract
Most of the 65536 16-bit patterns form valid machine code instructions for the DDP-516 computer, About 1000 of the remainder are investigated in this report and it is shown that combinations of functions can be assembled usefully into a single instruction. As an example of this the Algol "sign" function is coded more economically than is possible in DAP. Some comments on the hardware are included, particularly on the "skip on zero accumulator" instruction and a list of the variants of instructions concludes the report.
Contents
1. Introduction
Most small computers have a fixed instruction length, typically one word of from 12 to 18 bits, The DDP-516 computer under discussion in this report has a 16-bit word length and stores one instruction per word. Most of the 65536 bit patterns possible in a 16-bit word have a defined effect when used as instructions, but about 3250 patterns are not accounted for in the manuals. Since the PDP-8, 9 and 15 permit "micro coding" by combination of instructions, it was of interest to see if a similar technique was possible on the Honeywell DDP-516.
1.1 The Order Code of the DDP-516
The 16 instruction bits may be thought of as being divided up into 3 fields, a 2-bit flag field (F), a 4-bit op-code field (O) and a 10-bit address field (A).
FF0000AAAAAAAAAA
Where the op-code field takes on any of the values 101 to '13, '15 to '17 (using the apostrophe to signify an octal number), the instruction is a memory referencing one, and all possible combinations of FF and A....A are meaningful.
Likewise if the op-code is '14 a peripheral instruction is intended and all of these are, in principle,, defined.
If however the op-code is zero . the setting of the flag bits determines the type of the instruction as follows:
FF | |
---|---|
0 | Generic instruction Type B |
1 | Shift instruction |
2 | Skip instruction |
3 | Generic instruction Type A. |
The type A generic instruction is the most interesting but the others will be dealt with first.
2. The Skip Instructions
The general format of a skip instruction is (in binary)
1,0,0,0,0,0,R,P,M,E,Z,S1,S2,S3,S4,C
i.e. the first six bits are 100000 while the others describe the conditions under which the skip takes place.
2.1 The meaning of the bits
Leaving aside bit R, the other bits cause a skip if the following conditions are true.
P | accumulator positive |
M | memory parity error |
E | accumulator even |
Z | accumulator zero |
S1 | sense switch 1 reset |
S2 | sense switch 2 reset |
S3 | sense switch 3 reset |
S4 | sense switch 4 reset |
C | C-bit zero. |
The R bit, if set, reverses the skip decision. A few examples will make this clear.
SKP | unconditional skip is | '100000 |
NOP | no skip is | '101000 |
SRC | skip if C reset is | '100001 |
SSC | skip if C set is | '101001 |
2.2 Useful Micro coding
The instructions SSS (skip if any sense key set) and SSR (skip if all sense keys reset) are examples of microcoding included in the DAP repertoire. Their codes are '101036 and '100036 respectively. However any combination of the sense key bits may be used. E.g. '100030 will skip only if both sense keys 1 and 2 are reset, whereas '101030 will skip if either switch 1 or 2 is set.
This principle may be extended further, e.g. '100500 will skip only on positive even numbers while '101500 will skip on odd or negative numbers, '100041 will skip if both A and C are zero and '101041 will skip if either is non zero.
It is not possible to construct an instruction to skip if A > 0 or if A <= 0 due to the (rather careless) design of the logic, which is discussed in section 6.1.
3. The Shift Instructions
The format of the shift instructions is as follows:
0,1,0,0,0,0,C,C,C,C,L,L,L,L,L,L
where the code CCCC may take one of 12 values and the length of shift LLLLLL may range from 0 to 63.
The twelve types of legal shift are
CCCC | ||
---|---|---|
'X0 | LRL,LLL | long logical shift |
'X1 | LRS,LLS | long arithmetic shift |
'X2 | LRR,LLR | long rotate |
'X4 | LGR,LGL | logical shift |
'X5 | ARS,ALS | arithmetic shift |
'X6 | ARR,ALR | rotate |
where X = 0 for right and I for left, The remaining codes, '03, '13, '07 and '17, give a curious combination of arithmetic and cyclic shifts. In a left shift the bits shifted off the top, reappear at the bottom and C is set by overflow. In the long shifts the A and B registers are given a 31 bit cyclic shift and B1 is unaffected. In the case of the right shifts the bits disappearing off the bottom end are logically "or-ed" into the top end, with propagation of the sign bit. Again B1 does not take part in the long shifts.
It is not envisaged that these shifts will be useful.
4. The Generic Type B Instructions
These instructions have the general form
000000CCCCCCCCC1
except for the HLT instruction which has the last bit zero. The instructions do not form a tidy group, but their effects are as follows:
'5 | SGL | Enter Single Precision Mode |
'7 | DBL | Enter Double Precision Mode |
'11 | DXA | Disable Extended Mode |
'13 | EXA | Enter Extended Mode |
'21 | RMP | Reset Memory Parity Error |
'41 | SCA | Shift count to A |
'43 | INK | Input Keys |
'101 | NRM | Normalize |
'201 | IAB | Interchange A and B |
'401 | ENB | Enable Program Interrupt |
'1001 | INH | Inhibit Program Interrupt |
'1401 | ERM | Enter Restricted Mode |
Combinations are not likely to be useful, but the SCA instruction transfers the contents of the E-register (bits 11 - 16) to the A register. This same register is used as a transfer in the IAB instruction so that '241 transfers the A to B registers and 6 bits to the A register.
However this instruction does not work unless the High Speed Arithmetic option is fitted.
5. The Generic type A Instruction Set
The type A generic instructions have the general form
110000XCCCCCCCCC
where X .... C contain the bits specifying action to be taken on the A-register and C-bit. The seventh bit X has no effect on the operation of any instruction although it is often set in the "official" instructions of the DAP-16 set.
5.1 The DAP Instructions
The operations listed in the Programmers' Manual are as follows:
'140024 | CHS | complement A1 |
'140040 | CRA | clear A |
'140100 | SSP | set sign plus |
'140200 | RCB | reset C bit |
'140320 | CSA | copy sign and set plus |
'140401 | CMA | complement A |
'140407 | TCA | two's complement A |
'140500 | SSM | set sign minus |
'140600 | SCB | set C bit |
'141044 | CAR | clear A right half |
'141050 | CAL | clear A left half |
'141140 | ICL | interchange and clear left |
'141206 | AOA | add one to A |
'141216 | ACA | add C to A |
'141240 | ICR | interchange and clear right |
'141340 | ICA | interchange A |
To follow the operation of these instructions it is necessary to consider 3 of the 4 timing periods of the machine cycle, T2, T3 and T4. T1 and T2 are concerned with a standard setting up procedure such as fetching the instruction and moving on the P register by one.
The overlap between T2 and T3, known as TLATE may cause a one to be added into A if bit 15 is one and bit 13 is zero. (This will be written as (-13 & 15) in future). The ACA operation also may take place. Otherwise this period is spent in loading A into the adder. During T3, if bits 8 & 15 are both one then another half cycle (consisting of a repeat of T2 and T3) is called for. Normally this is only required for the TCA instruction, but other uses can be found for it, if (-9 & -11) then the D register is set to the complement of the A register during T3 otherwise it is set to A. via the adder.
Most of the action takes place during T4 when various parts of the D register are put back into the A register and C-bit, The actions are as follows:
bit(s) | |
---|---|
10 | clear A1 |
11 | clear A |
14 | D1 to A1 |
15 | D to A |
16 | D to A |
9 & -11 | overflow to C bit |
9 & 11 | D right to A left |
8 & 9 | set C bit |
8 & 10 | D1 to A1 |
10 & 11 | D left to A right |
10 & 12 | D1 to C |
11 & 13 | D to A right |
11 & 14 | D to A left |
These actions are obtained from the DDP-516 Instruction Manual Vol. 2. but, as will be seen later, the same effects may be arrived at by a variety of bit patterns, not all of them accountable for from the manual. Likewise there are some inconsistencies in the description of the signals which makes it difficult to follow the action of some instructions, For example ACA ('141216) has bits 7, 9, 13, 14, 15 set to one. The manual does not mention bits 7 and 14 in its description of the signals involved, and indeed the instruction still works if they are set to zero. The CRA instruction ('140040) uses bit 11 to clear A and the manual mentions this, however bit 15 is just as effective (i.e. '140002 clears A).
5.2 The Instruction Scan
A small program was written to see if any unusual bit patterns gave rise to useful instruction codes, Each of the 1024 instructions from '140000 to '141777 was applied to 15 bit patterns with the C bit both set and reset and any changes made were printed out via the on-line line printer, To ease the burden of examination of data, four symbols, 0,1,+,- were used to print out the binary, which was grouped in 4-bit bytes. A bit which remained unchanged was printed as 0 or 1, but a 0 to 1 transition was printed "+" and a 1 to 0 transition was printed "-". The CMA instruc- tion produced nothing but + and - symbols except in the C-bit, for example.
The first surprising result to emerge was the fact that bit 7 takes no part in the microcoding, although many DAP instructions use it. This anomaly was carefully and exhaustively checked by a program which cycled through all possible generic Group A instructions on all possible bit patterns in A and C.
The next result was the wealth of instructions which had the same effect. There are 32 SCB, 40 CMA and 46 RCB instructions (or twice those numbers if bit 7 is included). There is however only one for each of ICL and ICA.
5.3 Some Useful Microcoding
The potentially most useful instructions came from the set containing (8 & 15), i.e. the 1.5 cycle instructions. They will be given mnemonics as they are mentioned.
(a) | Load A register with 1, LD1 - with codes '140402, '140406, '140502 and '140506. This is a combination of CRA, of which '140002 is a variant, and the extra half cycle, which adds one. |
(b) | add two to A register - A2A - with codes '140442, '140446, These are variations on '140042 and 6 which perform AOA, without setting the overflow bit. The extra cycle causes the sequence to occur twice. |
(c) | Load A with the inverse of the C-bit - CM1 with codes '140012, '140016, '140112, '140116. Since an acceptable variant of ACA is '140212 it is evident that the missing bit 9 causes the adder to be set to -1 rather than A before the C bit is added in. |
(d) | Add twice C to A - A2C - with codes '140452, '140456. The codes '140052 and '140056 perform ACA without setting C to overflow, it appears that the operation is being done twice. |
(e) | Load A with 2C-1 - ISG (for inverse sign) - with codes '140412, '140416, '140512, '140516. This is an extra cycle version of CM1. It is one of the most useful, having a direct relevance to the "sign" function calculation, described in section 5.5. |
(f) | copy the sign bit of A into C - CPY - this is different from CSA since the sign bit is unaffected. There are many codes for this function, the one with fewest bits set is '140321. |
(g) | add one to A without setting overflow - AD1 - has a code of '140212 and 3 others. |
(h) | add C to A without setting overflow - ADC - has a
code of '140052 and 7 others, 6 of them 1.5 cycle
instructions, e.g. '140453. ('140053, incidentally is one
of the few "do nothing" instructions, of which there are
only 40 out of 512). It is possible to combine the action of some instructions, the following combinations may have some applications. The code given is the one with fewest bits set to 1. |
(a) | RCB/SSP | '140300 | reset C bit, set sign of A positive. |
(b) | CMA/ACA | '140413 | complement A then add C to A. This forms a one's complement or a two's complement according to the setting of C (1.5 cycle instruction). |
(c) | SCB/A2A | '140602 | set the C-bit, then add 2 to A. (1.5 cycle instruction), |
(d) | SCB/AOA | '140603 | set C bit, add I to A. (1.5 cycle) |
(e) | A2C/SCB | '140612 | add C twice, then set the C-bit. |
(f) | ACA/SCB | '140613 | add C , then set C-bit. |
5.4 Character handling microcodes
Nearly every conceivable character handling instruction is available, including a logical inclusive "or" function between the two 8-bit characters. The mnemonics used L for left, R for right, B for the logical "or" of left and right, T for "to", and C for "clearing".
(a) | RTL | copy right on to left | '140250 |
(b) | LTR | copy left on to right | '140144 |
(c) | BTL | "or" right on to left | '140241 |
(d) | BTR | "or" left on to right | '140141 |
(e) | BTB | "or" to both left and right | '140341 |
(f) | BCL | "or" to right, clearing left | '140150 |
(g) | BCR | "or" to left, clearing right | '140244 |
(h) | ICS | interchange, clear left and replace the original sign bit | '140540. |
Some of these, and the regular character handling codes may be combined with setting the C-bit.
(a) | ICR/SCB | '140640 |
(b) | RTL/SCB | '140650 |
(c) | BTB/SCB | '140741 |
Some instructions also perform an "or" function between the sign bit and the C-bit leaving the result in the C-bit. Denoting this function as ORC the following combinations have been noted.
(a) | CMA/ORC | in that order | '140121 |
(b) | CHS/ORC | in that order | '140124 |
The ORC instruction does not exist on its own.
Many other combinations exist, but it is unlikely that they will be useful. Most of them simply do too much in one go, As an example of this, consider '140570. Though only a single cycle instruction, it "or's" the left half to the right, clears the left half, restores the original sign bit and "or's" it on to the C-bit. Finally the instruction '140750 is so bizarre in its operation that it is left as an exercise for the reader.
5.5 The Algol "sign" function
The sign function is quite difficult to implement on a computer if there is not a specific instruction to perform it. It is defined as +1 if the argument is positive, 0 if it is zero and -1 if it is negative. The sequence CPY, SZE, ISG will produce the negative of the sign function by the following technique.
CPY | '140321 | copy the sign bit into the C-bit but leave the sign as it was (this caters for the case of '100000, which the CSA instruction would set to zero). |
SZE | '100040 | this avoids the following instruction which will load I or -1 when the argument is already zero. |
ISG | '140412 | this loads +1 into the A register if C is set, and -1 if C is not set. |
The sequence produces the "neg sign" function, if followed by a TCA the true "sign" function may be obtained. The TCA must not be applied first, since the case of '100000 would be wrongly dealt with. Programming the same function in DAP would require six instructions and one constant location, thus
SNZ JMP *+5 CSA LDA =1 SRC TCA
6. Comments on the hardware
6.1 The Skip Instructions
The SZE and SNZ instructions are the ones which prevent effective microcoding of the skips. The reason for this is that all of the other conditions being tested are independent of one another, A1 zero, A16 zero, C zero etc., but if the A register is zero then it must be both even and non negative, However, had the instruction '100040 been equivalent to a skip if non zero, then it would have been possible to skip on a strictly positive number or on one less than or equal to zero. This would have been of material assistance in the architecture of the PL-516 (and presumably FORTRAN) compilers. It is interesting to note that this fault does not occur on the H112 12-bit Honeywell computer, where the logic has been done "right way round".
6.2 The Group B Generic Instructions
These cater for such a wide variety of functions that it is difficult to see what other way they could be organized. The chief criticism is the lack of operations affecting the B-register. Operations such as copy A to B, copy B to A and logically "or" B to A would have been preferable to the IAB instruction and could have been provided within the existing logical framework.
6.3 The Group A Generic Instructions
The logic of the Group A instructions is, in part, quite difficult to follow. In particular the different usage of logical combinations of the function bits at the different cycle times. It is strange that potentially useful functions such as load 1, add 2, etc should not be mentioned in the manual, and a little frustrating that functions such as load -1, complement C-bit, copy inverse sign,, etc. should be excluded. The use of bit 7 could have led to simplifications in the logic.
Finally the shift instructions are excellent and represent the fullest range of shift types offered as standard on this size of small computer.
7. SUMMARY OF GENERIC INSTRUCTIONS, GROUP "A"
* CMA - COMPLEMENT A ******************** 140001 140003 140005 140007 140011 140013 140015 140017 140021 140022 140023 140025 140026 140027 140031 140032 140033 140035 140036 140037 140101 140103 140105 140107 140111 140113 140115 140117 140401 140405 140411 140415 140421 140425 140431 140435 140501 140505 140511 140515 * CRA - CLEAR A *************** 140002 140006 140040 140060 140102 140106 140440 140460 * SSM - SET SIGN MINUS ********************** 140004 140014 140104 140114 140404 140414 140500 140504 140510 140514 * CHS - CHANGE SIGN ******************* 140024 140034 140424 140434 * CAR - CLEAR A, RIGHT ********************** 140044 140064 140444 140464 * CAL - CLEAR A, LEFT ********************* 140050 140070 140450 140470 * SSP - SET SIGN PLUS ********************* 140100 140110 * ICL - INTERCHANGE AND CLEAR LEFT ********************************** 140140 * ICR - INTERCHANGE AND CLEAR RIGHT *********************************** 140240 140260 * RCB - RESET THE C BIT *********************** 140200 140201 140203 140204 140205 140207 140210 140211 140213 140214 140215 140217 140220 140221 140222 140223 140224 140225 140226 140227 140230 140231 140232 140233 140234 140235 140236 140237 140301 140303 140304 140305 140307 140311 140313 140314 140315 140317 * CSA - COPY SIGN AND SET PLUS ****************************** 140320 140330 * TCA - TWOS COMPLEMENT A ************************* 140403 140407 140422 140423 140426 140427 140503 140507 * ICA - INTERCHANGE A ********************* 140340 * SCB - SET THE C BIT ********************* 140600 140601 140604 140605 140610 140611 140614 140615 140620 140621 140624 140625 140630 140631 140634 140635 140700 140701 140704 140705 140710 140711 140714 140715 140720 140721 140724 140725 140730 140731 140734 140735 * AOA - ADD ONE TO A ******************** 140202 140206 140302 140306 * AD1 - ADD ONE WITHOUT SETTING OVERFLOW **************************************** 140042 140046 140443 140447 140462 140463 140466 140467 * ACA - ADD C TO A ****************** 140212 140216 140312 140316 * ADC - ADD C WITHOUT SETTING OVERFLOW ************************************** 140052 140056 140453 140457 140472 140473 140476 140477 * CM1 - LOAD C - 1 ****************** 140012 140016 140112 140116 * LTR - COPY LEFT TO RIGHT ************************** 140144 140544 * BTR - "OR" LEFT TO RIGHT ************************** 140141 140143 140145 140147 140151 140153 140154 140155 140157 140541 140545 140551 140554 140555 * BTL - "OR" RIGHT TO LEFT ************************** 140241 140243 140245 140247 140251 140253 140254 140255 140257 140261 140262 140263 140265 140266 140267 140271 140272 140273 140274 140275 140276 140277 * RTL - COPY RIGHT TO LEFT ************************** 140250 140270 * RCB/SSP ********* 140300 140310 * CPY - COPY SIGN ***************** 140321 140322 140323 140324 140325 140326 140327 140331 140332 140333 140334 140335 140336 140337 * BTB - "OR" TO BOTH HALVES *************************** 140341 140343 140345 140347 140351 140353 140354 140355 140357 * BCL - "OR" TO RIGHT., CLEARING LEFT ************************************* 140150 * BCR - "OR" TO LEFT., CLEARING RIGHT ************************************* 140244 140264 * LD1 - LOAD ONE INTO A REGISTER ******************************** 140402 140406 140502 140506 * ISG - LOAD 2*C - 1 INTO A REGISTER ************************************ 140412 140416 140512 140516 * CMA/ACA (Correction) ********* 140413 140417 140432 140433 140436 140437 140513 140517 140532 140533 140536 140537 * A2A - ADD TWO TO A ******************** 140442 140446 * A2C - ADD 2*C TO A ******************** 140452 140456 * ICS - INTERCHANGE, CLEAR LEFT., KEEP SIGN BIT *********************************************** 140540 * SCB/A2A ********* 140602 140606 140702 140706 * SCB/AOA ********* 140603 140607 140622 140623 140626 140627 140703 140707 140722 140723 149726 140727 * A2C/SCB ********* 143612 140616 140712 140716 * ACA/SCB ********* 140613 140617 140632 140633 140636 140637 140713 140717 140732 140733 140736 140737 * ICR/SCB ********* 140640 140660 * RTL/SCB ********* 140650 140670 * BTB/SCB ********* 140741 140745 140751 140754 140755 140761 140765 140771 140774 140775 * NOA - NO ACTION ***************** 140000 140010 140020 140030 140041 140043 140045 140047 140051 140053 140054 140055 140057 140061 140062 140063 140065 140066 140067 140071 140072 140073 140074 140075 140076 140077 140400 140410 140420 140430 140441 140445 140451 140454 140455 140461 140465 140471 140474 140475