Digital PDFs

DIGITAL-7-30-A

December 1965

17 pages

Original

0.6MB

Document:	FltPtPkg
Order Number:	DIGITAL-7-30-A
Revision:
Pages:	17
Original Filename:	http://bitsavers.org/pdf/dec/pdp7/DIGITAL-7-30-A_FltPtPkg.pdf

OCR Text

PDP-7 PROGRAM LIBRARY

NUMBER:

Digital 7-30-A

NPu1lJ!E:

Floating Point package

AUTHOR:

D. Fellows

DATE:

Revised

ABSTRACT:

A self-contained Scientific programming Syst·em

DEC

December 22, 1965

for the PDP-7 with Data and Results to 6 decimal
digit accuracy, optionally 9 decimal digit accuracy.

Instruction execution is interpretive.

Arithmetic is double precision normalized floating pOint.
CONTENTS:

1.0

·General Description of the System

2.0

Command List

3.0 . Subroutine Library
4.0

Operational Description

5.0

Illustrative program

PDP

LIBRARY

Digital-7-30-A
page 2

1.0

GENERAL DESCRIPTION OF THE SYSTEM
1.1 The Interpreter
The heart of ~, the interpretive instruction processor uses double precision floating point binary arithmetic;
data may be carried in one of two ways:
- a)

Sign plus packed 8 bit exponent plus 27-bit
magnitude, requiring two storage words.

Signed exponent and signed 35-bit magnitude,
requiring three words of storage.

The decimal precision possible with (a) is approximately 6 digits, with (b) 9 digits.
The interpreter is entered by the pseudo-instruction
ElM (enter interpretive mode) which initializes the interpretive program counter.

Instructions are executed sequen-

tially until a transfer of program control or the pseudo-instruction LIM (leave interpretive mode) is encountered.
(LIM initiates a return to the machine language program.)
The floating accumulator is never implicitly saved and in
particular is not preserved when r/o instructions areexecuted.

All interpretive instructions except those which

select data mode and the three indexing instructions, may
be executed with one level of indirect addressing.
PDP

LI BRARY

Digital-7-30-A
Page 3

1.2 Input/output
All entry of data is through a one-character-per-word
buffer.

Either teletype or paper tape may be used.

Allow-

able input is limited to numeric characters but alphanumeric
output is permitted. (see the HDG instruction.)

1.3 Iterative Operation
A primitive one level form of iterative operation is
possible in the interpretive mode (see SIX, EXI).

Any se-

quence of commands may be executed, iteratively, up to N
times (N ~ 81 91).

An address modification instruction (ADM)

is also provided which, when executed, increments the effective address of the instruction which precedes it in sequence by 2 or 3, depending on the current data mode (see
DMD).

The program must be re-initialized for a second run,

since the instructions themselves are changed (ADM).

PDP

LIBRARY

Digital-7-30-A
page 4

2.0

INTERPRETIVE COMMAND LIST
2.1 Mnemonics

DAC:

Deposit Floating Accumulator p. 6

JMS:

Subroutine JMP, Floating p. 5

INP:

Initiate Input p. 9

LAC:

Load Floating Accumulator p. 6

FCS:

Load Floating Accumulator with the Complement p. 6

ADD:

Add to the Floating Accumulator

FSB:

Subtract from the Floating Accumulator

FMP:

Multiply by the Floating Accumulator

FDV:

Divide into the Floating Accumulator

HDG:

Text Output p. 9

CAS:

Compare with Floating Accumulator p. 5

JMP:

Change Interpretive program Counter p. 5

OUT:

Initiate Output p. 9

NUM:

Indicates a Numeric Argument p. 8

laD:

I/O Device Indicator p. 7

DMD2:

Select two Word Data Mode p. 5

DMD3:

Select three Word Data Mode p. 5

SIX:

start Iteration p. 8

EXI:

End Iteration p. 8

ADM:

Modify Preceding Address p. 8

PDP

LIBRARY

Digltal-7-30-A
page 5

2.2 Control Transfer Instructions
Floating JMS,JMP
These instructions are logically the same as the
machine language JMS, 3MP.
ComEare Accumulator to storage:

CAS

The CA~ instruction requires an operand address, say A.
The logical transfer of control is to PC+1, PC+2, pc+3:

where

PC is the location of the ~ instruction and the condition of
transfer is that the signum of the quantity (C(FLOACC) - C(A))
be -1,0, +1 respectively.
allowable.

One level of indirect addressing is

For example, if the contents of the effective oper-

and address were zero, then the following program:
CAS A
JMP B (A > AC)
3MP C (A= AC)
JMP D (A < AC)

would execute "JMP B" when the floating accumulator is less
than zero: "JMP CIf when C(FLOACC) equals zero:

JMP D" when

C(FLOACC) is greater than zero.

2.3 Internal Data Mode Instructions
DMD2, DMD3 set the mode for the interpreter with respect to data handling.

The appropriate on) should be used

before any data is referenced.

No operand is used.

PDP

LIBRARY

Digital-7-30-A
page 6

Internal Data Mode Instructions

cont1d

DMD2 designates two word operands.

Considering the

extended word as bits 0-35, the operand is packed:
bit 0
= sign
bits 1-8 = two's exponent + 128
bits9-35 = binary fraction (magnitude)
Approximate range of data, 10±38.

Precision 6 deci-

mal digits.
DMD3 designates three word operands.

Considering

the extended word as bits 0-53:
bits 0-17 = signed two's exponent
bit 18
= sign
bits 19-53 = magnitude of binary fraction.
Usable range of data, 10±99.

Precision 9 decimal

digits.

2.4 Data Move Instructions
LAC, ~ will load or store respectively in accordance with the current value set by the data mode instruction.

FCS loads the floating accumulator with the comple-

ment of the effective operand.

2.5 Floating Point Arithmetic
Add, subtract, Multiply, Divide are provided. Exponentiation is provided as a library subroutine.
PDP

LIBRARY

Digital-7-30-A
page 7
Internal Arithmetic
The floating accumulator is a four register accumulator,
one register for the sign of the magnitude, one register for
the signed exponent in two's complement form and two registers
for the positive 35 bit magnitude held as a binary normalized
fraction.

The floating point accumulator is the same for both

two and three word data.

The result of an arithmetic operation

is normalized and unrounded.

The uninitiated should consult the

literature for an indication of the inherent pitfalls of this
approach to floating point arithmetic.

One excellent reference

is "Numerical Methods for Scientists and Engineers" by R. W.
Hamming; published by McGraw-Hill, 1962.
2.6 Red Tape Instructions
rOD indicates the I/O device referred to by an accompanying input or output control instruction.

It is not in-

terpreted by the interpreter but by the logical section of the
I/O subprogram.

Presently assigned device numbers are:
1

teletype

(keyboard or teleprinter)

paper tape

(reader or punch)

Numbers 3 thru 7 have not been assigned.

Reference to an

unassigned device will cause 765050 to appear in the AC lights
and an irrecoverable halt to occur in the program.
number+l00 S ) may be used to control format.

IOD+device

The occurence

of the added (octal) hundred implies that input (or output)
consists of (modulo) four-word strings.
Note that indirect addressing is not allowed with the IOD
instruction.
PDP

II BRARY

Digltal-7-30-A
page 8
Floating Point Arithmetic

cont'd

NUM

(numeric valuel
This pseudo-instruction is used with the I/O commands
and the iterative execution command §!!. When used with
INP/OUT it indicates the number of values expected; with

SIX, it indicates the number of times the following sequence of instructions is to be executed.

Indirect addres-

sing is allowable in the first use but not in the second.
ADM (address modification)
This instruction is used to modify, using a positive
increment, the instruction which precedes it in sequence.
In the two-word, (three word) data mode the increment is 2,

(3). The effective operand address (of the modified instruction) is the address modified.
2.7 Iterative Execution Instructions
These instructions take no operand.
SIX initiates an iterative operation. It is always
followed by a ~ command in the immediate address mode
whose argument is the number of passes desired.

The next

sequential address (i.e., the contents of the floating
mode program. counter) is saved for use by the ~ instruction. Only one level of iteration is allowed.
EXI reduces by one the counter (number of passes)
set by SIX and when it reaches zero falls through to the
next instruction in sequence. If the counter is not zero,
the program counter is reset and control is transferred to
that location (note that all interpretive instructions, including this one, operate in floating mode only). Multiple
exits are allowable, that is an iteration initiated by a

uc command may have several exit paths.

PDP

LI BRARY

Digital-7-30-A
page 9
2.8 I/O Instructions:

!!ill! (heading) is meaningful on output only and is used
to initiate output of prestored text. (see for reference the
assembler document with particular reference to the pseudoinstruction ~.) Formats are predetermined by the particular I/O device selected.

The effective address is the loca-

tion of the first word of the desired output.
sequence is:

The program

HDG (I) A
laD

INP/QUT are used to initiate data transfers only.
characters are (for teletype and paper tape)

The legal

SPACE
DASH
PERIOD
DECIMAL DIGIT

/~lus, the +sign .is not legal
lor minus sign
/decimal point

NULL/IDLE'
LINE FEED
CR

/ignored
/ignored
litem delimiters
litem delimiters

TAB

/0-9

If an illegal character is used "X" is typed and the buffer is
cleared.

The entire input item must be retyped to enter the num-

ber correctly.

Input format is flexible, output format rigid.

The normal

form for both input and output is:
± .DD ••• D ± EE

where the DiS are the decimal digits of a (decimally normalized) fraction and the E's represent a two digit decimal
exponent. Any meaningful variation of the normal form is
valid on input.

If the decimal point is omitted, the input

value is assumed to be integral; if the sign is omitted, it
is assumed to be positive; if the (decimal) exponent is
omitted, it is assumed to be zero. (e.g. 767.12-1 ~ 76.712;
76712 ~ 76712 etc).

If more than ten digits are input, the

PDP

LIBRARY

Digital-7-30-A
page 10
resulting value will not be true.

In the two-word mode, a

converted 35 bit value is truncated to 27 bits on input,
rounded to six digits on output. The effective address is
the first pick-up or store address and indexing is in 2 or

3 (storage) word increments in accordance with the current
data mode.
3.0 SUBROUTINE LIBRARY
subroutines are entered in floating mode with the floating ~ and the argument in the accumulator.

They exit in

floating mode with the result left in the accumulator.

The

following functions are an integral part of the ~ package.
The non-appearance of a version accurate to only 6 decimal
digits indicates that there was no essential temporal or
spatial advantage in providing a routine other than that
accurate to 9 decimal digits.

All of the functions except

Q!E (q.v.) require the following program sequence:
/argument to floating
/accumulator
/enter with floating jms

LAC (I) A
JMS

BBB

/return to program se/quence

3.1 Trigonometric FUnctions.
9 decimal digit accuracy with 6 decimal digits optionally available.
Sine/cosine (routine has optional entry, common exit).
Expected argument is in radians.

The quoted accuracy

falls off markedly when the argument is (in absolute value)
greater than 21r.
Arctangent

Call SIN

(cos).

9 decimal digits with 6 digits optionally
PDP

available.

Result is ± radians.

Call ATN.

LIBRARY

Digital-7-30-A
page 11

3.2 Natural Logarithm
The routine finds the logarithm to the base e of the
absolute value of the argument.

Accuracy is 9 decimal digits.

To find the logarithm to a different base, multiply the natural log by the log of e to the new base.

Call LOG

3.3 Exponential Functions
Normal Exponential Functio~
Result is e raised to the argument as a power.

9 de-

cimal digit accuracy with 6 decimal digits optionally
available.

Call EXP.

General Exponentiation
A compound function of ~ and LOG and hence relatively
slow.

Accuracy depends on the version of ~ in use.

calculate AtB call:
LAC (I) A
JMS
GXP
LAC
B

/direct addressing expected

3.4 Square Root

9 decimal digit accuracy if the value for which the
root was taken was exact. (e.g. 2).

In general, the preci-

sion of the result will be (necessarily) only one-half the
precision of the argument.

Call SQR.
PDP

LIBRARY

Digital-7-30-A
page 12

4.0 OPERATIONAL DESCRIPTION OF THE SYSTEM
4.1 Input/Output calling sequence
/(out(i)a)

INP (I) A
IOD
X
~M
Y

/(num i b, c(b)=y)

lA' (or its contents) is the initial data address.

IX' is the device number (1 or 2).

items to be processed.

'Y' is the number of

If 100 (octal) has been used (101

or 102) with IOD, the data will occur four words per line.
(the last line need not.)

Each data item must be termina-

ted by a tab or (line feed) carriage return.

4.2 Symbolic Tapes
1)

A definitions tape, defining for the assembler
the special symbols of the system.

The ~ system itself in four parts.

There

are two versions of part IV, one for normal
arithmetic, the other for use with machines
having an extended arithmetic element.
3)

TWo library subroutine tapes with 6 and 9 digit
accuracies.

A 'PUNDEFI request tape.
PDP

LIBRARY

Digital-7-30-A
page 13
4.3 Binary Tapes
No address assignment exists on the symbolic tapes.
The library version was assembled at 12000 (octal) for use
in 8K machines.

The system itself (with library) occupies

approximately 3000 (decimal) core locations.
is supplied.

A PUNDEF tape

Before assembling a program using disc, load

this tape through address 4 immediately after the assembler
is loaded.

4.4 Assembling from the symbolic tapes

Supply a title tape with the desired address
assignment. In lieu of thiS, the system will
be assembled at core locations 22 and ff.

Assemble together (and in order)
a
b
c
d
e

the title tape
the definitions tape
disc parts I, II, III
~desired part IV
the desired library

After punching the binary tape, get a symbol
print if you wish.

without restoring the assembler (start at 20)
assemble the PUNDEF request tape.

4.5 Assembling a Program to be used with Disc
Load'the PUNDEF tape binary before assembling, i.e." START
17770 to load assembler, START 4 to load PUNDEF tape then assemble
in the normal way.

To use, load the using program and the binary

Disc and start at the using programs beginning location.
are making use of the automatic start

binary Disc first.

If you

through the loader, load the

I~7DP
~ I

D1g1tal-7-30-A
page 14

5.0 ILLUSTRATIVE PROGRAM

DISC TEST
IPROBLEM TO CALCULATE THETA:EXP(-X) AND SIN (THETA)
IFOR X=A(B)C WHERE A IS STARTING VALUE, B IS INCREMENT
lAND C IS FI NAL VALUE
BAR 2
IENTER INTERPRETIVE MODE
BEGIN,
ElM
IDATA MODE TO 2 WORD
DMD2
IINITIAL STORE ADDRESS
INP ALPHAIX
IFROM TELETYPE
IOD 1
13 ITEMS
NUM 3
IPRECEDING INSTRUCTIONS READ IN A, S, C, TO ALPHAX, +1, +2
JMS INITAL
GAMMA,
LAC ALPHAX
/SET STARTING VALUE
DAC STARTIX
I-X
FCS STARTX
DAC MINUSIX
JMS EXP
DAC EXPIONX
JMS SIN
DAC SINIEX
JMS STOREX
LIM
ISZ COUNT
ElM
LAC STARTX
CAS GAMMAX
JMP ALPHA
JMP BETA
JMP BETA
ALPHA,
LAC ALPHAX
ADD BETAX
DAC ALPHAX
JMP GAMMA

BETA,

DELTA,

HDG MESS
IOD 1
LIM
LAC COUNT
RTL
ADD (NUr.,
DAe DELTA
ElM
OUT DATA
IOD 101
XX
JMP BEGIN

PDP

LIBRARY

Dig1tal-7-30-A
page 15

INITAL,

STOREX,

LIM
LAC (DAC DATA
DAC MODEX
DZM COUNT
ElM
JMP I INITAL
a
LIM
LAC (LAC STARTX
DAC INITEX
ElM
SIX
NUM 4

INITEX,

ADM

MODEX,

XX
ADM

ALPHAX,
BETAX,
GAMMAX,
STARTX,
MINUSX,
EXPONX,

EXI
JMP I STOREX
0

""
"0
""
"
ftJ

0
g

"
" 15 CHARS
/OUTPUT FORMAT
MESS,
TEXT/

SINEX,

-x

EXP(-X)

SINE

DATA,
DATA+1750/
START BEGIN

PDP

LIBRARY

D1g1tal-7-30-A
page 16

LINE 5:

Bar 2

tells the assembler to assign 2 words to multi-

word variables (~).
(

Later the program overrides the

variable assignment by providing explicit storage, (because the programmer decided to order his variable
storage in a definite way).
LINE 11:

comment '+1, +2', refers to

variable storage, not core

locations.
LINE 12:

This program stores four values for each value of 'x'
and then transfers the four values to array 'datal.
Subroutine 'inital' zeros the item count (for the output routine) and initializes the data storage address.
Note the shift from interpretive to machine language
and back again.

LINE 13:

'gamma' calculates and/or stores x, -x, exp (-x),
sin(exp(-x»).

LINE 21:

'storex' initializes the pick-up address, which has
been ~odified by 'ADM' then transfers the current
tabular values to the array 'data'.

LINE 30:

'alpha' increments x and continues.

LINE 34:

'beta' initiates printout on the teletype. The item
count is multiplied by four and set with 'NUM' in the
I/O sequence (delta). printout will be four items
per line (IOD 101).
PDP

LIBRARY

Dig1tal-7-30-A
page 17

The material reproduced below is one run through the
program with a=¢, b=1, c=5.

Note that the programmer misjudged

slightly when he prepared his text message.

o.
1.

-X

EXP(-X)

.000000 00 -.000000 00
.100000 01 -.100000 01
.200000 01 -.200000 01
.300000 01 -.300000 01
.400000 01 -.400000 01
.500000 01 -.500000 01

.100000 01
.367879-00
.135335-00
.497871-01
.183156-01
.673795-02

SINE

.841471 00
.359638-00
.134921-00
.497667-01
.183145-01
.673791-02

PDP

LIBRARY