Newsletter of the SR-52 Users Club
published at
9459 Taylorsville Road
Dayton, OH 45424
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
CROM Use Enhancements (58/59)
New/better uses of CROM code are starting to proliferate, and will
be shared in this space as they come to my attention. Priority will be
given to Master Library over applications CROMs.
Hyperbolic Trig Functions: Fred Fish (606) has found that ML-05
can be used to calculate the Sinh and Cosh functions: ...52 Pgm 5 C'...
for Sinh, and ...S2 Pgm 5 E'... for Cosh. If you need Reg 2 for ano-
ther purpose: ...Pgm 5 SBR 110... and Pgm 5 SBR 006... for Sinh and
Cosh respectively with the argument in the display, will also work,
and run a bit faster. The user is cautioned that the E' and SBR 006
routines leave the machine in radian mode.
Register Clearing Routine Extension: Fred has also found that
ML-01 Routine CLR can be extended to clear registers up through the
displayed address by calling it via SBR 012.
Bypassing Stores: Norman Herzberg (668) suggests an improvement
to user-program-called CROM programs which devote user defined labels
to storing inputs. It is more efficient when accessing a CROM program
under user program control to store the input via the program, since
the sequence: Pgm ab c takes 3 steps and STO de only 2.
Year Day with ML-20: The following routine converts month, day,
and year to year day (0-365 or 0-366): LA S1 R/S S2 R/S S3 Pgm 20
SBR 086 - n = R/S where n=722084 for 1977, 722449 for 1978, etc. To
run: Key month (1-12), press A; key day (1-31) press R/S; key year,
press R/S; see year day. To calculate n for year y, run this routine
for Jan 0, y with n=0 in the routine. For its intended use, tapping
ML-20 at step 086 allows separate integer inputs, and speeds processing.
PPC Cryptology
Larry Mayhew (145) suggests that the "unbreakable" trapdoor method
(or variations thereof) for coding messages described in Scientific
American (Aug 77 p 120-124) might be a stimulating challenge for users
of the more advanced PPCs. Anyone want to try ? I would think that
neither the coding nor decoding processes should take more than an
hour of program execution time, to be acceptable.
Another approach that has already been explored to some extent by
HP-65 users (65-Notes V3N1p11) uses a pseudo random number generator
(PRNG) as a sort of "one-time pad". For a given message, a specified
seed number and the PRNG algorithm are the only required hidden
quantities. For encoding, the PRNG is initialized with the specified
seed, and run to produce as many outputs as there are characters to
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
encode. Each output offsets a standard number code for one character.
The entire character string is decoded using the same seed and PRNG.
The degree of difficulty of breaking the code would appear to depend
upon the degree to which different seeds produce different output
strings, and whether there is a repeated cycle identifiable in the
PRNG output. Members with cryptological expertise are invited to
share their comments, ideas, approaches, etc.
Advanced Programming Techniques IV: Topological Sorting
There are important practical uses for the precedence-ordering of
certain objects... situations where it is easy enough to determine pair
by pair which of 2 elements of a large set needs to be "done" (processed,
defined, etc) before the other, but where it is not so easy to order
all the elements in one long precedence string such that each required
paired relationship is maintained, or to detect the presence of circu-
lar relationships where an element effectively precedes itself. In
Volume I of his "The Art of Computer Programming" Professor Donald
Knuth introduces a technique he calls topological sorting, which pro-
duces an overall precedence ordering consistent with all paired order-
ings, and which can detect and isolate circular relationships. Computer
implementation of this technique requires the establishment and con-
tinued update of inter-element relationships as input precedence pairs
are examined. After all inputs have been processed, the established
relationships are examined to see what order in which to output the
elements. A straightforward brute-force (but inefficient) approach
might be to assign tables of predecessor elements to each input element.
As each precedence pair, denoted here as j.k (meaning that element j
precedes element k), is examined, j would be added to k's predecessor
table. Elements with no predecessors would be output first. Then
successive passes through all the predecessor tables would be made,
deleting elements that have been output, outputting elements when their
predecessor tables have been emptied. But for large numbers of ele-
ments and complex interrelationships, this approach would take a lot of
memory, much of which would be wasted in order to assure that each pre-
cedence table is sufficiently large to handle all expected cases. Also,
such an approach would be slow, due to the requirement for a lot of
sequential searching. Fortunately, there are better algorithms, and
Knuth devised an elegant, highly efficient one (Algorithm T, page 262),
which illustrates the use of both sequential and linked lists, a queue
overlayed within a sequential list, and subscripted subscripting. A
linked list contains elements which may be scattered throughout a com-
puter's memory, but which are made effectively sequential through so-
called link fields: a part of each clement (or the element itself)
that points to the next element by containing (or being) its address.
For doubly linked lists, each element contains a forward and a back-
ward pointer. One advantage of linking is that insertions and dele-
tions do not require moving large numbers of elements; another, that
lists of unpredictable length may be constructed one element at a time
from a single memory pool. A queue is a collection of time-related
elements, such that the first in is the first out (FIFO), which con-
trasts with a stack, where the last in is the first out (LIFO).
Algorithm T builds successor lists for each input, drawing from
a common memory pool, and establishing forward linking. In a sequen-
tial list a count is kept for each input of how many successors it
has, the address of each element corresponding to the input identifier.
52-NOTES V2N11p2
During processing, as each successor counter goes to zero, it is con-
verted to an output queue link. The program that follows mechanizes
Algorithm T and shows how these programming concepts and techniques may
be translated into 59ese. Decimal fractions serve as link fields,
input identifiers correspond directly to the addresses of a register
block (1-30), and indirect and double-indirect (a pointer points to
another pointer) addressing handle subscripting and subscripted sub-
scripting. Required memory partitioning is done under program control,
with the machine left in a 239.89 partition upon program termination.
Algorithm T may be enhanced with 6 more steps (page 543) to
isolate circular relationships (not part of my program), and might be
further improved by inputing the number of input pairs, which could
then be used to efficiently partition data memory between the sequen-
tial list and the linked-list pool.
TI-59/PC-100A Program: Topological Sorting Ed
User Instructions:
1. Initialize: Key n, press E; n printed, 0 displayed.
2. Input Relational Pairs: Key j.k*: 0 < j,k < 31; press
R/S; pairs printed, current number of pairs displayed. Repeat for up
to 50 pairs.
3. Initiate Processing: Press A; see printed topological sor-
ting of inputs followed by the number of inputs remaining to be output
(should be zero). Circular relationships encountered cause output to
be aborted, and the number of inputs not yet output will be non-zero.
*Integers j and k identify n distinct objects in a set. Each j.k pair
(j≠k) specifies that j is an immediate predecessor of k. In the j.k
format, k must occupy 2 places (i.e. for j=5, k=2, key 5.02).
Program Listing:
000: LE x:t 9 Opl7 x:t CMs S34 39 S37 31 Op4 R34 Op6 Adv 2431 Op4 CLR
030: R/S Op6 - Int S32 = X 100 = S33 + R*37 INV Int = S*37 1 SUM*33
056: R*32 INV Int + R*37 INT = S*37 R37 ÷ 100 + R*32 Int = S*32 1 SUM
082: 37 SUM 38 R38 GTO 030 LA Adv Adv 324137 0p04 0 S38 1 SUM38 CP
109: R*38 Int INV x=t 130 R38 + R*35 INV Int = S*35 R38 S35 R34 x:t
133: R38 INV x=t 105 R0 S36 R36 CP x=t 221 0p06 1 INV SUM34 R*36 INV
158: Int X 100 = S37 CP x=t 213 R*37 S31 1 INV SUM*31 R*31 Int INV
182: x=t 199 R*37 Int + R*35 INV Int = S*35 Int S35 R31 INV Int X
204: 100 = S37 GTO 166 R*36 Int S36 GTO 143 Adv 31351730 Op4 R34 0p6
236: Adv Adv Adv R/S
- - - - - - - - - -
Some New SR-52 Tips and Discoveries
Productive minority member Larry Mayhew (145) continues to explore
SR-52 behavior, and notes the following:
1. The 0 ÷ 0 error state is only a special case of the more
general case in which any division by zero takes place. Example: 5 ÷
0 = CE 1 SUM 00 RCL 00 yields -1. All the properties mentioned in
V1N1p2, V1N2p2, and V1N7p4 hold. Dividing a register by 0 does not
set the special error state, but does cause a flashing display. The
following sequence provides a simple way to check for the special error
state: 0 PROD 20. If the display flashes, the machine is in the
special error state, otherwise not.
52-NOTES V2N11p3
2. Conditional tests behave differently with undefined labels
than they do with undefined absolute addresses. The sequence: ifzro C
(where C is undefined) will cause a flashing display regardless of
whether the display is zero. However, the sequence: ifzro 999 will
cause a flashing display only when the display is zero. (A special
case of this has been noted in V1N4p6 for testing flags.) Thus,
if one can use an error condition as a flag (V2N1p4), the sequence:
ifzro 999 sets a flag (error condition) only when the display is zero.
3. I have long had the impression that the only difference between
absolute and relative addressing is the amount of space and time they
take. However, there are special cases where absolute addressing can
be used to much advantage over relative addressing. For example,
beginning at 000 write: LA ifzro 008 R01 SUM00 HLT. This has the effect
of incrementing register 00 by 1 if the display is 0, but by the
amount in Reg 1 if the display is non-zero. A few days ago I found a
practical application for having an address branch to itself: Beginning
at step 099 write: ifzro 100 STO 00, which puts any non-zero value into
Reg 00, but which puts 100 into register 00 if the display is zero.
Such tricks save space only under special conditions, since they depend
upon being able to structure one's program just so.
4. If in doubt about using p73 (V2N7p4) one can write simply:
RCL p73, and the transfer always takes place without an error being
caused.
5. One should note that the short absolute branch mentioned in
V1N7p3 will work with unconditional as well as conditional transfers.
For example: GTO STO 0 causes a transfer to 000 (works with SBR as well).
Furthermore, it can be very useful to put the short conditional branch
at the beginning of program memory, as this has the effect of simply
skipping a certain number of steps if the test is met. For example:
at step 000 write: LA ifpos S9, and steps 5..8 are skipped when the
display is positive (though ofcourse an error condition is set).
Label Rules (58/59)
The SR-52 LBL LBL tricks won't work for the 58 or 59. Apparently
the Lbl code (76) following a conditional test instruction is not
recognized as the Lbl instruction during a label search.
Rusty Wright (581) notes an inconsistency in the owner's manual
concerning the use of Lbl as a label. Page IV-43 says it cannot be
used and page V-56 implies that it can. It appears that Lbl Lbl will
work only for the unconditional transfers: GTO Lbl and SBR Lbl, and
a further restriction is illustrated by the following sequence:
GTO Lbl B Lbl Lbl R/S. Execution starting at the GTO does as expected:
normal transfer to the R/S occurs. But a call to B behaves as though
B were undefined.
As the manual says: 2nd, LRN, Ins, Del, SST, BST, and Ind are not
valid labels; p82 (HIR) is the only pseudo that is. Although the manual
cautions the user not to use R/S as a label, the reason given is not the
primary one. As Jared Weinberger (221) discovered, a subroutine labeled
R/S placed near the top of memory is sometimes uncallable by SBR R/S ...
execution can be an effective GTO R/S. There appears to be a subtle
machine-state dependence affected sometimes by CLRs and RSTs. Further
investigation of this phenomenon is invited.
52-NOTES V2N11p4
On a conditional transfer to a labeled address, the 58/59 (unlike
the 52) make a label search only if the condition is met. This has 2
consequences: 1) execution is faster for an unmet condition, particu-
larly when the labeled address is far along in a program, and 2) if
the label is undefined, an error state is produced only if the condition
is met. This provides a handy way to produce a conditional halt with
flashing display.
Extending Data Memory via Efficient Packing
Many PPC users have at one time or another saved storage space
by putting 2 numbers in a single register using the mantissa decimal
point as a convenient separator; and for one and 2-digit positive int-
egers, it is easy to store 5 to 13 quantities per register by decimal
place separation. While these 2 common methods work, they are often
wasteful. An optimum approach should just comfortably handle all
expected data values, and minimize the execution time required to
pack and unpack each datum. Before deciding on a particular approach,
one should first determine the max and min data values that can be
expected. A scaling offset may be added to each datum to make all
data positive, and a multiplying factor applied to eliminate fractions.
The difference between the scaled max and min values determines how
many data will fit into one register. For data which are so-called
Boolean variables (they assume one of only 2 values or states), use
can be made of the mantissa and decapower signs. Thus at one extreme,
you could squeeze 45 (43 in the mantissa, and 1 for each sign) Boolean
variables into one register, and at the other, only one, where data
values span a range or more than a million. Between these 2 extremes,
it would be wasteful to have to reserve 2 decimal places for each
datum that falls into a scaled range of say 0-11, the limit being 6
data per register... the same as for a 0-99 range. But if use is
made of another radix (twelve is best for this example), the limit is
doubled (see an analogous radix sixteen application in V1N5p4). Follow-
ing along the lines of routines A and B in V1N5p4 we can (in 58/59ese)
pack and unpack up to 12 numbers in the 0-11 range in Reg 1 with:
LA + 12 Prd 01 0 = SUM 1 rtn LB R1 ÷ 12 - Int S1 = X 12 = fix 0 EE
INV EE rtn. Note that since the newer machines do 13-place display
arithmetic, there is no point in mechanizing routine C. The manner in
which packed data need to be addressed will impact the complexity of
addressing mechanisms. Suppose using a TI-59 in the above example, we
wish to pack up to 1080 numbers in the 0-11 range in Reg 0-89. If we
need only to pack sequentially and unpack in reverse order, also seq-
uentially only once, with no partially filled registers, the following
routines would do: LA 12 S98 L1' R/S + 12 Prd*99 0 = SUM*99 Dsz "98"
1' 1 SUM 99 GTO A LB 12 S98 L2' R*99 ÷ 12 - Int S*99 = X 12 = fix 0
EE INV EE R/S Dsz "98" 2' 1 INV SUM 99 GTO B where CMs A initializes
routine A, routine B is initialized by A, and the routines are run by:
A, R/S, R/S,... and B, R/R, R/S,... . But if we need to be able to
pack or unpack randomly and more than once, things get more complica-
ted. Members are invited to address this problem, and to share their
best mechanizations. Of particular practical use would be the effi-
cient storage of coded personnel information, or business transactions.
In some cases it will be advantageous to group data by ranges to make
packing more efficient.
52-NOTES V2N11p5
Post-printing exercise of the above routines revealed the following
qualifiers: there are specific sequences of 12 numbers in the 0-11
range that can cause packing overflow due to the non-rounding of the
13th place, and following the use of the latter routine A, Reg 99 needs
to be decremented by 2 for proper routine B initialization. In general,
it's best not to use more than 12 decimal places for packing. The
formula for determining how many digits (n) in radix r can be packed
into a d-place decimal number is: n=Int(d÷logr) and thus for d=12, the
max number of radix twelve digits should be 11, and for radix two
(Boolean): 39 plus the 2 that the signs can represent.
The consequences of machine inexactness are many, and often insid-
ious, as many have found from painful experience, but since inexactness
is a basic characteristic of digital machines, we might as well learn
to live with it. Oscar Louik (745) reports a related problem: In using
the decimal point separation method to get one TI-59 register to hold
2 3-character Op4 Op6 prompting words, he found that 10 yx 6 = as a
multiplier did not properly convert a six place fraction to an integer.
6 INV log is just as bad, and the problem results from the fact that
both sequences produce 999999.9999959, not 1000000, and that Op4 only
acts upon the integer part of the display. The sequence: ...EE 6 INV
EE ... will multiply a displayed fraction by exactly a million, and is
a better way to unpack the fractional part. As Oscar discovered, in
this application it is best to make the most commonly used prompting
words into the integer parts, since no multiplier is required, and Op4
even does its own integer extraction.
Routines (58/59)
Unit Step and Delta Functions: Roy Chardon (515) notes that the
signum function (Op1O) can be used to evaluate the Heaviside Unit Step
(H(x)=O if x < 0, H(x)=1 if x ≥ 0), and Dirac Delta (delta(x)=0 if
x≠O, delta(x)=1 if x=0) functions ... for H(x): (Op10 + 1) Op 10; and
for delta(x): (Op 10 ABS - 1) ABS.
Angle Mode Indicator: Jared Weinberger (221) found another use
for Op1O: LA 90 cos Op10 rtn returns 0 when in degree mode, -1 for
radian, and 1 for grad.
Digit Reversing (56,57,58,59): Jared also has a TI-59 digit
reversing routine that handles ten-digit positive or negative integers:
LA (CE ÷ 10 Prd 1 - Int S2) SUM 1 R2 INV x:t A (Exc 1 X 10) rtn, where
for the 56 and 57, LA is omitted, and the A following the x=t is re-
placed with the address of the first step, or for the 57, A is replaced
by an available label.
SR-56/TI-57 Program Exchange
Dave Johnston (5) has a new catalog listing 101 SR-56 programs, 6
of which have been translated for TI-57 use. Dave continues to provide
his exchange service at bare cost (5¢ per page), but even with a little
help from the Club I'm sure he appreciates extra contributions. Dave's
also looking for more inputs to his library: 56, 57, and 58 programs.
Membership Address Changes
219: Deceased; 671: c/o ARAMCO Box 9999 Dhahran, Saudi Arabia;
149: 1205 Akers Las Cruces, NM 68001; 697: 21 Kristin Dr #718
Schaumburg, IL 60195.
52-NOTES V2N11p6 (end)