71 & 75 Math ROMs, Series 80 Matrix ROM

[Valentin Albillo]

.
Hi, J-F:

My comment on your document:

Quote:This document compares the features of the HP-71B and HP-75C Math modules. The HP-75C Math pac is a 16 kB ROM module whereas the HP-71B pac is a 32 kB ROM module, from which only 27 kB are actually used.

This actually makes me see red. More below.

Also, the title of this document and your very first sentence states that this document compares the features of the HP-71B and HP-75C Math modules. I understand that this is only the first draft of it, because you've left out anything other than matrix-related functionality, and there are many (and big) differences re the rest of the function set. Waiting for a second draft or the finished document proper.

Quote:A quick look at the specifications of both modules shows quite similar features, which may be surprising since the HP-71B code is about 70% bigger.

Your explanation about the development team using direct pointer manipulations to access elements instead of indexing for speed at the expense of ROM space would explain the extra size in part but not entirely, I doubt that such a change would require anything near 11 additional Kb, i.e. almost 70% extra over the original 16 Kb ROM.

Quote:MAT A(..)=B(..) subarray copy [...] Note: The series 80 allows you to copy subarrays, such as: MAT B(3,1:3) = A(1,2:4) or MAT B(3,)=D. None of the 71 or 75 can do it.

This is a seriously needed feature criminally left out of the 71B Math ROM. Without it, you need to issue one or two (nested) FOR-NEXT loops to perform the copy/assignment, which is terribly inefficient and two orders of magnitue slower.

In the case of the 71B this is particularly noticeable because it's 5 times slower that the 75C to begin with, and it's CPU and OS are optimized for number crunching, not for branching or looping. The need for subarray handling arises very frequently and without that functionality you're left with RAM-wasting, time-wasting nested loops which run 100 times slower or more.

Also, you don't include as an important feature that the Series 80 Matrix ROM allows for the creation and handling of matrices having 0 rows and/or 0 columns without giving any errors. This allows for straightforward generalizations in many sophisticated algorithms without having to consider and code especial cases. Neither the 71B nor the 75C Math ROMs implement this.

Finally, MAT A=CON, =IDN and =ZER do admit redimensioning subscripts (i.e.: MAT A=CON(3,3)).

Quote:MAT A=B.C, MAT A=B/C, MAT A=(X)+B, MAT A=(X)-B, MAT A=(X)/B, MAT A=(X)*B+(Y)*C, MAT A=CROSS(B,C), MAT A=CSUM(B), MAT A=RSUM(B), CNORMCOL, RNORMROW, AMIN(A), AMINCOL, AMINROW, AMAX(A), AMAXCOL, AMAXROW, MINAB(A), MAXAB(A), MAXABCOL, MAXABROW, SUM(A), ABSUM(A)

That all those operations were left out of the 71B Math ROM while having no less than 5 Kb available which went to waste is unforgivable.

Most of these operations would take only a few extra bytes to implement (AMINROW, AMAXCOL, RNORMROW,...), and would be incredibly useful (SUM, MAT A=(X)*B+(Y)*C, ...) and would not only fit in the 5 Kb available, with room to spare, but any development team worth their salt could implement all of them in a few days, a week at most, with ample time to check and document them.

They were probably prevented from doing it for whatever marketing reasons. Meanwhile they were ordered to waste lots of time and effort in that useless abomination called CALC MODE, utterly wasting 5 Kb of the System ROM on it only because some <hard expletive> higher up decided that the 71B "had" to have some calculator mode, like the SHARPs and other much simpler calculators did.

Quote:Finally, the 71 only implements a basic matrix operation set.

How true and how sad. They could have implemented most of the fuller set using those 5 Kb left to waste (filled up with 0's !! can you believe it !!??)

Thanks and best regards.
V.
.