Note 1

This is a slightly different convention from that used in, for example, [nr], which defines\epsilon to be the smallest number for which 1\oplus\epsilon\neq1.

Note 2

For example, appendix E of [sunncg] displays `Kahan's summation formula', which allows you to perform a large sum without losing precision in the manner described above.

Note 3

There used to be an issue here in that all of K&R C's floating-point operations were defined to be done in double-precision -- this made things easy for compiler writers, at the expense of runtime. This is no longer true in ANSI C.

Note 4

Not impossible. Since someone has already done the hard work of implementing a BASIC interpreter in TeX (honestly!), you'd simply have to port your code to BASIC and let it rip.

Note 5

Note that, with pipelining, RISC chips can typically support some degree of on-chip parallelization, even for a single CPU.

Note 6

A statement like that can't be made without some qualification. Depending how you added it up, you could probably make a case that old Fortran dialects probably have more code actually running on CPUs, since many heavily-used libraries were written a long time ago.

Note 7

There are two versions of ps on Suns -- this example assumes you are using the /usr/ucb/ps version.

Note 8

However, there is no excuse for Bubble Sort!

Note 9

If you don't get a core file, it might be that you have your shell set to prevent it -- possibly as a (reasonable) precaution to avoid filling up filespace with `useless' core files. The command ulimit -c (sh-type shells only) will show the maximum size of core file: setting this to zero inhibits creating core files, and setting it to a very large number or to unlimited allows core files to be created. On csh-type shells, the corresponding command is limit coredumpsize unlimited

Note 10

A little known factoid for C enthusiasts: did you know that C's array reference syntax is commutative, since a[i] is defined to be equivalent to *(a+i) and is thus equal to i[a]? This means that a[3] is the fourth element of the array a, and so is 3[a]! Bizarrely enough, the latter is a legitimate array reference, but one you're probably best not including in your own code.

Note 11

IDL experts will know that the common IDL idiom for this is d=shift(dist(64),32,32), but have you ever actually compared dist with its documentation? The documentation for dist suggests that this idiom wouldn't work, but the function's actual behaviour seems to (substantially) depart from the claimed behaviour in exactly the right way.

Note 12

It's unfortunate that neither book is much of an advert for TeX's potentially beautiful typesetting -- both seem to be produced using a practically unmodified LaTeX book style.