Mike,

People are free to use right justify and left justify in different
ways.  I covered the cases of eliminating leftmost and rightmost
blanks using respectiively TRTE and TRTRE.

Chris,

I have some different numbers; but if the thrust of what you are
saying is that the behavior of millicoded instructions is
model-dependent, I agree strongly.  Worse, it appears to be [calendar]
time-dependent.  IBM apparently fiddles with the millicode
implementations of these instructions on a continuing basis and in
ways that affect their relative and absolute timings.

Moreover, cache effects, mostly deleterious ones, are more significant
for loops than for single instructions.  On the other hand your
situation with SYNCSORT may require you to be very largely
model-independent in your choices of instruction sequences.

I do not tend to avoid millicoded instructions because they are
millicoded.  Their deficiencies have, I think, been exaggerated.  They
get better over time, and they make use of facilities/instructions
that we cannot use.  They are not, that is, just stored code
sequences.

John Gilmore, Ashland, MA 01721 - USA

Reply via email to