Marcel Birthelmer wrote:
> Here's the example from Wikipedia (slightly modified for (my)
> readabilitiy):
>
>         register n=(count+7)/8;
> switch(count%8){
> case 0: do{ *to = *from++;
> case 7: *to = *from++;
> case 6: *to = *from++;
> case 5: *to = *from++;
> case 4: *to = *from++;
> case 3: *to = *from++;
> case 2: *to = *from++;
> case 1: *to = *from++;
> }while(--n>0);
> }

Hmm.  This seems rather dangerous to me.  It's definitely making some
assumptions about how the particular compiler implements things.  It's also
not clear this is really a win in machine code.  While the C source code
looks compact and possibly efficient at first glance, it could expand into
more than hoped.

The first thing that makes me wonder is what the effect is of entering a DO
loop in the middle.  Since in this case the decision is performed at the
end, maybe it's OK.  Does the C definition really promise that no hidden
initialization or whatever be written by the compiler at the beginning of
the DO?  I expect not, but am not a C whiz.  I can see most compilers
probably don't do anything special there, but I still suspect this is
compiler-specific code.

As for efficiency, the initial jump into the loop may not be so great.
Compilers typically use a jump table or successive IFs to jump to the
selected case.  The tradeoff depends on the number of cases and the overhead
of indexing into a jump table in the underlying machine language.  Eight
cases is probably enough to result in a jump table, but even that can have
significant overhead depending on the machine.  If you were writing this in
PIC assembler, you'd probably notice that each case is the same length and
do a relative jump forward by a multiple of the case size.  I rather doubt a
compiler writer would have bothered to check for the special instance of all
cases being the same size and sequential and then computing the relative
jump offset directly.

So while Duff's device may look cute in C, I don't think it's something you
really want to put in your code.  You're better off with a loop that does
blocks of 8 transfers followed by a loop or SWITCH that does the remaining
fraction.  Depending on the overhead of indexing into a jump table and what
the remaining fraction turns out to be at run time, it's not clear whether a
loop or a SWITCH would be faster for the remaining transfers.

> C's default fall-through in case statements has long been its most
> controversial single feature; Duff observed that "This code forms some
> sort of argument in that debate, but I'm not sure whether it's for or
> against."[1]

I've met Tom Duff, and he's a smart guy.  This shows that even he isn't so
sure this is a good thing to recommend.  I think he was working on a PDP-11
at the time.  He probably knew enough about the instruction set and his
particular compiler to know it would work well in his case.


********************************************************************
Embed Inc, Littleton Massachusetts, http://www.embedinc.com/products
(978) 742-9014.  Gold level PIC consultants since 2000.
-- 
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist