What does this code do?

[James Dow Allen]

How about this idea? Post fragments of C code which
seem fun, interesting or instructive. Puzzles can
be posed in various ways. (What does this do? Can you
see the bug? How to code this for better efficiency?
Code for better readability, etc.) Best, perhaps, would
be fragments from another's admirable application.

"What does this do?" puzzles might need comments removed
and symbols obfuscated, as I've done in both of the
following fragments.

Fragments #1 and #2 are unrelated to each other.
Please "reverse-engineer" them before Googl'ing for
exact provenance.

Fragment #1
int hw(uint64 x)
{
uint64 y;

y = x & (x >> H0);
if (y & (y >> 2 * H0))
return 1;
y = x & (x >> H1);
if (y & (y >> 2 * H1))
return 1;
y = x & (x >> H2);
if (y & (y >> 2 * H2))
return 1;
y = x & (x >> 1);
return (y & (y >> 2));
}


Fragment #2
void baz(int b, int c)
{
int a, d, e, f, g;

if (e = b, g = a = -1, c)
while (d = b) {
b = c-1, f = c = d;
while (f--)
foo(g -= a);
d = -e, e = a, a = d;
}
}


What does each of these fragments do?

James Dow Allen

 

[CBFalconer]

.... snip ...

Fragment #1
int hw(uint64 x) {
uint64 y;

y = x & (x >> H0);

nothing. H0 is undefined.

.... snip ...

Fragment #2
void baz(int b, int c) {
int a, d, e, f, g;

if (e = b, g = a = -1, c)
while (d = b) {

nothing. d is uninitialized.

 

[Anand Hariharan]

nothing. d is uninitialized.

'd' is initialised using input argument 'b' (it is not a comparison to 'b').

- Anand

 

[Richard Heathfield]

CBFalconer said:

nothing. H0 is undefined.

Chuck, don't you think James *knows* he hasn't defined H0? He's not exactly
J Random Newbie.

 

[Richard Bos]

Fragment #1
int hw(uint64 x)
{
uint64 y;

y = x & (x >> H0);
if (y & (y >> 2 * H0))
return 1;

Checks whether certain groups of 4 bits are all set. Which 4 bits
exactly depends on the values of H0/1/2, but the lowest 4 bits are
always involved because of the literal 1 and 2.

Fragment #2
void baz(int b, int c)
{
int a, d, e, f, g;

if (e = b, g = a = -1, c)
while (d = b) {
b = c-1, f = c = d;
while (f--)
foo(g -= a);
d = -e, e = a, a = d;
}
}

Outputs all numbers N from 0 <= N < b*c, in some shuffled order that I
can't be bothered to figure out.

What does each of these fragments do?

I'll tell you what both of them would do if I encountered them in real
code, though. They'd make me tell off their coders. There may be some
justification for the first, if H0/1/2 are given proper names and some
comments are added. There can be no justification for the latter,
outside the IOCCC.

Richard

 

[CBFalconer]

'd' is initialised using input argument 'b' (it is not a
comparison to 'b').

Woops - you're right.

 

[Kenny McCormack]

CBFalconer said:


Chuck, don't you think James *knows* he hasn't defined H0? He's not exactly
J Random Newbie.

Oh. The. Irony.

Note that recently the regs are getting more and more open (and openness
is a good thing - don't get me wrong!) about the fact that there are
different rules for the Clique than for, as RH so quaintly puts it, J
Random Newbie.

Note also that CBF's Clique membership seems to be in a probationary
status these days...

 

[Richard]

Oh. The. Irony.

Note that recently the regs are getting more and more open (and openness
is a good thing - don't get me wrong!) about the fact that there are
different rules for the Clique than for, as RH so quaintly puts it, J
Random Newbie.

Note also that CBF's Clique membership seems to be in a probationary
status these days...

"Chuck" is probably one of the most useless posters to a Usenet
technical group I have ever witnessed. No wonder his code is so awful -
I doubt if he was ever allowed to post in a real team as I have rarely
seen anyone so antisocial and up themselves. Better to killfile him,
much as a I abhor doing so.

 

[christophe.4.news]

Re: What does this code do?

Decrease productivity of newsgroup readers ?

 

[CBFalconer]

CBFalconer said:

Chuck, don't you think James *knows* he hasn't defined H0? He's
not exactly J Random Newbie.

I rarely pay any great attention to the writers identity. It's
still undefined.

 

[Richard Heathfield]

CBFalconer said:

I rarely pay any great attention to the writers identity. It's
still undefined.

Do you pay any great attention to the writer's article? Here's what he
wrote:

'How about this idea? Post fragments of C code which
seem fun, interesting or instructive. Puzzles can
be posed in various ways. (What does this do? Can you
see the bug? How to code this for better efficiency?
Code for better readability, etc.) Best, perhaps, would
be fragments from another's admirable application.

'"What does this do?" puzzles might need comments removed
and symbols obfuscated, as I've done in both of the
following fragments.'

Given the above, it is hardly surprising that some information was removed.
Note especially the phrase "fragments of C code".

 

[Kenny McCormack]

CBFalconer <[email protected]> tries desperately to re-assert
his position in the Clique thusly:
....

I rarely pay any great attention to the writers identity. It's
still undefined.

Oh. The. Irony.

 

[James Dow Allen]

CBFalconer said:

Do you pay any great attention to the writer's article?

Perhaps it was unwise to start with comment-less
"What does this do?" puzzles. Still, it seems
bizarre that responders complained about slight
obfuscation when the preface explained why
comments were removed!

Note especially the phrase "fragments of C code".

I suppose "fragment" isn't defined in the standard.
"Not-necessarily self-sufficient excerpt" is my
intent. Might be interesting to hear how Chuck
defines "fragment", if he can do so civilly.

Fragment #2 is "self-sufficient" in the sense
that "cc -c" won't complain, but needs the reader
to supply *very* trivial main() and foo() to
experiment. I'd expect this to be a 1 or 2
minute task.

[Fragment #2] Outputs all numbers N
from 0 <= N < b*c, in some shuffled order
that I can't be bothered to figure out.

No, it doesn't output; it visits with 'foo()'.
This lapse, and the fact that Bos got this
far, means he did indeed write the trivial
main() and foo() with the latter based on
printf(). Hard to believe no one else could
do this.

that I can't be bothered to figure out.

I don't want to impugn Bos -- we all have our
limitations -- but for a programmer he must
be peculiarly poor at arithmetic for the
order to be unclear. And, frankly, a bit
of a sore loser to blame his own inadequacy
on my puzzle!

Best wishes to all!
James

 

[Richard Harter]

CBFalconer said:

Do you pay any great attention to the writer's article? Here's what he
wrote:

On the evidence I've seen, CBF seldom pays any great attention to
the articles he is commenting on. There is no point in
mentioning this to him; he clearly isn't going to change.

Richard Harter, (e-mail address removed)
http://home.tiac.net/~cri, http://www.varinoma.com
Save the Earth now!!
It's the only planet with chocolate.

 

[Dave Hansen]

[Fragment #2] Outputs all numbers N
from 0 <= N < b*c, in some shuffled order
that I can't be bothered to figure out.

No, it doesn't output; it visits with 'foo()'.
This lapse, and the fact that Bos got this
far, means he did indeed write the trivial
main() and foo() with the latter based on
printf(). Hard to believe no one else could
do this.

that I can't be bothered to figure out.

I don't want to impugn Bos -- we all have our
limitations -- but for a programmer he must
be peculiarly poor at arithmetic for the
order to be unclear. And, frankly, a bit
of a sore loser to blame his own inadequacy
on my puzzle![/QUOTE]

He didn't say he couldn't, but that he couldn't be bothered.
Apparently he's decided it's beneath him.

But I like a puzzle. I'm thinking "spiral..."

Regards,

-=Dave

 

[Kaz Kylheku]

Fragments #1 and #2 are unrelated to each other.
Please "reverse-engineer" them before Googl'ing for
exact provenance.

Fragment #1
int hw(uint64 x)
{
    uint64    y;

    y = x & (x >> H0);

Regard x is a bitmask representing a set of integers ranging from 0 to
63. If a bit is on in a position, the corresponding integer is in the
set.

This first step means ``find all integers j in x, such that j + H0 is
also in j''. This set is called y.

Thus every integer which has a counterpart that is H0 greater is
retained.

    if (y & (y >> 2 * H0))
        return 1;

This step means ``find all integers j in y, such that j + 2*H0 is also
in y''.

So now out of these integers which have a counterpart at H1, integers
which have a counterpart at 2*H1 are retained. If this set is non-
empty, we return 1.

In other words, if any integer in the set has a counterpart offset by
H0, and also another counterpart at twice that distance in the same
direction, 1 is returned, otherwise the test continues with other
constants in place of H0.

The test looks for the presence of triplets of equally spaced
integers, where the spacing is H0.

    y = x & (x >> H1);
    if (y & (y >> 2 * H1))
        return 1;
    y = x & (x >> H2);
    if (y & (y >> 2 * H2))
        return 1;
    y = x & (x >> 1);
    return (y & (y >> 2));

}

The last test is the same, but with the constant 1. This effectively
detects whether there are any consecutive triplets in the set.

The result of y & (y >> 2) is itself a set, which gives the root
position of each triplet: an integer i is in the set if it is the root
of a triplet of consecutive integers.

However, this set doesn't necessarily fit into the return type
(assuming that the uint64 typedef name, in good faith, represents some
kind of 64 bit or perhaps wider type). This is a bug; perhaps the
intent was: return (y & (y >> 2)) != 0;

The last test would then be consistent with the others: return a 1
indication if there are triplets. And so zero is returned if there are
no triplets present with any of the spacings.

How would this triplet finding be applied in some algorithm? No clue
there. Error detection? A zero is returned if the word is free of
triplets at the given spacings. Could this constitute some kind of
error detection?

Suppose the words are used to communicate or store data. The bits of
data are encoded into the codeword such that straight triplets of 1
bits do not exist anywhere (but there can be arbitrary runs of 0).
Triplets at the two other spacings don't exist either. If the data is
retrieved or received containing triplets, it is erroneous.

But that's quite dumb since it doesn't treat 0's and 1's equally. Any
erasure of 1 to a 0 goes undetected.

However, this is just one function, and not necessarily the whole
scheme. The function could also be tried on a bitwise-inversion of the
input to test for triplets of 0 bits: the encoding scheme would be
that triplet runs of the same bit value are not allowed at various
spacings.

So for instance the data cannot be 000111000111. It must transition
between 0 and 1 more rapidly than every three.

Hmm.

 

[Kaz Kylheku]

I rarely pay any great attention to the writers identity.  It's
still undefined.

Unknown quantities named by symbols do not block thought.

(See mathematics and logic for instance).

 

[James Dow Allen]

He didn't say he couldn't, but that he couldn't be bothered.
Apparently he's decided it's beneath him.

But I like a puzzle. I'm thinking "spiral..."[/QUOTE]

Yes, with a nearby spiralling thread I'm surprised
only Dave got this. It would be interesting to
know why Bos disliked the code so much. (Omission
of comments was deliberate.) I don't generally pack
multiple statements onto one line, but did so
almost whimsically in the fragment. Avoiding
new lines seems not too egregious in simple idioms like
d = e, e = a, a = d; /* swap */

All other things being equal (though of course
they never are!) the succincter code is the better
code: one FOR is better than four FOR's. For this
reason, my spiraller seems OK.

James

 

[James Dow Allen]

The test looks for the presence of triplets of equally spaced
integers, where the spacing is H0.

Check again! First y finds pairs;
second y finds ... quadruplets!

In the original source code (from a famous benchmark)
each
if (R)
actually has the form
if ((R) != 0)
At the last moment before posting I deleted the
"unnecessary" ( ... != 0)
to reduce clutter. I was overzealous :-(

However, this set doesn't necessarily fit into the return type
(assuming that the uint64 typedef name, in good faith, represents some
kind of 64 bit or perhaps wider type). This is a bug; perhaps the
intent was: return (y & (y >> 2)) != 0;

Good catch, Kaz! I've been running this routine a lot
lately; but fortunately the real bug-free code, not
the zealously decluttered version I posted.

Thus the code checks for quadruplets aligned along
any of four axes. This could be a 4-D modelling but
in fact is a 2-D structure with diagonals, rows
and columns producing 4 axes. Application?
Hasbro's game Connect Four.

James

 

[Richard Bos]

Perhaps it was unwise to start with comment-less
"What does this do?" puzzles. Still, it seems
bizarre that responders complained about slight
obfuscation when the preface explained why
comments were removed!

Slight?

I suppose "fragment" isn't defined in the standard.

Oh, come off it. You're not Kenny.

[Fragment #2] Outputs all numbers N
from 0 <=3D N < b*c, in some shuffled order
that I can't be bothered to figure out.

No, it doesn't output; it visits with 'foo()'.
This lapse, and the fact that Bos got this
far, means he did indeed write the trivial
main() and foo() with the latter based on
printf(). Hard to believe no one else could
do this.

Caught. Actually, I did a dry run first, on paper, thought that it
probably called foo() with a shuffled list of 0..b*c-1, and decided to
let the computer do the busy-work of checking that for larger b and c.

I don't want to impugn Bos -- we all have our
limitations -- but for a programmer he must
be peculiarly poor at arithmetic for the
order to be unclear. And, frankly, a bit
of a sore loser to blame his own inadequacy
on my puzzle!

Get off your friggin' high horse, mister perfect-maths-wizard. You have
not the slightest idea how I tested your program wand why I could not be
arsed to do more of an effort. But since you asked for it, I'll tell you
now: I couldn't be arsed because it was crappy code, used in a crappy
puzzle, by a crappy programmer who should have known better than to
consider that crap even remotely acceptable. And in the future, I won't
even be making what little effort I have this time; your programming
inadequacy is not worth it.

Richard