"Salt_Peter " <pj*****@yahoo. comwrote in message
news:11******** *************@m 73g2000cwd.goog legroups.com...
>
Ganesh wrote:
>Hi,
This is a question that pertains to pointers in general (C or C++).
Which of the following is faster and why?
for (int i = 0; i < N; i++)
= ... a[i]...
(or)
for (int i = 0; i < N; i++)
= ...*a++....
'a' is a dynamically allocated array of any valid scalar type and of
size N. I have read somewhere that the latter is faster. But could not
understand why? Or is that dependent on architecture or compiler?
Probably because the first implies a cast.
Ironically, you'ld probably get "faster" with:
for( size_t i = 0; i < N; ++i)
{
// your choice
}
Huh? The "int" type is the "native" integral type on any machine. Why
would there be a cast using "int i" to index a dynamically allocated array
(or any array for that matter)? The question wasn't about vectors, but
about raw pointers and arrays. I'd think that a "conversion " from size_t to
int would be needed if using your code and indexing the array with it. But
I see no cast implied or required anywhere in the OP's code.
The "usual" assumption is that the latter method (using pointers and
incrementing) is faster, on the assumption that indexing requires
multiplying the index by the size of an element, and adding that as an
offset to the array start. The assumption is false. That's one way to
implement indexing, but compilers are not required to implement it that way.
Indeed, a decent optimizing compiler is likely to take advantage of whatever
native machine code would do the job best, whether that's by using code more
like the pointer version, or by pre-loading some registers and calling a
built-in machine instruction which handles loops with one simple command.
Or any other solution they see fit to use.
Compiler makers undoubtedly profile their solutions. You should do the same
if you want to know whether you need to change your code, and whether a
proposed solution is a good one for your particular platform.
-Howard
