memmove optimization for std.algorithm.copy

[dlangBugzillaToGithub]

tommitissari reported this on 2013-06-18T05:24:34Z

Transfered from https://issues.dlang.org/show_bug.cgi?id=10402

Description

Add an optimization for:

Range2 copy(Range1, Range2)(Range1 source, Range2 target);

...to use the c function memmove when it is safe to do so and the two ranges overlap each other (when they don't overlap, a vectorized copy should be faster). As far as I can tell, it is safe to use memmove for copying source over target if all of the following are true:

1) Both ranges are arrays (static or dynamic)
2) Both arrays have the same element type
3) hasElaborateAssign!T is false for the element type T

I haven't benchmarked anything, but all the good C++ standard library implementations of std::copy seem to do this optimization, and I trust that they know what they are doing.

[dlangBugzillaToGithub]

tommitissari commented on 2013-06-18T06:03:11Z

(In reply to comment #0)
> Add an optimization for:
> 
> Range2 copy(Range1, Range2)(Range1 source, Range2 target);
> 
> ...to use the c function memmove when it is safe to do so and the two ranges
> overlap each other (when they don't overlap, a vectorized copy should be
> faster). As far as I can tell, it is safe to use memmove for copying source
> over target if all of the following are true:
> 
> 1) Both ranges are arrays (static or dynamic)
> 2) Both arrays have the same element type
> 3) hasElaborateAssign!T is false for the element type T
> 
> I haven't benchmarked anything, but all the good C++ standard library
> implementations of std::copy seem to do this optimization, and I trust that
> they know what they are doing.

Also, the documentation for std.algorithm.copy should be changed to indicate that the ranges are allowed to overlap if the above conditions hold.