https://issues.dlang.org/show_bug.cgi?id=18899

Manu <turkeyman@gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |turkeyman@gmail.com

--- Comment #1 from Manu <turkeyman@gmail.com> ---
So, memcpy... as opposed to primitive assignment?

--

https://issues.dlang.org/show_bug.cgi?id=18899

Mike Franklin <slavo5150@yahoo.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |slavo5150@yahoo.com

--- Comment #2 from Mike Franklin <slavo5150@yahoo.com> ---
> dest[] = src[]

I'm pretty sure this gets lowered to memcpy by the compiler, but I haven't verified yet.

--

https://issues.dlang.org/show_bug.cgi?id=18899

--- Comment #3 from Mike Franklin <slavo5150@yahoo.com> ---
Yes, `dest[] = src[]` gets lowered to _d_arraycopy, which ultimately calls memcpy

View ASM here:  https://run.dlang.io/is/izCLp0
_d_arraycopy implementation here:
https://github.com/dlang/druntime/blob/2db828bd4f21807254b770b3ec304f14596a9805/src/rt/arraycat.d#L22-L29

--

https://issues.dlang.org/show_bug.cgi?id=18899

--- Comment #4 from Manu <turkeyman@gmail.com> ---
Yes, that's what I'm saying :) ... you're not happy with memcpy, do want element-copy?

--

https://issues.dlang.org/show_bug.cgi?id=18899

--- Comment #5 from Steven Schveighoffer <schveiguy@yahoo.com> ---
(In reply to Manu from comment #4)
> Yes, that's what I'm saying :) ... you're not happy with memcpy, do want element-copy?

It's more that I want to copy the one int that is in the struct than call an opaque function, no matter how great that function is implemented.

--

https://issues.dlang.org/show_bug.cgi?id=18899

--- Comment #6 from Manu <turkeyman@gmail.com> ---
memcpy should be an intrinsic, which is implemented using magic...

--

https://issues.dlang.org/show_bug.cgi?id=18899

--- Comment #7 from Steven Schveighoffer <schveiguy@yahoo.com> ---
I'm not sure that it is.

But we aren't calling memcpy anyway, we are calling _d_arraycopy, not inlined. See the generated AST from Mike's example.

--

https://issues.dlang.org/show_bug.cgi?id=18899

--- Comment #8 from Steven Schveighoffer <schveiguy@yahoo.com> ---
(In reply to Steven Schveighoffer from comment #7)
> See the generated AST

...generated *assembly*

--

https://issues.dlang.org/show_bug.cgi?id=18899

--- Comment #9 from Manu <turkeyman@gmail.com> ---
True.
Anyway, I'm just playing devils advocate. I agree, it should do an element copy
for small structs.

--

https://issues.dlang.org/show_bug.cgi?id=18899

Nick Treleaven <nick@geany.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |nick@geany.org

--- Comment #10 from Nick Treleaven <nick@geany.org> ---
For the record, Andrei wants to solve this by not blitting init for primitive
types:
https://github.com/dlang/druntime/pull/2126

> memcpy should be an intrinsic

Would that allow unrolling loops when the length is known at compile-time?

An interim improvement might be to have an array version of std.algorithm.copy in druntime (which phobos then uses internally).

--