Array append performance

[Walter Bright]

Lionello Lunesu wrote:
> 
> "Walter Bright" <newshound1 at digitalmars.com> wrote in message 
> news:g95mh3$2e5i$2 at digitalmars.com...
>> Essentially, I'm having a real hard time believing that for the 4 byte 
>> case, memcpy is faster. Take a look at the actual implementation of 
>> memcpy, vs the code generated by the compiler.
> 
> I agree with you, but I've noticed the byte[] = byte[] emits "rep movsb" 
> and that must be the slowest way there is to copy 4 bytes. In fact, from 
> my demo scene experience I've learned to not use those CISC instructions 
> like rep and movs* and loop. mov/dec/jnz always proved faster.

But for 4 bytes, how much slower could it be than memcpy, which executes 
a lot of setup code?


> However, I do think that it's a good idea to move the []=[] code into 
> the run-time library. That way specialized versions can be written, like 
> the other array operations. (Maybe it's already in the lib, but I 
> haven't found it. I've only seen mention of OPmemcpy in e2ir.c which I 
> assume represents the asm code for a memcpy.)

OPmemcpy is an operator supported by the compiler back end, it generates 
the movsb's.