Skip to main content
gamedev.net
🔒 Locked

memcpy

Started by Doomrat 2 replies 250+ views
Original Post
Doomrat
Doomrat
Is this function usually as fast as you could hope for without writing a custom asm routine?
Alimonster
Alimonster
Well, here's a test program:

  #include <cstring>

int main()
{
	int a[40];
	int b[40];
	memset(a, 0, sizeof(a));
	memset(b, 0, sizeof(b));

	memcpy(a, b, sizeof(a));

	for (int index = 0; index < 40; ++index)
	{
		if (a[index] != b[index])
			return - 1;
	}

	return 0;
}

(Note that the for loop is there so the entire main function isn't snipped out while optimising(!)):
	TITLE	C:\My Documents\Programs\C++\testBits\main.cpp
	.386P
include listing.inc
if @Version gt 510
.model FLAT
else
_TEXT	SEGMENT PARA USE32 PUBLIC 'CODE'
_TEXT	ENDS
_DATA	SEGMENT DWORD USE32 PUBLIC 'DATA'
_DATA	ENDS
CONST	SEGMENT DWORD USE32 PUBLIC 'CONST'
CONST	ENDS
_BSS	SEGMENT DWORD USE32 PUBLIC 'BSS'
_BSS	ENDS
_TLS	SEGMENT DWORD USE32 PUBLIC 'TLS'
_TLS	ENDS
;	COMDAT _main
_TEXT	SEGMENT PARA USE32 PUBLIC 'CODE'
_TEXT	ENDS
FLAT	GROUP _DATA, CONST, _BSS
	ASSUME	CS: FLAT, DS: FLAT, SS: FLAT
endif
PUBLIC	_main
;	COMDAT _main
_TEXT	SEGMENT
_a$ = -160
_b$ = -320
_main	PROC NEAR					; COMDAT

; 3    : {

	sub	esp, 320				; 00000140H

; 4    : 	int a[40];
; 5    : 	int b[40];
; 6    : 	memset(a, 0, sizeof(a));
; 7    : 	memset(b, 0, sizeof(b));

	mov	ecx, 40					; 00000028H
	xor	eax, eax
	push	esi
	push	edi
	lea	edi, DWORD PTR _b$[esp+328]

; 8    : 
; 9    : 	memcpy(a, b, sizeof(a));

	lea	esi, DWORD PTR _b$[esp+328]
	rep stosd
	mov	ecx, 40					; 00000028H
	lea	edi, DWORD PTR _a$[esp+328]
	rep movsd
$L485:

; 10   : 
; 11   : 	for (int index = 0; index < 40; ++index)
; 12   : 	{
; 13   : 		if (a[index] != b[index])

	mov	ecx, DWORD PTR _a$[esp+eax+328]
	mov	esi, DWORD PTR _b$[esp+eax+328]
	cmp	ecx, esi
	jne	SHORT $L494
	add	eax, 4
	cmp	eax, 160				; 000000a0H
	jl	SHORT $L485
	pop	edi

; 15   : 	}
; 16   : 
; 17   : 	return 0;

	xor	eax, eax
	pop	esi

; 18   : }

	add	esp, 320				; 00000140H
	ret	0
$L494:
	pop	edi

; 14   : 			return - 1;

	or	eax, -1
	pop	esi

; 18   : }

	add	esp, 320				; 00000140H
	ret	0
_main	ENDP
_TEXT	ENDS
END

That's the generated asm in the default Release mode settings. I'm sure it means something to someone...

It seems that rep stosd is being used. I remember reading somewhere that newer processors prefer simpler instructions instead, but stosd will probably perform fine here. I'm not an assembly guru, though, so that info may be somehow wrong...

[edited by - Alimonster on May 26, 2002 6:57:34 AM]
Dean Harding
Dean Harding
memcpy is usually implemented in assembly anyway (at least for MSVC++ it is) and it''s probably as fast as you''ll get it (copying in DWORDs wherever possible, etc).

Besides, I don''t think memcpy is your bottleneck (unless you can prove it to me with a profiler).

codeka.com - Just click it.
Doomrat
Doomrat
Thanks guys, that''s exactly what I needed to know.

Topic Locked

This topic has been locked by a moderator. New replies are not allowed.

Sign in to reply to this topic.