AMD x86 manual MOV ECX, Maxsize

AMD Athlon™ Processor x86 Code Optimization

Example 1 (Avoid):

22007E/0 — November 1999

int a[MAXSIZE], b[MAXSIZE], c[MAXSIZE], i;

for (i=0; i < MAXSIZE; i++) { c [i] = a[i] + b[i];

}

MOV	ECX, MAXSIZE	;initialize	loop counter
XOR	ESI, ESI	;initialize	offset into array a
XOR	EDI, EDI	;initialize	offset into array b
XOR	EBX, EBX	;initialize	offset into array c
$add_loop:
MOV	EAX, [ESI + a]	;get element a
MOV	EDX, [EDI + b]	;get element b
ADD	EAX, EDX	;a[i] + b[i]
MOV	[EBX + c], EAX	;write result to c
ADD	ESI, 4	;increment offset into a
ADD	EDI, 4	;increment offset into b
ADD	EBX, 4	;increment offset into c
DEC	ECX	;decrement loop count
JNZ	$add_loop	;until loop	count 0

Example 2 (Preferred):

int a[MAXSIZE], b[MAXSIZE], c[MAXSIZE], i;

for (i=0; i < MAXSIZE; i++) { c [i] = a[i] + b[i];

}

MOV ECX, MAXSIZE-1 ;initialize loop counter

$add_loop:

MOV	EAX, [ECX*4 + a]
MOV	EDX, [ECX*4 + b]
ADD	EAX, EDX
MOV	[ECX*4 + c], EAX
DEC	ECX
JNS	$add_loop

;get element a ;get element b ;a[i] + b[i] ;write result to c ;decrement index ;until index negative

Note that the code in example 2 traverses the arrays in a downward direction (i.e., from higher addresses to lower addresses), whereas the original code in example 1 traverses the arrays in an upward direction. Such a change in the direction of the traversal is possible if each loop iteration is completely independent of all other loop iterations, as is the case here.

In code where the direction of the array traversal can’t be switched, it is still possible to minimize pointer arithmetic by appropriately biasing base addresses and using an index

74 Minimize Pointer Arithmetic in Loops