Testing an odd number

[dedndave]

MUL is pretty fast on newer processors
and - this code has a lot of dependancies

Code Select Expand

        mov     ebx,eax
        shl     eax,2
        add     eax,ebx
        shl     eax,1
not to mention code bytes
in a loop, MUL might be the better choice

also interesting....
on a P4, MUL wants the register multiplier, EAX, and EDX "settled"
it's a good idea to perform some unrelated instruction just before the MUL
IMUL is not so sensitive - it even seems to like it   :P
also probably not true on newer cores

[bomz]

mul more universal. it multiply not only eax but put result to edx. also it mul not only 10. but in many cases we don't need mul edx and it possible simplify code

[bomz]

my code is not very correct. command CLD and STD get 48 ticks. PUSHF POPF too more than 50. may be thats true . of course CLD don't need so many ticks, but call some process.

[dedndave]

yah - those instructions are slow
i think the OS has to check for priviledge level changes or something for POPFD   :P
for the direction flag, it must keep track of the current state because API's expect it to be cleared
PUSHFD isn't too bad, because it does not alter flags
but CLD, STD, and POPFD are slow - about 100 cycles on my machine
CLC and STC are fast, at least

[bomz]

Code Select Expand


;std
;stosw
;cld

;mov word ptr[edi], ax
;sub edi, 2

stosw
sub edi, 4


[dedndave]

 :bg

Code Select Expand

        mov     [edi],ax

i suspect which is fastest depends on which generation of CPU it's running on

[bomz]

need mov     [edi], ax

[bomz]

I think STOS use special line

[bomz]

word ptr - seems no differences Pentium 4

STOS slowly

[bomz]

strange that all "combine" directives slowly. because than cpu execute code it never now the next command....may be my cpu very old

NEXT:

loop NEXT

slowly than 25%

NEXT:
sub ecx, 1
cmp ecx, 0
jnz NEXT

inc eax

slowly than

add eax, 1

in my code was interesting situation lea quickly than sub

scasb lodsb movsb stosb cmpsb - slow

[dedndave]

Code Select Expand

NEXT:
sub ecx, 1
cmp ecx, 0
jnz NEXT

no need to CMP
SUB will set or clear the zero flag   :P

Code Select Expand

NEXT:
sub ecx, 1
jnz NEXT

DEC works, too - and is a single-byte instruction for dword general registers

Code Select Expand

NEXT:
dec ecx
jnz NEXT

[bomz]

with cmp do the same loop