方法1
mov eax, 0mov ebx, 0mov ecx, 0mov edx, 0
方法2
xor eax, eaxxor edx, edxxor ecx, ecxxor edx, ecx
方法3
hmm...yep, here the stall is partially balanced by the non-change of the source argument (a work/branch avoided by the cpu), so in this specific case, the difference is minimised.
xor eax, eaxxor ecx, ecxmov ebx, eaxmov edx, ecx
technically the "result" stall is slower (but never enough to put a useless instruction ! even a nop !), there is also possible stall with the source argument (when r+i/r), but not here. so you don't need to change the register in the last line, you will obtain the benefit of a non-changed source (there is no dependency/alteration here) :
xor eax, eaxxor ecx, ecxmov ebx, eaxmov edx, eax
方法4
include /masm32/include/masm32rt.inc
.data?Null8 dd 8 dup (?)
.code
start: mov Null8[12], esp ; save stack pointer in a global variable mov esp, offset Null8 ; put stack pointer in front of 8 nullwords popad mov esp, Null8[12] ; restore stack pointer
getkey exit ; short form of invoke ExitProcess, 0
end start
方法5
对于AMD的CPU
mov al,99cpuid
说明：1、对于Intel Celeron M，ebx，ecx, edx会被置0
2.On Intel 64 processors, CPUID clears the high 32 bits of the RAX/RBX/RCX/RDX registers in all modes.This was info from: Intel 3-180 Vol. 2A CPUID-CPU Identification.I think this translate to the high 32 bits of said register are trashed,over-written with zeros.
方法6
fldz.xmmprefetchnta qword ptr es:[esp-8]push csverw word ptr es:[esp]pop ecxfistp qword ptr es:[esp-8]pauselock cmpxchg8b qword ptr es:[esp-8]cmovne ecx,eaxcmovnz ebx,ecx
方法7
sub esp,32mov edi,espxor eax,eaxmov ecx,8rep stosdpopad 方法8
push 32pop ecxsub esp, ecxmov edi, espxor eax, eaxrep stosbpopad
方法9
.data zer dd 0,0,0,0,0,0,0,0, zer
.codexchg esp, [zer+8*4]popadpop esp
00401004 87 25 E4 E4 42 00 xchg esp, dword ptr ds:[42E4E4h] 0040100A 61 popad 0040100B 5C pop esp ; just 8 bytes
测试代码：
Code:; ««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««« include /masm32/include/masm32rt.inc .686 include /masm32/macros/timers.asm; ««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««« .data .code; «««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««start:; ««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««« invoke Sleep, 4000
counter_begin 1000, HIGH_PRIORITY_CLASS REPEAT 100 xor eax, eax xor ecx, ecx mov ebx, eax mov edx, ecx ENDM counter_end print ustr$(eax),13,10
counter_begin 1000, HIGH_PRIORITY_CLASS REPEAT 100 xor eax, eax xor ebx, eax mov ecx, eax mov edx, eax ENDM counter_end print ustr$(eax),13,10,13,10
counter_begin 1000, HIGH_PRIORITY_CLASS REPEAT 100 add eax, 1 add ecx, 1 mov ebx, eax mov edx, ecx ENDM counter_end print ustr$(eax),13,10
counter_begin 1000, HIGH_PRIORITY_CLASS REPEAT 100 add eax, 1 xor ebx, eax mov ecx, eax mov edx, eax ENDM counter_end print ustr$(eax),13,10,13,10
inkey "Press any key to exit..." exit
; «««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««««end startRunning on my P3, I would expect to see a much larger effect from 100 or more stalls:
206208
207205
以上源自MASM32官方论坛的精彩讨论贴：http://www.masm32.com/board/index.php?topic=11138.15
