code timing , whats wrong ?

[kmtdk]

well
i have tryed to time my code, using the RDTSC instrcution
but each time i use the instruction, Extreamly many cycles have passed since the last time,
and i dont have any code in betewen

the source is attached ( it is compiled as a "com" program, i have tryed on os, no change)
the time keeps changeing, but the difference is far too high.

Kmt dk

[Combuster]

This looks wrong:

Code: Select all

mov [si],eax
mov [si+2],edx

[kmtdk]

the code stores the 64 bit integer at the variable ( i use si to point)
but have i done it reverse , so edx shuld be the first , and then eax.??


KMt dk

[djmauretto]

Did you mean:

Code: Select all

mov [si],eax
mov [si+4],edx

[Masterkiller]

it returns EDX:EAX, so EAX store less-significant bytes and they are 32-bit = 4 bytes, you store them between two bytes so rewrite the EAX. In the memory less-significant bytes are stored in lower addresses.

Edit: Lower address is the address that you do not add offset (since the offset is positive). EAX is less-significant, so it should be stored at lower address e.g. [si+0], not [si+4]

[kmtdk]

* DOH
i forgot i was using 32 bit instructions, so that was one Big error.( the reason is i have worked on 16 bit too much ... )
so the code shuld look like this:

Code: Select all

mov [si],edx
mov [si+4],eax

I will just test it ( and if i dont post/ edit, it means that i dont have tryed it..)

KMT dk

[djmauretto]

* DOH
i forgot i was using 32 bit instructions, so that was one Big error.( the reason is i have worked on 16 bit too much ... )
so the code shuld look like this:

Code: Select all

mov [si],edx
mov [si+4],eax

I will just test it ( and if i dont post/ edit, it means that i dont have tryed it..)

KMT dk

Code: Select all

mov [si],eax
mov [si+4],edx

[kmtdk]

Well
now i got it to work, but one thing shuld be told
this does not work 100 % under windows, but it does ( it counts wierd), but you can see a different
the code is attached , and use it as you want, but if you want to publise the code, ask for permission.
bye the way, the code is Not optimised .

happy timing ,:P

KMT dk

[Combuster]

I expect the TSC to jump large amounts when the OS decides to yield the CPU to another process. Also, unsynchronized counters can cause strange effects (going back in time, anyone?) when windows tosses the application between cores (windows' CPU affinity sucks big time, try forcing applications onto a single core and see whether it works better)