What kind of jump to a fixed address would be faster on most processors?

or


I was expecting second to be faster, but for some reason first seems to be at least ~5% faster.

Also, sometimes the part with first runs twice as fast, which never happens if i use the second one. Branch prediction/cache stuff?

jmp A is faster, i don't see why 2 instructions should be faster than just one.
even if push/ret is HW optimized, jmp too is optimized, then, jmp is faster

but if you want to make some code obfuscation, push, code, ret is a good solution

Either one could be faster, depending on the surrounding context...

i gues jmp address takes more bytes to encode...plus it may not be aligned.

Artlav, would be possible for you to share your speed testing code? Besides the fact PUSH/RET is harder for the processor to predict the target (if not impossible), you spoil the processor's return address buffer because the RET isn't paired with a CALL.

I think a more appropriate answer would be: If you can't measure any perceivable difference in your code then is does not matter which code sequence you use.

push A
ret

whould be not a bad trick for obfuscating the code for some noob reversers.

Hey! What do i have to do with it? Leave me alone!

According to "some" manuals that I have read "some years ago", call & ret are meant to work together. If you got one "call" instruction and then "ret" following the call (some instructions are allowed in between) then CPU will cache call+ret pair.

So if we are inside function that was called and executing some other functions:

Then "call" of the function that we are inside in will be paired with the "ret" belonging to a different function.

To which extend this can affect performance I don't know.

After I have typed this I can see that precise logic is way too complicated.