Very Computer

I have a simple question about the PUSH instruction on 386.
It is used to put values on the stack especially for passing
parameters in func call or saving registers. However, the
same effect can be achieved using the MOV instruction.
My question is: which instruction is faster.
It seems to me that PUSH has to push the value and decrement the stack
pointer, my guess it is slower(or at most as fast as MOV).
If my guess is true, I will try to modify the compiler to use the MOV.

I should have posted this to some assembly group, but I think someone
reading these groups I  am posting to can help me.

Thanks,
Dongxiao

To solve you problem have a better look at the Intel-docs and do not
forget to proof your substitution for really being faster than the
original way.

And please do not send such questions to comp.lang.c because they are
OFF-TOPIC

H.F. Steffani

Quote:>Thanks,
>Dongxiao

Quote:>I have a simple question about the PUSH instruction on 386.
>It is used to put values on the stack especially for passing
>parameters in func call or saving registers. However, the
>same effect can be achieved using the MOV instruction.
>My question is: which instruction is faster.
>It seems to me that PUSH has to push the value and decrement the stack
>pointer, my guess it is slower(or at most as fast as MOV).
>If my guess is true, I will try to modify the compiler to use the MOV.

>I should have posted this to some assembly group, but I think someone

Quote:>reading these groups I  am posting to can help me.

A true believer in Emily Postnews' advices :-)

I'm pretty sure someone reading alt.sex could help you.  Have you tried
there?

Dan
--
Dan Pop                       | The only reason God was able to make the
CERN, CN Division             | world in 7 days was he didn't have to remain

Mail:  CERN - PPE, Bat. 31 R-004, CH-1211 Geneve 23, Switzerland

[ Distribution of message limited to col.misc ]

: I have a simple question about the PUSH instruction on 386.
: It is used to put values on the stack especially for passing
: parameters in func call or saving registers. However, the
: same effect can be achieved using the MOV instruction.
: My question is: which instruction is faster.
: It seems to me that PUSH has to push the value and decrement the stack
: pointer, my guess it is slower(or at most as fast as MOV).
: If my guess is true, I will try to modify the compiler to use the MOV.

I assume you're referring to using:
         mov ax, [sp]   ; bad format - meant to put ax at location pointed
                        ; to by SP - my 386 assembly coding is fading :^)
as opposed to:
         push ax

On a 386, the former takes 4 clocks while the latter takes 2. Don't forget
for proper simulation of a push you should be adjusting SP, which is another
2 clocks. Either way you're still in a worse case than push. There's really
no reason to do a push manually if you don't have to.

Not to flame, but it's likely that if it were possible to make compiled
programs run better with such a trivial change then the compiler would
probably already do it :)

Matt

Quote:>Thanks,
>Dongxiao

: I have a simple question about the PUSH instruction on 386.
: It is used to put values on the stack especially for passing
: parameters in func call or saving registers. However, the
: same effect can be achieved using the MOV instruction.
: My question is: which instruction is faster.
: It seems to me that PUSH has to push the value and decrement the stack
: pointer, my guess it is slower(or at most as fast as MOV).
: If my guess is true, I will try to modify the compiler to use the MOV.

: I should have posted this to some assembly group, but I think someone
: reading these groups I  am posting to can help me.

: Thanks,
: Dongxiao
MOV is A LOT faster the push, As I beleave (Not sure thought) push is 14
cycles and MOV is I think 3 and Inc is 2 So even MOV + INC is faster then
PUSH, In case you are wondering PUSHA is I think 24 or something like that.

  "Possiblity for the C compiler to optimize C function calls"
  Subject: Can we make a C fucntion call  faster on  
           a CISC architecture using load/save operations?

Quote:>>I should have posted this to some assembly group, but I think someone
>>reading these groups I  am posting to can help me.
>First this group is absolutl wrong for this kind of questions.
>However you have to keep in mind that push/pop handles the stack-
>pointer. I am not sure that there is a instruction to increment/decrement
>the stack-pointer. But this is necessary for stacks! If there is

Quote:>such a instruction, I am not sure if the needed two instructions, the mov
>and the stack-ptr-instruction together as fast as PUSH/POP.
>Assuming that PUSH/POP, MOV and this anknown stack-ptr-instruction are
>as fast as possible your solution will be more slowly as one more
>decode-cycle is necessary.

Quote:>And please do not send such questions to comp.lang.c because they are
>OFF-TOPIC

  Thanks for all the help,
  Dongxiao

So, please go to alt.sex :-)

Dan
--
Dan Pop                       | The only reason God was able to make the
CERN, CN Division             | world in 7 days was he didn't have to remain

Mail:  CERN - PPE, Bat. 31 R-004, CH-1211 Geneve 23, Switzerland

Sorry, thank you for being our contestant.

-Dave
--
***   ***   ***   ***   ***   ***   ***   ***   ***   ***   ***   ***   ***
  *     *     *     *     *     *     *     *     *     *     *     *     *
 *     *     *     *     *     *     *     *     *     *     *     *     *