Thread: gcc inline assembler - rounding mode

I am trying to write a function named roundD that rounds its first argument to an integer value according to the mode specified by its second argument.

I will be writing the function in assembly language, using gcc’s inline assembler. I do not want to use any predefined functions.. preferably intels syntax
please keep in mind i have studied for a week on how to accomplish this task.


I also think i need to restore the original value of the Rounding Control field before roundD returns?

I am trying to figure out how i should accomplish this in the right order..

Should i use the Control Word of the FPU?

1. Do i load the value into the FPU?
2. Use the control word - BIT FIELDS to do the calculations?

like: The RC field (bits 11 and 10) or Rounding Control determines how the FPU will round results.
which in my case are the enum values..


I am going off this page.. Simply FPU Chap.1

is this a good page to go from? What else do i need to google for information? Any help is much appreciated.. I think i just need a detailed analysis of the steps i need to take to accomplish this..

This is what i have so far.. This should give you a good idea on what i want to accomplish.

Code:

#include <stdio.h>
#include <stdlib.h>

#define PRECISION           3
#define RND_CTL_BIT_SHIFT   10

// floating point rounding modes: IA-32 Manual, Vol. 1, p. 4-20
typedef enum {
ROUND_NEAREST_EVEN =    0 << RND_CTL_BIT_SHIFT,
ROUND_MINUS_INF =       1 << RND_CTL_BIT_SHIFT,
ROUND_PLUS_INF =        2 << RND_CTL_BIT_SHIFT,
ROUND_TOWARD_ZERO =     3 << RND_CTL_BIT_SHIFT
} RoundingMode;

double roundD (double n, RoundingMode roundingMode)
{

return n;


}

int main (int argc, char **argv)
{
double  n = 0.0;

printf("Rounding - Assembly");
if (argc > 1)
    n = atof(argv[1]);

printf("roundD even %.*f = %.*f\n",
       PRECISION, n, PRECISION, roundD(n, ROUND_NEAREST_EVEN));
printf("roundD down %.*f = %.*f\n",
       PRECISION, n, PRECISION, roundD(n, ROUND_MINUS_INF));
printf("roundD up   %.*f = %.*f\n",
       PRECISION, n, PRECISION, roundD(n, ROUND_PLUS_INF));
printf("roundD zero %.*f = %.*f\n",
       PRECISION, n, PRECISION, roundD(n, ROUND_TOWARD_ZERO));

return 0;
}

I understand that :
The RC field (bits 11 and 10) or Rounding Control determines how the FPU will round results in one of four ways:

00 = Round to nearest, or to even if equidistant (this is the initialized state)
01 = Round down (toward -infinity)
10 = Round up (toward +infinity)
11 = Truncate (toward 0)

so, my enum should serve the same purpose?


Do i need to do something like this?

Code:

push %ebp;
movl %ebp, %eax;
mov1 %eax[n- whatever enum rounding mode];    // double  n = 0.0;
fstcw ax;

I think i am confused on how to set the Rounding Control Field of the FPU control word accordance
with the second argument in the function..

Originally Posted by mrsirpoopsalot

I think i just need a detailed analysis of the steps i need to take to accomplish this.

I've had that thought a few times myself

This is a neat idea, if you are doing this out of an interest in assembly, but I dunno if that will really be advantageous, considering that rounding a number is a pretty simple, straightforward task that probably does not leave tons of room for optimization.

Speaking of which, if that is what you are after, I'd have two functions (roundDown, roundUp) with one parameter rather than one function with an extra "mode" parameter. That will save you the overhead of evaluating a conditional every time (since I presume your goal here is something that's going to get called thousands of times a second, so you want it as tight as possible).

Fiddling with the FPU control word goes against your goal of being fast. Reprogramming the FPU rounding mode has serious, negative consequences, like flushing the floating point pipeline. If you have to read or write the control word, you may end up anywhere from tens to hundreds of times slower than if you hadn't.

I was hoping someone could help me instead of telling me i should do it a different way.
I was planning on using a switch statement to handle the different enum values being passed in.

I am not really worried about optimization or being fast.
I would like to keep it the way i have it. I am trying to learn assembly.
is there a good book that could help me?
thank you

do i need to do something like

int num;

fld n ; //Push n' into st(0) of FPU stack?
fist num ; //Convert and store st(0) to integer?
fild num ; //Push integer to st(0) ?
fnstsw %ax;
mov %eax,[n - enum value];
fstcw %ax;

Originally Posted by mrsirpoopsalot

I was hoping someone could help me instead of telling me i should do it a different way.
I was planning on using a switch statement to handle the different enum values being passed in.

I am not really worried about optimization or being fast.
I would like to keep it the way i have it. I am trying to learn assembly.
is there a good book that could help me?
thank you

Seek out either Peter Nortons Guide to Intel Assembly Language or there are free Linux-based resources on the web for assembly....this is a C/C++ forum and while you are welcome to ask about your assy code, most are not assy coders that would respond which is why I recommend a reference volume..

Originally Posted by mrsirpoopsalot

I thought since its a mixture of C / assembly it whould be ok. Besides, dont c/c++ programmers need to know assembly to get their degrees?

Thank you for the book recommendation..

That implies you need a degree to be a C/C++ programmer. Assembly has nothing to do with either the C or C++ language, that's purely implementation specific.

Originally Posted by mrsirpoopsalot

I thought since its a mixture of C / assembly it whould be ok. Besides, dont c/c++ programmers need to know assembly to get their degrees?

Thank you for the book recommendation..

Maybe I don't know anymore. I picked up assembly back in 83 or so on the 6502...got to Intel circa 89 I think, realized it stunk on ice and never bothered with it again. It you are into it, that's your cross to bear...

Most degrees these days also seem to require some Java but I don't how far you would get asking about C/Java (JNI) here either...

Sorry, I assumed that you were using assembly for some practical purpose, not just trying to learn.

You actually have to load the current control word, flip the bits 10 and 11 to what you want, then store it back again. If you just store a word into the CR which has those bits sets, you'll toggle a bunch of other bits that you didn't intend to change.

Originally Posted by mrsirpoopsalot

I ordered the book.

Hooray! Good move.

I dunno any asm tho, so I can't comment further.