Hi everyone!


  I have an unsigned long long int (64bit) mac_category which is set to 0.
Now I have to set the i-th bit of mac_category to 1 (0<i<63). How to do
that using inline-assembly of linux gcc ?

for example, set the 59th bit of mac_category to 1:

....
....

  int i=59;
  unsigned long long int mac_category = 0;   // 64bit

  /* I have to use AT&T format assembly for gcc */

  asm (

	/* define a 64bit-long variable mydq and set it to zero */
	"mydq DQ, 0;"		

	/*
	 * do the shifting, %1 stands for the input operand,
	 * which is i in the c program;
	 */
	"shl %1, mydq;"

	/*
	 * copy the value of mydq to %0, the output operand,
	 * which is mac_category in the c program;
	 */
 	"mov mydq, %0;"

 	/* define the input and output operand */
	:"=r"(mac_category)
	:"r"(i)

	);

....
....


the above inline assembly code does NOT work, I don't know why.
(It seems that you cannot define any variables in inline-assembly of
linux programs, right?)

How to solve this problem, please help me!

Thank you!

X.H.
	

"Hao Xu" <spamtrap@crayne.org> wrote in message
news:ccd16u$1vkq$1@mail.cn99.com...
> Hi everyone!
>
>
>   I have an unsigned long long int (64bit) mac_category which is set to 0.
> Now I have to set the i-th bit of mac_category to 1 (0<i<63). How to do
> that using inline-assembly of linux gcc ?
>
> for example, set the 59th bit of mac_category to 1:
<snip>

First off, the assembly code you posted was not AT&T. It was masm assembly.
Second off, you can't define variables in the middle of your code in any
assembler -- the variable will be executed as code! Third, shl does not work
how you think it does.

Try this:

asm("shll %%cl, %0\n"
    "cmpl $32, %%cl\n"
    "cmovll %0, %1\n"
    "xorl %1, %0\n"
    "orl %0, %%edx\n"
    "orl %1, %%eax"
    : "A" (mac_category)
    : "r" (1), "r" (0), "c" (i));

I don't recall whether or not the first parameter becomes %0 even though
it's not gcc-allocated, so you may have to change %0 -> %1 and %1 -> %2.

-Matt

Hao Xu wrote:

> I have an unsigned long long int (64bit) mac_category which is set to 0.
> Now I have to set the i-th bit of mac_category to 1 (0<i<63). How to do
> that using inline-assembly of linux gcc ?
> 
> for example, set the 59th bit of mac_category to 1:
> 
> ...
> ...
> 
>   int i=59;
>   unsigned long long int mac_category = 0;   // 64bit

Why use assembly when C works fine?

#include <stdio.h>

int i = 59;

int main()
{
   unsigned long long int mac_category = 1ULL << i;
   printf("mac_category=%#llx\n", mac_category);
   return 0;
}

$ gcc -std=c99 -Wall -pedantic -O3 -S test.c

main:
         pushl   %ebp
         movl    %esp, %ebp
         subl    $8, %esp
         movl    i, %ecx
         movl    $1, %eax
         xorl    %edx, %edx
         shldl   %eax, %edx
         andl    $-16, %esp
         sall    %cl, %eax
         andl    $32, %ecx
         je      .L2
         movl    %eax, %edx
         xorl    %eax, %eax
..L2:
         pushl   %ecx
         pushl   %edx
         pushl   %eax
         pushl   $.LC0
         call    printf
         xorl    %eax, %eax
         leave
         ret

Hao Xu asked:

[about BIT-SET]
| for example, set the 59th bit of mac_category to 1:

I don't know AT&T syntax, but this (IA-docs form)

b8 3b 00 00 00          MOV eax,3Bh       ;dec 59
0f ab 05 mm mm mm mm    BTS [memptr],EAX  ;first test, then set bit
                        ...               ;carry-flag is previous bit-state

seems to be what you're searching for.

Even this isn't documented that way, it works for bit-offsets
in the range +-2^31 (if not out of limits) on my AMD K7 and on Celeron.

__
wolfgang

"wolfgang kern" <nowhere@nevernet.at> wrote in message
news:cch6gn$j42$1@newsreader1.utanet.at...
>
> Hao Xu asked:
>
> [about BIT-SET]
> | for example, set the 59th bit of mac_category to 1:
>
> I don't know AT&T syntax, but this (IA-docs form)
>
> b8 3b 00 00 00          MOV eax,3Bh       ;dec 59
> 0f ab 05 mm mm mm mm    BTS [memptr],EAX  ;first test, then set bit
>                         ...               ;carry-flag is previous
bit-state
>
> seems to be what you're searching for.
>
> Even this isn't documented that way, it works for bit-offsets
> in the range +-2^31 (if not out of limits) on my AMD K7 and on Celeron.

This behavior is documented. However, it's slow -- 9 cycles on K7.

-Matt

Hi Matt,

[about BIT-SET]
| > Even this isn't documented that way, it works for bit-offsets
| > in the range +-2^31 (if not out of limits) on my AMD K7 and on Celeron.
|
| This behavior is documented. However, it's slow -- 9 cycles on K7.

Must have missed this documentation, I saw only the 32-bit limits yet.
Nine cycles are somehow faster than the code Hao Xu's posted ;)
Yes, it can be done faster with (shifted) indexed offset and OR also,
with a few bytes more and one more register involved.

__
wolfgang

Oh, for reference, CC, I don't wan't my address spamtrapped.
(I also don't want date that my system put in the headers when 
I first shoved this message into the outside world overwritten.)


"Matt Taylor"  <spamtrap@crayne.org> writes:
> > b8 3b 00 00 00          MOV eax,3Bh       ;dec 59
> > 0f ab 05 mm mm mm mm    BTS [memptr],EAX  ;first test, then set bit
> >                         ...               ;carry-flag is previous
> bit-state

Some needs to fix their word-wrap behaviour (OE Quotefix?)

> >
> > seems to be what you're searching for.
> >
> > Even this isn't documented that way, it works for bit-offsets
> > in the range +-2^31 (if not out of limits) on my AMD K7 and on Celeron.
> 
> This behavior is documented. However, it's slow -- 9 cycles on K7.

As long as the result of the test part of the operation are not used
then surely this 9 cycles will be effectively almost cost free on an
aggressively OOO processor? Basically just invoke the write and forget 
about it?

Phil
-- 
1st bug in MS win2k source code found after 20 minutes: scanline.cpp
2nd and 3rd bug found after 10 more minutes: gethost.c
Both non-exploitable. (The 2nd/3rd ones might be, depending on the CRTL)

On Thu, 8 Jul 2004 10:01:29 +0000 (UTC)
Phil Carmody  <spamtrap@crayne.org> wrote:

:Oh, for reference, CC, I don't wan't my address spamtrapped.
:(I also don't want date that my system put in the headers when 
:I first shoved this message into the outside world overwritten.)

I have added "thefatphil_demunged@yahoo.co.uk" to the no spamtrap list,
but, unfortunately, I don't know how to fix the posting date problem.

-- Chuck