Re: postgreSQL-8.0.1 compilation with icc-8.1 on Itanium-2 gives "error: asm statements not supported"

Tom, Peter,

I have been able to compile and sucessfully run pgSQL after replacing
the asm statement in postgresql-8.0.1/src/include/storage/s_lock.h
with an equivalent intrinsic for the Itanium platform-

------------------------------------------BEGIN OLD
s_lock.h----------------------------------------------------------
#if defined(__ia64__) || defined(__ia64) /* __ia64 used by ICC compiler? */
#define HAS_TEST_AND_SET
typedef unsigned int slock_t;
#define TAS(lock) tas(lock)

static __inline__ int
tas(volatile slock_t *lock)
{
long int ret;

__asm__ __volatile__(
" xchg4 %0=%1,%2 \n"
: "=r"(ret), "+m"(*lock)
: "r"(1)
: "memory");
return (int) ret;
}
#endif /* __ia64__ || __ia64 */
-----------------------------------------------END OLD
s_lock.h----------------------------------------------------------

------------------------------------------BEGIN NEW
s_lock.h----------------------------------------------------------
#if defined(__ia64__) || defined(__ia64) /* __ia64 used by ICC compiler? */
#define HAS_TEST_AND_SET
typedef unsigned int slock_t;
#define TAS(lock) tas(lock)

static __inline__ int
tas(volatile slock_t *lock)
{
int ret;

ret = _InterlockedExchange(lock,1);

return ret;
}
#endif /* __ia64__ || __ia64 */
----------------------------------------------END NEW
s_lock.h----------------------------------------------------------

The binary appears to be stable and the tpc-H benchmark executed
successfully against it as well. I also ran the regression test but
the following tests failed, the reasons for which I haven't
investigated yet
(http://www.cse.psu.edu/~kalsi/files/regression.diffs)-

test create_function_1 ... FAILED
test create_type ... FAILED
test create_table ... FAILED
test create_function_2 ... FAILED
test triggers ... FAILED
test create_operator ... FAILED
test create_view ... FAILED
test transactions ... FAILED
test misc ... FAILED
test select_views ... FAILED
test rules ... FAILED
test plpgsql ... failed (ignored)
test copy2 ... FAILED
test rangefuncs ... FAILED
test conversion ... FAILED
test stats ... FAILED

The _InterlockedExchange() function is defined in ia64intrin.h header file

int _InterlockedExchange(volatile int *Target, long value)
Do an exchange operation atomically. Maps to the xchg4 instruction.

More information is available at
http://www.intel.com/software/products/compilers/clin/docs/ug_cpp/lin1072.htm

Also, some other points to note, _ICC wasn't defined on my
installation when I was using icc by setting env var CC=icc. So, when
I tried to put a "#if defined" for using asm() for gcc and
_InterlockedExchange(), it didn't work. So, after this change gcc
compilation fails.

As of now, I am trying to test the binary further to see if it is
stable. Would you be knowing some good way to test this change?

I am not aware of the procedure of building patches but if this
resolves this issue and you would like me to make some sort of a
patch, then please let me know.

Thanks,
-Vikram

On Thu, 3 Mar 2005 09:55:18 +0100, Peter Eisentraut <peter_e(at)gmx(dot)net> wrote:
> Tom Lane wrote:
> > #if defined(__GNUC__) || defined(__ICC)
> >
> > Can anyone say a reason why the above #if is not wrong ... ie,
> > are there any platforms where icc does handle gcc asm syntax,
> > and if so exactly which ones are they?
>
> I believe I added that a few releases ago. The platform is IA32.
> Evidently, the GCC compatibility on IA64 is not quite as far yet.
>
> --
> Peter Eisentraut
> http://developer.postgresql.org/~petere/
>