* Mathieu Desnoyers (mathieu.desnoyers@polymtl.ca) wrote:

quoted

What I generally dislike about this patch is that several fairly large
functions have been duplicated with only little change.

Yeah, I know. Until we find some way to share atomic operation code for
both operation on local and shared data, we have to duplicate this. We
could think about a header that would support multiple inclusion and
behave differently (different function prefix and LOCKing/memory
barriers) depending on defines set by the top level header.

Something like

asm/atomic.h
  #define ATOMIC_SHARED
  #include <asm/atomic-ops.h>  /* shared */
  #undef ATOMIC_SHARED
  #include <asm/atomic-ops.h>  /* local */

asm/atomic-ops.h 
  #ifdef ATOMIC_SHARED
  #define ATOMIC_PREFIX atomic
  #define ATOMIC_BARRIER() smp_mb()
  #define ATOMIC_TYPE atomic_t
  #define ATOMIC_VAR (v->counter)
  #else
  #define ATOMIC_PREFIX local
  #define ATOMIC_BARRIER()
  #define ATOMIC_TYPE local_t
  #define ATOMIC_VAR (v->a.counter)
  #endif
  
  static __inline__ ATOMIC_PREFIX##_add_return(int i, ATOMIC_TYPE *v)
  .....
  #undef ATOMIC_PREFIX
  #undef ATOMIC_BARRIER
  #undef ATOMIC_TYPE
  #undef ATOMIC_VAR

More prefisely, for this to work, we should change :
#define ATOMIC_PREFIX atomic
and
#define ATOMIC_PREFIX local
for

#define ATOMICF(fctname) atomic_##fctname
and
#define ATOMICF(fctname) local_##fctname


which would make de declaration look like :

static __inline__ ATOMICF(add_return)(int i, ATOMIC_TYPE *v)

Which starts to look a little odd to me, doesn't it ?

Regards,

Mathieu

-- 
OpenPGP public key:              http://krystal.dyndns.org:8080/key/compudj.gpg
Key fingerprint:     8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68