[dpdk-dev] [EXT] [PATCH v3 3/3] eal/stack: enable lock-free stack for aarch64

[Jerin Jacob Kollanukkaran]

> > Can we remove this all #ifdef clutter by adding the following
> >
> > $ git diff
> > diff --git a/lib/librte_stack/rte_stack_lf.h
> > b/lib/librte_stack/rte_stack_lf.h index f5581f0c2..46af08b83 100644
> > --- a/lib/librte_stack/rte_stack_lf.h
> > +++ b/lib/librte_stack/rte_stack_lf.h
> > @@ -5,7 +5,7 @@
> >  #ifndef _RTE_STACK_LF_H_
> >  #define _RTE_STACK_LF_H_
> >
> > -#ifdef RTE_USE_C11_MEM_MODEL
> > +#if defined (RTE_USE_C11_MEM_MODEL) &&
> defined(RTE_ARCH_X86_64)
> > &&
> > +defined(RTE_ARCH_ARM64)
> 
> I assume you meant (defined(RTE_ARCH_X86_64) ||
> defined(RTE_ARCH_ARM64))?

Yup.

> 
> >  #include "rte_stack_lf_c11.h"
> >  #else
> >  #include "rte_stack_lf_generic.h"
> >
> >
> 
> The ifdefs in those two headers prevent DPDK from trying to build
> rte_atomic128_cmp_exchange() on architectures that don't implement it. So
> the proposal wouldn't quite work, since rte_stack_lf_generic.h calls
> rte_atomic128_cmp_exchange().
> 
> Something like this could work:
> 
> #if !(defined(RTE_ARCH_X86_64) || defined(RTE_ARCH_ARM64)) #include
> rte_stack_lf_stubs.h #else #ifdef RTE_USE_C11_MEM_MODEL #include
> "rte_stack_lf_c11.h"
> #else
> #include "rte_stack_lf_generic.h"
> #endif
> #endif
> 
> Where rte_stack_lf_stubs.h is a new header containing stub
> implementations of __rte_stack_lf_count, __rte_stack_lf_push_elems, and
> __rte_stack_lf_pop_elems. It still has some ifdef clutter, but less overall.

Agree. I prefer to take this route to reduce the ifdef clutter across generic and c11 files.