Re: [eigen] Fix for ARM __builtin_prefetch fails

[Christoph Hertzberg <chtz@xxxxxxxxxxxxxxxxxxxxxxxx>]

On 19.11.2013 03:37, Rick Mann wrote:
> On Nov 18, 2013, at 15:23 , Christoph Hertzberg <chtz@xxxxxxxxxxxxxxxxxxxxxxxx> wrote:
> > Ok, so the issue is that __has_builtin(__builtin_prefetch)
> > is syntactically incorrect if __has_builtin is not defined. That means your patch resolves the issue only for GCC and the fix is to separate the && expression into two preprocessor lines.
> > [...]
>
> Yes, I think that might be better. The engineer who found this is not around to ask right now if this change would work for him.

Next alternative:

//This can be put in Core/util/*
#ifdef __has_builtin
#  define EIGEN_HAS_BUILTIN(x) __has_builtin
#else
#  define EIGEN_HAS_BUILTIN(x) 0
#endif


// only the first line needs to be changed:
#if EIGEN_HAS_BUILTIN(__builtin_prefetch) || defined(__GNUC__)
  #define EIGEN_ARM_PREFETCH(ADDR) __builtin_prefetch(ADDR);
//...


Does any compiler except LLVM support __has_builtin?

Christoph

--
----------------------------------------------
Dipl.-Inf., Dipl.-Math. Christoph Hertzberg
Cartesium 0.049
Universität Bremen
Enrique-Schmidt-Straße 5
28359 Bremen

Tel: +49 (421) 218-64252
----------------------------------------------