INLINE redefined by kernel with Intel (beignet)
hgomersall opened this issue · 5 comments
The generated source includes the preprocessor directive #define INLINE inline
. It seems that with Intel beignet (on Ubuntu 15.04), the INLINE macro is already defined, returning the warning (something like):
/tmp/widEMp.cl:19:13: warning: 'INLINE' macro redefined
The original definition of INLINE is as inline __attribute__((always_inline))
, which seems reasonable. Does it make sense to protect the INLINE definition inside an #if !define(INLINE)
?
Of course, this is set inside reikna/cluda/kernel.mako
.
This is defined inside beignet/include/ocl_types.h
in the beignet tree. Is Beignet the only system that defines that macro?
Is Beignet the only system that defines that macro?
I think so, I've never had any problems before. Could you check the header files and see what is INLINE resolved to? If it's the same as what Reikna defines (inline
), I'll just add an ifndef
.
It resolves to inline __attribute__((always_inline))
, as in ocl_types.h.
Thanks Bogdan.