Re: [PATCH] iio: tools: Use proper macro to detect _Float16 support

[Ivan A. Melnikov]

On Sat, Sep 19, 2026 at 11:39:45AM +0200, Joshua Crofts wrote:
> On Fri, 18 Sep 2026 20:13:02 +0400
> "Ivan A. Melnikov" <iv@xxxxxxxxxxxx> wrote:
> 
> > __FLT16_MAX__ is an internal compiler macro. GCC defines it if
> > _Float16 is supported by the target in general, but the support
> > may depend on the exact target options. For example, on i586
> > without -msse2 I get the following error from GCC15:
> > 
> >   iio_generic_buffer.c: In function 'print2byte':
> >   iio_generic_buffer.c:141:17: error: invalid conversion from type '_Float16' without option '-msse2'
> >     141 |                 printf("%05f ", ((float)converter.f + info->offset) * info->scale);
> >         |                 ^~~~~~
> > 
> > To fix that, use the public macro without double underscores.
> > 
> > Fixes: cdd445d4b2a9 ("iio: tools: Add support for floating-point types in buffer scan elements")
> > Signed-off-by: Ivan A. Melnikov <iv@xxxxxxxxxxxx>
> > ---
> >  tools/iio/iio_generic_buffer.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/tools/iio/iio_generic_buffer.c b/tools/iio/iio_generic_buffer.c
> > index 000193612aad..9ea93984e0e9 100644
> > --- a/tools/iio/iio_generic_buffer.c
> > +++ b/tools/iio/iio_generic_buffer.c
> > @@ -131,7 +131,7 @@ static void print2byte(uint16_t input, struct iio_channel_info *info)
> >  		printf("%05f ", ((float)input + info->offset) * info->scale);
> >  		break;
> >  	case 'f': {
> > -#if defined(__FLT16_MAX__)
> > +#if defined(FLT16_MAX)
> >  		union {
> >  			uint16_t u;
> >  			_Float16 f;
> 
> One Sashiko issue:
> 
> Does this unintentionally disable _Float16 support for all architectures?

Indeed, it does. Sorry, my patch is incorrect. I must have completely
messed up all my checks on Friday.

> Since the tools build environment uses gnu11 compilation flags without
> explicitly defining __STDC_WANT_IEC_60559_TYPES_EXT__, FLT16_MAX is
> undefined across all architectures. This causes the preprocessor to skip
> this block entirely, causing the tool to always print 
> <unsupported 2-byte float> for 16-bit float data, even on platforms where 
> native support previously worked.
> 
> If FLT16_MAX were actually defined, the compiler would still parse the
> _Float16 block and throw the same missing -msse2 compilation error mentioned
> in the commit message. Is there an alternative way to check for hardware
> support without completely disabling the feature?

Apparently, the only correct way to detect _Float16 is to try compiling
some code that uses it. For that, a feature can be added to tools/build.

On other hand, x86 without SSE2 is the only platform I could find where
__FLT16_MAX__ is defined, but _Float16 does not work, so a simple
platform-specific workaround would also work and would be much less
elaborate.

WDYT?

-- 
  wbr,
    iv m.