CPP macro for Byte-order

[Gabriel Dos Reis]

Joe Buck <jbuck@synopsys.com> writes:

| Gaby writes:
| >   I'm currently working on the numeric_limits<> support for floating
| > point types (mainly for things like NaNs, SNaNs, Infinities,
| > subnormals and such) in the C++ runtime system.  I would like to know
| > whether there is a CPP macro that defines the endianness selected for
| > compiling a given program under a given OS.  Any hints?
| 
| There's nothing that I know of that's provided standard by the OS.
| And clearly configure won't work because of cross-compilation.
| 
| On the other hand, the gcc backends define BITS_BIG_ENDIAN,
| BYTES_BIG_ENDIAN, and WORDS_BIG_ENDIAN (you would want BYTES_BIG_ENDIAN to
| solve the problem of detecting NaNs and the like).  Perhaps these
| definitions can just be collected, with appropriate ifdefs based on
| preprocessor symbols defined by gcc, so that correct endian-ness will be
| selected on all ports and OSes where we have a backend.

Joe,

  Thanks for the pointer.  Indeed, starting from the back-end I have
found useful macros and more info.

| If this approach is followed (define a macro based on the port, determine
| the port by preprocessor constants that gcc defines for that port), then
| it would be a good idea to cause a syntax error to be generated if someone
| tries to build it for an unsupported port.

I think that is doable.  We'll see what I come up with.  SPARC-V9
formally let's one specifies which endianness to use.  Is GCC
supporting or planned to support dual endianness on the same target?

| Another approach, that would impose a bit of runtime overhead, would be
| to have a runtime test.  Something like
| 
| const int __endian_detect = 1;
| 
| inline bool __little_endian() { return *(char *)&__endian_detect == 1;}
| 
| In principle the optimizer can constant-fold this, so perhaps there would
| not be any runtime overhead.  If so, it would be a good solution.
| 
| There would be an issue with ports in which int and char are the same
| size (the DSP targets: c4x and dsp16xx).

Ugh.

-- Gaby