[cxx-abi-dev] Mangling of ISO/IEC TR 18037 fixed-point types?

[David Vandevoorde]

On Dec 4, 2008, at 2:30 PM, Jason Merrill wrote:

> GCC supports fixed-point arithmetic along the lines of the above  
> standard, but there's currently no mangling specified in the ABI.  
> Anyone want to propose one?  Do other compilers have support for  
> these types?


We implemented these some time ago for C, but not C++ (so we didn't  
decide on mangling).

How about:

	DF [a|f] <kind>

with "a" for _Accum types, and "f" for _Fract types, and <kind>  
imitating the integer types:

	<kind> = s for "signed short"
	         t for "unsigned short"
	         i for "signed"
	         j for "unsigned"
	         l for "long"
	         m for "unsigned long"


E.g., "unsigned long _Accum" would be encoded as "DFam".

Alternatively, if we think a single platform may have different  
precision choices for these types, we could go with:

	DF <accumbits(opt)> <kind> <fractbits> _

where

	<accumbits(opt)> = nothing for _Fract types
	                   the number of explicit nonfractional bits for  
_Accum types (base 10 digits)
	<fractbits> = the number of fractional bits (base 10 digits)

E.g., "short _Fract" might get encoded as "DFs15_" if the compiler  
implements _Fract as a 16-bit type, or "DFs7_" if it uses an 8-bit  
type (e.g., to map on Intel MMX).

	Daveed