DW_AT_bit_size for 80-bit floating point

[Doug Kwan (關振德)]

Hi,

I am working on the LTO project in gcc and one of the problem I am
looking at is to emit type information for 80-bit extended floating
point format on x86.  My question is whether I can use DW_AT_bit_size
and DW_AT_bit_offset to encode the number of bits used.

The 80-bit format is stored in a 12-bit location so I need to emit
extra information. Striclty speaking DW_AT_bit_size and
DW_AT_bit_offset may not be appropriate.  The exponent and mantissa
part of an 80-bit float may or may not be contiguous.  I also need to
handle 80-bit extended based complex float. I'm sure the bits are not
contiguous for the complex case.  So DW_AT_bit_size and
DW_AT_bit_offset are kind of odd here.

I can use a new dwarf extension like DW_AT_GNU_precision but I want to
avoid that.

-Doug