[Dwarf-Discuss] DWARF piece questions

[Michael Eager]

On 01/27/2017 07:19 AM, Andreas Arnez wrote:
> On Thu, Jan 26 2017, Michael Eager wrote:
>
>> I don't understand the assertion that "most significant" can not be
>> applied to registers.  In the case where a register contains a single
>> value, this appears to be unambiguous.  When a register contains
>> multiple values (some architectures support 2 16-bit values in a
>> 32-bit register, for example), then additional clarification may be
>> needed.  The ABI provides this, usually.  The situation where a single
>> value is split across two registers is similar.  When discussing
>> compound values (such as floating point) it is common to say that the
>> sign bit is in the most-significant bit of the register, without
>> ambiguity.
>
> Hm, can you point to a formal definition?  The usual definition of
> "most/least significant bit" assumes that the underlying object is a
> single binary integer:
>
>    https://en.wikipedia.org/wiki/Least_significant_bit
>
> And I see how all this still makes sense for registers that can
> "naturally" contain full-sized integers.  But not in general.  You
> already list some cases that aren't that obvious, and there are many
> more.  When you say "the ABI provides this [clarification], usually",
> where did you see that?  For instance, when looking at the System V
> Application Binary Interface AMD64 Architecture Processor Supplement
> Draft Version 0.99.7, I don't see any definition related to DWARF
> pieces, nor a clarification of the terms "most/least signficant bit" of
> a register.  Did you see something like that in any other ABI?

There are architectures which support performing arithmetic on multiple
values in a register, say two 16-bit values in a 32-bit register.  The
ISA defines sign bits in each half, and the operations are defined so
that overflow in one half does not affect the other.

>
> Maybe I'm too picky, but IMO a standard document should rely on formally
> defined terms only.  And "least significant bit", when applied to a
> register, is not one of those, I think.

Perhaps we do need a comment that registers are viewed as a sequence
of bits.


-- 
Michael Eager	 eager at eagercon.com
1960 Park Blvd., Palo Alto, CA 94306  650-325-8077