Revised patch to remove integer output macros

Hans-Peter Nilsson hp@bitrange.com
Sat Dec 15 21:32:00 GMT 2001


On 14 Dec 2001, Richard Sandiford wrote:
> I know of three cases where things genuinely changed:

  ...

>    cris
> 	This target uses the same ops for unaligned and aligned data.
> 	It used to use ".quad" for unaligned data but not aligned data
> 	It will now use ".quad" for both.

Before, it seems it'd be split up in two ".dword":s.
That's ok.

>    mmix
> 	Now uses ".byte" exclusively, instead of using "BYTE" for
> 	some things and ".byte" for others.

That's not really good.  Now we lost compatibility with Knuth's
mmixal assembler/linker for assembly generated from simple C
code.  C++ was hopelessly lost before too, but the generated
assembly wasn't really readable anyway. :-)

I'll have to check whether using BYTE consistently is possible.
It has some constrained syntax and semantics.

Can't you have two different byte-size asm directives, default
the same?  Perhaps I get the same effect through the new target
hook?

With your patch, are numbers always emitted as decimal, or are
there cases (for C code) where numbers are emitted as 0xNNNN?
I ask because the "0xNNNN" syntax isn't allowed in mmixal
syntax; it has to be decimal, or "#NNNN" for hex.

I guess I'll find out.

brgds, H-P



More information about the Gcc-patches mailing list