Re: Patch to implement rotates by zero

[Paul Koning <pkoning at equallogic dot com>]

>>>>> "Richard" == Richard Sandiford <rsandifo@redhat.com> writes:

 Richard> Hmm, having had a bit more chance to think about it, I'd
 Richard> like to stick with the original patch...

 Richard> cgd@broadcom.com writes:
 >> it just makes the binutils code simpler.  People who do rotate by
 >> 0 and expect an optimization should lose.  it wouldn't happen if
 >> they had real rotate instructions (i.e., they'd get an actual
 >> instruction).  So, they might as well get the 'normal' rotate
 >> macro expansion.

I think the point is this: it's an across-the-board rule in MIPS that
32 bit operands in 64-bit registers MUST be properly sign-extended or
the result is unpredictable.

There's just one single exception to that, which is sll.  It seems
that this special exception was created to shorten the common code
sequence:
	dsll rx,rx,32
	dsra rx,rx,32
which you will see all over the place when compilers are making sure
that 32-bit operands are properly formed.

It seems logical for a macro to obey the standard data format rule,
which is that 32-bit data must be properly sign-extended in 64 bit
registers.   

	     paul