[Fortran, committed to trunk] Fixed powerpc "unsigned" printing of

[Toon Moene]

[ Moved the gnats and patches mailing list out of the loop ]

I wrote:
> 
> John David Anglin wrote:

> > I did a liitle more digging on the pa.  It beginning to look like some
> > kind of aliasing problem.  It important to note that the test passes
> > at -O0 but not at -O1.  At -O0, there are insns that write the i*2
> > values of j and ja to virtual stack slots.  However, at -O1, it appears
> > the write for ja is deleted and as a result ka is read from an unitialized
> > slot.  I got as far as ffestd_stmt_pass_ in trying to locate the problem.

On the powerpc, it looks slightly different.  Here's the smallest
program I could construct that still shows the problem:

      integer*2 j
      integer*1 k
      j = -9
      k = j
      if (k .ne. j) call abort
      print*,j
      end

Note the necessity of `print*,j' - if you remove it the abort will not
happen.

Now for the rtl (c.f.00.rtl)

...
(insn 12 10 42 (set (reg:SI 121)
	(const_int -9 [0xfffffffffffffff7])) -1 (nil)
    (nil))
		Wonder why we need a 64-bit constant to set a 32-bit reg
(insn 42 12 43 (set (reg:HI 122)
	(subreg:HI (reg:SI 121) 2)) -1 (nil)
    (nil))
		So far, so good: constant -9 loaded into HI register (j)
(insn 43 42 14 (set (mem/f:HI (reg/f:SI 111 virtual-stack-vars) [0 S2
A8]
	(reg:HI 122)) -1 (nil)
    (nil))
		Store j on stack ?  Because of `print*,j' ?
...
(insn 46 15 17 (set (reg:QI 123)
	(mem/f:QI (plus:SI (reg/f:SI 111 virtual-stack-vars)
		(const_int 3 [0x3])) [0 S1 A8])) -1 (nil)
    (nil))
		So the byte k is offset 3 away from the memory address
		where we stored the short j ?  Howzdat ?  Shouldn't this
		be (const_int 1 ...) (that's what you get when compiling
		without -O) ?
...

I get the feeling this isn't a Fortran frontend problem ...

Which optimisation passes run *before* 00.rtl is dumped ?  I thought
this was the first thing done ...

Confusedly yours,

-- 
Toon Moene - mailto:toon@moene.indiv.nluug.nl - phoneto: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
Maintainer, GNU Fortran 77: http://gcc.gnu.org/onlinedocs/g77_news.html
Join GNU Fortran 95: http://g95.sourceforge.net/ (under construction)