This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [Patch Darwin 2/2] fix PPC64 ABI

On 24 Jul 2010, at 16:10, David Edelsohn wrote:

On Sat, Jul 24, 2010 at 11:06 AM, IainS
<> wrote:
Mike, David,

On 24 Jul 2010, at 10:45, Mike Stump wrote:

Also, as memory serves there is one more fix from last year that fixes the
abi so that it is self consistent from last year. I think it fixes a
failing struct abi testcase as I recall.

OK, I found it (the struct hack is needed after all, in fact).

FWIW, The fail of is misleading - I think
that is a situation where the test-case needs some tweaking.

The darwin64-abi.c check from 4.2.1 (or LLVM) is the right thing to use...
... but I have no idea what license it is covered by and, therefore,
whether we can import it... so for now I'm using it outside the gcc tree.


There will follow a part #3 (I can hack a fix... now I need to find the
Right Way).

Part 3 will apply on top of parts 1 & 2 .. so those patches stand.

Darwin-specific patches approved by Mike are fine with me

part 1 is r162567
part 2 (as amended with Nathan's suggestion, copied below) is r162568 (changelog for #2 is 162569)


amended part 2/2

/* We need to deal with the fact that the darwin ppc64 ABI is defined by an
earlier version of gcc, with the property that it always applied alignment
adjustments to the va-args (even for zero-sized types). The cheapest way
to deal with this is to replicate the effect of the part of
std_gimplify_va_arg_expr that carries out the align adjust, for the case
of relevance.
We don't need to check for pass-by-reference because of the test above.
We can return a simplifed answer, since we know there's no offset to add. */

      && rs6000_darwin64_abi
      && integer_zerop (TYPE_SIZE (type)))
      unsigned HOST_WIDE_INT align, boundary;
      tree valist_tmp = get_initialized_tmp_var (valist, pre_p, NULL);
      boundary = FUNCTION_ARG_BOUNDARY (TYPE_MODE (type), type);
      boundary /= BITS_PER_UNIT;
      if (boundary > align)
	  tree t ;
	  /* This updates arg ptr by the amount that would be necessary
	     to align the zero-sized (but not zero-alignment) item.  */
	  t = build2 (MODIFY_EXPR, TREE_TYPE (valist), valist_tmp,
		  fold_build2 (POINTER_PLUS_EXPR,
			       TREE_TYPE (valist),
			       valist_tmp, size_int (boundary - 1)));
	  gimplify_and_add (t, pre_p);

t = fold_convert (sizetype, valist_tmp);
t = build2 (MODIFY_EXPR, TREE_TYPE (valist), valist_tmp,
fold_convert (TREE_TYPE (valist),
fold_build2 (BIT_AND_EXPR, sizetype, t,
size_int (-boundary))));
t = build2 (MODIFY_EXPR, TREE_TYPE (valist), valist, t);
gimplify_and_add (t, pre_p);
/* Since it is zero-sized there's no increment for the item itself. */
valist_tmp = fold_convert (build_pointer_type (type), valist_tmp);
return build_va_arg_indirect_ref (valist_tmp);

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]