[PATCH, rs6000, 4.8/4.9] Fix ELFv2 homogeneous float aggregate ABI bug

["Ulrich Weigand" <uweigand at de dot ibm dot com>]

Hello,

this is the variant intended for the 4.8/4.9 branches of the patch:
https://gcc.gnu.org/ml/gcc-patches/2014-07/msg00994.html

As discussed, it does *not* actually change ABI, but only warn when
encountering a situation where the ABI will change in a future GCC.
(Avoiding the specific term "GCC 4.10" here since I'm not certain
whether the next GCC release will in fact be called that ...)

Tested on powerpc64-linux and powerpc64le-linux; also verified using
the ABI compat suite (against an unpatched GCC) that this patch does
not change the ABI.

OK for 4.8/4.9 once the mainline patch is in?

Bye,
Ulrich


gcc/ChangeLog:

	* config/rs6000/rs6000.c (rs6000_function_arg): If a float argument
	does not fit fully into floating-point registers, and there is still
	space in the register parameter area, issue -Wpsabi note that the ABI
	will change in a future GCC release.

gcc/testsuite/ChangeLog:

	* gcc.target/powerpc/ppc64-abi-warn-1.c: New test.


Index: gcc-4_9-branch/gcc/config/rs6000/rs6000.c
===================================================================
--- gcc-4_9-branch.orig/gcc/config/rs6000/rs6000.c
+++ gcc-4_9-branch/gcc/config/rs6000/rs6000.c
@@ -10225,6 +10225,7 @@ rs6000_function_arg (cumulative_args_t c
 	  rtx r, off;
 	  int i, k = 0;
 	  unsigned long n_fpreg = (GET_MODE_SIZE (elt_mode) + 7) >> 3;
+	  int fpr_words;
 
 	  /* Do we also need to pass this argument in the parameter
 	     save area?  */
@@ -10253,6 +10254,37 @@ rs6000_function_arg (cumulative_args_t c
 	      rvec[k++] = gen_rtx_EXPR_LIST (VOIDmode, r, off);
 	    }
 
+	  /* If there were not enough FPRs to hold the argument, the rest
+	     usually goes into memory.  However, if the current position
+	     is still within the register parameter area, a portion may
+	     actually have to go into GPRs.
+
+	     Note that it may happen that the portion of the argument
+	     passed in the first "half" of the first GPR was already
+	     passed in the last FPR as well.
+
+	     For unnamed arguments, we already set up GPRs to cover the
+	     whole argument in rs6000_psave_function_arg, so there is
+	     nothing further to do at this point.
+
+	     GCC 4.8/4.9 Note: This was implemented incorrectly in earlier
+	     GCC releases.  To avoid any ABI change on the release branch,
+	     we retain that original implementation here, but warn if we
+	     encounter a case where the ABI will change in the future.  */
+	  fpr_words = (i * GET_MODE_SIZE (elt_mode)) / (TARGET_32BIT ? 4 : 8);
+	  if (i < n_elts && align_words + fpr_words < GP_ARG_NUM_REG
+	      && cum->nargs_prototype > 0)
+            {
+	      static bool warned;
+	      if (!warned && warn_psabi)
+		{
+		  warned = true;
+		  inform (input_location,
+			  "the ABI of passing homogeneous float aggregates"
+			  " will change in a future GCC release");
+		}
+	    }
+
 	  return rs6000_finish_function_arg (mode, rvec, k);
 	}
       else if (align_words < GP_ARG_NUM_REG)
Index: gcc-4_9-branch/gcc/testsuite/gcc.target/powerpc/ppc64-abi-warn-1.c
===================================================================
--- /dev/null
+++ gcc-4_9-branch/gcc/testsuite/gcc.target/powerpc/ppc64-abi-warn-1.c
@@ -0,0 +1,12 @@
+/* { dg-do compile { target { powerpc*-*-linux* && lp64 } } } */
+/* { dg-options "-mabi=elfv2" } */
+
+struct f8
+  {
+    float x[8];
+  };
+
+void test (struct f8 a, struct f8 b) /* { dg-message "note: the ABI of passing homogeneous float aggregates will change" } */
+{
+}
+
-- 
  Dr. Ulrich Weigand
  GNU/Linux compilers and toolchain
  Ulrich.Weigand@de.ibm.com