SUBREG_PROMOTED_VAR_P handling in expand_direct_optab_fn

This is needed by the later SVE LAST reductions, where an 8-bit
or 16-bit result is zero- rather than sign-extended to 32 bits.
I think it could occur in other situations too.

2017-09-19  Richard Sandiford  <richard.sandiford@linaro.org>
	    Alan Hayward  <alan.hayward@arm.com>
	    David Sherwood  <david.sherwood@arm.com>

gcc/
	* internal-fn.c (expand_direct_optab_fn): Don't assign directly
	to a SUBREG_PROMOTED_VAR.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>

From-SVN: r253992

gcc/ChangeLog

@@ -2,6 +2,13 @@
 	    Alan Hayward  <alan.hayward@arm.com>
 	    David Sherwood  <david.sherwood@arm.com>
 
+	* internal-fn.c (expand_direct_optab_fn): Don't assign directly
+	to a SUBREG_PROMOTED_VAR.
+
+2017-10-22  Richard Sandiford  <richard.sandiford@linaro.org>
+	    Alan Hayward  <alan.hayward@arm.com>
+	    David Sherwood  <david.sherwood@arm.com>
+
 	* cfgexpand.c (expand_debug_expr): Use GET_MODE_UNIT_PRECISION.
 	(expand_debug_source_expr): Likewise.
 	* combine.c (combine_simplify_rtx): Likewise.

gcc/internal-fn.c

@@ -2606,7 +2606,15 @@ expand_direct_optab_fn (internal_fn fn, gcall *stmt, direct_optab optab,
   tree lhs = gimple_call_lhs (stmt);
   tree lhs_type = TREE_TYPE (lhs);
   rtx lhs_rtx = expand_expr (lhs, NULL_RTX, VOIDmode, EXPAND_WRITE);
-  create_output_operand (&ops[0], lhs_rtx, insn_data[icode].operand[0].mode);
+
+  /* Do not assign directly to a promoted subreg, since there is no
+     guarantee that the instruction will leave the upper bits of the
+     register in the state required by SUBREG_PROMOTED_SIGN.  */
+  rtx dest = lhs_rtx;
+  if (GET_CODE (dest) == SUBREG && SUBREG_PROMOTED_VAR_P (dest))
+    dest = NULL_RTX;
+
+  create_output_operand (&ops[0], dest, insn_data[icode].operand[0].mode);
 
   for (unsigned int i = 0; i < nargs; ++i)
     {