combine: Another hard register problem (PR85805)

The current code in reg_nonzero_bits_for_combine allows using the
reg_stat info when last_set_mode is a different integer mode.  This is
completely wrong for non-pseudos.  For example, as in the PR, a value
in a DImode hard register is set by eight writes to its constituent
QImode parts.  The value written to the DImode is not the same as that
written to the lowest-numbered QImode!


	PR rtl-optimization/85805
	* combine.c (reg_nonzero_bits_for_combine): Only use the last set
	value for hard registers if that was written in the same mode.

From-SVN: r262994

gcc/ChangeLog

+2018-07-26  Segher Boessenkool  <segher@kernel.crashing.org>
+
+	PR rtl-optimization/85805
+	* combine.c (reg_nonzero_bits_for_combine): Only use the last set
+	value for hard registers if that was written in the same mode.
+
 2018-07-26  Martin Liska  <mliska@suse.cz>
 
-        PR gcov-profile/86536
+	PR gcov-profile/86536
 	* gcov.c (format_gcov): Use printf format %.*f directly
-        and do not handle special values.
+	and do not handle special values.
 
 2018-07-25  Claudiu Zissulescu  <claziss@synopsys.com>
 

gcc/combine.c

@@ -10169,7 +10169,8 @@ reg_nonzero_bits_for_combine (const_rtx x, scalar_int_mode xmode,
   rsp = &reg_stat[REGNO (x)];
   if (rsp->last_set_value != 0
       && (rsp->last_set_mode == mode
-	  || (GET_MODE_CLASS (rsp->last_set_mode) == MODE_INT
+	  || (REGNO (x) >= FIRST_PSEUDO_REGISTER
+	      && GET_MODE_CLASS (rsp->last_set_mode) == MODE_INT
 	      && GET_MODE_CLASS (mode) == MODE_INT))
       && ((rsp->last_set_label >= label_tick_ebb_start
 	   && rsp->last_set_label < label_tick)