public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* patch for PR64342
@ 2015-03-05 19:48 Vladimir Makarov
  0 siblings, 0 replies; only message in thread
From: Vladimir Makarov @ 2015-03-05 19:48 UTC (permalink / raw)
  To: gcc-patches

[-- Attachment #1: Type: text/plain, Size: 480 bytes --]

  The following patch fixes a bad code generation for avx512f-kandnw-1.c
reported in

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64342

  The patch was bootstrapped and tested on x86-64.

  Committed as rev. 221223.

2015-03-05  Vladimir Makarov  <vmakarov@redhat.com>

        PR target/64342
        * lra-assigns.c (find_hard_regno_for): Rename to
        find_hard_regno_for_1.  Add a new parameter.
        (find_hard_regno_for): New function using find_hard_regno_for_1.


[-- Attachment #2: pr64342.patch --]
[-- Type: text/x-patch, Size: 2637 bytes --]

Index: lra-assigns.c
===================================================================
--- lra-assigns.c	(revision 220916)
+++ lra-assigns.c	(working copy)
@@ -491,10 +491,13 @@ adjust_hard_regno_cost (int hard_regno,
    pseudos in complicated situations where pseudo sizes are different.
 
    If TRY_ONLY_HARD_REGNO >= 0, consider only that hard register,
-   otherwise consider all hard registers in REGNO's class.  */
+   otherwise consider all hard registers in REGNO's class.
+
+   If REGNO_SET is not empty, only hard registers from the set are
+   considered.  */
 static int
-find_hard_regno_for (int regno, int *cost, int try_only_hard_regno,
-		     bool first_p)
+find_hard_regno_for_1 (int regno, int *cost, int try_only_hard_regno,
+		       bool first_p, HARD_REG_SET regno_set)
 {
   HARD_REG_SET conflict_set;
   int best_cost = INT_MAX, best_priority = INT_MIN, best_usage = INT_MAX;
@@ -509,7 +512,13 @@ find_hard_regno_for (int regno, int *cos
   bool *rclass_intersect_p;
   HARD_REG_SET impossible_start_hard_regs, available_regs;
 
-  COPY_HARD_REG_SET (conflict_set, lra_no_alloc_regs);
+  if (hard_reg_set_empty_p (regno_set))
+    COPY_HARD_REG_SET (conflict_set, lra_no_alloc_regs);
+  else
+    {
+      COMPL_HARD_REG_SET (conflict_set, regno_set);
+      IOR_HARD_REG_SET (conflict_set, lra_no_alloc_regs);
+    }
   rclass = regno_allocno_class_array[regno];
   rclass_intersect_p = ira_reg_classes_intersect_p[rclass];
   curr_hard_regno_costs_check++;
@@ -680,6 +689,33 @@ find_hard_regno_for (int regno, int *cos
   return best_hard_regno;
 }
 
+/* A wrapper for find_hard_regno_for_1 (see comments for that function
+   description).  This function tries to find a hard register for
+   preferred class first if it is worth.  */
+static int
+find_hard_regno_for (int regno, int *cost, int try_only_hard_regno, bool first_p)
+{
+  int hard_regno;
+  HARD_REG_SET regno_set;
+
+  /* Only original pseudos can have a different preferred class.  */
+  if (try_only_hard_regno < 0 && regno < lra_new_regno_start)
+    {
+      enum reg_class pref_class = reg_preferred_class (regno);
+      
+      if (regno_allocno_class_array[regno] != pref_class)
+	{
+	  hard_regno = find_hard_regno_for_1 (regno, cost, -1, first_p,
+					      reg_class_contents[pref_class]);
+	  if (hard_regno >= 0)
+	    return hard_regno;
+	}
+    }
+  CLEAR_HARD_REG_SET (regno_set);
+  return find_hard_regno_for_1 (regno, cost, try_only_hard_regno, first_p,
+				regno_set);
+}
+
 /* Current value used for checking elements in
    update_hard_regno_preference_check.	*/
 static int curr_update_hard_regno_preference_check;

^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2015-03-05 19:48 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-03-05 19:48 patch for PR64342 Vladimir Makarov

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).