public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* [PATCH] Fix compile time complexity of search_type_for_mask (PR tree-optimization/72866)
@ 2016-08-29 19:57 Jakub Jelinek
  2016-08-30  4:42 ` Richard Biener
  0 siblings, 1 reply; 2+ messages in thread
From: Jakub Jelinek @ 2016-08-29 19:57 UTC (permalink / raw)
  To: Richard Biener; +Cc: gcc-patches

Hi!

As the testcase shows, with deep chains of COND_EXPRs with the same
SSA_NAMEs appearing more than once in the various operands we can
hang in search_type_for_mask.  Similar check_bool_pattern function
uses a hash_set to handle stmts that have been checked already, this patch
uses a hash_map in which we cache the return values once we've handled some
stmt already.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

2016-08-29  Jakub Jelinek  <jakub@redhat.com>

	PR tree-optimization/72866
	* tree-vect-patterns.c (search_type_for_mask): Turn into
	a small wrapper, move all code to ...
	(search_type_for_mask_1): ... this new function.  Add caching
	and adjust recursive calls.

	* gcc.dg/vect/pr72866.c: New test.

--- gcc/tree-vect-patterns.c.jj	2016-07-21 08:59:55.000000000 +0200
+++ gcc/tree-vect-patterns.c	2016-08-29 19:26:12.807164641 +0200
@@ -3459,13 +3459,11 @@ adjust_bool_stmts (hash_set <gimple *> &
   return gimple_assign_lhs (pattern_stmt);
 }
 
-/* Return the proper type for converting bool VAR into
-   an integer value or NULL_TREE if no such type exists.
-   The type is chosen so that converted value has the
-   same number of elements as VAR's vector type.  */
+/* Helper for search_type_for_mask.  */
 
 static tree
-search_type_for_mask (tree var, vec_info *vinfo)
+search_type_for_mask_1 (tree var, vec_info *vinfo,
+			hash_map<gimple *, tree> &cache)
 {
   gimple *def_stmt;
   enum vect_def_type dt;
@@ -3490,6 +3488,10 @@ search_type_for_mask (tree var, vec_info
   if (!is_gimple_assign (def_stmt))
     return NULL_TREE;
 
+  tree *c = cache.get (def_stmt);
+  if (c)
+    return *c;
+
   rhs_code = gimple_assign_rhs_code (def_stmt);
   rhs1 = gimple_assign_rhs1 (def_stmt);
 
@@ -3498,14 +3500,15 @@ search_type_for_mask (tree var, vec_info
     case SSA_NAME:
     case BIT_NOT_EXPR:
     CASE_CONVERT:
-      res = search_type_for_mask (rhs1, vinfo);
+      res = search_type_for_mask_1 (rhs1, vinfo, cache);
       break;
 
     case BIT_AND_EXPR:
     case BIT_IOR_EXPR:
     case BIT_XOR_EXPR:
-      res = search_type_for_mask (rhs1, vinfo);
-      res2 = search_type_for_mask (gimple_assign_rhs2 (def_stmt), vinfo);
+      res = search_type_for_mask_1 (rhs1, vinfo, cache);
+      res2 = search_type_for_mask_1 (gimple_assign_rhs2 (def_stmt), vinfo,
+				     cache);
       if (!res || (res2 && TYPE_PRECISION (res) > TYPE_PRECISION (res2)))
 	res = res2;
       break;
@@ -3517,8 +3520,9 @@ search_type_for_mask (tree var, vec_info
 
 	  if (TREE_CODE (TREE_TYPE (rhs1)) == BOOLEAN_TYPE)
 	    {
-	      res = search_type_for_mask (rhs1, vinfo);
-	      res2 = search_type_for_mask (gimple_assign_rhs2 (def_stmt), vinfo);
+	      res = search_type_for_mask_1 (rhs1, vinfo, cache);
+	      res2 = search_type_for_mask_1 (gimple_assign_rhs2 (def_stmt),
+					     vinfo, cache);
 	      if (!res || (res2 && TYPE_PRECISION (res) > TYPE_PRECISION (res2)))
 		res = res2;
 	      break;
@@ -3526,12 +3530,18 @@ search_type_for_mask (tree var, vec_info
 
 	  comp_vectype = get_vectype_for_scalar_type (TREE_TYPE (rhs1));
 	  if (comp_vectype == NULL_TREE)
-	    return NULL_TREE;
+	    {
+	      res = NULL_TREE;
+	      break;
+	    }
 
 	  mask_type = get_mask_type_for_scalar_type (TREE_TYPE (rhs1));
 	  if (!mask_type
 	      || !expand_vec_cmp_expr_p (comp_vectype, mask_type))
-	    return NULL_TREE;
+	    {
+	      res = NULL_TREE;
+	      break;
+	    }
 
 	  if (TREE_CODE (TREE_TYPE (rhs1)) != INTEGER_TYPE
 	      || !TYPE_UNSIGNED (TREE_TYPE (rhs1)))
@@ -3544,9 +3554,21 @@ search_type_for_mask (tree var, vec_info
 	}
     }
 
+  cache.put (def_stmt, res);
   return res;
 }
 
+/* Return the proper type for converting bool VAR into
+   an integer value or NULL_TREE if no such type exists.
+   The type is chosen so that converted value has the
+   same number of elements as VAR's vector type.  */
+
+static tree
+search_type_for_mask (tree var, vec_info *vinfo)
+{
+  hash_map<gimple *, tree> cache;
+  return search_type_for_mask_1 (var, vinfo, cache);
+}
 
 /* Function vect_recog_bool_pattern
 
--- gcc/testsuite/gcc.dg/vect/pr72866.c.jj	2016-08-29 19:28:22.453646025 +0200
+++ gcc/testsuite/gcc.dg/vect/pr72866.c	2016-08-29 19:28:17.000000000 +0200
@@ -0,0 +1,19 @@
+/* PR tree-optimization/72866 */
+/* { dg-do compile } */
+
+unsigned int dl;
+int rx, lb;
+
+void
+fo (int jv, int be)
+{
+  const unsigned int xw = 16;
+  unsigned int ya, wo;
+
+  for (ya = 0; ya < 2; ++ya)
+    for (wo = 0; wo < xw; ++wo)
+      {
+	dl += (jv ? be : rx);
+	rx += ((lb == 0) + 1);
+      }
+}

	Jakub

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [PATCH] Fix compile time complexity of search_type_for_mask (PR tree-optimization/72866)
  2016-08-29 19:57 [PATCH] Fix compile time complexity of search_type_for_mask (PR tree-optimization/72866) Jakub Jelinek
@ 2016-08-30  4:42 ` Richard Biener
  0 siblings, 0 replies; 2+ messages in thread
From: Richard Biener @ 2016-08-30  4:42 UTC (permalink / raw)
  To: Jakub Jelinek; +Cc: gcc-patches

On August 29, 2016 9:57:00 PM GMT+02:00, Jakub Jelinek <jakub@redhat.com> wrote:
>Hi!
>
>As the testcase shows, with deep chains of COND_EXPRs with the same
>SSA_NAMEs appearing more than once in the various operands we can
>hang in search_type_for_mask.  Similar check_bool_pattern function
>uses a hash_set to handle stmts that have been checked already, this
>patch
>uses a hash_map in which we cache the return values once we've handled
>some
>stmt already.
>
>Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

OK.

Thanks,
Richard.

>2016-08-29  Jakub Jelinek  <jakub@redhat.com>
>
>	PR tree-optimization/72866
>	* tree-vect-patterns.c (search_type_for_mask): Turn into
>	a small wrapper, move all code to ...
>	(search_type_for_mask_1): ... this new function.  Add caching
>	and adjust recursive calls.
>
>	* gcc.dg/vect/pr72866.c: New test.
>
>--- gcc/tree-vect-patterns.c.jj	2016-07-21 08:59:55.000000000 +0200
>+++ gcc/tree-vect-patterns.c	2016-08-29 19:26:12.807164641 +0200
>@@ -3459,13 +3459,11 @@ adjust_bool_stmts (hash_set <gimple *> &
>   return gimple_assign_lhs (pattern_stmt);
> }
> 
>-/* Return the proper type for converting bool VAR into
>-   an integer value or NULL_TREE if no such type exists.
>-   The type is chosen so that converted value has the
>-   same number of elements as VAR's vector type.  */
>+/* Helper for search_type_for_mask.  */
> 
> static tree
>-search_type_for_mask (tree var, vec_info *vinfo)
>+search_type_for_mask_1 (tree var, vec_info *vinfo,
>+			hash_map<gimple *, tree> &cache)
> {
>   gimple *def_stmt;
>   enum vect_def_type dt;
>@@ -3490,6 +3488,10 @@ search_type_for_mask (tree var, vec_info
>   if (!is_gimple_assign (def_stmt))
>     return NULL_TREE;
> 
>+  tree *c = cache.get (def_stmt);
>+  if (c)
>+    return *c;
>+
>   rhs_code = gimple_assign_rhs_code (def_stmt);
>   rhs1 = gimple_assign_rhs1 (def_stmt);
> 
>@@ -3498,14 +3500,15 @@ search_type_for_mask (tree var, vec_info
>     case SSA_NAME:
>     case BIT_NOT_EXPR:
>     CASE_CONVERT:
>-      res = search_type_for_mask (rhs1, vinfo);
>+      res = search_type_for_mask_1 (rhs1, vinfo, cache);
>       break;
> 
>     case BIT_AND_EXPR:
>     case BIT_IOR_EXPR:
>     case BIT_XOR_EXPR:
>-      res = search_type_for_mask (rhs1, vinfo);
>-      res2 = search_type_for_mask (gimple_assign_rhs2 (def_stmt),
>vinfo);
>+      res = search_type_for_mask_1 (rhs1, vinfo, cache);
>+      res2 = search_type_for_mask_1 (gimple_assign_rhs2 (def_stmt),
>vinfo,
>+				     cache);
>    if (!res || (res2 && TYPE_PRECISION (res) > TYPE_PRECISION (res2)))
> 	res = res2;
>       break;
>@@ -3517,8 +3520,9 @@ search_type_for_mask (tree var, vec_info
> 
> 	  if (TREE_CODE (TREE_TYPE (rhs1)) == BOOLEAN_TYPE)
> 	    {
>-	      res = search_type_for_mask (rhs1, vinfo);
>-	      res2 = search_type_for_mask (gimple_assign_rhs2 (def_stmt),
>vinfo);
>+	      res = search_type_for_mask_1 (rhs1, vinfo, cache);
>+	      res2 = search_type_for_mask_1 (gimple_assign_rhs2 (def_stmt),
>+					     vinfo, cache);
>	      if (!res || (res2 && TYPE_PRECISION (res) > TYPE_PRECISION
>(res2)))
> 		res = res2;
> 	      break;
>@@ -3526,12 +3530,18 @@ search_type_for_mask (tree var, vec_info
> 
> 	  comp_vectype = get_vectype_for_scalar_type (TREE_TYPE (rhs1));
> 	  if (comp_vectype == NULL_TREE)
>-	    return NULL_TREE;
>+	    {
>+	      res = NULL_TREE;
>+	      break;
>+	    }
> 
> 	  mask_type = get_mask_type_for_scalar_type (TREE_TYPE (rhs1));
> 	  if (!mask_type
> 	      || !expand_vec_cmp_expr_p (comp_vectype, mask_type))
>-	    return NULL_TREE;
>+	    {
>+	      res = NULL_TREE;
>+	      break;
>+	    }
> 
> 	  if (TREE_CODE (TREE_TYPE (rhs1)) != INTEGER_TYPE
> 	      || !TYPE_UNSIGNED (TREE_TYPE (rhs1)))
>@@ -3544,9 +3554,21 @@ search_type_for_mask (tree var, vec_info
> 	}
>     }
> 
>+  cache.put (def_stmt, res);
>   return res;
> }
> 
>+/* Return the proper type for converting bool VAR into
>+   an integer value or NULL_TREE if no such type exists.
>+   The type is chosen so that converted value has the
>+   same number of elements as VAR's vector type.  */
>+
>+static tree
>+search_type_for_mask (tree var, vec_info *vinfo)
>+{
>+  hash_map<gimple *, tree> cache;
>+  return search_type_for_mask_1 (var, vinfo, cache);
>+}
> 
> /* Function vect_recog_bool_pattern
> 
>--- gcc/testsuite/gcc.dg/vect/pr72866.c.jj	2016-08-29
>19:28:22.453646025 +0200
>+++ gcc/testsuite/gcc.dg/vect/pr72866.c	2016-08-29 19:28:17.000000000
>+0200
>@@ -0,0 +1,19 @@
>+/* PR tree-optimization/72866 */
>+/* { dg-do compile } */
>+
>+unsigned int dl;
>+int rx, lb;
>+
>+void
>+fo (int jv, int be)
>+{
>+  const unsigned int xw = 16;
>+  unsigned int ya, wo;
>+
>+  for (ya = 0; ya < 2; ++ya)
>+    for (wo = 0; wo < xw; ++wo)
>+      {
>+	dl += (jv ? be : rx);
>+	rx += ((lb == 0) + 1);
>+      }
>+}
>
>	Jakub


^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2016-08-30  4:42 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-08-29 19:57 [PATCH] Fix compile time complexity of search_type_for_mask (PR tree-optimization/72866) Jakub Jelinek
2016-08-30  4:42 ` Richard Biener

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).