public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc r13-5986] Improve VN PHI hash table handling
@ 2023-02-14 15:19 Richard Biener
  0 siblings, 0 replies; only message in thread
From: Richard Biener @ 2023-02-14 15:19 UTC (permalink / raw)
  To: gcc-cvs

https://gcc.gnu.org/g:e72c2770b6d913c82a56d45a306c4cb2ed88fba5

commit r13-5986-ge72c2770b6d913c82a56d45a306c4cb2ed88fba5
Author: Richard Biener <rguenther@suse.de>
Date:   Tue Feb 14 14:51:48 2023 +0100

    Improve VN PHI hash table handling
    
    The hash function of PHIs is weak since we want to be able to CSE
    them even across basic-blocks in some cases.  The following avoids
    weakening the hash for cases we are never going to CSE, reducing
    the number of collisions and avoiding redundant work in the
    hash and equality functions.
    
            * tree-ssa-sccvn.cc (vn_phi_compute_hash): Key skipping
            basic block index hashing on the availability of ->cclhs.
            (vn_phi_eq): Avoid re-doing sanity checks for CSE but
            rely on ->cclhs availability.
            (vn_phi_lookup): Set ->cclhs only when we are eventually
            going to CSE the PHI.
            (vn_phi_insert): Likewise.

Diff:
---
 gcc/tree-ssa-sccvn.cc | 77 ++++++++++++++++++++++++++++-----------------------
 1 file changed, 43 insertions(+), 34 deletions(-)

diff --git a/gcc/tree-ssa-sccvn.cc b/gcc/tree-ssa-sccvn.cc
index e5bb278196a..8ee77fd2b78 100644
--- a/gcc/tree-ssa-sccvn.cc
+++ b/gcc/tree-ssa-sccvn.cc
@@ -4629,9 +4629,9 @@ vn_phi_compute_hash (vn_phi_t vp1)
     case 1:
       break;
     case 2:
-      if (vp1->block->loop_father->header == vp1->block)
-	;
-      else
+      /* When this is a PHI node subject to CSE for different blocks
+	 avoid hashing the block index.  */
+      if (vp1->cclhs)
 	break;
       /* Fallthru.  */
     default:
@@ -4715,32 +4715,33 @@ vn_phi_eq (const_vn_phi_t const vp1, const_vn_phi_t const vp2)
 
 	case 2:
 	  {
-	    /* Rule out backedges into the PHI.  */
-	    if (vp1->block->loop_father->header == vp1->block
-		|| vp2->block->loop_father->header == vp2->block)
+	    /* Make sure both PHIs are classified as CSEable.  */
+	    if (! vp1->cclhs || ! vp2->cclhs)
 	      return false;
 
+	    /* Rule out backedges into the PHI.  */
+	    gcc_checking_assert
+	      (vp1->block->loop_father->header != vp1->block
+	       && vp2->block->loop_father->header != vp2->block);
+
 	    /* If the PHI nodes do not have compatible types
 	       they are not the same.  */
 	    if (!types_compatible_p (vp1->type, vp2->type))
 	      return false;
 
+	    /* If the immediate dominator end in switch stmts multiple
+	       values may end up in the same PHI arg via intermediate
+	       CFG merges.  */
 	    basic_block idom1
 	      = get_immediate_dominator (CDI_DOMINATORS, vp1->block);
 	    basic_block idom2
 	      = get_immediate_dominator (CDI_DOMINATORS, vp2->block);
-	    /* If the immediate dominator end in switch stmts multiple
-	       values may end up in the same PHI arg via intermediate
-	       CFG merges.  */
-	    if (EDGE_COUNT (idom1->succs) != 2
-		|| EDGE_COUNT (idom2->succs) != 2)
-	      return false;
+	    gcc_checking_assert (EDGE_COUNT (idom1->succs) == 2
+				 && EDGE_COUNT (idom2->succs) == 2);
 
 	    /* Verify the controlling stmt is the same.  */
-	    gcond *last1 = safe_dyn_cast <gcond *> (last_stmt (idom1));
-	    gcond *last2 = safe_dyn_cast <gcond *> (last_stmt (idom2));
-	    if (! last1 || ! last2)
-	      return false;
+	    gcond *last1 = as_a <gcond *> (last_stmt (idom1));
+	    gcond *last2 = as_a <gcond *> (last_stmt (idom2));
 	    bool inverted_p;
 	    if (! cond_stmts_equal_p (last1, vp1->cclhs, vp1->ccrhs,
 				      last2, vp2->cclhs, vp2->ccrhs,
@@ -4835,15 +4836,19 @@ vn_phi_lookup (gimple *phi, bool backedges_varying_p)
   /* Extract values of the controlling condition.  */
   vp1->cclhs = NULL_TREE;
   vp1->ccrhs = NULL_TREE;
-  basic_block idom1 = get_immediate_dominator (CDI_DOMINATORS, vp1->block);
-  if (EDGE_COUNT (idom1->succs) == 2)
-    if (gcond *last1 = safe_dyn_cast <gcond *> (last_stmt (idom1)))
-      {
-	/* ???  We want to use SSA_VAL here.  But possibly not
-	   allow VN_TOP.  */
-	vp1->cclhs = vn_valueize (gimple_cond_lhs (last1));
-	vp1->ccrhs = vn_valueize (gimple_cond_rhs (last1));
-      }
+  if (EDGE_COUNT (vp1->block->preds) == 2
+      && vp1->block->loop_father->header != vp1->block)
+    {
+      basic_block idom1 = get_immediate_dominator (CDI_DOMINATORS, vp1->block);
+      if (EDGE_COUNT (idom1->succs) == 2)
+	if (gcond *last1 = safe_dyn_cast <gcond *> (last_stmt (idom1)))
+	  {
+	    /* ???  We want to use SSA_VAL here.  But possibly not
+	       allow VN_TOP.  */
+	    vp1->cclhs = vn_valueize (gimple_cond_lhs (last1));
+	    vp1->ccrhs = vn_valueize (gimple_cond_rhs (last1));
+	  }
+    }
   vp1->hashcode = vn_phi_compute_hash (vp1);
   slot = valid_info->phis->find_slot_with_hash (vp1, vp1->hashcode, NO_INSERT);
   if (!slot)
@@ -4885,15 +4890,19 @@ vn_phi_insert (gimple *phi, tree result, bool backedges_varying_p)
   /* Extract values of the controlling condition.  */
   vp1->cclhs = NULL_TREE;
   vp1->ccrhs = NULL_TREE;
-  basic_block idom1 = get_immediate_dominator (CDI_DOMINATORS, vp1->block);
-  if (EDGE_COUNT (idom1->succs) == 2)
-    if (gcond *last1 = safe_dyn_cast <gcond *> (last_stmt (idom1)))
-      {
-	/* ???  We want to use SSA_VAL here.  But possibly not
-	   allow VN_TOP.  */
-	vp1->cclhs = vn_valueize (gimple_cond_lhs (last1));
-	vp1->ccrhs = vn_valueize (gimple_cond_rhs (last1));
-      }
+  if (EDGE_COUNT (vp1->block->preds) == 2
+      && vp1->block->loop_father->header != vp1->block)
+    {
+      basic_block idom1 = get_immediate_dominator (CDI_DOMINATORS, vp1->block);
+      if (EDGE_COUNT (idom1->succs) == 2)
+	if (gcond *last1 = safe_dyn_cast <gcond *> (last_stmt (idom1)))
+	  {
+	    /* ???  We want to use SSA_VAL here.  But possibly not
+	       allow VN_TOP.  */
+	    vp1->cclhs = vn_valueize (gimple_cond_lhs (last1));
+	    vp1->ccrhs = vn_valueize (gimple_cond_rhs (last1));
+	  }
+    }
   vp1->result = result;
   vp1->hashcode = vn_phi_compute_hash (vp1);

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

only message in thread, other threads:[~2023-02-14 15:19 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-02-14 15:19 [gcc r13-5986] Improve VN PHI hash table handling 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).