7070134: Hotspot crashes with sigsegv from PorterStemmer

Summary: Do not move data nodes which are attached to a predicate test to a dominating test.
Reviewed-by: never

src/share/vm/opto/ifnode.cpp

@@ -1009,6 +1009,13 @@ void IfNode::dominated_by( Node *prev_dom, PhaseIterGVN *igvn ) {
   int prev_op = prev_dom->Opcode();
   Node *top = igvn->C->top(); // Shortcut to top
 
+  // Loop predicates may have depending checks which should not
+  // be skipped. For example, range check predicate has two checks
+  // for lower and upper bounds.
+  ProjNode* unc_proj = proj_out(1 - prev_dom->as_Proj()->_con)->as_Proj();
+  if (PhaseIdealLoop::is_uncommon_trap_proj(unc_proj, Deoptimization::Reason_predicate))
+    prev_dom = idom;
+
   // Now walk the current IfNode's projections.
   // Loop ends when 'this' has no more uses.
   for (DUIterator_Last imin, i = last_outs(imin); i >= imin; --i) {
@@ -1019,9 +1026,9 @@ void IfNode::dominated_by( Node *prev_dom, PhaseIterGVN *igvn ) {
     // or TOP if the dominating projection is of opposite type.
     // Data-target will be used as the new control edge for the non-CFG
     // nodes like Casts and Loads.
-    Node *data_target = (ifp->Opcode() == prev_op ) ? prev_dom : top;
+    Node *data_target = (ifp->Opcode() == prev_op) ? prev_dom : top;
     // Control-target is just the If's immediate dominator or TOP.
-    Node *ctrl_target = (ifp->Opcode() == prev_op ) ?     idom : top;
+    Node *ctrl_target = (ifp->Opcode() == prev_op) ?     idom : top;
 
     // For each child of an IfTrue/IfFalse projection, reroute.
     // Loop ends when projection has no more uses.

src/share/vm/opto/loopPredicate.cpp

@@ -817,6 +817,10 @@ bool PhaseIdealLoop::loop_predication_impl(IdealLoopTree *loop) {
     cl = head->as_CountedLoop();
     // do nothing for iteration-splitted loops
     if (!cl->is_normal_loop()) return false;
+    // Avoid RCE if Counted loop's test is '!='.
+    BoolTest::mask bt = cl->loopexit()->test_trip();
+    if (bt != BoolTest::lt && bt != BoolTest::gt)
+      cl = NULL;
   }
 
   Node* entry = head->in(LoopNode::EntryControl);
@@ -958,7 +962,7 @@ bool PhaseIdealLoop::loop_predication_impl(IdealLoopTree *loop) {
       if (TraceLoopPredicate) tty->print_cr("lower bound check if: %d", lower_bound_iff->_idx);
 
       // Test the upper bound
-      Node* upper_bound_bol = rc_predicate(loop, ctrl, scale, offset, init, limit, stride, rng, true);
+      Node* upper_bound_bol = rc_predicate(loop, lower_bound_proj, scale, offset, init, limit, stride, rng, true);
       IfNode* upper_bound_iff = upper_bound_proj->in(0)->as_If();
       _igvn.hash_delete(upper_bound_iff);
       upper_bound_iff->set_req(1, upper_bound_bol);

src/share/vm/opto/loopnode.hpp

@@ -1009,7 +1009,7 @@ public:
   Node *has_local_phi_input( Node *n );
   // Mark an IfNode as being dominated by a prior test,
   // without actually altering the CFG (and hence IDOM info).
-  void dominated_by( Node *prevdom, Node *iff, bool flip = false );
+  void dominated_by( Node *prevdom, Node *iff, bool flip = false, bool exclude_loop_predicate = false );
 
   // Split Node 'n' through merge point
   Node *split_thru_region( Node *n, Node *region );

src/share/vm/opto/loopopts.cpp

@@ -194,7 +194,7 @@ Node *PhaseIdealLoop::split_thru_phi( Node *n, Node *region, int policy ) {
 // Replace the dominated test with an obvious true or false.  Place it on the
 // IGVN worklist for later cleanup.  Move control-dependent data Nodes on the
 // live path up to the dominating control.
-void PhaseIdealLoop::dominated_by( Node *prevdom, Node *iff, bool flip ) {
+void PhaseIdealLoop::dominated_by( Node *prevdom, Node *iff, bool flip, bool exclude_loop_predicate ) {
 #ifndef PRODUCT
   if (VerifyLoopOptimizations && PrintOpto) tty->print_cr("dominating test");
 #endif
@@ -228,7 +228,16 @@ void PhaseIdealLoop::dominated_by( Node *prevdom, Node *iff, bool flip ) {
   // Make control-dependent data Nodes on the live path (path that will remain
   // once the dominated IF is removed) become control-dependent on the
   // dominating projection.
-  Node* dp = ((IfNode*)iff)->proj_out(pop == Op_IfTrue);
+  Node* dp = iff->as_If()->proj_out(pop == Op_IfTrue);
+
+  // Loop predicates may have depending checks which should not
+  // be skipped. For example, range check predicate has two checks
+  // for lower and upper bounds.
+  ProjNode* unc_proj = iff->as_If()->proj_out(1 - dp->as_Proj()->_con)->as_Proj();
+  if (exclude_loop_predicate &&
+      is_uncommon_trap_proj(unc_proj, Deoptimization::Reason_predicate))
+    return; // Let IGVN transformation change control dependence.
+
   IdealLoopTree *old_loop = get_loop(dp);
 
   for (DUIterator_Fast imax, i = dp->fast_outs(imax); i < imax; i++) {
@@ -859,7 +868,7 @@ void PhaseIdealLoop::split_if_with_blocks_post( Node *n ) {
           // Replace the dominated test with an obvious true or false.
           // Place it on the IGVN worklist for later cleanup.
           C->set_major_progress();
-          dominated_by( prevdom, n );
+          dominated_by( prevdom, n, false, true );
 #ifndef PRODUCT
           if( VerifyLoopOptimizations ) verify();
 #endif

test/compiler/7070134/Stemmer.java

+/**
+ * @test
+ * @bug 7070134
+ * @summary Hotspot crashes with sigsegv from PorterStemmer
+ *
+ * @run shell Test7070134.sh
+ */
+
+/*
+
+   Porter stemmer in Java. The original paper is in
+
+       Porter, 1980, An algorithm for suffix stripping, Program, Vol. 14,
+       no. 3, pp 130-137,
+
+   See also http://www.tartarus.org/~martin/PorterStemmer
+
+   History:
+
+   Release 1
+
+   Bug 1 (reported by Gonzalo Parra 16/10/99) fixed as marked below.
+   The words 'aed', 'eed', 'oed' leave k at 'a' for step 3, and b[k-1]
+   is then out outside the bounds of b.
+
+   Release 2
+
+   Similarly,
+
+   Bug 2 (reported by Steve Dyrdahl 22/2/00) fixed as marked below.
+   'ion' by itself leaves j = -1 in the test for 'ion' in step 5, and
+   b[j] is then outside the bounds of b.
+
+   Release 3
+
+   Considerably revised 4/9/00 in the light of many helpful suggestions
+   from Brian Goetz of Quiotix Corporation (brian@quiotix.com).
+
+   Release 4
+
+*/
+
+import java.io.*;
+
+/**
+  * Stemmer, implementing the Porter Stemming Algorithm
+  *
+  * The Stemmer class transforms a word into its root form.  The input
+  * word can be provided a character at time (by calling add()), or at once
+  * by calling one of the various stem(something) methods.
+  */
+
+class Stemmer
+{  private char[] b;
+   private int i,     /* offset into b */
+               i_end, /* offset to end of stemmed word */
+               j, k;
+   private static final int INC = 50;
+                     /* unit of size whereby b is increased */
+   public Stemmer()
+   {  b = new char[INC];
+      i = 0;
+      i_end = 0;
+   }
+
+   /**
+    * Add a character to the word being stemmed.  When you are finished
+    * adding characters, you can call stem(void) to stem the word.
+    */
+
+   public void add(char ch)
+   {  if (i == b.length)
+      {  char[] new_b = new char[i+INC];
+         for (int c = 0; c < i; c++) new_b[c] = b[c];
+         b = new_b;
+      }
+      b[i++] = ch;
+   }
+
+
+   /** Adds wLen characters to the word being stemmed contained in a portion
+    * of a char[] array. This is like repeated calls of add(char ch), but
+    * faster.
+    */
+
+   public void add(char[] w, int wLen)
+   {  if (i+wLen >= b.length)
+      {  char[] new_b = new char[i+wLen+INC];
+         for (int c = 0; c < i; c++) new_b[c] = b[c];
+         b = new_b;
+      }
+      for (int c = 0; c < wLen; c++) b[i++] = w[c];
+   }
+
+   /**
+    * After a word has been stemmed, it can be retrieved by toString(),
+    * or a reference to the internal buffer can be retrieved by getResultBuffer
+    * and getResultLength (which is generally more efficient.)
+    */
+   public String toString() { return new String(b,0,i_end); }
+
+   /**
+    * Returns the length of the word resulting from the stemming process.
+    */
+   public int getResultLength() { return i_end; }
+
+   /**
+    * Returns a reference to a character buffer containing the results of
+    * the stemming process.  You also need to consult getResultLength()
+    * to determine the length of the result.
+    */
+   public char[] getResultBuffer() { return b; }
+
+   /* cons(i) is true <=> b[i] is a consonant. */
+
+   private final boolean cons(int i)
+   {  switch (b[i])
+      {  case 'a': case 'e': case 'i': case 'o': case 'u': return false;
+         case 'y': return (i==0) ? true : !cons(i-1);
+         default: return true;
+      }
+   }
+
+   /* m() measures the number of consonant sequences between 0 and j. if c is
+      a consonant sequence and v a vowel sequence, and <..> indicates arbitrary
+      presence,
+
+         <c><v>       gives 0
+         <c>vc<v>     gives 1
+         <c>vcvc<v>   gives 2
+         <c>vcvcvc<v> gives 3
+         ....
+   */
+
+   private final int m()
+   {  int n = 0;
+      int i = 0;
+      while(true)
+      {  if (i > j) return n;
+         if (! cons(i)) break; i++;
+      }
+      i++;
+      while(true)
+      {  while(true)
+         {  if (i > j) return n;
+               if (cons(i)) break;
+               i++;
+         }
+         i++;
+         n++;
+         while(true)
+         {  if (i > j) return n;
+            if (! cons(i)) break;
+            i++;
+         }
+         i++;
+       }
+   }
+
+   /* vowelinstem() is true <=> 0,...j contains a vowel */
+
+   private final boolean vowelinstem()
+   {  int i; for (i = 0; i <= j; i++) if (! cons(i)) return true;
+      return false;
+   }
+
+   /* doublec(j) is true <=> j,(j-1) contain a double consonant. */
+
+   private final boolean doublec(int j)
+   {  if (j < 1) return false;
+      if (b[j] != b[j-1]) return false;
+      return cons(j);
+   }
+
+   /* cvc(i) is true <=> i-2,i-1,i has the form consonant - vowel - consonant
+      and also if the second c is not w,x or y. this is used when trying to
+      restore an e at the end of a short word. e.g.
+
+         cav(e), lov(e), hop(e), crim(e), but
+         snow, box, tray.
+
+   */
+
+   private final boolean cvc(int i)
+   {  if (i < 2 || !cons(i) || cons(i-1) || !cons(i-2)) return false;
+      {  int ch = b[i];
+         if (ch == 'w' || ch == 'x' || ch == 'y') return false;
+      }
+      return true;
+   }
+
+   private final boolean ends(String s)
+   {  int l = s.length();
+      int o = k-l+1;
+      if (o < 0) return false;
+      for (int i = 0; i < l; i++) if (b[o+i] != s.charAt(i)) return false;
+      j = k-l;
+      return true;
+   }
+
+   /* setto(s) sets (j+1),...k to the characters in the string s, readjusting
+      k. */
+
+   private final void setto(String s)
+   {  int l = s.length();
+      int o = j+1;
+      for (int i = 0; i < l; i++) b[o+i] = s.charAt(i);
+      k = j+l;
+   }
+
+   /* r(s) is used further down. */
+
+   private final void r(String s) { if (m() > 0) setto(s); }
+
+   /* step1() gets rid of plurals and -ed or -ing. e.g.
+
+          caresses  ->  caress
+          ponies    ->  poni
+          ties      ->  ti
+          caress    ->  caress
+          cats      ->  cat
+
+          feed      ->  feed
+          agreed    ->  agree
+          disabled  ->  disable
+
+          matting   ->  mat
+          mating    ->  mate
+          meeting   ->  meet
+          milling   ->  mill
+          messing   ->  mess
+
+          meetings  ->  meet
+
+   */
+
+   private final void step1()
+   {  if (b[k] == 's')
+      {  if (ends("sses")) k -= 2; else
+         if (ends("ies")) setto("i"); else
+         if (b[k-1] != 's') k--;
+      }
+      if (ends("eed")) { if (m() > 0) k--; } else
+      if ((ends("ed") || ends("ing")) && vowelinstem())
+      {  k = j;
+         if (ends("at")) setto("ate"); else
+         if (ends("bl")) setto("ble"); else
+         if (ends("iz")) setto("ize"); else
+         if (doublec(k))
+         {  k--;
+            {  int ch = b[k];
+               if (ch == 'l' || ch == 's' || ch == 'z') k++;
+            }
+         }
+         else if (m() == 1 && cvc(k)) setto("e");
+     }
+   }
+
+   /* step2() turns terminal y to i when there is another vowel in the stem. */
+
+   private final void step2() { if (ends("y") && vowelinstem()) b[k] = 'i'; }
+
+   /* step3() maps double suffices to single ones. so -ization ( = -ize plus
+      -ation) maps to -ize etc. note that the string before the suffix must give
+      m() > 0. */
+
+   private final void step3() { if (k == 0) return; /* For Bug 1 */ switch (b[k-1])
+   {
+       case 'a': if (ends("ational")) { r("ate"); break; }
+                 if (ends("tional")) { r("tion"); break; }
+                 break;
+       case 'c': if (ends("enci")) { r("ence"); break; }
+                 if (ends("anci")) { r("ance"); break; }
+                 break;
+       case 'e': if (ends("izer")) { r("ize"); break; }
+                 break;
+       case 'l': if (ends("bli")) { r("ble"); break; }
+                 if (ends("alli")) { r("al"); break; }
+                 if (ends("entli")) { r("ent"); break; }
+                 if (ends("eli")) { r("e"); break; }
+                 if (ends("ousli")) { r("ous"); break; }
+                 break;
+       case 'o': if (ends("ization")) { r("ize"); break; }
+                 if (ends("ation")) { r("ate"); break; }
+                 if (ends("ator")) { r("ate"); break; }
+                 break;
+       case 's': if (ends("alism")) { r("al"); break; }
+                 if (ends("iveness")) { r("ive"); break; }
+                 if (ends("fulness")) { r("ful"); break; }
+                 if (ends("ousness")) { r("ous"); break; }
+                 break;
+       case 't': if (ends("aliti")) { r("al"); break; }
+                 if (ends("iviti")) { r("ive"); break; }
+                 if (ends("biliti")) { r("ble"); break; }
+                 break;
+       case 'g': if (ends("logi")) { r("log"); break; }
+   } }
+
+   /* step4() deals with -ic-, -full, -ness etc. similar strategy to step3. */
+
+   private final void step4() { switch (b[k])
+   {
+       case 'e': if (ends("icate")) { r("ic"); break; }
+                 if (ends("ative")) { r(""); break; }
+                 if (ends("alize")) { r("al"); break; }
+                 break;
+       case 'i': if (ends("iciti")) { r("ic"); break; }
+                 break;
+       case 'l': if (ends("ical")) { r("ic"); break; }
+                 if (ends("ful")) { r(""); break; }
+                 break;
+       case 's': if (ends("ness")) { r(""); break; }
+                 break;
+   } }
+
+   /* step5() takes off -ant, -ence etc., in context <c>vcvc<v>. */
+
+   private final void step5()
+   {   if (k == 0) return; /* for Bug 1 */ switch (b[k-1])
+       {  case 'a': if (ends("al")) break; return;
+          case 'c': if (ends("ance")) break;
+                    if (ends("ence")) break; return;
+          case 'e': if (ends("er")) break; return;
+          case 'i': if (ends("ic")) break; return;
+          case 'l': if (ends("able")) break;
+                    if (ends("ible")) break; return;
+          case 'n': if (ends("ant")) break;
+                    if (ends("ement")) break;
+                    if (ends("ment")) break;
+                    /* element etc. not stripped before the m */
+                    if (ends("ent")) break; return;
+          case 'o': if (ends("ion") && j >= 0 && (b[j] == 's' || b[j] == 't')) break;
+                                    /* j >= 0 fixes Bug 2 */
+                    if (ends("ou")) break; return;
+                    /* takes care of -ous */
+          case 's': if (ends("ism")) break; return;
+          case 't': if (ends("ate")) break;
+                    if (ends("iti")) break; return;
+          case 'u': if (ends("ous")) break; return;
+          case 'v': if (ends("ive")) break; return;
+          case 'z': if (ends("ize")) break; return;
+          default: return;
+       }
+       if (m() > 1) k = j;
+   }
+
+   /* step6() removes a final -e if m() > 1. */
+
+   private final void step6()
+   {  j = k;
+      if (b[k] == 'e')
+      {  int a = m();
+         if (a > 1 || a == 1 && !cvc(k-1)) k--;
+      }
+      if (b[k] == 'l' && doublec(k) && m() > 1) k--;
+   }
+
+   /** Stem the word placed into the Stemmer buffer through calls to add().
+    * Returns true if the stemming process resulted in a word different
+    * from the input.  You can retrieve the result with
+    * getResultLength()/getResultBuffer() or toString().
+    */
+   public void stem()
+   {  k = i - 1;
+      if (k > 1) { step1(); step2(); step3(); step4(); step5(); step6(); }
+      i_end = k+1; i = 0;
+   }
+
+   /** Test program for demonstrating the Stemmer.  It reads text from a
+    * a list of files, stems each word, and writes the result to standard
+    * output. Note that the word stemmed is expected to be in lower case:
+    * forcing lower case must be done outside the Stemmer class.
+    * Usage: Stemmer file-name file-name ...
+    */
+   public static void main(String[] args)
+   {
+      char[] w = new char[501];
+      Stemmer s = new Stemmer();
+      for (int i = 0; i < args.length; i++)
+      try
+      {
+         FileInputStream in = new FileInputStream(args[i]);
+
+         try
+         { while(true)
+
+           {  int ch = in.read();
+              if (Character.isLetter((char) ch))
+              {
+                 int j = 0;
+                 while(true)
+                 {  ch = Character.toLowerCase((char) ch);
+                    w[j] = (char) ch;
+                    if (j < 500) j++;
+                    ch = in.read();
+                    if (!Character.isLetter((char) ch))
+                    {
+                       /* to test add(char ch) */
+                       for (int c = 0; c < j; c++) s.add(w[c]);
+
+                       /* or, to test add(char[] w, int j) */
+                       /* s.add(w, j); */
+
+                       s.stem();
+                       {  String u;
+
+                          /* and now, to test toString() : */
+                          u = s.toString();
+
+                          /* to test getResultBuffer(), getResultLength() : */
+                          /* u = new String(s.getResultBuffer(), 0, s.getResultLength()); */
+
+                          System.out.print(u);
+                       }
+                       break;
+                    }
+                 }
+              }
+              if (ch < 0) break;
+              System.out.print((char)ch);
+           }
+         }
+         catch (IOException e)
+         {  System.out.println("error reading " + args[i]);
+            break;
+         }
+      }
+      catch (FileNotFoundException e)
+      {  System.out.println("file " + args[i] + " not found");
+         break;
+      }
+   }
+}

test/compiler/7070134/Test7070134.sh

+#!/bin/sh
+# 
+# Copyright (c) 2011, Oracle and/or its affiliates. All rights reserved.
+# DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+# 
+# This code is free software; you can redistribute it and/or modify it
+# under the terms of the GNU General Public License version 2 only, as
+# published by the Free Software Foundation.
+# 
+# This code is distributed in the hope that it will be useful, but WITHOUT
+# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+# FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+# version 2 for more details (a copy is included in the LICENSE file that
+# accompanied this code).
+# 
+# You should have received a copy of the GNU General Public License version
+# 2 along with this work; if not, write to the Free Software Foundation,
+# Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+# 
+# Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+# or visit www.oracle.com if you need additional information or have any
+# questions.
+# 
+# 
+
+if [ "${TESTSRC}" = "" ]
+then
+  echo "TESTSRC not set.  Test cannot execute.  Failed."
+  exit 1
+fi
+echo "TESTSRC=${TESTSRC}"
+if [ "${TESTJAVA}" = "" ]
+then
+  echo "TESTJAVA not set.  Test cannot execute.  Failed."
+  exit 1
+fi
+echo "TESTJAVA=${TESTJAVA}"
+if [ "${TESTCLASSES}" = "" ]
+then
+  echo "TESTCLASSES not set.  Test cannot execute.  Failed."
+  exit 1
+fi
+echo "TESTCLASSES=${TESTCLASSES}"
+echo "CLASSPATH=${CLASSPATH}"
+
+set -x
+
+cp ${TESTSRC}/Stemmer.java .
+cp ${TESTSRC}/words .
+
+${TESTJAVA}/bin/javac -d . Stemmer.java
+
+${TESTJAVA}/bin/java ${TESTVMOPTS} -Xbatch Stemmer words > test.out 2>&1
+
+exit $?
+