Merge

hotspot/agent/make/Makefile

@@ -257,7 +257,7 @@ RMIC = $(JDK_HOME)/bin/rmic
 all: filelist
 	@mkdir -p $(OUTPUT_DIR)
 	@echo "$(SA_BUILD_VERSION_PROP)" > $(SA_PROPERTIES)
-	$(JAVAC) -source 1.4 -classpath $(CLASSPATH) -deprecation -sourcepath $(SRC_DIR) -g -d $(OUTPUT_DIR) @filelist
+	$(JAVAC) -classpath $(CLASSPATH) -deprecation -sourcepath $(SRC_DIR) -g -d $(OUTPUT_DIR) @filelist
 	$(RMIC) -classpath $(OUTPUT_DIR) -d $(OUTPUT_DIR) sun.jvm.hotspot.debugger.remote.RemoteDebuggerServer
 	rm -f $(OUTPUT_DIR)/sun/jvm/hotspot/utilities/soql/sa.js
 	cp $(SRC_DIR)/sun/jvm/hotspot/utilities/soql/sa.js $(OUTPUT_DIR)/sun/jvm/hotspot/utilities/soql
@@ -269,7 +269,7 @@ all: filelist
 allprof: filelist
 	@mkdir -p $(OUTPUT_DIR)
 	@echo "$(SA_BUILD_VERSION_PROP)" > $(SA_PROPERTIES)
-	$(JAVAC) -source 1.4 -J-Xprof -classpath $(CLASSPATH) -deprecation -sourcepath $(SRC_DIR) -g -d $(OUTPUT_DIR) @filelist
+	$(JAVAC) -J-Xprof -classpath $(CLASSPATH) -deprecation -sourcepath $(SRC_DIR) -g -d $(OUTPUT_DIR) @filelist
 	$(RMIC) -classpath $(OUTPUT_DIR) -d $(OUTPUT_DIR) sun.jvm.hotspot.debugger.remote.RemoteDebuggerServer
 	rm -f $(OUTPUT_DIR)/sun/jvm/hotspot/utilities/soql/sa.js
 	cp $(SRC_DIR)/sun/jvm/hotspot/utilities/soql/sa.js $(OUTPUT_DIR)/sun/jvm/hotspot/utilities/soql

hotspot/agent/src/os/solaris/proc/libproc.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2002, 2003, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2002, 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
@@ -420,7 +420,22 @@ extern uintptr_t Ppltdest(struct ps_prochandle *, uintptr_t, int *);
 /*
  * Stack frame iteration interface.
  */
+#ifdef SOLARIS_11_B159_OR_LATER
+/* building on Nevada-B159 or later so define the new callback */
+typedef int proc_stack_f(
+    void *,             /* the cookie given to Pstack_iter() */
+    const prgregset_t,  /* the frame's registers */
+    uint_t,             /* argc for the frame's function */
+    const long *,       /* argv for the frame's function */
+    int,                /* bitwise flags describing the frame (see below) */
+    int);               /* a signal number */
+
+#define PR_SIGNAL_FRAME    1    /* called by a signal handler */
+#define PR_FOUND_SIGNAL    2    /* we found the corresponding signal number */
+#else
+/* building on Nevada-B158 or earlier so define the old callback */
 typedef int proc_stack_f(void *, const prgregset_t, uint_t, const long *);
+#endif
 
 extern int Pstack_iter(struct ps_prochandle *,
     const prgregset_t, proc_stack_f *, void *);

hotspot/agent/src/os/solaris/proc/salibproc.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2005, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 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
@@ -101,7 +101,23 @@ extern  int     Pstop(struct ps_prochandle *, uint_t);
 /*
  * Stack frame iteration interface.
  */
+#ifdef SOLARIS_11_B159_OR_LATER
+/* building on Nevada-B159 or later so define the new callback */
+typedef int proc_stack_f(
+    void *,             /* the cookie given to Pstack_iter() */
+    const prgregset_t,  /* the frame's registers */
+    uint_t,             /* argc for the frame's function */
+    const long *,       /* argv for the frame's function */
+    int,                /* bitwise flags describing the frame (see below) */
+    int);               /* a signal number */
+
+#define PR_SIGNAL_FRAME    1    /* called by a signal handler */
+#define PR_FOUND_SIGNAL    2    /* we found the corresponding signal number */
+#else
+/* building on Nevada-B158 or earlier so define the old callback */
 typedef int proc_stack_f(void *, const prgregset_t, uint_t, const long *);
+#endif
+
 extern int Pstack_iter(struct ps_prochandle *,
     const prgregset_t, proc_stack_f *, void *);
 

hotspot/agent/src/os/solaris/proc/saproc.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2002, 2009, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2002, 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
@@ -24,6 +24,9 @@
 
 #include "salibproc.h"
 #include "sun_jvm_hotspot_debugger_proc_ProcDebuggerLocal.h"
+#ifndef SOLARIS_11_B159_OR_LATER
+#include <sys/utsname.h>
+#endif
 #include <thread_db.h>
 #include <strings.h>
 #include <limits.h>
@@ -40,8 +43,22 @@
 #define SYMBOL_BUF_SIZE  256
 #define ERR_MSG_SIZE     (PATH_MAX + 256)
 
-// debug mode
+// debug modes
 static int _libsaproc_debug = 0;
+#ifndef SOLARIS_11_B159_OR_LATER
+static bool _Pstack_iter_debug = false;
+
+static void dprintf_2(const char* format,...) {
+  if (_Pstack_iter_debug) {
+    va_list alist;
+
+    va_start(alist, format);
+    fputs("Pstack_iter DEBUG: ", stderr);
+    vfprintf(stderr, format, alist);
+    va_end(alist);
+  }
+}
+#endif // !SOLARIS_11_B159_OR_LATER
 
 static void print_debug(const char* format,...) {
   if (_libsaproc_debug) {
@@ -450,6 +467,7 @@ fill_load_object_list(void *cd, const prmap_t* pmp, const char* obj_name) {
   return 0;
 }
 
+// Pstack_iter() proc_stack_f callback prior to Nevada-B159
 static int
 fill_cframe_list(void *cd, const prgregset_t regs, uint_t argc, const long *argv) {
   DebuggerWith2Objects* dbgo2 = (DebuggerWith2Objects*) cd;
@@ -472,6 +490,14 @@ fill_cframe_list(void *cd, const prgregset_t regs, uint_t argc, const long *argv
   return 0;
 }
 
+// Pstack_iter() proc_stack_f callback in Nevada-B159 or later
+/*ARGSUSED*/
+static int
+wrapper_fill_cframe_list(void *cd, const prgregset_t regs, uint_t argc,
+                         const long *argv, int frame_flags, int sig) {
+  return(fill_cframe_list(cd, regs, argc, argv));
+}
+
 // part of the class sharing workaround
 
 // FIXME: !!HACK ALERT!!
@@ -970,6 +996,11 @@ JNIEXPORT void JNICALL Java_sun_jvm_hotspot_debugger_proc_ProcDebuggerLocal_fill
                    TD_THR_ANY_STATE, TD_THR_LOWEST_PRIORITY, TD_SIGNO_MASK, TD_THR_ANY_USER_FLAGS);
 }
 
+#ifndef SOLARIS_11_B159_OR_LATER
+// building on Nevada-B158 or earlier so more hoops to jump through
+static bool has_newer_Pstack_iter = false;  // older version by default
+#endif
+
 /*
  * Class:       sun_jvm_hotspot_debugger_proc_ProcDebuggerLocal
  * Method:      fillCFrameList0
@@ -997,7 +1028,24 @@ JNIEXPORT jobject JNICALL Java_sun_jvm_hotspot_debugger_proc_ProcDebuggerLocal_f
 
   env->ReleaseLongArrayElements(regsArray, ptr, JNI_ABORT);
   CHECK_EXCEPTION_(0);
-  Pstack_iter((struct ps_prochandle*) p_ps_prochandle, gregs, fill_cframe_list, &dbgo2);
+
+#ifdef SOLARIS_11_B159_OR_LATER
+  // building on Nevada-B159 or later so use the new callback
+  Pstack_iter((struct ps_prochandle*) p_ps_prochandle, gregs,
+              wrapper_fill_cframe_list, &dbgo2);
+#else
+  // building on Nevada-B158 or earlier so figure out which callback to use
+
+  if (has_newer_Pstack_iter) {
+    // Since we're building on Nevada-B158 or earlier, we have to
+    // cast wrapper_fill_cframe_list to make the compiler happy.
+    Pstack_iter((struct ps_prochandle*) p_ps_prochandle, gregs,
+                (proc_stack_f *)wrapper_fill_cframe_list, &dbgo2);
+  } else {
+    Pstack_iter((struct ps_prochandle*) p_ps_prochandle, gregs,
+                fill_cframe_list, &dbgo2);
+  }
+#endif // SOLARIS_11_B159_OR_LATER
   return dbgo2.obj;
 }
 
@@ -1218,6 +1266,102 @@ JNIEXPORT jstring JNICALL Java_sun_jvm_hotspot_debugger_proc_ProcDebuggerLocal_d
   return res;
 }
 
+#ifndef SOLARIS_11_B159_OR_LATER
+// Determine if the OS we're running on has the newer version
+// of libproc's Pstack_iter.
+//
+// Set env var PSTACK_ITER_DEBUG=true to debug this logic.
+// Set env var PSTACK_ITER_DEBUG_RELEASE to simulate a 'release' value.
+// Set env var PSTACK_ITER_DEBUG_VERSION to simulate a 'version' value.
+//
+// frankenputer 'uname -r -v': 5.10 Generic_141445-09
+// jurassic 'uname -r -v':     5.11 snv_164
+// lonepeak 'uname -r -v':     5.11 snv_127
+//
+static void set_has_newer_Pstack_iter(JNIEnv *env) {
+  static bool done_set = false;
+
+  if (done_set) {
+    // already set has_newer_Pstack_iter
+    return;
+  }
+
+  struct utsname name;
+  if (uname(&name) == -1) {
+    THROW_NEW_DEBUGGER_EXCEPTION("uname() failed!");
+  }
+  dprintf_2("release='%s'  version='%s'\n", name.release, name.version);
+
+  if (_Pstack_iter_debug) {
+    char *override = getenv("PSTACK_ITER_DEBUG_RELEASE");
+    if (override != NULL) {
+      strncpy(name.release, override, SYS_NMLN - 1);
+      name.release[SYS_NMLN - 2] = '\0';
+      dprintf_2("overriding with release='%s'\n", name.release);
+    }
+    override = getenv("PSTACK_ITER_DEBUG_VERSION");
+    if (override != NULL) {
+      strncpy(name.version, override, SYS_NMLN - 1);
+      name.version[SYS_NMLN - 2] = '\0';
+      dprintf_2("overriding with version='%s'\n", name.version);
+    }
+  }
+
+  // the major number corresponds to the old SunOS major number
+  int major = atoi(name.release);
+  if (major >= 6) {
+    dprintf_2("release is SunOS 6 or later\n");
+    has_newer_Pstack_iter = true;
+    done_set = true;
+    return;
+  }
+  if (major < 5) {
+    dprintf_2("release is SunOS 4 or earlier\n");
+    done_set = true;
+    return;
+  }
+
+  // some SunOS 5.* build so now check for Solaris versions
+  char *dot = strchr(name.release, '.');
+  int minor = 0;
+  if (dot != NULL) {
+    // release is major.minor format
+    *dot = NULL;
+    minor = atoi(dot + 1);
+  }
+
+  if (minor <= 10) {
+    dprintf_2("release is Solaris 10 or earlier\n");
+    done_set = true;
+    return;
+  } else if (minor >= 12) {
+    dprintf_2("release is Solaris 12 or later\n");
+    has_newer_Pstack_iter = true;
+    done_set = true;
+    return;
+  }
+
+  // some Solaris 11 build so now check for internal build numbers
+  if (strncmp(name.version, "snv_", 4) != 0) {
+    dprintf_2("release is Solaris 11 post-GA or later\n");
+    has_newer_Pstack_iter = true;
+    done_set = true;
+    return;
+  }
+
+  // version begins with "snv_" so a pre-GA build of Solaris 11
+  int build = atoi(&name.version[4]);
+  if (build >= 159) {
+    dprintf_2("release is Nevada-B159 or later\n");
+    has_newer_Pstack_iter = true;
+  } else {
+    dprintf_2("release is Nevada-B158 or earlier\n");
+  }
+
+  done_set = true;
+}
+#endif // !SOLARIS_11_B159_OR_LATER
+
 /*
  * Class:       sun_jvm_hotspot_debugger_proc_ProcDebuggerLocal
  * Method:      initIDs
@@ -1237,6 +1381,14 @@ JNIEXPORT void JNICALL Java_sun_jvm_hotspot_debugger_proc_ProcDebuggerLocal_init
   if (libproc_handle == 0)
      THROW_NEW_DEBUGGER_EXCEPTION("can't load libproc.so, if you are using Solaris 5.7 or below, copy libproc.so from 5.8!");
 
+#ifndef SOLARIS_11_B159_OR_LATER
+  _Pstack_iter_debug = getenv("PSTACK_ITER_DEBUG") != NULL;
+
+  set_has_newer_Pstack_iter(env);
+  CHECK_EXCEPTION;
+  dprintf_2("has_newer_Pstack_iter=%d\n", has_newer_Pstack_iter);
+#endif
+
   p_ps_prochandle_ID = env->GetFieldID(clazz, "p_ps_prochandle", "J");
   CHECK_EXCEPTION;
 

hotspot/make/altsrc.make

@@ -24,7 +24,8 @@
 
 # This file defines variables and macros which are used in the makefiles to 
 # allow distributions to augment or replace common hotspot code with 
-# distribution-specific source files.
+# distribution-specific source files. This capability is disabled when
+# an OPENJDK build is requested, unless HS_ALT_SRC_REL has been set externally.
 
 # Requires: GAMMADIR
 # Provides:
@@ -33,14 +34,17 @@
 
 HS_COMMON_SRC_REL=src
 
-# This needs to be changed to a more generic location, but we keep it as this 
+ifneq ($(OPENJDK),true)
-# for now for compatibility
+  # This needs to be changed to a more generic location, but we keep it 
-HS_ALT_SRC_REL=src/closed
+  # as this for now for compatibility
+  HS_ALT_SRC_REL=src/closed
+else
+  HS_ALT_SRC_REL=NO_SUCH_PATH
+endif
 
 HS_COMMON_SRC=$(GAMMADIR)/$(HS_COMMON_SRC_REL)
 HS_ALT_SRC=$(GAMMADIR)/$(HS_ALT_SRC_REL)
 
-
 ## altsrc-equiv 
 # 
 # Convert a common source path to an alternative source path

hotspot/make/jdk6_hotspot_distro

@@ -0,0 +1,32 @@
+# 
+# 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.
+# 
+
+#
+# This file format must remain compatible with both
+# GNU Makefile and Microsoft nmake formats.
+#
+
+# Don't put quotes (fail windows build).
+HOTSPOT_VM_DISTRO=Java HotSpot(TM)
+COMPANY_NAME=Sun Microsystems, Inc.
+PRODUCT_NAME=Java(TM) Platform SE

hotspot/make/linux/makefiles/gcc.make

@@ -205,7 +205,7 @@ SONAMEFLAG = -Xlinker -soname=SONAME
 SHARED_FLAG = -shared
 
 # Keep symbols even they are not used
-AOUT_FLAGS += -export-dynamic
+AOUT_FLAGS += -Xlinker -export-dynamic
 
 #------------------------------------------------------------------------
 # Debug flags

hotspot/make/linux/makefiles/vm.make

@@ -102,6 +102,10 @@ CFLAGS += $(CFLAGS/NOEX)
 CFLAGS += $(EXTRA_CFLAGS)
 LFLAGS += $(EXTRA_CFLAGS)
 
+# Don't set excutable bit on stack segment
+# the same could be done by separate execstack command
+LFLAGS += -Xlinker -z -Xlinker noexecstack
+
 LIBS += -lm -ldl -lpthread
 
 # By default, link the *.o into the library, not the executable.

hotspot/make/solaris/makefiles/saproc.make

@@ -1,5 +1,5 @@
 #
-# Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+# Copyright (c) 2005, 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
@@ -56,6 +56,30 @@ else
 SA_LFLAGS += -mt -xnolib -norunpath
 endif
 
+# The libproc Pstack_iter() interface changed in Nevada-B159.
+# This logic needs to match
+# agent/src/os/solaris/proc/saproc.cpp: set_has_newer_Pstack_iter():
+#   - skip SunOS 4 or older
+#   - skip Solaris 10 or older
+#   - skip two digit Nevada builds
+#   - skip three digit Nevada builds thru 149
+#   - skip Nevada builds 150-158
+SOLARIS_11_B159_OR_LATER := \
+$(shell uname -r -v \
+    | sed -n ' \
+          /^[0-3]\. /b \
+          /^5\.[0-9] /b \
+          /^5\.10 /b \
+          / snv_[0-9][0-9]$/b \
+          / snv_[01][0-4][0-9]$/b \
+          / snv_15[0-8]$/b \
+          s/.*/-DSOLARIS_11_B159_OR_LATER/p \
+          ')
+
+# Uncomment the following to simulate building on Nevada-B159 or later
+# when actually building on Nevada-B158 or earlier:
+#SOLARIS_11_B159_OR_LATER=-DSOLARIS_11_B159_OR_LATER
+
 $(LIBSAPROC): $(SASRCFILES) $(SAMAPFILE)
 	$(QUIETLY) if [ "$(BOOT_JAVA_HOME)" = "" ]; then \
 	  echo "ALT_BOOTDIR, BOOTDIR or JAVA_HOME needs to be defined to build SA"; \
@@ -68,6 +92,7 @@ $(LIBSAPROC): $(SASRCFILES) $(SAMAPFILE)
 	           -I$(GENERATED)                                       \
 	           -I$(BOOT_JAVA_HOME)/include                          \
 	           -I$(BOOT_JAVA_HOME)/include/$(Platform_os_family)    \
+	           $(SOLARIS_11_B159_OR_LATER)                          \
 	           $(SASRCFILES)                                        \
 	           $(SA_LFLAGS)                                         \
 	           -o $@                                                \

hotspot/make/solaris/makefiles/sparcWorks.make

@@ -100,11 +100,6 @@ JVM_CHECK_SYMBOLS = $(NM) -u -p $(LIBJVM.o) | \
 
 LINK_LIB.CC/PRE_HOOK += $(JVM_CHECK_SYMBOLS) || exit 1;
 
-# Some interfaces (_lwp_create) changed with LP64 and Solaris 7
-SOLARIS_7_OR_LATER := \
-$(shell uname -r | awk -F. '{ if ($$2 >= 7) print "-DSOLARIS_7_OR_LATER"; }')
-CFLAGS += ${SOLARIS_7_OR_LATER}
-
 # New architecture options started in SS12 (5.9), we need both styles to build.
 #   The older arch options for SS11 (5.8) or older and also for /usr/ccs/bin/as.
 #   Note: default for 32bit sparc is now the same as v8plus, so the

hotspot/make/windows/build.make

@@ -1,5 +1,5 @@
 #
-# Copyright (c) 1998, 2010, Oracle and/or its affiliates. All rights reserved.
+# Copyright (c) 1998, 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
@@ -125,7 +125,25 @@ VARIANT_TEXT=Kernel
 # or make/hotspot_distro.
 !ifndef HOTSPOT_VM_DISTRO
 !if exists($(WorkSpace)\src\closed)
+
+# if the build is for JDK6 or earlier version, it should include jdk6_hotspot_distro,
+# instead of hotspot_distro.
+JDK6_OR_EARLIER=0
+!if "$(JDK_MAJOR_VERSION)" != "" && "$(JDK_MINOR_VERSION)" != "" && "$(JDK_MICRO_VERSION)" != ""
+!if $(JDK_MAJOR_VERSION) == 1 && $(JDK_MINOR_VERSION) < 7
+JDK6_OR_EARLIER=1
+!endif
+!else
+!if $(JDK_MAJOR_VER) == 1 && $(JDK_MINOR_VER) < 7
+JDK6_OR_EARLIER=1
+!endif
+!endif
+
+!if $(JDK6_OR_EARLIER) == 1
+!include $(WorkSpace)\make\jdk6_hotspot_distro
+!else
 !include $(WorkSpace)\make\hotspot_distro
+!endif
 !else
 !include $(WorkSpace)\make\openjdk_distro
 !endif

hotspot/src/cpu/sparc/vm/frame_sparc.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 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
@@ -806,3 +806,34 @@ intptr_t* frame::interpreter_frame_tos_at(jint offset) const {
   int index = (Interpreter::expr_offset_in_bytes(offset)/wordSize) - 1;
   return &interpreter_frame_tos_address()[index];
 }
+
+
+#ifdef ASSERT
+
+#define DESCRIBE_FP_OFFSET(name) \
+  values.describe(-1, fp() + frame::name##_offset, #name)
+
+void frame::describe_pd(FrameValues& values, int frame_no) {
+  for (int w = 0; w < frame::register_save_words; w++) {
+    values.describe(frame_no, sp() + w, err_msg("register save area word %d", w), 1);
+  }
+
+  if (is_interpreted_frame()) {
+    DESCRIBE_FP_OFFSET(interpreter_frame_d_scratch_fp);
+    DESCRIBE_FP_OFFSET(interpreter_frame_l_scratch_fp);
+    DESCRIBE_FP_OFFSET(interpreter_frame_padding);
+    DESCRIBE_FP_OFFSET(interpreter_frame_oop_temp);
+  }
+
+  if (!is_compiled_frame()) {
+    if (frame::callee_aggregate_return_pointer_words != 0) {
+      values.describe(frame_no, sp() + frame::callee_aggregate_return_pointer_sp_offset, "callee_aggregate_return_pointer_word");
+    }
+    for (int w = 0; w < frame::callee_register_argument_save_area_words; w++) {
+      values.describe(frame_no, sp() + frame::callee_register_argument_save_area_sp_offset + w,
+                      err_msg("callee_register_argument_save_area_words %d", w));
+    }
+  }
+}
+
+#endif

hotspot/src/cpu/sparc/vm/methodHandles_sparc.cpp

@@ -350,8 +350,9 @@ void MethodHandles::remove_arg_slots(MacroAssembler* _masm,
 #ifndef PRODUCT
 extern "C" void print_method_handle(oop mh);
 void trace_method_handle_stub(const char* adaptername,
-                              oopDesc* mh) {
+                              oopDesc* mh,
-  printf("MH %s mh="INTPTR_FORMAT"\n", adaptername, (intptr_t) mh);
+                              intptr_t* saved_sp) {
+  tty->print_cr("MH %s mh="INTPTR_FORMAT " saved_sp=" INTPTR_FORMAT, adaptername, (intptr_t) mh, saved_sp);
   print_method_handle(mh);
 }
 void MethodHandles::trace_method_handle(MacroAssembler* _masm, const char* adaptername) {
@@ -361,6 +362,7 @@ void MethodHandles::trace_method_handle(MacroAssembler* _masm, const char* adapt
   __ save_frame(16);
   __ set((intptr_t) adaptername, O0);
   __ mov(G3_method_handle, O1);
+  __ mov(I5_savedSP, O2);
   __ mov(G3_method_handle, L3);
   __ mov(Gargs, L4);
   __ mov(G5_method_type, L5);
@@ -643,9 +645,10 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
 
       // Live at this point:
       // - G5_klass        :  klass required by the target method
+      // - O0_argslot      :  argslot index in vmarg; may be required in the failing path
       // - O1_scratch      :  argument klass to test
       // - G3_method_handle:  adapter method handle
-      __ check_klass_subtype(O1_scratch, G5_klass, O0_argslot, O2_scratch, done);
+      __ check_klass_subtype(O1_scratch, G5_klass, O2_scratch, O3_scratch, done);
 
       // If we get here, the type check failed!
       __ load_heap_oop(G3_amh_argument,        O2_required);  // required class

hotspot/src/cpu/sparc/vm/templateInterpreter_sparc.cpp

@@ -1698,35 +1698,21 @@ int AbstractInterpreter::layout_activation(methodOop method,
                      popframe_extra_args;
 
     int local_words = method->max_locals() * Interpreter::stackElementWords;
-    int parm_words  = method->size_of_parameters() * Interpreter::stackElementWords;
-    NEEDS_CLEANUP;
     intptr_t* locals;
-    if (caller->is_interpreted_frame()) {
-      // Can force the locals area to end up properly overlapping the top of the expression stack.
-      intptr_t* Lesp_ptr = caller->interpreter_frame_tos_address() - 1;
-      // Note that this computation means we replace size_of_parameters() values from the caller
-      // interpreter frame's expression stack with our argument locals
-      locals = Lesp_ptr + parm_words;
-      int delta = local_words - parm_words;
-      int computed_sp_adjustment = (delta > 0) ? round_to(delta, WordsPerLong) : 0;
-      *interpreter_frame->register_addr(I5_savedSP)    = (intptr_t) (fp + computed_sp_adjustment) - STACK_BIAS;
-    } else {
-      assert(caller->is_compiled_frame() || caller->is_entry_frame(), "only possible cases");
-      // Don't have Lesp available; lay out locals block in the caller
-      // adjacent to the register window save area.
-      //
-      // Compiled frames do not allocate a varargs area which is why this if
-      // statement is needed.
-      //
     if (caller->is_compiled_frame()) {
+      // Compiled frames do not allocate a varargs area so place them
+      // next to the register save area.
       locals = fp + frame::register_save_words + local_words - 1;
-      } else {
-        locals = fp + frame::memory_parameter_word_sp_offset + local_words - 1;
-      }
-      if (!caller->is_entry_frame()) {
       // Caller wants his own SP back
       int caller_frame_size = caller->cb()->frame_size();
       *interpreter_frame->register_addr(I5_savedSP) = (intptr_t)(caller->fp() - caller_frame_size) - STACK_BIAS;
+    } else {
+      assert(caller->is_interpreted_frame() || caller->is_entry_frame(), "only possible cases");
+      // The entry and interpreter frames are laid out like normal C
+      // frames so place the locals adjacent to the varargs area.
+      locals = fp + frame::memory_parameter_word_sp_offset + local_words - 1;
+      if (caller->is_interpreted_frame()) {
+        *interpreter_frame->register_addr(I5_savedSP)    = (intptr_t) (fp + rounded_cls) - STACK_BIAS;
       }
     }
     if (TraceDeoptimization) {

hotspot/src/cpu/x86/vm/assembler_x86.cpp

@@ -6039,6 +6039,43 @@ void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register
   call_VM_leaf(entry_point, 3);
 }
 
+void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
+  pass_arg0(this, arg_0);
+  MacroAssembler::call_VM_leaf_base(entry_point, 1);
+}
+
+void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
+
+  LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg"));
+  pass_arg1(this, arg_1);
+  pass_arg0(this, arg_0);
+  MacroAssembler::call_VM_leaf_base(entry_point, 2);
+}
+
+void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
+  LP64_ONLY(assert(arg_0 != c_rarg2, "smashed arg"));
+  LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
+  pass_arg2(this, arg_2);
+  LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg"));
+  pass_arg1(this, arg_1);
+  pass_arg0(this, arg_0);
+  MacroAssembler::call_VM_leaf_base(entry_point, 3);
+}
+
+void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
+  LP64_ONLY(assert(arg_0 != c_rarg3, "smashed arg"));
+  LP64_ONLY(assert(arg_1 != c_rarg3, "smashed arg"));
+  LP64_ONLY(assert(arg_2 != c_rarg3, "smashed arg"));
+  pass_arg3(this, arg_3);
+  LP64_ONLY(assert(arg_0 != c_rarg2, "smashed arg"));
+  LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
+  pass_arg2(this, arg_2);
+  LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg"));
+  pass_arg1(this, arg_1);
+  pass_arg0(this, arg_0);
+  MacroAssembler::call_VM_leaf_base(entry_point, 4);
+}
+
 void MacroAssembler::check_and_handle_earlyret(Register java_thread) {
 }
 

hotspot/src/cpu/x86/vm/assembler_x86.hpp

@@ -1655,6 +1655,14 @@ class MacroAssembler: public Assembler {
   void call_VM_leaf(address entry_point,
                     Register arg_1, Register arg_2, Register arg_3);
 
+  // These always tightly bind to MacroAssembler::call_VM_leaf_base
+  // bypassing the virtual implementation
+  void super_call_VM_leaf(address entry_point);
+  void super_call_VM_leaf(address entry_point, Register arg_1);
+  void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
+  void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
+  void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4);
+
   // last Java Frame (fills frame anchor)
   void set_last_Java_frame(Register thread,
                            Register last_java_sp,

hotspot/src/cpu/x86/vm/frame_x86.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 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
@@ -669,3 +669,23 @@ intptr_t* frame::interpreter_frame_tos_at(jint offset) const {
   int index = (Interpreter::expr_offset_in_bytes(offset)/wordSize);
   return &interpreter_frame_tos_address()[index];
 }
+
+#ifdef ASSERT
+
+#define DESCRIBE_FP_OFFSET(name) \
+  values.describe(-1, fp() + frame::name##_offset, #name)
+
+void frame::describe_pd(FrameValues& values, int frame_no) {
+  if (is_interpreted_frame()) {
+    DESCRIBE_FP_OFFSET(interpreter_frame_sender_sp);
+    DESCRIBE_FP_OFFSET(interpreter_frame_last_sp);
+    DESCRIBE_FP_OFFSET(interpreter_frame_method);
+    DESCRIBE_FP_OFFSET(interpreter_frame_mdx);
+    DESCRIBE_FP_OFFSET(interpreter_frame_cache);
+    DESCRIBE_FP_OFFSET(interpreter_frame_locals);
+    DESCRIBE_FP_OFFSET(interpreter_frame_bcx);
+    DESCRIBE_FP_OFFSET(interpreter_frame_initial_sp);
+  }
+
+}
+#endif

hotspot/src/cpu/x86/vm/interp_masm_x86_32.cpp

@@ -383,32 +383,6 @@ void InterpreterMacroAssembler::store_ptr(int n, Register val) {
   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
 }
 
-void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point) {
-  MacroAssembler::call_VM_leaf_base(entry_point, 0);
-}
-
-
-void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point, Register arg_1) {
-  push(arg_1);
-  MacroAssembler::call_VM_leaf_base(entry_point, 1);
-}
-
-
-void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2) {
-  push(arg_2);
-  push(arg_1);
-  MacroAssembler::call_VM_leaf_base(entry_point, 2);
-}
-
-
-void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3) {
-  push(arg_3);
-  push(arg_2);
-  push(arg_1);
-  MacroAssembler::call_VM_leaf_base(entry_point, 3);
-}
-
-
 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
   // set sender sp
   lea(rsi, Address(rsp, wordSize));

hotspot/src/cpu/x86/vm/interp_masm_x86_32.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 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
@@ -124,12 +124,6 @@ class InterpreterMacroAssembler: public MacroAssembler {
   void load_ptr(int n, Register val);
   void store_ptr(int n, Register val);
 
-  // Super call_VM calls - correspond to MacroAssembler::call_VM(_leaf) calls
-  void super_call_VM_leaf(address entry_point);
-  void super_call_VM_leaf(address entry_point, Register arg_1);
-  void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
-  void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
-
   // Generate a subtype check: branch to ok_is_subtype if sub_klass is
   // a subtype of super_klass.  EAX holds the super_klass.  Blows ECX
   // and EDI.  Register sub_klass cannot be any of the above.

hotspot/src/cpu/x86/vm/interp_masm_x86_64.cpp

@@ -381,56 +381,6 @@ void InterpreterMacroAssembler::store_ptr(int n, Register val) {
 }
 
 
-void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point) {
-  MacroAssembler::call_VM_leaf_base(entry_point, 0);
-}
-
-
-void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
-                                                   Register arg_1) {
-  if (c_rarg0 != arg_1) {
-    mov(c_rarg0, arg_1);
-  }
-  MacroAssembler::call_VM_leaf_base(entry_point, 1);
-}
-
-
-void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
-                                                   Register arg_1,
-                                                   Register arg_2) {
-  assert(c_rarg0 != arg_2, "smashed argument");
-  assert(c_rarg1 != arg_1, "smashed argument");
-  if (c_rarg0 != arg_1) {
-    mov(c_rarg0, arg_1);
-  }
-  if (c_rarg1 != arg_2) {
-    mov(c_rarg1, arg_2);
-  }
-  MacroAssembler::call_VM_leaf_base(entry_point, 2);
-}
-
-void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
-                                                   Register arg_1,
-                                                   Register arg_2,
-                                                   Register arg_3) {
-  assert(c_rarg0 != arg_2, "smashed argument");
-  assert(c_rarg0 != arg_3, "smashed argument");
-  assert(c_rarg1 != arg_1, "smashed argument");
-  assert(c_rarg1 != arg_3, "smashed argument");
-  assert(c_rarg2 != arg_1, "smashed argument");
-  assert(c_rarg2 != arg_2, "smashed argument");
-  if (c_rarg0 != arg_1) {
-    mov(c_rarg0, arg_1);
-  }
-  if (c_rarg1 != arg_2) {
-    mov(c_rarg1, arg_2);
-  }
-  if (c_rarg2 != arg_3) {
-    mov(c_rarg2, arg_3);
-  }
-  MacroAssembler::call_VM_leaf_base(entry_point, 3);
-}
-
 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
   // set sender sp
   lea(r13, Address(rsp, wordSize));

hotspot/src/cpu/x86/vm/interp_masm_x86_64.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 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
@@ -136,13 +136,6 @@ class InterpreterMacroAssembler: public MacroAssembler {
   void load_ptr(int n, Register val);
   void store_ptr(int n, Register val);
 
-  // Super call_VM calls - correspond to MacroAssembler::call_VM(_leaf) calls
-  void super_call_VM_leaf(address entry_point);
-  void super_call_VM_leaf(address entry_point, Register arg_1);
-  void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
-  void super_call_VM_leaf(address entry_point,
-                          Register arg_1, Register arg_2, Register arg_3);
-
   // Generate a subtype check: branch to ok_is_subtype if sub_klass is
   // a subtype of super_klass.
   void gen_subtype_check( Register sub_klass, Label &ok_is_subtype );

hotspot/src/cpu/x86/vm/methodHandles_x86.cpp

@@ -315,56 +315,38 @@ void MethodHandles::remove_arg_slots(MacroAssembler* _masm,
 #ifndef PRODUCT
 extern "C" void print_method_handle(oop mh);
 void trace_method_handle_stub(const char* adaptername,
-                              oop mh,
-                              intptr_t* saved_regs,
-                              intptr_t* entry_sp,
                               intptr_t* saved_sp,
-                              intptr_t* saved_bp) {
+                              oop mh,
+                              intptr_t* sp) {
   // called as a leaf from native code: do not block the JVM!
-  intptr_t* last_sp = (intptr_t*) saved_bp[frame::interpreter_frame_last_sp_offset];
+  intptr_t* entry_sp = sp + LP64_ONLY(16) NOT_LP64(8);
-  intptr_t* base_sp = (intptr_t*) saved_bp[frame::interpreter_frame_monitor_block_top_offset];
+  tty->print_cr("MH %s mh="INTPTR_FORMAT" sp="INTPTR_FORMAT" saved_sp="INTPTR_FORMAT")",
-  printf("MH %s mh="INTPTR_FORMAT" sp=("INTPTR_FORMAT"+"INTX_FORMAT") stack_size="INTX_FORMAT" bp="INTPTR_FORMAT"\n",
+                adaptername, (intptr_t)mh, (intptr_t)entry_sp, saved_sp);
-         adaptername, (intptr_t)mh, (intptr_t)entry_sp, (intptr_t)(saved_sp - entry_sp), (intptr_t)(base_sp - last_sp), (intptr_t)saved_bp);
-  if (last_sp != saved_sp && last_sp != NULL)
-    printf("*** last_sp="INTPTR_FORMAT"\n", (intptr_t)last_sp);
   if (Verbose) {
-    printf(" reg dump: ");
-    int saved_regs_count = (entry_sp-1) - saved_regs;
-    // 32 bit: rdi rsi rbp rsp; rbx rdx rcx (*) rax
-    int i;
-    for (i = 0; i <= saved_regs_count; i++) {
-      if (i > 0 && i % 4 == 0 && i != saved_regs_count)
-        printf("\n   + dump: ");
-      printf(" %d: "INTPTR_FORMAT, i, saved_regs[i]);
-    }
-    printf("\n");
-    int stack_dump_count = 16;
-    if (stack_dump_count < (int)(saved_bp + 2 - saved_sp))
-      stack_dump_count = (int)(saved_bp + 2 - saved_sp);
-    if (stack_dump_count > 64)  stack_dump_count = 48;
-    for (i = 0; i < stack_dump_count; i += 4) {
-      printf(" dump at SP[%d] "INTPTR_FORMAT": "INTPTR_FORMAT" "INTPTR_FORMAT" "INTPTR_FORMAT" "INTPTR_FORMAT"\n",
-             i, (intptr_t) &entry_sp[i+0], entry_sp[i+0], entry_sp[i+1], entry_sp[i+2], entry_sp[i+3]);
-    }
     print_method_handle(mh);
   }
 }
 void MethodHandles::trace_method_handle(MacroAssembler* _masm, const char* adaptername) {
   if (!TraceMethodHandles)  return;
   BLOCK_COMMENT("trace_method_handle {");
-  __ push(rax);
-  __ lea(rax, Address(rsp, wordSize*6)); // entry_sp
   __ pusha();
+#ifdef _LP64
+  // Pass arguments carefully since the registers overlap with the calling convention.
+  // rcx: method handle
+  // r13: saved sp
+  __ mov(c_rarg2, rcx); // mh
+  __ mov(c_rarg1, r13); // saved sp
+  __ mov(c_rarg3, rsp); // sp
+  __ movptr(c_rarg0, (intptr_t) adaptername);
+  __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, trace_method_handle_stub), c_rarg0, c_rarg1, c_rarg2, c_rarg3);
+#else
   // arguments:
-  __ push(rbp);               // interpreter frame pointer
+  // rcx: method handle
-  __ push(rsi);               // saved_sp
+  // rsi: saved sp
-  __ push(rax);               // entry_sp
+  __ movptr(rbx, (intptr_t) adaptername);
-  __ push(rcx);               // mh
+  __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, trace_method_handle_stub), rbx, rsi, rcx, rsp);
-  __ push(rcx);
+#endif
-  __ movptr(Address(rsp, 0), (intptr_t) adaptername);
-  __ call_VM_leaf(CAST_FROM_FN_PTR(address, trace_method_handle_stub), 5);
   __ popa();
-  __ pop(rax);
   BLOCK_COMMENT("} trace_method_handle");
 }
 #endif //PRODUCT

hotspot/src/cpu/x86/vm/x86_32.ad

@@ -12988,6 +12988,53 @@ instruct maxI_eReg(eRegI dst, eRegI src, eFlagsReg flags) %{
   ins_pipe( pipe_slow );
 %}
 
+// ============================================================================
+// Counted Loop limit node which represents exact final iterator value.
+// Note: the resulting value should fit into integer range since
+// counted loops have limit check on overflow.
+instruct loopLimit_eReg(eAXRegI limit, nadxRegI init, immI stride, eDXRegI limit_hi, nadxRegI tmp, eFlagsReg flags) %{
+  match(Set limit (LoopLimit (Binary init limit) stride));
+  effect(TEMP limit_hi, TEMP tmp, KILL flags);
+  ins_cost(300);
+
+  format %{ "loopLimit $init,$limit,$stride  # $limit = $init + $stride *( $limit - $init + $stride -1)/ $stride, kills $limit_hi" %}
+  ins_encode %{
+    int strd = (int)$stride$$constant;
+    assert(strd != 1 && strd != -1, "sanity");
+    int m1 = (strd > 0) ? 1 : -1;
+    // Convert limit to long (EAX:EDX)
+    __ cdql();
+    // Convert init to long (init:tmp)
+    __ movl($tmp$$Register, $init$$Register);
+    __ sarl($tmp$$Register, 31);
+    // $limit - $init
+    __ subl($limit$$Register, $init$$Register);
+    __ sbbl($limit_hi$$Register, $tmp$$Register);
+    // + ($stride - 1)
+    if (strd > 0) {
+      __ addl($limit$$Register, (strd - 1));
+      __ adcl($limit_hi$$Register, 0);
+      __ movl($tmp$$Register, strd);
+    } else {
+      __ addl($limit$$Register, (strd + 1));
+      __ adcl($limit_hi$$Register, -1);
+      __ lneg($limit_hi$$Register, $limit$$Register);
+      __ movl($tmp$$Register, -strd);
+    }
+    // signed devision: (EAX:EDX) / pos_stride
+    __ idivl($tmp$$Register);
+    if (strd < 0) {
+      // restore sign
+      __ negl($tmp$$Register);
+    }
+    // (EAX) * stride
+    __ mull($tmp$$Register);
+    // + init (ignore upper bits)
+    __ addl($limit$$Register, $init$$Register);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 // ============================================================================
 // Branch Instructions
 // Jump Table

hotspot/src/os/linux/vm/globals_linux.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2005, 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
@@ -47,7 +47,7 @@
 // Defines Linux-specific default values. The flags are available on all
 // platforms, but they may have different default values on other platforms.
 //
-define_pd_global(bool, UseLargePages, false);
+define_pd_global(bool, UseLargePages, true);
 define_pd_global(bool, UseLargePagesIndividualAllocation, false);
 define_pd_global(bool, UseOSErrorReporting, false);
 define_pd_global(bool, UseThreadPriorities, true) ;

hotspot/src/os/linux/vm/os_linux.cpp

@@ -2914,17 +2914,22 @@ static void set_coredump_filter(void) {
 
 static size_t _large_page_size = 0;
 
-bool os::large_page_init() {
+void os::large_page_init() {
   if (!UseLargePages) {
     UseHugeTLBFS = false;
     UseSHM = false;
-    return false;
+    return;
   }
 
   if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
-    // Our user has not expressed a preference, so we'll try both.
+    // If UseLargePages is specified on the command line try both methods,
+    // if it's default, then try only HugeTLBFS.
+    if (FLAG_IS_DEFAULT(UseLargePages)) {
+      UseHugeTLBFS = true;
+    } else {
       UseHugeTLBFS = UseSHM = true;
     }
+  }
 
   if (LargePageSizeInBytes) {
     _large_page_size = LargePageSizeInBytes;
@@ -2978,7 +2983,6 @@ bool os::large_page_init() {
     _page_sizes[1] = default_page_size;
     _page_sizes[2] = 0;
   }
-
   UseHugeTLBFS = UseHugeTLBFS &&
                  Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
 
@@ -2988,12 +2992,6 @@ bool os::large_page_init() {
   UseLargePages = UseHugeTLBFS || UseSHM;
 
   set_coredump_filter();
-
-  // Large page support is available on 2.6 or newer kernel, some vendors
-  // (e.g. Redhat) have backported it to their 2.4 based distributions.
-  // We optimistically assume the support is available. If later it turns out
-  // not true, VM will automatically switch to use regular page size.
-  return true;
 }
 
 #ifndef SHM_HUGETLB
@@ -4118,7 +4116,7 @@ jint os::init_2(void)
 #endif
   }
 
-  FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
+  os::large_page_init();
 
   // initialize suspend/resume support - must do this before signal_sets_init()
   if (SR_initialize() != 0) {

hotspot/src/os/solaris/vm/os_solaris.cpp

@@ -3336,11 +3336,11 @@ bool os::Solaris::mpss_sanity_check(bool warn, size_t * page_size) {
   return true;
 }
 
-bool os::large_page_init() {
+void os::large_page_init() {
   if (!UseLargePages) {
     UseISM = false;
     UseMPSS = false;
-    return false;
+    return;
   }
 
   // print a warning if any large page related flag is specified on command line
@@ -3361,7 +3361,6 @@ bool os::large_page_init() {
             Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
 
   UseLargePages = UseISM || UseMPSS;
-  return UseLargePages;
 }
 
 bool os::Solaris::set_mpss_range(caddr_t start, size_t bytes, size_t align) {
@@ -4992,7 +4991,7 @@ jint os::init_2(void) {
 #endif
 }
 
-  FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
+  os::large_page_init();
 
   // Check minimum allowable stack size for thread creation and to initialize
   // the java system classes, including StackOverflowError - depends on page

hotspot/src/os/windows/vm/os_windows.cpp

@@ -2762,8 +2762,8 @@ static void cleanup_after_large_page_init() {
   _hToken = NULL;
 }
 
-bool os::large_page_init() {
+void os::large_page_init() {
-  if (!UseLargePages) return false;
+  if (!UseLargePages) return;
 
   // print a warning if any large page related flag is specified on command line
   bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages) ||
@@ -2808,7 +2808,7 @@ bool os::large_page_init() {
   }
 
   cleanup_after_large_page_init();
-  return success;
+  UseLargePages = success;
 }
 
 // On win32, one cannot release just a part of reserved memory, it's an
@@ -3561,7 +3561,7 @@ jint os::init_2(void) {
 #endif
 }
 
-  FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
+  os::large_page_init();
 
   // Setup Windows Exceptions
 

hotspot/src/os_cpu/linux_x86/vm/orderAccess_linux_x86.inline.hpp

@@ -93,7 +93,7 @@ inline void     OrderAccess::release_store_ptr(volatile void*     p, void*    v)
 
 inline void     OrderAccess::store_fence(jbyte*  p, jbyte  v) {
   __asm__ volatile (  "xchgb (%2),%0"
-                    : "=r" (v)
+                    : "=q" (v)
                     : "0" (v), "r" (p)
                     : "memory");
 }
@@ -155,7 +155,7 @@ inline void     OrderAccess::store_ptr_fence(void**    p, void*    v) {
 // Must duplicate definitions instead of calling store_fence because we don't want to cast away volatile.
 inline void     OrderAccess::release_store_fence(volatile jbyte*  p, jbyte  v) {
   __asm__ volatile (  "xchgb (%2),%0"
-                    : "=r" (v)
+                    : "=q" (v)
                     : "0" (v), "r" (p)
                     : "memory");
 }

hotspot/src/share/vm/gc_implementation/g1/concurrentMark.cpp

@@ -826,6 +826,14 @@ public:
 void ConcurrentMark::checkpointRootsInitialPost() {
   G1CollectedHeap*   g1h = G1CollectedHeap::heap();
 
+  // If we force an overflow during remark, the remark operation will
+  // actually abort and we'll restart concurrent marking. If we always
+  // force an oveflow during remark we'll never actually complete the
+  // marking phase. So, we initilize this here, at the start of the
+  // cycle, so that at the remaining overflow number will decrease at
+  // every remark and we'll eventually not need to cause one.
+  force_overflow_stw()->init();
+
   // For each region note start of marking.
   NoteStartOfMarkHRClosure startcl;
   g1h->heap_region_iterate(&startcl);
@@ -893,27 +901,37 @@ void ConcurrentMark::checkpointRootsInitial() {
 }
 
 /*
-   Notice that in the next two methods, we actually leave the STS
+ * Notice that in the next two methods, we actually leave the STS
-   during the barrier sync and join it immediately afterwards. If we
+ * during the barrier sync and join it immediately afterwards. If we
-   do not do this, this then the following deadlock can occur: one
+ * do not do this, the following deadlock can occur: one thread could
-   thread could be in the barrier sync code, waiting for the other
+ * be in the barrier sync code, waiting for the other thread to also
-   thread to also sync up, whereas another one could be trying to
+ * sync up, whereas another one could be trying to yield, while also
-   yield, while also waiting for the other threads to sync up too.
+ * waiting for the other threads to sync up too.
-
+ *
-   Because the thread that does the sync barrier has left the STS, it
+ * Note, however, that this code is also used during remark and in
-   is possible to be suspended for a Full GC or an evacuation pause
+ * this case we should not attempt to leave / enter the STS, otherwise
-   could occur. This is actually safe, since the entering the sync
+ * we'll either hit an asseert (debug / fastdebug) or deadlock
-   barrier is one of the last things do_marking_step() does, and it
+ * (product). So we should only leave / enter the STS if we are
-   doesn't manipulate any data structures afterwards.
+ * operating concurrently.
-*/
+ *
+ * Because the thread that does the sync barrier has left the STS, it
+ * is possible to be suspended for a Full GC or an evacuation pause
+ * could occur. This is actually safe, since the entering the sync
+ * barrier is one of the last things do_marking_step() does, and it
+ * doesn't manipulate any data structures afterwards.
+ */
 
 void ConcurrentMark::enter_first_sync_barrier(int task_num) {
   if (verbose_low())
     gclog_or_tty->print_cr("[%d] entering first barrier", task_num);
 
+  if (concurrent()) {
     ConcurrentGCThread::stsLeave();
+  }
   _first_overflow_barrier_sync.enter();
+  if (concurrent()) {
     ConcurrentGCThread::stsJoin();
+  }
   // at this point everyone should have synced up and not be doing any
   // more work
 
@@ -923,7 +941,12 @@ void ConcurrentMark::enter_first_sync_barrier(int task_num) {
   // let task 0 do this
   if (task_num == 0) {
     // task 0 is responsible for clearing the global data structures
-    clear_marking_state();
+    // We should be here because of an overflow. During STW we should
+    // not clear the overflow flag since we rely on it being true when
+    // we exit this method to abort the pause and restart concurent
+    // marking.
+    clear_marking_state(concurrent() /* clear_overflow */);
+    force_overflow()->update();
 
     if (PrintGC) {
       gclog_or_tty->date_stamp(PrintGCDateStamps);
@@ -940,15 +963,45 @@ void ConcurrentMark::enter_second_sync_barrier(int task_num) {
   if (verbose_low())
     gclog_or_tty->print_cr("[%d] entering second barrier", task_num);
 
+  if (concurrent()) {
     ConcurrentGCThread::stsLeave();
+  }
   _second_overflow_barrier_sync.enter();
+  if (concurrent()) {
     ConcurrentGCThread::stsJoin();
+  }
   // at this point everything should be re-initialised and ready to go
 
   if (verbose_low())
     gclog_or_tty->print_cr("[%d] leaving second barrier", task_num);
 }
 
+#ifndef PRODUCT
+void ForceOverflowSettings::init() {
+  _num_remaining = G1ConcMarkForceOverflow;
+  _force = false;
+  update();
+}
+
+void ForceOverflowSettings::update() {
+  if (_num_remaining > 0) {
+    _num_remaining -= 1;
+    _force = true;
+  } else {
+    _force = false;
+  }
+}
+
+bool ForceOverflowSettings::should_force() {
+  if (_force) {
+    _force = false;
+    return true;
+  } else {
+    return false;
+  }
+}
+#endif // !PRODUCT
+
 void ConcurrentMark::grayRoot(oop p) {
   HeapWord* addr = (HeapWord*) p;
   // We can't really check against _heap_start and _heap_end, since it
@@ -1117,6 +1170,7 @@ void ConcurrentMark::markFromRoots() {
   _restart_for_overflow = false;
 
   size_t active_workers = MAX2((size_t) 1, parallel_marking_threads());
+  force_overflow_conc()->init();
   set_phase(active_workers, true /* concurrent */);
 
   CMConcurrentMarkingTask markingTask(this, cmThread());
@@ -1845,7 +1899,7 @@ void ConcurrentMark::completeCleanup() {
   while (!_cleanup_list.is_empty()) {
     HeapRegion* hr = _cleanup_list.remove_head();
     assert(hr != NULL, "the list was not empty");
-    hr->rem_set()->clear();
+    hr->par_clear();
     tmp_free_list.add_as_tail(hr);
 
     // Instead of adding one region at a time to the secondary_free_list,
@@ -2703,12 +2757,16 @@ void ConcurrentMark::oops_do(OopClosure* cl) {
 
 }
 
-void ConcurrentMark::clear_marking_state() {
+void ConcurrentMark::clear_marking_state(bool clear_overflow) {
   _markStack.setEmpty();
   _markStack.clear_overflow();
   _regionStack.setEmpty();
   _regionStack.clear_overflow();
+  if (clear_overflow) {
     clear_has_overflown();
+  } else {
+    assert(has_overflown(), "pre-condition");
+  }
   _finger = _heap_start;
 
   for (int i = 0; i < (int)_max_task_num; ++i) {
@@ -4279,6 +4337,15 @@ void CMTask::do_marking_step(double time_target_ms,
     }
   }
 
+  // If we are about to wrap up and go into termination, check if we
+  // should raise the overflow flag.
+  if (do_termination && !has_aborted()) {
+    if (_cm->force_overflow()->should_force()) {
+      _cm->set_has_overflown();
+      regular_clock_call();
+    }
+  }
+
   // We still haven't aborted. Now, let's try to get into the
   // termination protocol.
   if (do_termination && !has_aborted()) {

hotspot/src/share/vm/gc_implementation/g1/concurrentMark.hpp

@@ -316,6 +316,19 @@ public:
   void setEmpty()   { _index = 0; clear_overflow(); }
 };
 
+class ForceOverflowSettings VALUE_OBJ_CLASS_SPEC {
+private:
+#ifndef PRODUCT
+  uintx _num_remaining;
+  bool _force;
+#endif // !defined(PRODUCT)
+
+public:
+  void init() PRODUCT_RETURN;
+  void update() PRODUCT_RETURN;
+  bool should_force() PRODUCT_RETURN_( return false; );
+};
+
 // this will enable a variety of different statistics per GC task
 #define _MARKING_STATS_       0
 // this will enable the higher verbose levels
@@ -462,6 +475,9 @@ protected:
 
   WorkGang* _parallel_workers;
 
+  ForceOverflowSettings _force_overflow_conc;
+  ForceOverflowSettings _force_overflow_stw;
+
   void weakRefsWork(bool clear_all_soft_refs);
 
   void swapMarkBitMaps();
@@ -470,7 +486,7 @@ protected:
   // task local ones; should be called during initial mark.
   void reset();
   // It resets all the marking data structures.
-  void clear_marking_state();
+  void clear_marking_state(bool clear_overflow = true);
 
   // It should be called to indicate which phase we're in (concurrent
   // mark or remark) and how many threads are currently active.
@@ -547,6 +563,22 @@ protected:
   void enter_first_sync_barrier(int task_num);
   void enter_second_sync_barrier(int task_num);
 
+  ForceOverflowSettings* force_overflow_conc() {
+    return &_force_overflow_conc;
+  }
+
+  ForceOverflowSettings* force_overflow_stw() {
+    return &_force_overflow_stw;
+  }
+
+  ForceOverflowSettings* force_overflow() {
+    if (concurrent()) {
+      return force_overflow_conc();
+    } else {
+      return force_overflow_stw();
+    }
+  }
+
 public:
   // Manipulation of the global mark stack.
   // Notice that the first mark_stack_push is CAS-based, whereas the

hotspot/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp

@@ -3975,6 +3975,9 @@ void G1CollectedHeap::drain_evac_failure_scan_stack() {
 oop
 G1CollectedHeap::handle_evacuation_failure_par(OopsInHeapRegionClosure* cl,
                                                oop old) {
+  assert(obj_in_cs(old),
+         err_msg("obj: "PTR_FORMAT" should still be in the CSet",
+                 (HeapWord*) old));
   markOop m = old->mark();
   oop forward_ptr = old->forward_to_atomic(old);
   if (forward_ptr == NULL) {
@@ -3997,7 +4000,13 @@ G1CollectedHeap::handle_evacuation_failure_par(OopsInHeapRegionClosure* cl,
     }
     return old;
   } else {
-    // Someone else had a place to copy it.
+    // Forward-to-self failed. Either someone else managed to allocate
+    // space for this object (old != forward_ptr) or they beat us in
+    // self-forwarding it (old == forward_ptr).
+    assert(old == forward_ptr || !obj_in_cs(forward_ptr),
+           err_msg("obj: "PTR_FORMAT" forwarded to: "PTR_FORMAT" "
+                   "should not be in the CSet",
+                   (HeapWord*) old, (HeapWord*) forward_ptr));
     return forward_ptr;
   }
 }
@@ -4308,12 +4317,11 @@ template <class T> void G1ParCopyHelper::mark_forwardee(T* p) {
   T heap_oop = oopDesc::load_heap_oop(p);
   if (!oopDesc::is_null(heap_oop)) {
     oop obj = oopDesc::decode_heap_oop(heap_oop);
-    assert((_g1->evacuation_failed()) || (!_g1->obj_in_cs(obj)),
-           "shouldn't still be in the CSet if evacuation didn't fail.");
     HeapWord* addr = (HeapWord*)obj;
-    if (_g1->is_in_g1_reserved(addr))
+    if (_g1->is_in_g1_reserved(addr)) {
       _cm->grayRoot(oop(addr));
     }
+  }
 }
 
 oop G1ParCopyHelper::copy_to_survivor_space(oop old) {
@@ -4961,36 +4969,45 @@ public:
 
 #ifndef PRODUCT
 class G1VerifyCardTableCleanup: public HeapRegionClosure {
+  G1CollectedHeap* _g1h;
   CardTableModRefBS* _ct_bs;
 public:
-  G1VerifyCardTableCleanup(CardTableModRefBS* ct_bs)
+  G1VerifyCardTableCleanup(G1CollectedHeap* g1h, CardTableModRefBS* ct_bs)
-    : _ct_bs(ct_bs) { }
+    : _g1h(g1h), _ct_bs(ct_bs) { }
   virtual bool doHeapRegion(HeapRegion* r) {
-    MemRegion mr(r->bottom(), r->end());
     if (r->is_survivor()) {
-      _ct_bs->verify_dirty_region(mr);
+      _g1h->verify_dirty_region(r);
     } else {
-      _ct_bs->verify_clean_region(mr);
+      _g1h->verify_not_dirty_region(r);
     }
     return false;
   }
 };
 
-void G1CollectedHeap::verify_dirty_young_list(HeapRegion* head) {
+void G1CollectedHeap::verify_not_dirty_region(HeapRegion* hr) {
-  CardTableModRefBS* ct_bs = (CardTableModRefBS*) (barrier_set());
+  // All of the region should be clean.
-  for (HeapRegion* hr = head; hr != NULL; hr = hr->get_next_young_region()) {
+  CardTableModRefBS* ct_bs = (CardTableModRefBS*)barrier_set();
+  MemRegion mr(hr->bottom(), hr->end());
+  ct_bs->verify_not_dirty_region(mr);
+}
+
+void G1CollectedHeap::verify_dirty_region(HeapRegion* hr) {
   // We cannot guarantee that [bottom(),end()] is dirty.  Threads
   // dirty allocated blocks as they allocate them. The thread that
   // retires each region and replaces it with a new one will do a
   // maximal allocation to fill in [pre_dummy_top(),end()] but will
   // not dirty that area (one less thing to have to do while holding
   // a lock). So we can only verify that [bottom(),pre_dummy_top()]
-    // is dirty. Also note that verify_dirty_region() requires
+  // is dirty.
-    // mr.start() and mr.end() to be card aligned and pre_dummy_top()
+  CardTableModRefBS* ct_bs = (CardTableModRefBS*) barrier_set();
-    // is not guaranteed to be.
+  MemRegion mr(hr->bottom(), hr->pre_dummy_top());
-    MemRegion mr(hr->bottom(),
-                 ct_bs->align_to_card_boundary(hr->pre_dummy_top()));
   ct_bs->verify_dirty_region(mr);
+}
+
+void G1CollectedHeap::verify_dirty_young_list(HeapRegion* head) {
+  CardTableModRefBS* ct_bs = (CardTableModRefBS*) barrier_set();
+  for (HeapRegion* hr = head; hr != NULL; hr = hr->get_next_young_region()) {
+    verify_dirty_region(hr);
   }
 }
 
@@ -5033,7 +5050,7 @@ void G1CollectedHeap::cleanUpCardTable() {
   g1_policy()->record_clear_ct_time( elapsed * 1000.0);
 #ifndef PRODUCT
   if (G1VerifyCTCleanup || VerifyAfterGC) {
-    G1VerifyCardTableCleanup cleanup_verifier(ct_bs);
+    G1VerifyCardTableCleanup cleanup_verifier(this, ct_bs);
     heap_region_iterate(&cleanup_verifier);
   }
 #endif

hotspot/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp

@@ -970,6 +970,8 @@ public:
   // The number of regions available for "regular" expansion.
   size_t expansion_regions() { return _expansion_regions; }
 
+  void verify_not_dirty_region(HeapRegion* hr) PRODUCT_RETURN;
+  void verify_dirty_region(HeapRegion* hr) PRODUCT_RETURN;
   void verify_dirty_young_list(HeapRegion* head) PRODUCT_RETURN;
   void verify_dirty_young_regions() PRODUCT_RETURN;
 

hotspot/src/share/vm/gc_implementation/g1/g1RemSet.cpp

@@ -157,7 +157,6 @@ public:
   void set_try_claimed() { _try_claimed = true; }
 
   void scanCard(size_t index, HeapRegion *r) {
-    _cards_done++;
     DirtyCardToOopClosure* cl =
       r->new_dcto_closure(_oc,
                          CardTableModRefBS::Precise,
@@ -168,17 +167,14 @@ public:
     HeapWord* card_start = _bot_shared->address_for_index(index);
     HeapWord* card_end = card_start + G1BlockOffsetSharedArray::N_words;
     Space *sp = SharedHeap::heap()->space_containing(card_start);
-    MemRegion sm_region;
+    MemRegion sm_region = sp->used_region_at_save_marks();
-    if (ParallelGCThreads > 0) {
-      // first find the used area
-      sm_region = sp->used_region_at_save_marks();
-    } else {
-      // The closure is not idempotent.  We shouldn't look at objects
-      // allocated during the GC.
-      sm_region = sp->used_region_at_save_marks();
-    }
     MemRegion mr = sm_region.intersection(MemRegion(card_start,card_end));
-    if (!mr.is_empty()) {
+    if (!mr.is_empty() && !_ct_bs->is_card_claimed(index)) {
+      // We make the card as "claimed" lazily (so races are possible
+      // but they're benign), which reduces the number of duplicate
+      // scans (the rsets of the regions in the cset can intersect).
+      _ct_bs->set_card_claimed(index);
+      _cards_done++;
       cl->do_MemRegion(mr);
     }
   }
@@ -199,6 +195,9 @@ public:
     HeapRegionRemSet* hrrs = r->rem_set();
     if (hrrs->iter_is_complete()) return false; // All done.
     if (!_try_claimed && !hrrs->claim_iter()) return false;
+    // If we ever free the collection set concurrently, we should also
+    // clear the card table concurrently therefore we won't need to
+    // add regions of the collection set to the dirty cards region.
     _g1h->push_dirty_cards_region(r);
     // If we didn't return above, then
     //   _try_claimed || r->claim_iter()
@@ -231,23 +230,16 @@ public:
       }
 
       // If the card is dirty, then we will scan it during updateRS.
-      if (!card_region->in_collection_set() && !_ct_bs->is_card_dirty(card_index)) {
+      if (!card_region->in_collection_set() &&
-        // We make the card as "claimed" lazily (so races are possible but they're benign),
+          !_ct_bs->is_card_dirty(card_index)) {
-        // which reduces the number of duplicate scans (the rsets of the regions in the cset
-        // can intersect).
-        if (!_ct_bs->is_card_claimed(card_index)) {
-          _ct_bs->set_card_claimed(card_index);
         scanCard(card_index, card_region);
       }
     }
-    }
     if (!_try_claimed) {
       hrrs->set_iter_complete();
     }
     return false;
   }
-  // Set all cards back to clean.
-  void cleanup() {_g1h->cleanUpCardTable();}
   size_t cards_done() { return _cards_done;}
   size_t cards_looked_up() { return _cards;}
 };
@@ -567,7 +559,8 @@ public:
     HeapWord* stop_point =
       r->oops_on_card_seq_iterate_careful(scanRegion,
                                           &filter_then_update_rs_cset_oop_cl,
-                                        false /* filter_young */);
+                                          false /* filter_young */,
+                                          NULL  /* card_ptr */);
 
     // Since this is performed in the event of an evacuation failure, we
     // we shouldn't see a non-null stop point
@@ -735,12 +728,6 @@ bool G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i,
                                 (OopClosure*)&mux :
                                 (OopClosure*)&update_rs_oop_cl));
 
-  // Undirty the card.
-  *card_ptr = CardTableModRefBS::clean_card_val();
-  // We must complete this write before we do any of the reads below.
-  OrderAccess::storeload();
-  // And process it, being careful of unallocated portions of TLAB's.
-
   // The region for the current card may be a young region. The
   // current card may have been a card that was evicted from the
   // card cache. When the card was inserted into the cache, we had
@@ -749,7 +736,7 @@ bool G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i,
   // and tagged as young.
   //
   // We wish to filter out cards for such a region but the current
-  // thread, if we're running conucrrently, may "see" the young type
+  // thread, if we're running concurrently, may "see" the young type
   // change at any time (so an earlier "is_young" check may pass or
   // fail arbitrarily). We tell the iteration code to perform this
   // filtering when it has been determined that there has been an actual
@@ -759,7 +746,8 @@ bool G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i,
   HeapWord* stop_point =
     r->oops_on_card_seq_iterate_careful(dirtyRegion,
                                         &filter_then_update_rs_oop_cl,
-                                        filter_young);
+                                        filter_young,
+                                        card_ptr);
 
   // If stop_point is non-null, then we encountered an unallocated region
   // (perhaps the unfilled portion of a TLAB.)  For now, we'll dirty the

hotspot/src/share/vm/gc_implementation/g1/g1_globals.hpp

@@ -311,7 +311,11 @@
                                                                             \
   develop(bool, G1ExitOnExpansionFailure, false,                            \
           "Raise a fatal VM exit out of memory failure in the event "       \
-          " that heap expansion fails due to running out of swap.")
+          " that heap expansion fails due to running out of swap.")         \
+                                                                            \
+  develop(uintx, G1ConcMarkForceOverflow, 0,                                \
+          "The number of times we'll force an overflow during "             \
+          "concurrent marking")
 
 G1_FLAGS(DECLARE_DEVELOPER_FLAG, DECLARE_PD_DEVELOPER_FLAG, DECLARE_PRODUCT_FLAG, DECLARE_PD_PRODUCT_FLAG, DECLARE_DIAGNOSTIC_FLAG, DECLARE_EXPERIMENTAL_FLAG, DECLARE_NOTPRODUCT_FLAG, DECLARE_MANAGEABLE_FLAG, DECLARE_PRODUCT_RW_FLAG)
 

hotspot/src/share/vm/gc_implementation/g1/heapRegion.cpp

@@ -376,6 +376,17 @@ void HeapRegion::hr_clear(bool par, bool clear_space) {
   if (clear_space) clear(SpaceDecorator::Mangle);
 }
 
+void HeapRegion::par_clear() {
+  assert(used() == 0, "the region should have been already cleared");
+  assert(capacity() == (size_t) HeapRegion::GrainBytes,
+         "should be back to normal");
+  HeapRegionRemSet* hrrs = rem_set();
+  hrrs->clear();
+  CardTableModRefBS* ct_bs =
+                   (CardTableModRefBS*)G1CollectedHeap::heap()->barrier_set();
+  ct_bs->clear(MemRegion(bottom(), end()));
+}
+
 // <PREDICTION>
 void HeapRegion::calc_gc_efficiency() {
   G1CollectedHeap* g1h = G1CollectedHeap::heap();
@@ -600,7 +611,15 @@ HeapWord*
 HeapRegion::
 oops_on_card_seq_iterate_careful(MemRegion mr,
                                  FilterOutOfRegionClosure* cl,
-                                 bool filter_young) {
+                                 bool filter_young,
+                                 jbyte* card_ptr) {
+  // Currently, we should only have to clean the card if filter_young
+  // is true and vice versa.
+  if (filter_young) {
+    assert(card_ptr != NULL, "pre-condition");
+  } else {
+    assert(card_ptr == NULL, "pre-condition");
+  }
   G1CollectedHeap* g1h = G1CollectedHeap::heap();
 
   // If we're within a stop-world GC, then we might look at a card in a
@@ -626,6 +645,15 @@ oops_on_card_seq_iterate_careful(MemRegion mr,
 
   assert(!is_young(), "check value of filter_young");
 
+  // We can only clean the card here, after we make the decision that
+  // the card is not young. And we only clean the card if we have been
+  // asked to (i.e., card_ptr != NULL).
+  if (card_ptr != NULL) {
+    *card_ptr = CardTableModRefBS::clean_card_val();
+    // We must complete this write before we do any of the reads below.
+    OrderAccess::storeload();
+  }
+
   // We used to use "block_start_careful" here.  But we're actually happy
   // to update the BOT while we do this...
   HeapWord* cur = block_start(mr.start());

hotspot/src/share/vm/gc_implementation/g1/heapRegion.hpp

@@ -584,6 +584,7 @@ class HeapRegion: public G1OffsetTableContigSpace {
 
   // Reset HR stuff to default values.
   void hr_clear(bool par, bool clear_space);
+  void par_clear();
 
   void initialize(MemRegion mr, bool clear_space, bool mangle_space);
 
@@ -802,12 +803,16 @@ class HeapRegion: public G1OffsetTableContigSpace {
   HeapWord*
   object_iterate_mem_careful(MemRegion mr, ObjectClosure* cl);
 
-  // In this version - if filter_young is true and the region
+  // filter_young: if true and the region is a young region then we
-  // is a young region then we skip the iteration.
+  // skip the iteration.
+  // card_ptr: if not NULL, and we decide that the card is not young
+  // and we iterate over it, we'll clean the card before we start the
+  // iteration.
   HeapWord*
   oops_on_card_seq_iterate_careful(MemRegion mr,
                                    FilterOutOfRegionClosure* cl,
-                                   bool filter_young);
+                                   bool filter_young,
+                                   jbyte* card_ptr);
 
   // A version of block start that is guaranteed to find *some* block
   // boundary at or before "p", but does not object iteration, and may

hotspot/src/share/vm/gc_implementation/parallelScavenge/psOldGen.cpp

@@ -224,6 +224,12 @@ void PSOldGen::expand(size_t bytes) {
   const size_t alignment = virtual_space()->alignment();
   size_t aligned_bytes  = align_size_up(bytes, alignment);
   size_t aligned_expand_bytes = align_size_up(MinHeapDeltaBytes, alignment);
+
+  if (UseNUMA) {
+    // With NUMA we use round-robin page allocation for the old gen. Expand by at least
+    // providing a page per lgroup. Alignment is larger or equal to the page size.
+    aligned_expand_bytes = MAX2(aligned_expand_bytes, alignment * os::numa_get_groups_num());
+  }
   if (aligned_bytes == 0){
     // The alignment caused the number of bytes to wrap.  An expand_by(0) will
     // return true with the implication that and expansion was done when it

hotspot/src/share/vm/interpreter/linkResolver.cpp

@@ -327,6 +327,7 @@ void LinkResolver::resolve_method(methodHandle& resolved_method, KlassHandle res
 
   // 1. check if klass is not interface
   if (resolved_klass->is_interface()) {
+    ResourceMark rm(THREAD);
     char buf[200];
     jio_snprintf(buf, sizeof(buf), "Found interface %s, but class was expected", Klass::cast(resolved_klass())->external_name());
     THROW_MSG(vmSymbols::java_lang_IncompatibleClassChangeError(), buf);
@@ -413,6 +414,7 @@ void LinkResolver::resolve_interface_method(methodHandle& resolved_method,
 
  // check if klass is interface
   if (!resolved_klass->is_interface()) {
+    ResourceMark rm(THREAD);
     char buf[200];
     jio_snprintf(buf, sizeof(buf), "Found class %s, but interface was expected", Klass::cast(resolved_klass())->external_name());
     THROW_MSG(vmSymbols::java_lang_IncompatibleClassChangeError(), buf);
@@ -534,6 +536,7 @@ void LinkResolver::resolve_field(FieldAccessInfo& result, constantPoolHandle poo
 
   // check for errors
   if (is_static != fd.is_static()) {
+    ResourceMark rm(THREAD);
     char msg[200];
     jio_snprintf(msg, sizeof(msg), "Expected %s field %s.%s", is_static ? "static" : "non-static", Klass::cast(resolved_klass())->external_name(), fd.name()->as_C_string());
     THROW_MSG(vmSymbols::java_lang_IncompatibleClassChangeError(), msg);
@@ -631,6 +634,7 @@ void LinkResolver::linktime_resolve_static_method(methodHandle& resolved_method,
 
   // check if static
   if (!resolved_method->is_static()) {
+    ResourceMark rm(THREAD);
     char buf[200];
     jio_snprintf(buf, sizeof(buf), "Expected static method %s", methodOopDesc::name_and_sig_as_C_string(Klass::cast(resolved_klass()),
                                                       resolved_method->name(),
@@ -671,6 +675,7 @@ void LinkResolver::linktime_resolve_special_method(methodHandle& resolved_method
 
   // check if not static
   if (resolved_method->is_static()) {
+    ResourceMark rm(THREAD);
     char buf[200];
     jio_snprintf(buf, sizeof(buf),
                  "Expecting non-static method %s",
@@ -717,6 +722,7 @@ void LinkResolver::runtime_resolve_special_method(CallInfo& result, methodHandle
 
   // check if not static
   if (sel_method->is_static()) {
+    ResourceMark rm(THREAD);
     char buf[200];
     jio_snprintf(buf, sizeof(buf), "Expecting non-static method %s", methodOopDesc::name_and_sig_as_C_string(Klass::cast(resolved_klass()),
                                                                                                              resolved_method->name(),
@@ -757,6 +763,7 @@ void LinkResolver::linktime_resolve_virtual_method(methodHandle &resolved_method
 
   // check if not static
   if (resolved_method->is_static()) {
+    ResourceMark rm(THREAD);
     char buf[200];
     jio_snprintf(buf, sizeof(buf), "Expecting non-static method %s", methodOopDesc::name_and_sig_as_C_string(Klass::cast(resolved_klass()),
                                                                                                              resolved_method->name(),
@@ -873,6 +880,7 @@ void LinkResolver::runtime_resolve_interface_method(CallInfo& result, methodHand
 
   // check if receiver klass implements the resolved interface
   if (!recv_klass->is_subtype_of(resolved_klass())) {
+    ResourceMark rm(THREAD);
     char buf[200];
     jio_snprintf(buf, sizeof(buf), "Class %s does not implement the requested interface %s",
                  (Klass::cast(recv_klass()))->external_name(),

hotspot/src/share/vm/memory/allocation.cpp

@@ -44,6 +44,14 @@ void* CHeapObj::operator new(size_t size){
   return (void *) AllocateHeap(size, "CHeapObj-new");
 }
 
+void* CHeapObj::operator new (size_t size, const std::nothrow_t&  nothrow_constant) {
+  char* p = (char*) os::malloc(size);
+#ifdef ASSERT
+  if (PrintMallocFree) trace_heap_malloc(size, "CHeapObj-new", p);
+#endif
+  return p;
+}
+
 void CHeapObj::operator delete(void* p){
  FreeHeap(p);
 }

hotspot/src/share/vm/memory/allocation.hpp

@@ -34,6 +34,8 @@
 #include "opto/c2_globals.hpp"
 #endif
 
+#include <new>
+
 #define ARENA_ALIGN_M1 (((size_t)(ARENA_AMALLOC_ALIGNMENT)) - 1)
 #define ARENA_ALIGN_MASK (~((size_t)ARENA_ALIGN_M1))
 #define ARENA_ALIGN(x) ((((size_t)(x)) + ARENA_ALIGN_M1) & ARENA_ALIGN_MASK)
@@ -99,6 +101,7 @@ class AllocatedObj {
 class CHeapObj ALLOCATION_SUPER_CLASS_SPEC {
  public:
   void* operator new(size_t size);
+  void* operator new (size_t size, const std::nothrow_t&  nothrow_constant);
   void  operator delete(void* p);
   void* new_array(size_t size);
 };

hotspot/src/share/vm/memory/cardTableModRefBS.cpp

@@ -652,43 +652,37 @@ void CardTableModRefBS::verify() {
 }
 
 #ifndef PRODUCT
-class GuaranteeNotModClosure: public MemRegionClosure {
+void CardTableModRefBS::verify_region(MemRegion mr,
-  CardTableModRefBS* _ct;
+                                      jbyte val, bool val_equals) {
-public:
+  jbyte* start    = byte_for(mr.start());
-  GuaranteeNotModClosure(CardTableModRefBS* ct) : _ct(ct) {}
+  jbyte* end      = byte_for(mr.last());
-  void do_MemRegion(MemRegion mr) {
+  bool   failures = false;
-    jbyte* entry = _ct->byte_for(mr.start());
+  for (jbyte* curr = start; curr <= end; ++curr) {
-    guarantee(*entry != CardTableModRefBS::clean_card,
+    jbyte curr_val = *curr;
-              "Dirty card in region that should be clean");
+    bool failed = (val_equals) ? (curr_val != val) : (curr_val == val);
+    if (failed) {
+      if (!failures) {
+        tty->cr();
+        tty->print_cr("== CT verification failed: ["PTR_FORMAT","PTR_FORMAT"]");
+        tty->print_cr("==   %sexpecting value: %d",
+                      (val_equals) ? "" : "not ", val);
+        failures = true;
       }
-};
+      tty->print_cr("==   card "PTR_FORMAT" ["PTR_FORMAT","PTR_FORMAT"], "
-
+                    "val: %d", curr, addr_for(curr),
-void CardTableModRefBS::verify_clean_region(MemRegion mr) {
+                    (HeapWord*) (((size_t) addr_for(curr)) + card_size),
-  GuaranteeNotModClosure blk(this);
+                    (int) curr_val);
-  non_clean_card_iterate_serial(mr, &blk);
+    }
+  }
+  guarantee(!failures, "there should not have been any failures");
 }
 
-// To verify a MemRegion is entirely dirty this closure is passed to
+void CardTableModRefBS::verify_not_dirty_region(MemRegion mr) {
-// dirty_card_iterate. If the region is dirty do_MemRegion will be
+  verify_region(mr, dirty_card, false /* val_equals */);
-// invoked only once with a MemRegion equal to the one being
+}
-// verified.
-class GuaranteeDirtyClosure: public MemRegionClosure {
-  CardTableModRefBS* _ct;
-  MemRegion _mr;
-  bool _result;
-public:
-  GuaranteeDirtyClosure(CardTableModRefBS* ct, MemRegion mr)
-    : _ct(ct), _mr(mr), _result(false) {}
-  void do_MemRegion(MemRegion mr) {
-    _result = _mr.equals(mr);
-  }
-  bool result() const { return _result; }
-};
 
 void CardTableModRefBS::verify_dirty_region(MemRegion mr) {
-  GuaranteeDirtyClosure blk(this, mr);
+  verify_region(mr, dirty_card, true /* val_equals */);
-  dirty_card_iterate(mr, &blk);
-  guarantee(blk.result(), "Non-dirty cards in region that should be dirty");
 }
 #endif
 

hotspot/src/share/vm/memory/cardTableModRefBS.hpp

@@ -475,7 +475,10 @@ public:
   void verify();
   void verify_guard();
 
-  void verify_clean_region(MemRegion mr) PRODUCT_RETURN;
+  // val_equals -> it will check that all cards covered by mr equal val
+  // !val_equals -> it will check that all cards covered by mr do not equal val
+  void verify_region(MemRegion mr, jbyte val, bool val_equals) PRODUCT_RETURN;
+  void verify_not_dirty_region(MemRegion mr) PRODUCT_RETURN;
   void verify_dirty_region(MemRegion mr) PRODUCT_RETURN;
 
   static size_t par_chunk_heapword_alignment() {

hotspot/src/share/vm/memory/collectorPolicy.cpp

@@ -265,8 +265,6 @@ void TwoGenerationCollectorPolicy::initialize_flags() {
   MaxHeapSize = align_size_up(MaxHeapSize, max_alignment());
 
   always_do_update_barrier = UseConcMarkSweepGC;
-  BlockOffsetArrayUseUnallocatedBlock =
-      BlockOffsetArrayUseUnallocatedBlock || ParallelGCThreads > 0;
 
   // Check validity of heap flags
   assert(OldSize     % min_alignment() == 0, "old space alignment");

hotspot/src/share/vm/memory/genOopClosures.hpp

@@ -175,7 +175,7 @@ class VerifyOopClosure: public OopClosure {
  protected:
   template <class T> inline void do_oop_work(T* p) {
     oop obj = oopDesc::load_decode_heap_oop(p);
-    guarantee(obj->is_oop_or_null(), err_msg("invalid oop: " INTPTR_FORMAT, obj));
+    guarantee(obj->is_oop_or_null(), err_msg("invalid oop: " INTPTR_FORMAT, (oopDesc*) obj));
   }
  public:
   virtual void do_oop(oop* p);

hotspot/src/share/vm/memory/modRefBarrierSet.hpp

@@ -100,12 +100,6 @@ public:
   // Pass along the argument to the superclass.
   ModRefBarrierSet(int max_covered_regions) :
     BarrierSet(max_covered_regions) {}
-
-#ifndef PRODUCT
-  // Verifies that the given region contains no modified references.
-  virtual void verify_clean_region(MemRegion mr) = 0;
-#endif
-
 };
 
 #endif // SHARE_VM_MEMORY_MODREFBARRIERSET_HPP

hotspot/src/share/vm/oops/methodDataOop.hpp

@@ -1194,7 +1194,7 @@ private:
   // Whole-method sticky bits and flags
 public:
   enum {
-    _trap_hist_limit    = 16,   // decoupled from Deoptimization::Reason_LIMIT
+    _trap_hist_limit    = 17,   // decoupled from Deoptimization::Reason_LIMIT
     _trap_hist_mask     = max_jubyte,
     _extra_data_count   = 4     // extra DataLayout headers, for trap history
   }; // Public flag values

hotspot/src/share/vm/opto/bytecodeInfo.cpp

@@ -310,13 +310,14 @@ const char* InlineTree::try_to_inline(ciMethod* callee_method, ciMethod* caller_
     return "inlining too deep";
   }
 
-  // We need to detect recursive inlining of method handle targets: if
+  // detect direct and indirect recursive inlining
-  // the current method is a method handle adapter and one of the
+  {
-  // callers is the same method as the callee, we bail out if
+    // count the current method and the callee
-  // MaxRecursiveInlineLevel is hit.
+    int inline_level = (method() == callee_method) ? 1 : 0;
-  if (method()->is_method_handle_adapter()) {
+    if (inline_level > MaxRecursiveInlineLevel)
+      return "recursively inlining too deep";
+    // count callers of current method and callee
     JVMState* jvms = caller_jvms();
-    int inline_level = 0;
     while (jvms != NULL && jvms->has_method()) {
       if (jvms->method() == callee_method) {
         inline_level++;
@@ -327,10 +328,6 @@ const char* InlineTree::try_to_inline(ciMethod* callee_method, ciMethod* caller_
     }
   }
 
-  if (method() == callee_method && inline_depth() > MaxRecursiveInlineLevel) {
-    return "recursively inlining too deep";
-  }
-
   int size = callee_method->code_size();
 
   if (UseOldInlining && ClipInlining
@@ -376,7 +373,6 @@ bool pass_initial_checks(ciMethod* caller_method, int caller_bci, ciMethod* call
   return true;
 }
 
-#ifndef PRODUCT
 //------------------------------print_inlining---------------------------------
 // Really, the failure_msg can be a success message also.
 void InlineTree::print_inlining(ciMethod* callee_method, int caller_bci, const char* failure_msg) const {
@@ -388,7 +384,6 @@ void InlineTree::print_inlining(ciMethod* callee_method, int caller_bci, const c
     tty->print("  bcs: %d+%d  invoked: %d", top->count_inline_bcs(), callee_method->code_size(), callee_method->interpreter_invocation_count());
   }
 }
-#endif
 
 //------------------------------ok_to_inline-----------------------------------
 WarmCallInfo* InlineTree::ok_to_inline(ciMethod* callee_method, JVMState* jvms, ciCallProfile& profile, WarmCallInfo* initial_wci) {

hotspot/src/share/vm/opto/c2_globals.hpp

@@ -183,6 +183,21 @@
   develop(bool, TraceLoopOpts, false,                                       \
           "Trace executed loop optimizations")                              \
                                                                             \
+  diagnostic(bool, LoopLimitCheck, true,                                    \
+          "Generate a loop limits check for overflow")                      \
+                                                                            \
+  develop(bool, TraceLoopLimitCheck, false,                                 \
+          "Trace generation of loop limits checks")                         \
+                                                                            \
+  diagnostic(bool, RangeLimitCheck, true,                                   \
+          "Additional overflow checks during range check elimination")      \
+                                                                            \
+  develop(bool, TraceRangeLimitCheck, false,                                \
+          "Trace additional overflow checks in RCE")                        \
+                                                                            \
+  diagnostic(bool, UnrollLimitCheck, true,                                  \
+          "Additional overflow checks during loop unroll")                  \
+                                                                            \
   product(bool, OptimizeFill, false,                                        \
           "convert fill/copy loops into intrinsic")                         \
                                                                             \

hotspot/src/share/vm/opto/cfgnode.cpp

@@ -1373,7 +1373,7 @@ static void split_once(PhaseIterGVN *igvn, Node *phi, Node *val, Node *n, Node *
 
   // Clone loop predicates
   if (predicate_proj != NULL) {
-    newn = igvn->clone_loop_predicates(predicate_proj, newn);
+    newn = igvn->clone_loop_predicates(predicate_proj, newn, !n->is_CountedLoop());
   }
 
   // Now I can point to the new node.

hotspot/src/share/vm/opto/classes.hpp

@@ -156,6 +156,7 @@ macro(Lock)
 macro(LogD)
 macro(Log10D)
 macro(Loop)
+macro(LoopLimit)
 macro(Mach)
 macro(MachProj)
 macro(MaxI)

hotspot/src/share/vm/opto/graphKit.cpp

@@ -3378,6 +3378,10 @@ void GraphKit::add_predicate(int nargs) {
   if (UseLoopPredicate) {
     add_predicate_impl(Deoptimization::Reason_predicate, nargs);
   }
+  // loop's limit check predicate should be near the loop.
+  if (LoopLimitCheck) {
+    add_predicate_impl(Deoptimization::Reason_loop_limit_check, nargs);
+  }
 }
 
 //----------------------------- store barriers ----------------------------

hotspot/src/share/vm/opto/ifnode.cpp

@@ -236,6 +236,7 @@ static Node* split_if(IfNode *iff, PhaseIterGVN *igvn) {
   }
   Node* predicate_c = NULL;
   Node* predicate_x = NULL;
+  bool counted_loop = r->is_CountedLoop();
 
   Node *region_c = new (igvn->C, req_c + 1) RegionNode(req_c + 1);
   Node *phi_c    = con1;
@@ -294,16 +295,16 @@ static Node* split_if(IfNode *iff, PhaseIterGVN *igvn) {
   if (predicate_c != NULL) {
     assert(predicate_x == NULL, "only one predicate entry expected");
     // Clone loop predicates to each path
-    iff_c_t = igvn->clone_loop_predicates(predicate_c, iff_c_t);
+    iff_c_t = igvn->clone_loop_predicates(predicate_c, iff_c_t, !counted_loop);
-    iff_c_f = igvn->clone_loop_predicates(predicate_c, iff_c_f);
+    iff_c_f = igvn->clone_loop_predicates(predicate_c, iff_c_f, !counted_loop);
   }
   Node *iff_x_t = phase->transform(new (igvn->C, 1) IfTrueNode (iff_x));
   Node *iff_x_f = phase->transform(new (igvn->C, 1) IfFalseNode(iff_x));
   if (predicate_x != NULL) {
     assert(predicate_c == NULL, "only one predicate entry expected");
     // Clone loop predicates to each path
-    iff_x_t = igvn->clone_loop_predicates(predicate_x, iff_x_t);
+    iff_x_t = igvn->clone_loop_predicates(predicate_x, iff_x_t, !counted_loop);
-    iff_x_f = igvn->clone_loop_predicates(predicate_x, iff_x_f);
+    iff_x_f = igvn->clone_loop_predicates(predicate_x, iff_x_f, !counted_loop);
   }
 
   // Merge the TRUE paths
@@ -545,6 +546,7 @@ static void adjust_check(Node* proj, Node* range, Node* index,
   Node *new_bol = gvn->transform( new (gvn->C, 2) BoolNode( new_cmp, bol->as_Bool()->_test._test ) );
   igvn->hash_delete( iff );
   iff->set_req_X( 1, new_bol, igvn );
+  igvn->_worklist.push( iff );
 }
 
 //------------------------------up_one_dom-------------------------------------

hotspot/src/share/vm/opto/library_call.cpp

@@ -867,12 +867,10 @@ Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1, Node* cnt1
   Node* str1_offset  = make_load(no_ctrl, str1_offseta, TypeInt::INT, T_INT, string_type->add_offset(offset_offset));
   Node* str1_start   = array_element_address(str1_value, str1_offset, T_CHAR);
 
-  // Pin loads from String::equals() argument since it could be NULL.
-  Node* str2_ctrl = (opcode == Op_StrEquals) ? control() : no_ctrl;
   Node* str2_valuea  = basic_plus_adr(str2, str2, value_offset);
-  Node* str2_value   = make_load(str2_ctrl, str2_valuea, value_type, T_OBJECT, string_type->add_offset(value_offset));
+  Node* str2_value   = make_load(no_ctrl, str2_valuea, value_type, T_OBJECT, string_type->add_offset(value_offset));
   Node* str2_offseta = basic_plus_adr(str2, str2, offset_offset);
-  Node* str2_offset  = make_load(str2_ctrl, str2_offseta, TypeInt::INT, T_INT, string_type->add_offset(offset_offset));
+  Node* str2_offset  = make_load(no_ctrl, str2_offseta, TypeInt::INT, T_INT, string_type->add_offset(offset_offset));
   Node* str2_start   = array_element_address(str2_value, str2_offset, T_CHAR);
 
   Node* result = NULL;
@@ -1012,14 +1010,15 @@ bool LibraryCallKit::inline_string_equals() {
   if (!stopped()) {
     // Properly cast the argument to String
     argument = _gvn.transform(new (C, 2) CheckCastPPNode(control(), argument, string_type));
+    // This path is taken only when argument's type is String:NotNull.
+    argument = cast_not_null(argument, false);
 
     // Get counts for string and argument
     Node* receiver_cnta = basic_plus_adr(receiver, receiver, count_offset);
     receiver_cnt  = make_load(no_ctrl, receiver_cnta, TypeInt::INT, T_INT, string_type->add_offset(count_offset));
 
-    // Pin load from argument string since it could be NULL.
     Node* argument_cnta = basic_plus_adr(argument, argument, count_offset);
-    argument_cnt  = make_load(control(), argument_cnta, TypeInt::INT, T_INT, string_type->add_offset(count_offset));
+    argument_cnt  = make_load(no_ctrl, argument_cnta, TypeInt::INT, T_INT, string_type->add_offset(count_offset));
 
     // Check for receiver count != argument count
     Node* cmp = _gvn.transform( new(C, 3) CmpINode(receiver_cnt, argument_cnt) );

hotspot/src/share/vm/opto/loopPredicate.cpp

@@ -341,7 +341,7 @@ ProjNode* PhaseIdealLoop::move_predicate(ProjNode* predicate_proj, Node* new_ent
   // Cut predicate from old place.
   Node* old = predicate_proj;
   igvn->_worklist.push(old);
-  for (DUIterator_Last imin, i = old->last_outs(imin); i >= imin; ) {
+  for (DUIterator_Last imin, i = old->last_outs(imin); i >= imin;) {
     Node* use = old->last_out(i);  // for each use...
     igvn->hash_delete(use);
     igvn->_worklist.push(use);
@@ -384,24 +384,25 @@ ProjNode* PhaseIdealLoop::move_predicate(ProjNode* predicate_proj, Node* new_ent
 
 //--------------------------clone_loop_predicates-----------------------
 // Interface from IGVN
-Node* PhaseIterGVN::clone_loop_predicates(Node* old_entry, Node* new_entry) {
+Node* PhaseIterGVN::clone_loop_predicates(Node* old_entry, Node* new_entry, bool clone_limit_check) {
-  return PhaseIdealLoop::clone_loop_predicates(old_entry, new_entry, false, NULL, this);
+  return PhaseIdealLoop::clone_loop_predicates(old_entry, new_entry, false, clone_limit_check, NULL, this);
 }
-Node* PhaseIterGVN::move_loop_predicates(Node* old_entry, Node* new_entry) {
+Node* PhaseIterGVN::move_loop_predicates(Node* old_entry, Node* new_entry, bool clone_limit_check) {
-  return PhaseIdealLoop::clone_loop_predicates(old_entry, new_entry, true, NULL, this);
+  return PhaseIdealLoop::clone_loop_predicates(old_entry, new_entry, true, clone_limit_check, NULL, this);
 }
 
 // Interface from PhaseIdealLoop
-Node* PhaseIdealLoop::clone_loop_predicates(Node* old_entry, Node* new_entry) {
+Node* PhaseIdealLoop::clone_loop_predicates(Node* old_entry, Node* new_entry, bool clone_limit_check) {
-  return clone_loop_predicates(old_entry, new_entry, false, this, &this->_igvn);
+  return clone_loop_predicates(old_entry, new_entry, false, clone_limit_check, this, &this->_igvn);
 }
-Node* PhaseIdealLoop::move_loop_predicates(Node* old_entry, Node* new_entry) {
+Node* PhaseIdealLoop::move_loop_predicates(Node* old_entry, Node* new_entry, bool clone_limit_check) {
-  return clone_loop_predicates(old_entry, new_entry, true, this, &this->_igvn);
+  return clone_loop_predicates(old_entry, new_entry, true, clone_limit_check, this, &this->_igvn);
 }
 
 // Clone loop predicates to cloned loops (peeled, unswitched, split_if).
 Node* PhaseIdealLoop::clone_loop_predicates(Node* old_entry, Node* new_entry,
                                                 bool move_predicates,
+                                                bool clone_limit_check,
                                                 PhaseIdealLoop* loop_phase,
                                                 PhaseIterGVN* igvn) {
 #ifdef ASSERT
@@ -413,10 +414,16 @@ Node* PhaseIdealLoop::clone_loop_predicates(Node* old_entry, Node* new_entry,
 #endif
   // Search original predicates
   Node* entry = old_entry;
+  ProjNode* limit_check_proj = NULL;
+  if (LoopLimitCheck) {
+    limit_check_proj = find_predicate_insertion_point(entry, Deoptimization::Reason_loop_limit_check);
+    if (limit_check_proj != NULL) {
+      entry = entry->in(0)->in(0);
+    }
+  }
   if (UseLoopPredicate) {
     ProjNode* predicate_proj = find_predicate_insertion_point(entry, Deoptimization::Reason_predicate);
     if (predicate_proj != NULL) { // right pattern that can be used by loop predication
-      assert(entry->in(0)->in(1)->in(1)->Opcode()==Op_Opaque1, "must be");
       if (move_predicates) {
         new_entry =  move_predicate(predicate_proj, new_entry,
                                     Deoptimization::Reason_predicate,
@@ -435,11 +442,37 @@ Node* PhaseIdealLoop::clone_loop_predicates(Node* old_entry, Node* new_entry,
       }
     }
   }
+  if (limit_check_proj != NULL && clone_limit_check) {
+    // Clone loop limit check last to insert it before loop.
+    // Don't clone a limit check which was already finalized
+    // for this counted loop (only one limit check is needed).
+    if (move_predicates) {
+      new_entry =  move_predicate(limit_check_proj, new_entry,
+                                  Deoptimization::Reason_loop_limit_check,
+                                  loop_phase, igvn);
+      assert(new_entry == limit_check_proj, "old limit check fall through projection");
+    } else {
+      new_entry = clone_predicate(limit_check_proj, new_entry,
+                                  Deoptimization::Reason_loop_limit_check,
+                                  loop_phase, igvn);
+      assert(new_entry != NULL && new_entry->is_Proj(), "IfTrue or IfFalse after clone limit check");
+    }
+    if (TraceLoopLimitCheck) {
+      tty->print_cr("Loop Limit Check %s: ", move_predicates ? "moved" : "cloned");
+      debug_only( new_entry->in(0)->dump(); )
+    }
+  }
   return new_entry;
 }
 
 //--------------------------eliminate_loop_predicates-----------------------
 void PhaseIdealLoop::eliminate_loop_predicates(Node* entry) {
+  if (LoopLimitCheck) {
+    Node* predicate = find_predicate_insertion_point(entry, Deoptimization::Reason_loop_limit_check);
+    if (predicate != NULL) {
+      entry = entry->in(0)->in(0);
+    }
+  }
   if (UseLoopPredicate) {
     ProjNode* predicate_proj = find_predicate_insertion_point(entry, Deoptimization::Reason_predicate);
     if (predicate_proj != NULL) { // right pattern that can be used by loop predication
@@ -456,10 +489,15 @@ void PhaseIdealLoop::eliminate_loop_predicates(Node* entry) {
 // Skip related predicates.
 Node* PhaseIdealLoop::skip_loop_predicates(Node* entry) {
   Node* predicate = NULL;
+  if (LoopLimitCheck) {
+    predicate = find_predicate_insertion_point(entry, Deoptimization::Reason_loop_limit_check);
+    if (predicate != NULL) {
+      entry = entry->in(0)->in(0);
+    }
+  }
   if (UseLoopPredicate) {
     predicate = find_predicate_insertion_point(entry, Deoptimization::Reason_predicate);
     if (predicate != NULL) { // right pattern that can be used by loop predication
-      assert(entry->is_Proj() && entry->in(0)->in(1)->in(1)->Opcode()==Op_Opaque1, "must be");
       IfNode* iff = entry->in(0)->as_If();
       ProjNode* uncommon_proj = iff->proj_out(1 - entry->as_Proj()->_con);
       Node* rgn = uncommon_proj->unique_ctrl_out();
@@ -491,10 +529,15 @@ ProjNode* PhaseIdealLoop::find_predicate_insertion_point(Node* start_c, Deoptimi
 // Find a predicate
 Node* PhaseIdealLoop::find_predicate(Node* entry) {
   Node* predicate = NULL;
+  if (LoopLimitCheck) {
+    predicate = find_predicate_insertion_point(entry, Deoptimization::Reason_loop_limit_check);
+    if (predicate != NULL) { // right pattern that can be used by loop predication
+      return entry;
+    }
+  }
   if (UseLoopPredicate) {
     predicate = find_predicate_insertion_point(entry, Deoptimization::Reason_predicate);
     if (predicate != NULL) { // right pattern that can be used by loop predication
-      assert(entry->in(0)->in(1)->in(1)->Opcode()==Op_Opaque1, "must be");
       return entry;
     }
   }
@@ -658,7 +701,7 @@ bool IdealLoopTree::is_range_check_if(IfNode *iff, PhaseIdealLoop *phase, Invari
   Node* range = cmp->in(2);
   if (range->Opcode() != Op_LoadRange) {
     const TypeInt* tint = phase->_igvn.type(range)->isa_int();
-    if (!OptimizeFill || tint == NULL || tint->empty() || tint->_lo < 0) {
+    if (tint == NULL || tint->empty() || tint->_lo < 0) {
       // Allow predication on positive values that aren't LoadRanges.
       // This allows optimization of loops where the length of the
       // array is a known value and doesn't need to be loaded back
@@ -696,36 +739,49 @@ bool IdealLoopTree::is_range_check_if(IfNode *iff, PhaseIdealLoop *phase, Invari
 //   max(scale*i + offset) = scale*(limit-stride) + offset
 // (2) stride*scale < 0
 //   max(scale*i + offset) = scale*init + offset
-BoolNode* PhaseIdealLoop::rc_predicate(Node* ctrl,
+BoolNode* PhaseIdealLoop::rc_predicate(IdealLoopTree *loop, Node* ctrl,
                                        int scale, Node* offset,
                                        Node* init, Node* limit, Node* stride,
                                        Node* range, bool upper) {
-  DEBUG_ONLY(ttyLocker ttyl);
+  stringStream* predString = NULL;
-  if (TraceLoopPredicate) tty->print("rc_predicate ");
+  if (TraceLoopPredicate) {
+    predString = new stringStream();
+    predString->print("rc_predicate ");
+  }
 
   Node* max_idx_expr  = init;
   int stride_con = stride->get_int();
   if ((stride_con > 0) == (scale > 0) == upper) {
+    if (LoopLimitCheck) {
+      // With LoopLimitCheck limit is not exact.
+      // Calculate exact limit here.
+      // Note, counted loop's test is '<' or '>'.
+      limit = exact_limit(loop);
       max_idx_expr = new (C, 3) SubINode(limit, stride);
       register_new_node(max_idx_expr, ctrl);
-    if (TraceLoopPredicate) tty->print("(limit - stride) ");
+      if (TraceLoopPredicate) predString->print("(limit - stride) ");
     } else {
-    if (TraceLoopPredicate) tty->print("init ");
+      max_idx_expr = new (C, 3) SubINode(limit, stride);
+      register_new_node(max_idx_expr, ctrl);
+      if (TraceLoopPredicate) predString->print("(limit - stride) ");
+    }
+  } else {
+    if (TraceLoopPredicate) predString->print("init ");
   }
 
   if (scale != 1) {
     ConNode* con_scale = _igvn.intcon(scale);
     max_idx_expr = new (C, 3) MulINode(max_idx_expr, con_scale);
     register_new_node(max_idx_expr, ctrl);
-    if (TraceLoopPredicate) tty->print("* %d ", scale);
+    if (TraceLoopPredicate) predString->print("* %d ", scale);
   }
 
   if (offset && (!offset->is_Con() || offset->get_int() != 0)){
     max_idx_expr = new (C, 3) AddINode(max_idx_expr, offset);
     register_new_node(max_idx_expr, ctrl);
     if (TraceLoopPredicate)
-      if (offset->is_Con()) tty->print("+ %d ", offset->get_int());
+      if (offset->is_Con()) predString->print("+ %d ", offset->get_int());
-      else tty->print("+ offset ");
+      else predString->print("+ offset ");
   }
 
   CmpUNode* cmp = new (C, 3) CmpUNode(max_idx_expr, range);
@@ -733,7 +789,10 @@ BoolNode* PhaseIdealLoop::rc_predicate(Node* ctrl,
   BoolNode* bol = new (C, 2) BoolNode(cmp, BoolTest::lt);
   register_new_node(bol, ctrl);
 
-  if (TraceLoopPredicate) tty->print_cr("<u range");
+  if (TraceLoopPredicate) {
+    predString->print_cr("<u range");
+    tty->print(predString->as_string());
+  }
   return bol;
 }
 
@@ -746,29 +805,36 @@ bool PhaseIdealLoop::loop_predication_impl(IdealLoopTree *loop) {
     // Could be a simple region when irreducible loops are present.
     return false;
   }
+  LoopNode* head = loop->_head->as_Loop();
 
-  if (loop->_head->unique_ctrl_out()->Opcode() == Op_NeverBranch) {
+  if (head->unique_ctrl_out()->Opcode() == Op_NeverBranch) {
     // do nothing for infinite loops
     return false;
   }
 
   CountedLoopNode *cl = NULL;
-  if (loop->_head->is_CountedLoop()) {
+  if (head->is_CountedLoop()) {
-    cl = loop->_head->as_CountedLoop();
+    cl = head->as_CountedLoop();
     // do nothing for iteration-splitted loops
     if (!cl->is_normal_loop()) return false;
   }
 
-  LoopNode *lpn  = loop->_head->as_Loop();
+  Node* entry = head->in(LoopNode::EntryControl);
-  Node* entry = lpn->in(LoopNode::EntryControl);
+  ProjNode *predicate_proj = NULL;
+  // Loop limit check predicate should be near the loop.
+  if (LoopLimitCheck) {
+    predicate_proj = find_predicate_insertion_point(entry, Deoptimization::Reason_loop_limit_check);
+    if (predicate_proj != NULL)
+      entry = predicate_proj->in(0)->in(0);
+  }
 
-  ProjNode *predicate_proj = find_predicate_insertion_point(entry, Deoptimization::Reason_predicate);
+  predicate_proj = find_predicate_insertion_point(entry, Deoptimization::Reason_predicate);
   if (!predicate_proj) {
 #ifndef PRODUCT
     if (TraceLoopPredicate) {
       tty->print("missing predicate:");
       loop->dump_head();
-      lpn->dump(1);
+      head->dump(1);
     }
 #endif
     return false;
@@ -782,7 +848,6 @@ bool PhaseIdealLoop::loop_predication_impl(IdealLoopTree *loop) {
   // Create list of if-projs such that a newer proj dominates all older
   // projs in the list, and they all dominate loop->tail()
   Node_List if_proj_list(area);
-  LoopNode *head  = loop->_head->as_Loop();
   Node *current_proj = loop->tail(); //start from tail
   while (current_proj != head) {
     if (loop == get_loop(current_proj) && // still in the loop ?
@@ -856,8 +921,8 @@ bool PhaseIdealLoop::loop_predication_impl(IdealLoopTree *loop) {
       const Node*    cmp    = bol->in(1)->as_Cmp();
       Node*          idx    = cmp->in(1);
       assert(!invar.is_invariant(idx), "index is variant");
-      assert(cmp->in(2)->Opcode() == Op_LoadRange || OptimizeFill, "must be");
       Node* rng = cmp->in(2);
+      assert(rng->Opcode() == Op_LoadRange || _igvn.type(rng)->is_int() >= 0, "must be");
       assert(invar.is_invariant(rng), "range must be invariant");
       int scale    = 1;
       Node* offset = zero;
@@ -886,14 +951,14 @@ bool PhaseIdealLoop::loop_predication_impl(IdealLoopTree *loop) {
       }
 
       // Test the lower bound
-      Node*  lower_bound_bol = rc_predicate(ctrl, scale, offset, init, limit, stride, rng, false);
+      Node*  lower_bound_bol = rc_predicate(loop, ctrl, scale, offset, init, limit, stride, rng, false);
       IfNode* lower_bound_iff = lower_bound_proj->in(0)->as_If();
       _igvn.hash_delete(lower_bound_iff);
       lower_bound_iff->set_req(1, lower_bound_bol);
       if (TraceLoopPredicate) tty->print_cr("lower bound check if: %d", lower_bound_iff->_idx);
 
       // Test the upper bound
-      Node* upper_bound_bol = rc_predicate(ctrl, scale, offset, init, limit, stride, rng, true);
+      Node* upper_bound_bol = rc_predicate(loop, ctrl, 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);
@@ -957,4 +1022,3 @@ bool IdealLoopTree::loop_predication( PhaseIdealLoop *phase) {
 
   return hoisted;
 }
-

hotspot/src/share/vm/opto/loopTransform.cpp

@@ -83,7 +83,7 @@ void IdealLoopTree::compute_exact_trip_count( PhaseIdealLoop *phase ) {
 #ifdef ASSERT
   BoolTest::mask bt = cl->loopexit()->test_trip();
   assert(bt == BoolTest::lt || bt == BoolTest::gt ||
-         bt == BoolTest::ne, "canonical test is expected");
+         (bt == BoolTest::ne && !LoopLimitCheck), "canonical test is expected");
 #endif
 
   Node* init_n = cl->init_trip();
@@ -510,7 +510,7 @@ void PhaseIdealLoop::do_peeling( IdealLoopTree *loop, Node_List &old_new ) {
   //         the pre-loop with only 1 user (the new peeled iteration), but the
   //         peeled-loop backedge has 2 users.
   Node* new_exit_value = old_new[head->in(LoopNode::LoopBackControl)->_idx];
-  new_exit_value = move_loop_predicates(entry, new_exit_value);
+  new_exit_value = move_loop_predicates(entry, new_exit_value, !counted_loop);
   _igvn.hash_delete(head);
   head->set_req(LoopNode::EntryControl, new_exit_value);
   for (DUIterator_Fast jmax, j = head->fast_outs(jmax); j < jmax; j++) {
@@ -593,6 +593,12 @@ bool IdealLoopTree::policy_maximally_unroll( PhaseIdealLoop *phase ) const {
     return false;
   }
 
+  // Fully unroll a loop with few iterations regardless next
+  // conditions since following loop optimizations will split
+  // such loop anyway (pre-main-post).
+  if (trip_count <= 3)
+    return true;
+
   // Take into account that after unroll conjoined heads and tails will fold,
   // otherwise policy_unroll() may allow more unrolling than max unrolling.
   uint new_body_size = EMPTY_LOOP_SIZE + (body_size - EMPTY_LOOP_SIZE) * trip_count;
@@ -605,15 +611,6 @@ bool IdealLoopTree::policy_maximally_unroll( PhaseIdealLoop *phase ) const {
     return false;
   }
 
-  // Currently we don't have policy to optimize one iteration loops.
-  // Maximally unrolling transformation is used for that:
-  // it is peeled and the original loop become non reachable (dead).
-  // Also fully unroll a loop with few iterations regardless next
-  // conditions since following loop optimizations will split
-  // such loop anyway (pre-main-post).
-  if (trip_count <= 3)
-    return true;
-
   // Do not unroll a loop with String intrinsics code.
   // String intrinsics are large and have loops.
   for (uint k = 0; k < _body.size(); k++) {
@@ -632,6 +629,8 @@ bool IdealLoopTree::policy_maximally_unroll( PhaseIdealLoop *phase ) const {
 }
 
 
+#define MAX_UNROLL 16 // maximum number of unrolls for main loop
+
 //------------------------------policy_unroll----------------------------------
 // Return TRUE or FALSE if the loop should be unrolled or not.  Unroll if
 // the loop is a CountedLoop and the body is small enough.
@@ -643,13 +642,15 @@ bool IdealLoopTree::policy_unroll( PhaseIdealLoop *phase ) const {
   if (!cl->is_valid_counted_loop())
     return false; // Malformed counted loop
 
-  // protect against over-unrolling
+  // Protect against over-unrolling.
-  if (cl->trip_count() <= 1) return false;
+  // After split at least one iteration will be executed in pre-loop.
-
+  if (cl->trip_count() <= (uint)(cl->is_normal_loop() ? 2 : 1)) return false;
-  // Check for stride being a small enough constant
-  if (abs(cl->stride_con()) > (1<<3)) return false;
 
   int future_unroll_ct = cl->unrolled_count() * 2;
+  if (future_unroll_ct > MAX_UNROLL) return false;
+
+  // Check for initial stride being a small enough constant
+  if (abs(cl->stride_con()) > (1<<2)*future_unroll_ct) return false;
 
   // Don't unroll if the next round of unrolling would push us
   // over the expected trip count of the loop.  One is subtracted
@@ -675,6 +676,7 @@ bool IdealLoopTree::policy_unroll( PhaseIdealLoop *phase ) const {
 
   Node *init_n = cl->init_trip();
   Node *limit_n = cl->limit();
+  int stride_con = cl->stride_con();
   // Non-constant bounds.
   // Protect against over-unrolling when init or/and limit are not constant
   // (so that trip_count's init value is maxint) but iv range is known.
@@ -684,7 +686,7 @@ bool IdealLoopTree::policy_unroll( PhaseIdealLoop *phase ) const {
     if (phi != NULL) {
       assert(phi->is_Phi() && phi->in(0) == _head, "Counted loop should have iv phi.");
       const TypeInt* iv_type = phase->_igvn.type(phi)->is_int();
-      int next_stride = cl->stride_con() * 2; // stride after this unroll
+      int next_stride = stride_con * 2; // stride after this unroll
       if (next_stride > 0) {
         if (iv_type->_lo + next_stride <= iv_type->_lo || // overflow
             iv_type->_lo + next_stride >  iv_type->_hi) {
@@ -699,15 +701,19 @@ bool IdealLoopTree::policy_unroll( PhaseIdealLoop *phase ) const {
     }
   }
 
+  // After unroll limit will be adjusted: new_limit = limit-stride.
+  // Bailout if adjustment overflow.
+  const TypeInt* limit_type = phase->_igvn.type(limit_n)->is_int();
+  if (stride_con > 0 && ((limit_type->_hi - stride_con) >= limit_type->_hi) ||
+      stride_con < 0 && ((limit_type->_lo - stride_con) <= limit_type->_lo))
+    return false;  // overflow
+
   // Adjust body_size to determine if we unroll or not
   uint body_size = _body.size();
-  // Key test to unroll CaffeineMark's Logic test
-  int xors_in_loop = 0;
   // Also count ModL, DivL and MulL which expand mightly
   for (uint k = 0; k < _body.size(); k++) {
     Node* n = _body.at(k);
     switch (n->Opcode()) {
-      case Op_XorI: xors_in_loop++; break; // CaffeineMark's Logic test
       case Op_ModL: body_size += 30; break;
       case Op_DivL: body_size += 30; break;
       case Op_MulL: body_size += 10; break;
@@ -724,7 +730,6 @@ bool IdealLoopTree::policy_unroll( PhaseIdealLoop *phase ) const {
 
   // Check for being too big
   if (body_size > (uint)LoopUnrollLimit) {
-    if (xors_in_loop >= 4 && body_size < (uint)LoopUnrollLimit*4) return true;
      // Normal case: loop too big
     return false;
   }
@@ -747,28 +752,31 @@ bool IdealLoopTree::policy_align( PhaseIdealLoop *phase ) const {
 // Return TRUE or FALSE if the loop should be range-check-eliminated.
 // Actually we do iteration-splitting, a more powerful form of RCE.
 bool IdealLoopTree::policy_range_check( PhaseIdealLoop *phase ) const {
-  if( !RangeCheckElimination ) return false;
+  if (!RangeCheckElimination) return false;
 
   CountedLoopNode *cl = _head->as_CountedLoop();
   // If we unrolled with no intention of doing RCE and we later
   // changed our minds, we got no pre-loop.  Either we need to
   // make a new pre-loop, or we gotta disallow RCE.
-  if( cl->is_main_no_pre_loop() ) return false; // Disallowed for now.
+  if (cl->is_main_no_pre_loop()) return false; // Disallowed for now.
   Node *trip_counter = cl->phi();
 
   // Check loop body for tests of trip-counter plus loop-invariant vs
   // loop-invariant.
-  for( uint i = 0; i < _body.size(); i++ ) {
+  for (uint i = 0; i < _body.size(); i++) {
     Node *iff = _body[i];
-    if( iff->Opcode() == Op_If ) { // Test?
+    if (iff->Opcode() == Op_If) { // Test?
 
       // Comparing trip+off vs limit
       Node *bol = iff->in(1);
-      if( bol->req() != 2 ) continue; // dead constant test
+      if (bol->req() != 2) continue; // dead constant test
       if (!bol->is_Bool()) {
         assert(UseLoopPredicate && bol->Opcode() == Op_Conv2B, "predicate check only");
         continue;
       }
+      if (bol->as_Bool()->_test._test == BoolTest::ne)
+        continue; // not RC
+
       Node *cmp = bol->in(1);
 
       Node *rc_exp = cmp->in(1);
@@ -1064,6 +1072,7 @@ void PhaseIdealLoop::insert_pre_post_loops( IdealLoopTree *loop, Node_List &old_
   // negative stride use >
 
   if (pre_end->in(CountedLoopEndNode::TestValue)->as_Bool()->_test._test == BoolTest::ne) {
+    assert(!LoopLimitCheck, "only canonical tests (lt or gt) are expected");
 
     BoolTest::mask new_test = (main_end->stride_con() > 0) ? BoolTest::lt : BoolTest::gt;
     // Modify pre loop end condition
@@ -1090,6 +1099,9 @@ void PhaseIdealLoop::insert_pre_post_loops( IdealLoopTree *loop, Node_List &old_
   main_head->set_main_loop();
   if( peel_only ) main_head->set_main_no_pre_loop();
 
+  // Subtract a trip count for the pre-loop.
+  main_head->set_trip_count(main_head->trip_count() - 1);
+
   // It's difficult to be precise about the trip-counts
   // for the pre/post loops.  They are usually very short,
   // so guess that 4 trips is a reasonable value.
@@ -1141,7 +1153,8 @@ void PhaseIdealLoop::do_unroll( IdealLoopTree *loop, Node_List &old_new, bool ad
   Node *stride = loop_head->stride();
 
   Node *opaq = NULL;
-  if( adjust_min_trip ) {       // If not maximally unrolling, need adjustment
+  if (adjust_min_trip) {       // If not maximally unrolling, need adjustment
+    // Search for zero-trip guard.
     assert( loop_head->is_main_loop(), "" );
     assert( ctrl->Opcode() == Op_IfTrue || ctrl->Opcode() == Op_IfFalse, "" );
     Node *iff = ctrl->in(0);
@@ -1151,15 +1164,160 @@ void PhaseIdealLoop::do_unroll( IdealLoopTree *loop, Node_List &old_new, bool ad
     Node *cmp = bol->in(1);
     assert( cmp->Opcode() == Op_CmpI, "" );
     opaq = cmp->in(2);
-    // Occasionally it's possible for a pre-loop Opaque1 node to be
+    // Occasionally it's possible for a zero-trip guard Opaque1 node to be
     // optimized away and then another round of loop opts attempted.
     // We can not optimize this particular loop in that case.
-    if( opaq->Opcode() != Op_Opaque1 )
+    if (opaq->Opcode() != Op_Opaque1)
-      return;                   // Cannot find pre-loop!  Bail out!
+      return; // Cannot find zero-trip guard!  Bail out!
+    // Zero-trip test uses an 'opaque' node which is not shared.
+    assert(opaq->outcnt() == 1 && opaq->in(1) == limit, "");
   }
 
   C->set_major_progress();
 
+  Node* new_limit = NULL;
+  if (UnrollLimitCheck) {
+    int stride_con = stride->get_int();
+    int stride_p = (stride_con > 0) ? stride_con : -stride_con;
+    uint old_trip_count = loop_head->trip_count();
+    // Verify that unroll policy result is still valid.
+    assert(old_trip_count > 1 &&
+           (!adjust_min_trip || stride_p <= (1<<3)*loop_head->unrolled_count()), "sanity");
+
+    // Adjust loop limit to keep valid iterations number after unroll.
+    // Use (limit - stride) instead of (((limit - init)/stride) & (-2))*stride
+    // which may overflow.
+    if (!adjust_min_trip) {
+      assert(old_trip_count > 1 && (old_trip_count & 1) == 0,
+             "odd trip count for maximally unroll");
+      // Don't need to adjust limit for maximally unroll since trip count is even.
+    } else if (loop_head->has_exact_trip_count() && init->is_Con()) {
+      // Loop's limit is constant. Loop's init could be constant when pre-loop
+      // become peeled iteration.
+      long init_con = init->get_int();
+      // We can keep old loop limit if iterations count stays the same:
+      //   old_trip_count == new_trip_count * 2
+      // Note: since old_trip_count >= 2 then new_trip_count >= 1
+      // so we also don't need to adjust zero trip test.
+      long limit_con  = limit->get_int();
+      // (stride_con*2) not overflow since stride_con <= 8.
+      int new_stride_con = stride_con * 2;
+      int stride_m    = new_stride_con - (stride_con > 0 ? 1 : -1);
+      long trip_count = (limit_con - init_con + stride_m)/new_stride_con;
+      // New trip count should satisfy next conditions.
+      assert(trip_count > 0 && (julong)trip_count < (julong)max_juint/2, "sanity");
+      uint new_trip_count = (uint)trip_count;
+      adjust_min_trip = (old_trip_count != new_trip_count*2);
+    }
+
+    if (adjust_min_trip) {
+      // Step 2: Adjust the trip limit if it is called for.
+      // The adjustment amount is -stride. Need to make sure if the
+      // adjustment underflows or overflows, then the main loop is skipped.
+      Node* cmp = loop_end->cmp_node();
+      assert(cmp->in(2) == limit, "sanity");
+      assert(opaq != NULL && opaq->in(1) == limit, "sanity");
+
+      // Verify that policy_unroll result is still valid.
+      const TypeInt* limit_type = _igvn.type(limit)->is_int();
+      assert(stride_con > 0 && ((limit_type->_hi - stride_con) < limit_type->_hi) ||
+             stride_con < 0 && ((limit_type->_lo - stride_con) > limit_type->_lo), "sanity");
+
+      if (limit->is_Con()) {
+        // The check in policy_unroll and the assert above guarantee
+        // no underflow if limit is constant.
+        new_limit = _igvn.intcon(limit->get_int() - stride_con);
+        set_ctrl(new_limit, C->root());
+      } else {
+        // Limit is not constant.
+        {
+          // Separate limit by Opaque node in case it is an incremented
+          // variable from previous loop to avoid using pre-incremented
+          // value which could increase register pressure.
+          // Otherwise reorg_offsets() optimization will create a separate
+          // Opaque node for each use of trip-counter and as result
+          // zero trip guard limit will be different from loop limit.
+          assert(has_ctrl(opaq), "should have it");
+          Node* opaq_ctrl = get_ctrl(opaq);
+          limit = new (C, 2) Opaque2Node( C, limit );
+          register_new_node( limit, opaq_ctrl );
+        }
+        if (stride_con > 0 && ((limit_type->_lo - stride_con) < limit_type->_lo) ||
+                   stride_con < 0 && ((limit_type->_hi - stride_con) > limit_type->_hi)) {
+          // No underflow.
+          new_limit = new (C, 3) SubINode(limit, stride);
+        } else {
+          // (limit - stride) may underflow.
+          // Clamp the adjustment value with MININT or MAXINT:
+          //
+          //   new_limit = limit-stride
+          //   if (stride > 0)
+          //     new_limit = (limit < new_limit) ? MININT : new_limit;
+          //   else
+          //     new_limit = (limit > new_limit) ? MAXINT : new_limit;
+          //
+          BoolTest::mask bt = loop_end->test_trip();
+          assert(bt == BoolTest::lt || bt == BoolTest::gt, "canonical test is expected");
+          Node* adj_max = _igvn.intcon((stride_con > 0) ? min_jint : max_jint);
+          set_ctrl(adj_max, C->root());
+          Node* old_limit = NULL;
+          Node* adj_limit = NULL;
+          Node* bol = limit->is_CMove() ? limit->in(CMoveNode::Condition) : NULL;
+          if (loop_head->unrolled_count() > 1 &&
+              limit->is_CMove() && limit->Opcode() == Op_CMoveI &&
+              limit->in(CMoveNode::IfTrue) == adj_max &&
+              bol->as_Bool()->_test._test == bt &&
+              bol->in(1)->Opcode() == Op_CmpI &&
+              bol->in(1)->in(2) == limit->in(CMoveNode::IfFalse)) {
+            // Loop was unrolled before.
+            // Optimize the limit to avoid nested CMove:
+            // use original limit as old limit.
+            old_limit = bol->in(1)->in(1);
+            // Adjust previous adjusted limit.
+            adj_limit = limit->in(CMoveNode::IfFalse);
+            adj_limit = new (C, 3) SubINode(adj_limit, stride);
+          } else {
+            old_limit = limit;
+            adj_limit = new (C, 3) SubINode(limit, stride);
+          }
+          assert(old_limit != NULL && adj_limit != NULL, "");
+          register_new_node( adj_limit, ctrl ); // adjust amount
+          Node* adj_cmp = new (C, 3) CmpINode(old_limit, adj_limit);
+          register_new_node( adj_cmp, ctrl );
+          Node* adj_bool = new (C, 2) BoolNode(adj_cmp, bt);
+          register_new_node( adj_bool, ctrl );
+          new_limit = new (C, 4) CMoveINode(adj_bool, adj_limit, adj_max, TypeInt::INT);
+        }
+        register_new_node(new_limit, ctrl);
+      }
+      assert(new_limit != NULL, "");
+      // Replace in loop test.
+      _igvn.hash_delete(cmp);
+      cmp->set_req(2, new_limit);
+
+      // Step 3: Find the min-trip test guaranteed before a 'main' loop.
+      // Make it a 1-trip test (means at least 2 trips).
+
+      // Guard test uses an 'opaque' node which is not shared.  Hence I
+      // can edit it's inputs directly.  Hammer in the new limit for the
+      // minimum-trip guard.
+      assert(opaq->outcnt() == 1, "");
+      _igvn.hash_delete(opaq);
+      opaq->set_req(1, new_limit);
+    }
+
+    // Adjust max trip count. The trip count is intentionally rounded
+    // down here (e.g. 15-> 7-> 3-> 1) because if we unwittingly over-unroll,
+    // the main, unrolled, part of the loop will never execute as it is protected
+    // by the min-trip test.  See bug 4834191 for a case where we over-unrolled
+    // and later determined that part of the unrolled loop was dead.
+    loop_head->set_trip_count(old_trip_count / 2);
+
+    // Double the count of original iterations in the unrolled loop body.
+    loop_head->double_unrolled_count();
+
+  } else { // LoopLimitCheck
+
     // Adjust max trip count. The trip count is intentionally rounded
     // down here (e.g. 15-> 7-> 3-> 1) because if we unwittingly over-unroll,
     // the main, unrolled, part of the loop will never execute as it is protected
@@ -1186,12 +1344,12 @@ void PhaseIdealLoop::do_unroll( IdealLoopTree *loop, Node_List &old_new, bool ad
     register_new_node( rond, ctrl );
     Node *spn2 = new (C, 3) MulINode( rond, stride );
     register_new_node( spn2, ctrl );
-  Node *lim2 = new (C, 3) AddINode( spn2, init );
+    new_limit = new (C, 3) AddINode( spn2, init );
-  register_new_node( lim2, ctrl );
+    register_new_node( new_limit, ctrl );
 
     // Hammer in the new limit
     Node *ctrl2 = loop_end->in(0);
-  Node *cmp2 = new (C, 3) CmpINode( loop_head->incr(), lim2 );
+    Node *cmp2 = new (C, 3) CmpINode( loop_head->incr(), new_limit );
     register_new_node( cmp2, ctrl2 );
     Node *bol2 = new (C, 2) BoolNode( cmp2, loop_end->test_trip() );
     register_new_node( bol2, ctrl2 );
@@ -1201,13 +1359,15 @@ void PhaseIdealLoop::do_unroll( IdealLoopTree *loop, Node_List &old_new, bool ad
     // Step 3: Find the min-trip test guaranteed before a 'main' loop.
     // Make it a 1-trip test (means at least 2 trips).
     if( adjust_min_trip ) {
+      assert( new_limit != NULL, "" );
       // Guard test uses an 'opaque' node which is not shared.  Hence I
       // can edit it's inputs directly.  Hammer in the new limit for the
       // minimum-trip guard.
       assert( opaq->outcnt() == 1, "" );
       _igvn.hash_delete(opaq);
-    opaq->set_req(1, lim2);
+      opaq->set_req(1, new_limit);
     }
+  } // LoopLimitCheck
 
   // ---------
   // Step 4: Clone the loop body.  Move it inside the loop.  This loop body
@@ -1263,6 +1423,7 @@ void PhaseIdealLoop::do_unroll( IdealLoopTree *loop, Node_List &old_new, bool ad
 
 void PhaseIdealLoop::do_maximally_unroll( IdealLoopTree *loop, Node_List &old_new ) {
   CountedLoopNode *cl = loop->_head->as_CountedLoop();
+  assert(cl->has_exact_trip_count(), "trip count is not exact");
   assert(cl->trip_count() > 0, "");
 #ifndef PRODUCT
   if (TraceLoopOpts) {
@@ -1279,6 +1440,7 @@ void PhaseIdealLoop::do_maximally_unroll( IdealLoopTree *loop, Node_List &old_ne
   // Now its tripping an even number of times remaining.  Double loop body.
   // Do not adjust pre-guards; they are not needed and do not exist.
   if (cl->trip_count() > 0) {
+    assert((cl->trip_count() & 1) == 0, "missed peeling");
     do_unroll(loop, old_new, false);
   }
 }
@@ -1292,22 +1454,13 @@ bool IdealLoopTree::dominates_backedge(Node* ctrl) {
 }
 
 //------------------------------add_constraint---------------------------------
-// Constrain the main loop iterations so the condition:
+// Constrain the main loop iterations so the conditions:
-//    scale_con * I + offset  <  limit
+//    low_limit <= scale_con * I + offset  <  upper_limit
 // always holds true.  That is, either increase the number of iterations in
 // the pre-loop or the post-loop until the condition holds true in the main
 // loop.  Stride, scale, offset and limit are all loop invariant.  Further,
 // stride and scale are constants (offset and limit often are).
-void PhaseIdealLoop::add_constraint( int stride_con, int scale_con, Node *offset, Node *limit, Node *pre_ctrl, Node **pre_limit, Node **main_limit ) {
+void PhaseIdealLoop::add_constraint( int stride_con, int scale_con, Node *offset, Node *low_limit, Node *upper_limit, Node *pre_ctrl, Node **pre_limit, Node **main_limit ) {
-
-  // Compute "I :: (limit-offset)/scale_con"
-  Node *con = new (C, 3) SubINode( limit, offset );
-  register_new_node( con, pre_ctrl );
-  Node *scale = _igvn.intcon(scale_con);
-  set_ctrl(scale, C->root());
-  Node *X = new (C, 3) DivINode( 0, con, scale );
-  register_new_node( X, pre_ctrl );
-
   // For positive stride, the pre-loop limit always uses a MAX function
   // and the main loop a MIN function.  For negative stride these are
   // reversed.
@@ -1316,48 +1469,143 @@ void PhaseIdealLoop::add_constraint( int stride_con, int scale_con, Node *offset
   // pre-loop must check for underflow and the post-loop for overflow.
   // Negative stride*scale reverses this; pre-loop checks for overflow and
   // post-loop for underflow.
-  if( stride_con*scale_con > 0 ) {
+  if (stride_con*scale_con > 0) {
-    // Compute I < (limit-offset)/scale_con
+    // The overflow limit: scale*I+offset < upper_limit
-    // Adjust main-loop last iteration to be MIN/MAX(main_loop,X)
+    // For main-loop compute
-    *main_limit = (stride_con > 0)
+    //   ( if (scale > 0) /* and stride > 0 */
-      ? (Node*)(new (C, 3) MinINode( *main_limit, X ))
+    //       I < (upper_limit-offset)/scale
-      : (Node*)(new (C, 3) MaxINode( *main_limit, X ));
+    //     else /* scale < 0 and stride < 0 */
-    register_new_node( *main_limit, pre_ctrl );
+    //       I > (upper_limit-offset)/scale
+    //   )
+    //
+    // (upper_limit-offset) may overflow when offset < 0.
+    // But it is fine since main loop will either have
+    // less iterations or will be skipped in such case.
+    Node *con = new (C, 3) SubINode(upper_limit, offset);
+    register_new_node(con, pre_ctrl);
+    Node *scale = _igvn.intcon(scale_con);
+    set_ctrl(scale, C->root());
+    Node *X = new (C, 3) DivINode(0, con, scale);
+    register_new_node(X, pre_ctrl);
 
+    // Adjust main-loop last iteration
+    Node *loop_limit = *main_limit;
+    loop_limit = (stride_con > 0) // scale > 0
+      ? (Node*)(new (C, 3) MinINode(loop_limit, X))
+      : (Node*)(new (C, 3) MaxINode(loop_limit, X));
+    register_new_node(loop_limit, pre_ctrl);
+    *main_limit = loop_limit;
+
+    // The underflow limit: low_limit <= scale*I+offset.
+    // For pre-loop compute
+    //   NOT(scale*I+offset >= low_limit)
+    //   scale*I+offset < low_limit
+    //   ( if (scale > 0) /* and stride > 0 */
+    //       I < (low_limit-offset)/scale
+    //     else /* scale < 0 and stride < 0 */
+    //       I > (low_limit-offset)/scale
+    //   )
+
+    if (low_limit->get_int() == -max_jint) {
+      if (!RangeLimitCheck) return;
+      // We need this guard when scale*pre_limit+offset >= limit
+      // due to underflow so we need execute pre-loop until
+      // scale*I+offset >= min_int. But (low_limit-offset) will
+      // underflow when offset > 0 and X will be > original_limit.
+      // To avoid it we replace offset = offset > 0 ? 0 : offset
+      // and add min(pre_limit, original_limit).
+      Node* shift = _igvn.intcon(31);
+      set_ctrl(shift, C->root());
+      Node *neg_off = new (C, 3) RShiftINode(offset, shift);
+      register_new_node(neg_off, pre_ctrl);
+      offset = new (C, 3) AndINode(offset, neg_off);
+      register_new_node(offset, pre_ctrl);
     } else {
-    // Compute (limit-offset)/scale_con + SGN(-scale_con) <= I
+      assert(low_limit->get_int() == 0, "wrong low limit for range check");
-    // Add the negation of the main-loop constraint to the pre-loop.
+      // The only problem we have here when offset == min_int
-    // See footnote [++] below for a derivation of the limit expression.
+      // since (0-min_int) == min_int. It may be fine for scale > 0
-    Node *incr = _igvn.intcon(scale_con > 0 ? -1 : 1);
+      // but for scale < 0 X will be < original_limit.
-    set_ctrl(incr, C->root());
+    }
-    Node *adj = new (C, 3) AddINode( X, incr );
+    con = new (C, 3) SubINode(low_limit, offset);
-    register_new_node( adj, pre_ctrl );
+    register_new_node(con, pre_ctrl);
-    *pre_limit = (scale_con > 0)
+    scale = _igvn.intcon(scale_con);
-      ? (Node*)new (C, 3) MinINode( *pre_limit, adj )
+    set_ctrl(scale, C->root());
-      : (Node*)new (C, 3) MaxINode( *pre_limit, adj );
+    X = new (C, 3) DivINode(0, con, scale);
-    register_new_node( *pre_limit, pre_ctrl );
+    register_new_node(X, pre_ctrl);
+
+    // Adjust pre-loop last iteration
+    loop_limit = *pre_limit;
+    loop_limit = (stride_con > 0) // scale > 0
+      ? (Node*)(new (C, 3) MaxINode(loop_limit, X))
+      : (Node*)(new (C, 3) MinINode(loop_limit, X));
+    register_new_node( loop_limit, pre_ctrl );
+    *pre_limit = loop_limit;
+
+  } else { // stride_con*scale_con < 0
+    // For negative stride*scale pre-loop checks for overflow and
+    // post-loop for underflow.
+    //
+    // The underflow limit: low_limit <= scale*I+offset.
+    // For main-loop compute
+    //   scale*I+offset+1 > low_limit
+    //   ( if (scale < 0) /* and stride > 0 */
+    //       I < (low_limit-(offset+1))/scale
+    //     else /* scale < 0 and stride < 0 */
+    //       I > (low_limit-(offset+1))/scale
+    //   )
+
+    if (low_limit->get_int() == -max_jint) {
+      if (!RangeLimitCheck) return;
+    } else {
+      assert(low_limit->get_int() == 0, "wrong low limit for range check");
+    }
+
+    Node *one  = _igvn.intcon(1);
+    set_ctrl(one, C->root());
+    Node *plus_one = new (C, 3) AddINode(offset, one);
+    register_new_node( plus_one, pre_ctrl );
+    Node *con = new (C, 3) SubINode(low_limit, plus_one);
+    register_new_node(con, pre_ctrl);
+    Node *scale = _igvn.intcon(scale_con);
+    set_ctrl(scale, C->root());
+    Node *X = new (C, 3) DivINode(0, con, scale);
+    register_new_node(X, pre_ctrl);
+
+    // Adjust main-loop last iteration
+    Node *loop_limit = *main_limit;
+    loop_limit = (stride_con > 0) // scale < 0
+      ? (Node*)(new (C, 3) MinINode(loop_limit, X))
+      : (Node*)(new (C, 3) MaxINode(loop_limit, X));
+    register_new_node(loop_limit, pre_ctrl);
+    *main_limit = loop_limit;
+
+    // The overflow limit: scale*I+offset < upper_limit
+    // For pre-loop compute
+    //   NOT(scale*I+offset < upper_limit)
+    //   scale*I+offset >= upper_limit
+    //   scale*I+offset+1 > upper_limit
+    //   ( if (scale < 0) /* and stride > 0 */
+    //       I < (upper_limit-(offset+1))/scale
+    //     else /* scale < 0 and stride < 0 */
+    //       I > (upper_limit-(offset+1))/scale
+    //   )
+    plus_one = new (C, 3) AddINode(offset, one);
+    register_new_node( plus_one, pre_ctrl );
+    con = new (C, 3) SubINode(upper_limit, plus_one);
+    register_new_node(con, pre_ctrl);
+    scale = _igvn.intcon(scale_con);
+    set_ctrl(scale, C->root());
+    X = new (C, 3) DivINode(0, con, scale);
+    register_new_node(X, pre_ctrl);
+
+    // Adjust pre-loop last iteration
+    loop_limit = *pre_limit;
+    loop_limit = (stride_con > 0) // scale < 0
+      ? (Node*)(new (C, 3) MaxINode(loop_limit, X))
+      : (Node*)(new (C, 3) MinINode(loop_limit, X));
+    register_new_node( loop_limit, pre_ctrl );
+    *pre_limit = loop_limit;
 
-//   [++] Here's the algebra that justifies the pre-loop limit expression:
-//
-//   NOT( scale_con * I + offset  <  limit )
-//      ==
-//   scale_con * I + offset  >=  limit
-//      ==
-//   SGN(scale_con) * I  >=  (limit-offset)/|scale_con|
-//      ==
-//   (limit-offset)/|scale_con|   <=  I * SGN(scale_con)
-//      ==
-//   (limit-offset)/|scale_con|-1  <  I * SGN(scale_con)
-//      ==
-//   ( if (scale_con > 0) /*common case*/
-//       (limit-offset)/scale_con - 1  <  I
-//     else
-//       (limit-offset)/scale_con + 1  >  I
-//    )
-//   ( if (scale_con > 0) /*common case*/
-//       (limit-offset)/scale_con + SGN(-scale_con)  <  I
-//     else
-//       (limit-offset)/scale_con + SGN(-scale_con)  >  I
   }
 }
 
@@ -1488,7 +1736,7 @@ void PhaseIdealLoop::do_range_check( IdealLoopTree *loop, Node_List &old_new ) {
   Node *cmpzm = bolzm->in(1);
   assert(cmpzm->is_Cmp(), "");
   Node *opqzm = cmpzm->in(2);
-  // Can not optimize a loop if pre-loop Opaque1 node is optimized
+  // Can not optimize a loop if zero-trip Opaque1 node is optimized
   // away and then another round of loop opts attempted.
   if (opqzm->Opcode() != Op_Opaque1)
     return;
@@ -1523,8 +1771,11 @@ void PhaseIdealLoop::do_range_check( IdealLoopTree *loop, Node_List &old_new ) {
   int stride_con = cl->stride_con();
   Node *zero = _igvn.intcon(0);
   Node *one  = _igvn.intcon(1);
+  // Use symmetrical int range [-max_jint,max_jint]
+  Node *mini = _igvn.intcon(-max_jint);
   set_ctrl(zero, C->root());
   set_ctrl(one,  C->root());
+  set_ctrl(mini, C->root());
 
   // Range checks that do not dominate the loop backedge (ie.
   // conditionally executed) can lengthen the pre loop limit beyond
@@ -1599,7 +1850,12 @@ void PhaseIdealLoop::do_range_check( IdealLoopTree *loop, Node_List &old_new ) {
       if( offset_c == ctrl ) {
         continue; // Don't rce this check but continue looking for other candidates.
       }
-
+#ifdef ASSERT
+      if (TraceRangeLimitCheck) {
+        tty->print_cr("RC bool node%s", flip ? " flipped:" : ":");
+        bol->dump(2);
+      }
+#endif
       // At this point we have the expression as:
       //   scale_con * trip_counter + offset :: limit
       // where scale_con, offset and limit are loop invariant.  Trip_counter
@@ -1610,17 +1866,16 @@ void PhaseIdealLoop::do_range_check( IdealLoopTree *loop, Node_List &old_new ) {
       // Adjust pre and main loop limits to guard the correct iteration set
       if( cmp->Opcode() == Op_CmpU ) {// Unsigned compare is really 2 tests
         if( b_test._test == BoolTest::lt ) { // Range checks always use lt
-          // The overflow limit: scale*I+offset < limit
+          // The underflow and overflow limits: 0 <= scale*I+offset < limit
-          add_constraint( stride_con, scale_con, offset, limit, pre_ctrl, &pre_limit, &main_limit );
+          add_constraint( stride_con, scale_con, offset, zero, limit, pre_ctrl, &pre_limit, &main_limit );
-          // The underflow limit: 0 <= scale*I+offset.
-          // Some math yields: -scale*I-(offset+1) < 0
-          Node *plus_one = new (C, 3) AddINode( offset, one );
-          register_new_node( plus_one, pre_ctrl );
-          Node *neg_offset = new (C, 3) SubINode( zero, plus_one );
-          register_new_node( neg_offset, pre_ctrl );
-          add_constraint( stride_con, -scale_con, neg_offset, zero, pre_ctrl, &pre_limit, &main_limit );
           if (!conditional_rc) {
             conditional_rc = !loop->dominates_backedge(iff);
+            // It is also needed if offset->_lo == min_int since
+            // (0-min_int) == min_int. It may be fine for stride > 0
+            // but for stride < 0 pre_limit will be < original_limit.
+            const TypeInt* offset_t = _igvn.type(offset)->is_int();
+            conditional_rc |= RangeLimitCheck && (offset_t->_lo == min_jint) &&
+                              (scale_con<0) && (stride_con<0);
           }
         } else {
 #ifndef PRODUCT
@@ -1631,21 +1886,35 @@ void PhaseIdealLoop::do_range_check( IdealLoopTree *loop, Node_List &old_new ) {
         }
       } else {                  // Otherwise work on normal compares
         switch( b_test._test ) {
-        case BoolTest::ge:      // Convert X >= Y to -X <= -Y
+        case BoolTest::gt:
+          // Fall into GE case
+        case BoolTest::ge:
+          // Convert (I*scale+offset) >= Limit to (I*(-scale)+(-offset)) <= -Limit
           scale_con = -scale_con;
           offset = new (C, 3) SubINode( zero, offset );
           register_new_node( offset, pre_ctrl );
           limit  = new (C, 3) SubINode( zero, limit  );
           register_new_node( limit, pre_ctrl );
           // Fall into LE case
-        case BoolTest::le:      // Convert X <= Y to X < Y+1
+        case BoolTest::le:
+          if (b_test._test != BoolTest::gt) {
+            // Convert X <= Y to X < Y+1
             limit = new (C, 3) AddINode( limit, one );
             register_new_node( limit, pre_ctrl );
+          }
           // Fall into LT case
         case BoolTest::lt:
-          add_constraint( stride_con, scale_con, offset, limit, pre_ctrl, &pre_limit, &main_limit );
+          // The underflow and overflow limits: MIN_INT <= scale*I+offset < limit
+          add_constraint( stride_con, scale_con, offset, mini, limit, pre_ctrl, &pre_limit, &main_limit );
           if (!conditional_rc) {
             conditional_rc = !loop->dominates_backedge(iff);
+            // It is also needed if scale*pre_limit+offset >= limit
+            // due to underflow so we need execute pre-loop until
+            // scale*I+offset >= min_int. But (low_limit-offset) will
+            // underflow when offset > 0 and X will be > original_limit.
+            const TypeInt* offset_t = _igvn.type(offset)->is_int();
+            conditional_rc |= RangeLimitCheck && (offset_t->_hi > 0) &&
+                              (scale_con>0) && (stride_con>0);
           }
           break;
         default:
@@ -1696,7 +1965,8 @@ void PhaseIdealLoop::do_range_check( IdealLoopTree *loop, Node_List &old_new ) {
 
   // Note:: we are making the main loop limit no longer precise;
   // need to round up based on stride.
-  if( stride_con != 1 && stride_con != -1 ) { // Cutout for common case
+  cl->set_nonexact_trip_count();
+  if (!LoopLimitCheck && stride_con != 1 && stride_con != -1) { // Cutout for common case
     // "Standard" round-up logic:  ([main_limit-init+(y-1)]/y)*y+init
     // Hopefully, compiler will optimize for powers of 2.
     Node *ctrl = get_ctrl(main_limit);
@@ -1876,7 +2146,19 @@ bool IdealLoopTree::policy_do_remove_empty_loop( PhaseIdealLoop *phase ) {
   // iteration.  Then the CountedLoopEnd will collapse (backedge never
   // taken) and all loop-invariant uses of the exit values will be correct.
   Node *phi = cl->phi();
-  Node *final = new (phase->C, 3) SubINode( cl->limit(), cl->stride() );
+  Node *exact_limit = phase->exact_limit(this);
+  if (exact_limit != cl->limit()) {
+    // We also need to replace the original limit to collapse loop exit.
+    Node* cmp = cl->loopexit()->cmp_node();
+    assert(cl->limit() == cmp->in(2), "sanity");
+    phase->_igvn._worklist.push(cmp->in(2)); // put limit on worklist
+    phase->_igvn.hash_delete(cmp);
+    cmp->set_req(2, exact_limit);
+    phase->_igvn._worklist.push(cmp);        // put cmp on worklist
+  }
+  // Note: the final value after increment should not overflow since
+  // counted loop has limit check predicate.
+  Node *final = new (phase->C, 3) SubINode( exact_limit, cl->stride() );
   phase->register_new_node(final,cl->in(LoopNode::EntryControl));
   phase->_igvn.replace_node(phi,final);
   phase->C->set_major_progress();

hotspot/src/share/vm/opto/loopUnswitch.cpp

@@ -130,6 +130,11 @@ void PhaseIdealLoop::do_unswitching (IdealLoopTree *loop, Node_List &old_new) {
   Node* uniqc = proj_true->unique_ctrl_out();
   Node* entry = head->in(LoopNode::EntryControl);
   Node* predicate = find_predicate(entry);
+  if (predicate != NULL && LoopLimitCheck && UseLoopPredicate) {
+    // We may have two predicates, find first.
+    entry = find_predicate(entry->in(0)->in(0));
+    if (entry != NULL) predicate = entry;
+  }
   if (predicate != NULL) predicate = predicate->in(0);
   assert(proj_true->is_IfTrue() &&
          (predicate == NULL && uniqc == head ||
@@ -217,6 +222,7 @@ void PhaseIdealLoop::do_unswitching (IdealLoopTree *loop, Node_List &old_new) {
 ProjNode* PhaseIdealLoop::create_slow_version_of_loop(IdealLoopTree *loop,
                                                       Node_List &old_new) {
   LoopNode* head  = loop->_head->as_Loop();
+  bool counted_loop = head->is_CountedLoop();
   Node*     entry = head->in(LoopNode::EntryControl);
   _igvn.hash_delete(entry);
   _igvn._worklist.push(entry);
@@ -242,14 +248,14 @@ ProjNode* PhaseIdealLoop::create_slow_version_of_loop(IdealLoopTree *loop,
   assert(old_new[head->_idx]->is_Loop(), "" );
 
   // Fast (true) control
-  Node* iffast_pred = clone_loop_predicates(entry, iffast);
+  Node* iffast_pred = clone_loop_predicates(entry, iffast, !counted_loop);
   _igvn.hash_delete(head);
   head->set_req(LoopNode::EntryControl, iffast_pred);
   set_idom(head, iffast_pred, dom_depth(head));
   _igvn._worklist.push(head);
 
   // Slow (false) control
-  Node* ifslow_pred = move_loop_predicates(entry, ifslow);
+  Node* ifslow_pred = move_loop_predicates(entry, ifslow, !counted_loop);
   LoopNode* slow_head = old_new[head->_idx]->as_Loop();
   _igvn.hash_delete(slow_head);
   slow_head->set_req(LoopNode::EntryControl, ifslow_pred);

hotspot/src/share/vm/opto/loopnode.cpp

@@ -206,7 +206,7 @@ bool PhaseIdealLoop::is_counted_loop( Node *x, IdealLoopTree *loop ) {
   // Get backedge compare
   Node *cmp = test->in(1);
   int cmp_op = cmp->Opcode();
-  if( cmp_op != Op_CmpI )
+  if (cmp_op != Op_CmpI)
     return false;                // Avoid pointer & float compares
 
   // Find the trip-counter increment & limit.  Limit must be loop invariant.
@@ -259,7 +259,8 @@ bool PhaseIdealLoop::is_counted_loop( Node *x, IdealLoopTree *loop ) {
   }
   // Stride must be constant
   int stride_con = stride->get_int();
-  assert(stride_con != 0, "missed some peephole opt");
+  if (stride_con == 0)
+    return false; // missed some peephole opt
 
   if (!xphi->is_Phi())
     return false; // Too much math on the trip counter
@@ -319,7 +320,7 @@ bool PhaseIdealLoop::is_counted_loop( Node *x, IdealLoopTree *loop ) {
       // Count down loop rolls through MAXINT
       (bt == BoolTest::le || bt == BoolTest::lt) && stride_con < 0 ||
       // Count up loop rolls through MININT
-      (bt == BoolTest::ge || bt == BoolTest::gt) && stride_con > 0 ) {
+      (bt == BoolTest::ge || bt == BoolTest::gt) && stride_con > 0) {
     return false; // Bail out
   }
 
@@ -341,12 +342,137 @@ bool PhaseIdealLoop::is_counted_loop( Node *x, IdealLoopTree *loop ) {
   //
   assert(x->Opcode() == Op_Loop, "regular loops only");
   C->print_method("Before CountedLoop", 3);
-#ifndef PRODUCT
+
-  if (TraceLoopOpts) {
+  Node *hook = new (C, 6) Node(6);
-    tty->print("Counted      ");
+
+  if (LoopLimitCheck) {
+
+  // ===================================================
+  // Generate loop limit check to avoid integer overflow
+  // in cases like next (cyclic loops):
+  //
+  // for (i=0; i <= max_jint; i++) {}
+  // for (i=0; i <  max_jint; i+=2) {}
+  //
+  //
+  // Limit check predicate depends on the loop test:
+  //
+  // for(;i != limit; i++)       --> limit <= (max_jint)
+  // for(;i <  limit; i+=stride) --> limit <= (max_jint - stride + 1)
+  // for(;i <= limit; i+=stride) --> limit <= (max_jint - stride    )
+  //
+
+  // Check if limit is excluded to do more precise int overflow check.
+  bool incl_limit = (bt == BoolTest::le || bt == BoolTest::ge);
+  int stride_m  = stride_con - (incl_limit ? 0 : (stride_con > 0 ? 1 : -1));
+
+  // If compare points directly to the phi we need to adjust
+  // the compare so that it points to the incr. Limit have
+  // to be adjusted to keep trip count the same and the
+  // adjusted limit should be checked for int overflow.
+  if (phi_incr != NULL) {
+    stride_m  += stride_con;
+  }
+
+  if (limit->is_Con()) {
+    int limit_con = limit->get_int();
+    if ((stride_con > 0 && limit_con > (max_jint - stride_m)) ||
+        (stride_con < 0 && limit_con < (min_jint - stride_m))) {
+      // Bailout: it could be integer overflow.
+      return false;
+    }
+  } else if ((stride_con > 0 && limit_t->_hi <= (max_jint - stride_m)) ||
+             (stride_con < 0 && limit_t->_lo >= (min_jint - stride_m))) {
+      // Limit's type may satisfy the condition, for example,
+      // when it is an array length.
+  } else {
+    // Generate loop's limit check.
+    // Loop limit check predicate should be near the loop.
+    ProjNode *limit_check_proj = find_predicate_insertion_point(init_control, Deoptimization::Reason_loop_limit_check);
+    if (!limit_check_proj) {
+      // The limit check predicate is not generated if this method trapped here before.
+#ifdef ASSERT
+      if (TraceLoopLimitCheck) {
+        tty->print("missing loop limit check:");
         loop->dump_head();
+        x->dump(1);
       }
 #endif
+      return false;
+    }
+
+    IfNode* check_iff = limit_check_proj->in(0)->as_If();
+    Node* cmp_limit;
+    Node* bol;
+
+    if (stride_con > 0) {
+      cmp_limit = new (C, 3) CmpINode(limit, _igvn.intcon(max_jint - stride_m));
+      bol = new (C, 2) BoolNode(cmp_limit, BoolTest::le);
+    } else {
+      cmp_limit = new (C, 3) CmpINode(limit, _igvn.intcon(min_jint - stride_m));
+      bol = new (C, 2) BoolNode(cmp_limit, BoolTest::ge);
+    }
+    cmp_limit = _igvn.register_new_node_with_optimizer(cmp_limit);
+    bol = _igvn.register_new_node_with_optimizer(bol);
+    set_subtree_ctrl(bol);
+
+    // Replace condition in original predicate but preserve Opaque node
+    // so that previous predicates could be found.
+    assert(check_iff->in(1)->Opcode() == Op_Conv2B &&
+           check_iff->in(1)->in(1)->Opcode() == Op_Opaque1, "");
+    Node* opq = check_iff->in(1)->in(1);
+    _igvn.hash_delete(opq);
+    opq->set_req(1, bol);
+    // Update ctrl.
+    set_ctrl(opq, check_iff->in(0));
+    set_ctrl(check_iff->in(1), check_iff->in(0));
+
+#ifndef PRODUCT
+    // report that the loop predication has been actually performed
+    // for this loop
+    if (TraceLoopLimitCheck) {
+      tty->print_cr("Counted Loop Limit Check generated:");
+      debug_only( bol->dump(2); )
+    }
+#endif
+  }
+
+  if (phi_incr != NULL) {
+    // If compare points directly to the phi we need to adjust
+    // the compare so that it points to the incr. Limit have
+    // to be adjusted to keep trip count the same and we
+    // should avoid int overflow.
+    //
+    //   i = init; do {} while(i++ < limit);
+    // is converted to
+    //   i = init; do {} while(++i < limit+1);
+    //
+    limit = gvn->transform(new (C, 3) AddINode(limit, stride));
+  }
+
+  // Now we need to canonicalize loop condition.
+  if (bt == BoolTest::ne) {
+    assert(stride_con == 1 || stride_con == -1, "simple increment only");
+    bt = (stride_con > 0) ? BoolTest::lt : BoolTest::gt;
+  }
+
+  if (incl_limit) {
+    // The limit check guaranties that 'limit <= (max_jint - stride)' so
+    // we can convert 'i <= limit' to 'i < limit+1' since stride != 0.
+    //
+    Node* one = (stride_con > 0) ? gvn->intcon( 1) : gvn->intcon(-1);
+    limit = gvn->transform(new (C, 3) AddINode(limit, one));
+    if (bt == BoolTest::le)
+      bt = BoolTest::lt;
+    else if (bt == BoolTest::ge)
+      bt = BoolTest::gt;
+    else
+      ShouldNotReachHere();
+  }
+  set_subtree_ctrl( limit );
+
+  } else { // LoopLimitCheck
+
   // If compare points to incr, we are ok.  Otherwise the compare
   // can directly point to the phi; in this case adjust the compare so that
   // it points to the incr by adjusting the limit.
@@ -359,7 +485,6 @@ bool PhaseIdealLoop::is_counted_loop( Node *x, IdealLoopTree *loop ) {
   Node *one_m = gvn->intcon(-1);
 
   Node *trip_count = NULL;
-  Node *hook = new (C, 6) Node(6);
   switch( bt ) {
   case BoolTest::eq:
     ShouldNotReachHere();
@@ -441,6 +566,8 @@ bool PhaseIdealLoop::is_counted_loop( Node *x, IdealLoopTree *loop ) {
   limit = gvn->transform(new (C, 3) AddINode(span,init_trip));
   set_subtree_ctrl( limit );
 
+  } // LoopLimitCheck
+
   // Check for SafePoint on backedge and remove
   Node *sfpt = x->in(LoopNode::LoopBackControl);
   if (sfpt->Opcode() == Op_SafePoint && is_deleteable_safept(sfpt)) {
@@ -531,7 +658,7 @@ bool PhaseIdealLoop::is_counted_loop( Node *x, IdealLoopTree *loop ) {
 
   // Check for immediately preceding SafePoint and remove
   Node *sfpt2 = le->in(0);
-  if( sfpt2->Opcode() == Op_SafePoint && is_deleteable_safept(sfpt2))
+  if (sfpt2->Opcode() == Op_SafePoint && is_deleteable_safept(sfpt2))
     lazy_replace( sfpt2, sfpt2->in(TypeFunc::Control));
 
   // Free up intermediate goo
@@ -541,12 +668,56 @@ bool PhaseIdealLoop::is_counted_loop( Node *x, IdealLoopTree *loop ) {
   assert(l->is_valid_counted_loop(), "counted loop shape is messed up");
   assert(l == loop->_head && l->phi() == phi && l->loopexit() == lex, "" );
 #endif
+#ifndef PRODUCT
+  if (TraceLoopOpts) {
+    tty->print("Counted      ");
+    loop->dump_head();
+  }
+#endif
 
   C->print_method("After CountedLoop", 3);
 
   return true;
 }
 
+//----------------------exact_limit-------------------------------------------
+Node* PhaseIdealLoop::exact_limit( IdealLoopTree *loop ) {
+  assert(loop->_head->is_CountedLoop(), "");
+  CountedLoopNode *cl = loop->_head->as_CountedLoop();
+
+  if (!LoopLimitCheck || ABS(cl->stride_con()) == 1 ||
+      cl->limit()->Opcode() == Op_LoopLimit) {
+    // Old code has exact limit (it could be incorrect in case of int overflow).
+    // Loop limit is exact with stride == 1. And loop may already have exact limit.
+    return cl->limit();
+  }
+  Node *limit = NULL;
+#ifdef ASSERT
+  BoolTest::mask bt = cl->loopexit()->test_trip();
+  assert(bt == BoolTest::lt || bt == BoolTest::gt, "canonical test is expected");
+#endif
+  if (cl->has_exact_trip_count()) {
+    // Simple case: loop has constant boundaries.
+    // Use longs to avoid integer overflow.
+    int stride_con = cl->stride_con();
+    long  init_con = cl->init_trip()->get_int();
+    long limit_con = cl->limit()->get_int();
+    julong trip_cnt = cl->trip_count();
+    long final_con = init_con + trip_cnt*stride_con;
+    final_con -= stride_con;
+    int final_int = (int)final_con;
+    // The final value should be in integer range since the loop
+    // is counted and the limit was checked for overflow.
+    assert(final_con == (long)final_int, "final value should be integer");
+    limit = _igvn.intcon(final_int);
+  } else {
+    // Create new LoopLimit node to get exact limit (final iv value).
+    limit = new (C, 4) LoopLimitNode(C, cl->init_trip(), cl->limit(), cl->stride());
+    register_new_node(limit, cl->in(LoopNode::EntryControl));
+  }
+  assert(limit != NULL, "sanity");
+  return limit;
+}
 
 //------------------------------Ideal------------------------------------------
 // Return a node which is more "ideal" than the current node.
@@ -572,14 +743,12 @@ Node *CountedLoopNode::Ideal(PhaseGVN *phase, bool can_reshape) {
 #ifndef PRODUCT
 void CountedLoopNode::dump_spec(outputStream *st) const {
   LoopNode::dump_spec(st);
-  if( stride_is_con() ) {
+  if (stride_is_con()) {
     st->print("stride: %d ",stride_con());
-  } else {
-    st->print("stride: not constant ");
   }
-  if( is_pre_loop () ) st->print("pre of N%d" , _main_idx );
+  if (is_pre_loop ()) st->print("pre of N%d" , _main_idx);
-  if( is_main_loop() ) st->print("main of N%d", _idx );
+  if (is_main_loop()) st->print("main of N%d", _idx);
-  if( is_post_loop() ) st->print("post of N%d", _main_idx );
+  if (is_post_loop()) st->print("post of N%d", _main_idx);
 }
 #endif
 
@@ -588,7 +757,130 @@ int CountedLoopEndNode::stride_con() const {
   return stride()->bottom_type()->is_int()->get_con();
 }
 
+//=============================================================================
+//------------------------------Value-----------------------------------------
+const Type *LoopLimitNode::Value( PhaseTransform *phase ) const {
+  const Type* init_t   = phase->type(in(Init));
+  const Type* limit_t  = phase->type(in(Limit));
+  const Type* stride_t = phase->type(in(Stride));
+  // Either input is TOP ==> the result is TOP
+  if (init_t   == Type::TOP) return Type::TOP;
+  if (limit_t  == Type::TOP) return Type::TOP;
+  if (stride_t == Type::TOP) return Type::TOP;
 
+  int stride_con = stride_t->is_int()->get_con();
+  if (stride_con == 1)
+    return NULL;  // Identity
+
+  if (init_t->is_int()->is_con() && limit_t->is_int()->is_con()) {
+    // Use longs to avoid integer overflow.
+    long init_con   =  init_t->is_int()->get_con();
+    long limit_con  = limit_t->is_int()->get_con();
+    int  stride_m   = stride_con - (stride_con > 0 ? 1 : -1);
+    long trip_count = (limit_con - init_con + stride_m)/stride_con;
+    long final_con  = init_con + stride_con*trip_count;
+    int final_int = (int)final_con;
+    // The final value should be in integer range since the loop
+    // is counted and the limit was checked for overflow.
+    assert(final_con == (long)final_int, "final value should be integer");
+    return TypeInt::make(final_int);
+  }
+
+  return bottom_type(); // TypeInt::INT
+}
+
+//------------------------------Ideal------------------------------------------
+// Return a node which is more "ideal" than the current node.
+Node *LoopLimitNode::Ideal(PhaseGVN *phase, bool can_reshape) {
+  if (phase->type(in(Init))   == Type::TOP ||
+      phase->type(in(Limit))  == Type::TOP ||
+      phase->type(in(Stride)) == Type::TOP)
+    return NULL;  // Dead
+
+  int stride_con = phase->type(in(Stride))->is_int()->get_con();
+  if (stride_con == 1)
+    return NULL;  // Identity
+
+  if (in(Init)->is_Con() && in(Limit)->is_Con())
+    return NULL;  // Value
+
+  // Delay following optimizations until all loop optimizations
+  // done to keep Ideal graph simple.
+  if (!can_reshape || phase->C->major_progress())
+    return NULL;
+
+  const TypeInt* init_t  = phase->type(in(Init) )->is_int();
+  const TypeInt* limit_t = phase->type(in(Limit))->is_int();
+  int stride_p;
+  long lim, ini;
+  julong max;
+  if (stride_con > 0) {
+    stride_p = stride_con;
+    lim = limit_t->_hi;
+    ini = init_t->_lo;
+    max = (julong)max_jint;
+  } else {
+    stride_p = -stride_con;
+    lim = init_t->_hi;
+    ini = limit_t->_lo;
+    max = (julong)min_jint;
+  }
+  julong range = lim - ini + stride_p;
+  if (range <= max) {
+    // Convert to integer expression if it is not overflow.
+    Node* stride_m = phase->intcon(stride_con - (stride_con > 0 ? 1 : -1));
+    Node *range = phase->transform(new (phase->C, 3) SubINode(in(Limit), in(Init)));
+    Node *bias  = phase->transform(new (phase->C, 3) AddINode(range, stride_m));
+    Node *trip  = phase->transform(new (phase->C, 3) DivINode(0, bias, in(Stride)));
+    Node *span  = phase->transform(new (phase->C, 3) MulINode(trip, in(Stride)));
+    return new (phase->C, 3) AddINode(span, in(Init)); // exact limit
+  }
+
+  if (is_power_of_2(stride_p) ||                // divisor is 2^n
+      !Matcher::has_match_rule(Op_LoopLimit)) { // or no specialized Mach node?
+    // Convert to long expression to avoid integer overflow
+    // and let igvn optimizer convert this division.
+    //
+    Node*   init   = phase->transform( new (phase->C, 2) ConvI2LNode(in(Init)));
+    Node*  limit   = phase->transform( new (phase->C, 2) ConvI2LNode(in(Limit)));
+    Node* stride   = phase->longcon(stride_con);
+    Node* stride_m = phase->longcon(stride_con - (stride_con > 0 ? 1 : -1));
+
+    Node *range = phase->transform(new (phase->C, 3) SubLNode(limit, init));
+    Node *bias  = phase->transform(new (phase->C, 3) AddLNode(range, stride_m));
+    Node *span;
+    if (stride_con > 0 && is_power_of_2(stride_p)) {
+      // bias >= 0 if stride >0, so if stride is 2^n we can use &(-stride)
+      // and avoid generating rounding for division. Zero trip guard should
+      // guarantee that init < limit but sometimes the guard is missing and
+      // we can get situation when init > limit. Note, for the empty loop
+      // optimization zero trip guard is generated explicitly which leaves
+      // only RCE predicate where exact limit is used and the predicate
+      // will simply fail forcing recompilation.
+      Node* neg_stride   = phase->longcon(-stride_con);
+      span = phase->transform(new (phase->C, 3) AndLNode(bias, neg_stride));
+    } else {
+      Node *trip  = phase->transform(new (phase->C, 3) DivLNode(0, bias, stride));
+      span = phase->transform(new (phase->C, 3) MulLNode(trip, stride));
+    }
+    // Convert back to int
+    Node *span_int = phase->transform(new (phase->C, 2) ConvL2INode(span));
+    return new (phase->C, 3) AddINode(span_int, in(Init)); // exact limit
+  }
+
+  return NULL;    // No progress
+}
+
+//------------------------------Identity---------------------------------------
+// If stride == 1 return limit node.
+Node *LoopLimitNode::Identity( PhaseTransform *phase ) {
+  int stride_con = phase->type(in(Stride))->is_int()->get_con();
+  if (stride_con == 1 || stride_con == -1)
+    return in(Limit);
+  return this;
+}
+
+//=============================================================================
 //----------------------match_incr_with_optional_truncation--------------------
 // Match increment with optional truncation:
 // CHAR: (i+1)&0x7fff, BYTE: ((i+1)<<8)>>8, or SHORT: ((i+1)<<16)>>16
@@ -870,7 +1162,7 @@ void IdealLoopTree::split_outer_loop( PhaseIdealLoop *phase ) {
   outer = igvn.register_new_node_with_optimizer(outer, _head);
   phase->set_created_loop_node();
 
-  Node* pred = phase->clone_loop_predicates(ctl, outer);
+  Node* pred = phase->clone_loop_predicates(ctl, outer, true);
   // Outermost loop falls into '_head' loop
   _head->set_req(LoopNode::EntryControl, pred);
   _head->del_req(outer_idx);
@@ -1440,9 +1732,16 @@ void IdealLoopTree::dump_head( ) const {
     tty->print("  ");
   tty->print("Loop: N%d/N%d ",_head->_idx,_tail->_idx);
   if (_irreducible) tty->print(" IRREDUCIBLE");
+  Node* entry = _head->in(LoopNode::EntryControl);
+  if (LoopLimitCheck) {
+    Node* predicate = PhaseIdealLoop::find_predicate_insertion_point(entry, Deoptimization::Reason_loop_limit_check);
+    if (predicate != NULL ) {
+      tty->print(" limit_check");
+      entry = entry->in(0)->in(0);
+    }
+  }
   if (UseLoopPredicate) {
-    Node* entry = PhaseIdealLoop::find_predicate_insertion_point(_head->in(LoopNode::EntryControl),
+    entry = PhaseIdealLoop::find_predicate_insertion_point(entry, Deoptimization::Reason_predicate);
-                                                                 Deoptimization::Reason_predicate);
     if (entry != NULL) {
       tty->print(" predicated");
     }
@@ -1528,10 +1827,15 @@ void PhaseIdealLoop::collect_potentially_useful_predicates(
       !loop->tail()->is_top()) {
     LoopNode* lpn = loop->_head->as_Loop();
     Node* entry = lpn->in(LoopNode::EntryControl);
-    Node* predicate_proj = find_predicate(entry);
+    Node* predicate_proj = find_predicate(entry); // loop_limit_check first
     if (predicate_proj != NULL ) { // right pattern that can be used by loop predication
       assert(entry->in(0)->in(1)->in(1)->Opcode() == Op_Opaque1, "must be");
       useful_predicates.push(entry->in(0)->in(1)->in(1)); // good one
+      entry = entry->in(0)->in(0);
+    }
+    predicate_proj = find_predicate(entry); // Predicate
+    if (predicate_proj != NULL ) {
+      useful_predicates.push(entry->in(0)->in(1)->in(1)); // good one
     }
   }
 
@@ -1542,6 +1846,8 @@ void PhaseIdealLoop::collect_potentially_useful_predicates(
 
 //------------------------eliminate_useless_predicates-----------------------------
 // Eliminate all inserted predicates if they could not be used by loop predication.
+// Note: it will also eliminates loop limits check predicate since it also uses
+// Opaque1 node (see Parse::add_predicate()).
 void PhaseIdealLoop::eliminate_useless_predicates() {
   if (C->predicate_count() == 0)
     return; // no predicate left
@@ -1731,7 +2037,7 @@ void PhaseIdealLoop::build_and_optimize(bool do_split_ifs) {
   // Some parser-inserted loop predicates could never be used by loop
   // predication or they were moved away from loop during some optimizations.
   // For example, peeling. Eliminate them before next loop optimizations.
-  if (UseLoopPredicate) {
+  if (UseLoopPredicate || LoopLimitCheck) {
     eliminate_useless_predicates();
   }
 

hotspot/src/share/vm/opto/loopnode.hpp

@@ -289,6 +289,28 @@ inline Node *CountedLoopNode::limit() const { return loopexit() ? loopexit()->li
 inline Node *CountedLoopNode::incr() const { return loopexit() ? loopexit()->incr() : NULL; }
 inline Node *CountedLoopNode::phi() const { return loopexit() ? loopexit()->phi() : NULL; }
 
+//------------------------------LoopLimitNode-----------------------------
+// Counted Loop limit node which represents exact final iterator value:
+// trip_count = (limit - init_trip + stride - 1)/stride
+// final_value= trip_count * stride + init_trip.
+// Use HW instructions to calculate it when it can overflow in integer.
+// Note, final_value should fit into integer since counted loop has
+// limit check: limit <= max_int-stride.
+class LoopLimitNode : public Node {
+  enum { Init=1, Limit=2, Stride=3 };
+ public:
+  LoopLimitNode( Compile* C, Node *init, Node *limit, Node *stride ) : Node(0,init,limit,stride) {
+    // Put it on the Macro nodes list to optimize during macro nodes expansion.
+    init_flags(Flag_is_macro);
+    C->add_macro_node(this);
+  }
+  virtual int Opcode() const;
+  virtual const Type *bottom_type() const { return TypeInt::INT; }
+  virtual uint ideal_reg() const { return Op_RegI; }
+  virtual const Type *Value( PhaseTransform *phase ) const;
+  virtual Node *Ideal(PhaseGVN *phase, bool can_reshape);
+  virtual Node *Identity( PhaseTransform *phase );
+};
 
 // -----------------------------IdealLoopTree----------------------------------
 class IdealLoopTree : public ResourceObj {
@@ -775,6 +797,8 @@ public:
 
   bool is_counted_loop( Node *x, IdealLoopTree *loop );
 
+  Node* exact_limit( IdealLoopTree *loop );
+
   // Return a post-walked LoopNode
   IdealLoopTree *get_loop( Node *n ) const {
     // Dead nodes have no loop, so return the top level loop instead
@@ -836,7 +860,6 @@ public:
   // Return true if exp is a scaled induction var plus (or minus) constant
   bool is_scaled_iv_plus_offset(Node* exp, Node* iv, int* p_scale, Node** p_offset, int depth = 0);
 
-  // Return true if proj is for "proj->[region->..]call_uct"
   // Return true if proj is for "proj->[region->..]call_uct"
   static bool is_uncommon_trap_proj(ProjNode* proj, Deoptimization::DeoptReason reason);
   // Return true for    "if(test)-> proj -> ...
@@ -860,10 +883,11 @@ public:
                                    PhaseIterGVN* igvn);
   static Node* clone_loop_predicates(Node* old_entry, Node* new_entry,
                                          bool move_predicates,
+                                         bool clone_limit_check,
                                          PhaseIdealLoop* loop_phase,
                                          PhaseIterGVN* igvn);
-  Node* clone_loop_predicates(Node* old_entry, Node* new_entry);
+  Node* clone_loop_predicates(Node* old_entry, Node* new_entry, bool clone_limit_check);
-  Node*  move_loop_predicates(Node* old_entry, Node* new_entry);
+  Node*  move_loop_predicates(Node* old_entry, Node* new_entry, bool clone_limit_check);
 
   void eliminate_loop_predicates(Node* entry);
   static Node* skip_loop_predicates(Node* entry);
@@ -873,7 +897,7 @@ public:
   // Find a predicate
   static Node* find_predicate(Node* entry);
   // Construct a range check for a predicate if
-  BoolNode* rc_predicate(Node* ctrl,
+  BoolNode* rc_predicate(IdealLoopTree *loop, Node* ctrl,
                          int scale, Node* offset,
                          Node* init, Node* limit, Node* stride,
                          Node* range, bool upper);
@@ -903,11 +927,11 @@ public:
 
   // Range Check Elimination uses this function!
   // Constrain the main loop iterations so the affine function:
-  //    scale_con * I + offset  <  limit
+  //    low_limit <= scale_con * I + offset  <  upper_limit
   // always holds true.  That is, either increase the number of iterations in
   // the pre-loop or the post-loop until the condition holds true in the main
   // loop.  Scale_con, offset and limit are all loop invariant.
-  void add_constraint( int stride_con, int scale_con, Node *offset, Node *limit, Node *pre_ctrl, Node **pre_limit, Node **main_limit );
+  void add_constraint( int stride_con, int scale_con, Node *offset, Node *low_limit, Node *upper_limit, Node *pre_ctrl, Node **pre_limit, Node **main_limit );
 
   // Partially peel loop up through last_peel node.
   bool partial_peel( IdealLoopTree *loop, Node_List &old_new );

hotspot/src/share/vm/opto/macro.cpp

@@ -2154,6 +2154,11 @@ bool PhaseMacroExpand::expand_macro_nodes() {
       debug_only(int old_macro_count = C->macro_count(););
       if (n->is_AbstractLock()) {
         success = eliminate_locking_node(n->as_AbstractLock());
+      } else if (n->Opcode() == Op_LoopLimit) {
+        // Remove it from macro list and put on IGVN worklist to optimize.
+        C->remove_macro_node(n);
+        _igvn._worklist.push(n);
+        success = true;
       } else if (n->Opcode() == Op_Opaque1 || n->Opcode() == Op_Opaque2) {
         _igvn.replace_node(n, n->in(1));
         success = true;

hotspot/src/share/vm/opto/matcher.cpp

@@ -2086,6 +2086,13 @@ void Matcher::find_shared( Node *n ) {
         n->del_req(3);
         break;
       }
+      case Op_LoopLimit: {
+        Node *pair1 = new (C, 3) BinaryNode(n->in(1),n->in(2));
+        n->set_req(1,pair1);
+        n->set_req(2,n->in(3));
+        n->del_req(3);
+        break;
+      }
       case Op_StrEquals: {
         Node *pair1 = new (C, 3) BinaryNode(n->in(2),n->in(3));
         n->set_req(2,pair1);

hotspot/src/share/vm/opto/parse.hpp

@@ -70,7 +70,7 @@ protected:
   const char* try_to_inline(ciMethod* callee_method, ciMethod* caller_method, int caller_bci, ciCallProfile& profile, WarmCallInfo* wci_result);
   const char* shouldInline(ciMethod* callee_method, ciMethod* caller_method, int caller_bci, ciCallProfile& profile, WarmCallInfo* wci_result) const;
   const char* shouldNotInline(ciMethod* callee_method, ciMethod* caller_method, WarmCallInfo* wci_result) const;
-  void        print_inlining(ciMethod *callee_method, int caller_bci, const char *failure_msg) const PRODUCT_RETURN;
+  void        print_inlining(ciMethod *callee_method, int caller_bci, const char *failure_msg) const;
 
   InlineTree *caller_tree()       const { return _caller_tree;  }
   InlineTree* callee_at(int bci, ciMethod* m) const;

hotspot/src/share/vm/opto/parse1.cpp

@@ -638,7 +638,7 @@ void Parse::do_all_blocks() {
         ensure_phis_everywhere();
 
         if (block->is_SEL_head() &&
-            UseLoopPredicate) {
+            (UseLoopPredicate || LoopLimitCheck)) {
           // Add predicate to single entry (not irreducible) loop head.
           assert(!block->has_merged_backedge(), "only entry paths should be merged for now");
           // Need correct bci for predicate.

hotspot/src/share/vm/opto/phaseX.hpp

@@ -472,8 +472,8 @@ public:
   }
 
   // Clone loop predicates. Defined in loopTransform.cpp.
-  Node* clone_loop_predicates(Node* old_entry, Node* new_entry);
+  Node* clone_loop_predicates(Node* old_entry, Node* new_entry, bool clone_limit_check);
-  Node*  move_loop_predicates(Node* old_entry, Node* new_entry);
+  Node*  move_loop_predicates(Node* old_entry, Node* new_entry, bool clone_limit_check);
   // Create a new if below new_entry for the predicate to be cloned
   ProjNode* create_new_if_for_predicate(ProjNode* cont_proj, Node* new_entry,
                                         Deoptimization::DeoptReason reason);

hotspot/src/share/vm/opto/subnode.cpp

@@ -1222,21 +1222,6 @@ bool BoolNode::is_counted_loop_exit_test() {
   return false;
 }
 
-//=============================================================================
-//------------------------------NegNode----------------------------------------
-Node *NegFNode::Ideal(PhaseGVN *phase, bool can_reshape) {
-  if( in(1)->Opcode() == Op_SubF )
-    return new (phase->C, 3) SubFNode( in(1)->in(2), in(1)->in(1) );
-  return NULL;
-}
-
-Node *NegDNode::Ideal(PhaseGVN *phase, bool can_reshape) {
-  if( in(1)->Opcode() == Op_SubD )
-    return new (phase->C, 3) SubDNode( in(1)->in(2), in(1)->in(1) );
-  return NULL;
-}
-
-
 //=============================================================================
 //------------------------------Value------------------------------------------
 // Compute sqrt

hotspot/src/share/vm/opto/subnode.hpp

@@ -377,7 +377,6 @@ class NegFNode : public NegNode {
 public:
   NegFNode( Node *in1 ) : NegNode(in1) {}
   virtual int Opcode() const;
-  virtual Node *Ideal(PhaseGVN *phase, bool can_reshape);
   const Type *bottom_type() const { return Type::FLOAT; }
   virtual uint ideal_reg() const { return Op_RegF; }
 };
@@ -391,7 +390,6 @@ class NegDNode : public NegNode {
 public:
   NegDNode( Node *in1 ) : NegNode(in1) {}
   virtual int Opcode() const;
-  virtual Node *Ideal(PhaseGVN *phase, bool can_reshape);
   const Type *bottom_type() const { return Type::DOUBLE; }
   virtual uint ideal_reg() const { return Op_RegD; }
 };

hotspot/src/share/vm/prims/jvmti.xml

@@ -280,10 +280,8 @@
    <!ELEMENT externallink (#PCDATA|jvmti|code|i|b|tm)*>
    <!ATTLIST externallink id CDATA #REQUIRED>
 
-   <!ELEMENT vmspeclink EMPTY>
+   <!ELEMENT vmspec EMPTY>
-   <!ATTLIST vmspeclink id CDATA #IMPLIED>
+   <!ATTLIST vmspec chapter CDATA #IMPLIED>
-   <!ATTLIST vmspeclink name CDATA #IMPLIED>
-   <!ATTLIST vmspeclink preposition CDATA #IMPLIED>
 
    <!ELEMENT internallink (#PCDATA|jvmti|code|i|b)*>
    <!ATTLIST internallink id CDATA #REQUIRED>
@@ -2285,9 +2283,8 @@ jvmtiEnv *jvmti;
         Stack frames are referenced by depth.
         The frame at depth zero is the current frame.
         <p/>
-        Stack frames are as described in the 
+        Stack frames are as described in
-        <vmspeclink id="Overview.doc.html#17257"
+        <vmspec chapter="3.6"/>,
-                    name="Frames section"/>.  
         That is, they correspond to method 
         invocations (including native methods) but do not correspond to platform native or 
         VM internal frames.
@@ -2627,7 +2624,7 @@ err = (*jvmti)-&gt;Deallocate(jvmti, stack_info);
         <param id="use_java_stack">
 	  <jboolean/>
 	  <description>
-	    Return the stack showing the <vmspeclink/>
+	    Return the stack showing <vmspec/>
 	    model of the stack; 
 	    otherwise, show the internal representation of the stack with
 	    inlined and optimized methods missing.  If the virtual machine
@@ -2707,7 +2704,7 @@ err = (*jvmti)-&gt;Deallocate(jvmti, stack_info);
 	When the thread is resumed, the execution 
 	state of the thread is reset to the state
 	immediately before the called method was invoked.
-	That is (using the <vmspeclink/> terminology):
+	That is (using <vmspec/> terminology):
 	  <ul>
 	    <li>the current frame is discarded as the previous frame becomes the current one</li>
 	    <li>the operand stack is restored--the argument values are added back
@@ -2868,9 +2865,8 @@ err = (*jvmti)-&gt;Deallocate(jvmti, stack_info);
       to return at any point during its execution.
       The method which will return early is referred to as the <i>called method</i>.
       The called method is the current method
-      (as defined by the 
+      (as defined by
-      <vmspeclink id="Overview.doc.html#17257"
+      <vmspec chapter="3.6"/>) 
-                  name="Frames section"/>) 
       for the specified thread at
       the time the function is called.
       <p/>
@@ -3576,10 +3572,8 @@ class C2 extends C1 implements I2 {
 	<field id="index">
 	  <jint/>
 	  <description>	    
-	    The index into the constant pool of the class. See the
+	    The index into the constant pool of the class. See the description in 
-            <vmspeclink id="ClassFile.doc.html#20080"
+      <vmspec chapter="4.4"/>.
-                        name="Constant Pool section"/>
-	    description.
 	  </description>
 	</field>
       </typedef>
@@ -5006,9 +5000,8 @@ class C2 extends C1 implements I2 {
 	    For references of this kind the <code>referrer_index</code>
             parameter to the <internallink id="jvmtiObjectReferenceCallback">
             jvmtiObjectReferenceCallback</internallink> is the index into
-            constant pool table of the class, starting at 1. See the
+            constant pool table of the class, starting at 1. See
-            <vmspeclink id="ClassFile.doc.html#20080"
+            <vmspec chapter="4.4"/>.
-                        name="Constant Pool section"/>
 	  </constant>
 	</constants>
 
@@ -6441,9 +6434,7 @@ class C2 extends C1 implements I2 {
 	been recorded as an initiating loader. Each 
 	class in the returned array was created by this class loader, 
 	either by defining it directly or by delegation to another class loader.
-        See the 
+	See <vmspec chapter="5.3"/>.
-        <vmspeclink id="ConstantPool.doc.html#72007"
-                    name="Creation and Loading section"/>.
 	<p/>
 	For JDK version 1.1 implementations that don't
 	recognize the distinction between initiating and defining class loaders,
@@ -6626,9 +6617,7 @@ class C2 extends C1 implements I2 {
 	For the class indicated by <code>klass</code>, return the access
 	flags
 	via <code>modifiers_ptr</code>.
-	Access flags are defined in the 
+	Access flags are defined in <vmspec chapter="4"/>.
-        <vmspeclink id="ClassFile.doc.html"
-                    name="Class File Format chapter"/>.
 	<p/>
 	If the class is an array class, then its public, private, and protected 
 	modifiers are the same as those of its component type. For arrays of 
@@ -6794,9 +6783,8 @@ class C2 extends C1 implements I2 {
       <description>
         For the class indicated by <code>klass</code>, 
         return the minor and major version numbers,
-        as defined in the
+        as defined in
-        <vmspeclink id="ClassFile.doc.html"
+        <vmspec chapter="4"/>. 
-                        name="Class File Format chapter"/>.
       </description>
       <origin>new</origin>
       <capabilities>
@@ -6839,10 +6827,8 @@ class C2 extends C1 implements I2 {
       <description>
 	For the class indicated by <code>klass</code>, 
         return the raw bytes of the constant pool in the format of the
-        <code>constant_pool</code> item of the 
+        <code>constant_pool</code> item of 
-        <vmspeclink id="ClassFile.doc.html"
+        <vmspec chapter="4"/>.
-                    name="Class File Format"
-                    preposition="in"/>.
         The format of the constant pool may differ between versions
         of the Class File Format, so, the 
         <functionlink id="GetClassVersionNumbers">minor and major 
@@ -7286,9 +7272,7 @@ class C2 extends C1 implements I2 {
 	<field id="class_bytes">
 	  <inbuf incount="class_byte_count"><uchar/></inbuf>
 	  <description>
-            Bytes defining class (in the 
+            Bytes defining class (in <vmspec chapter="4"/>)
-            <vmspeclink id="ClassFile.doc.html"
-                        name="Class File Format"/>)
 	  </description>
 	</field>
       </typedef>
@@ -7611,10 +7595,8 @@ class C2 extends C1 implements I2 {
 	<paramlink id="signature_ptr"/>.
 	<p/>
         Field signatures are defined in the JNI Specification and 
-        are referred to as 
+        are referred to as <code>field descriptors</code> in
-        <vmspeclink id="ClassFile.doc.html#14152"
+        <vmspec chapter="4.3.2"/>.
-                    name="field descriptors"
-                    preposition="in"/>.
       </description>
       <origin>jvmdiClone</origin>
       <capabilities>
@@ -7709,9 +7691,7 @@ class C2 extends C1 implements I2 {
       <description>
 	For the field indicated by <code>klass</code> and <code>field</code>
 	return the access flags via <code>modifiers_ptr</code>.
-	Access flags are defined in the 
+	Access flags are defined in <vmspec chapter="4"/>.
-        <vmspeclink id="ClassFile.doc.html"
-                    name="Class File Format chapter"/>.
       </description>
       <origin>jvmdi</origin>
       <capabilities>
@@ -7810,10 +7790,9 @@ class C2 extends C1 implements I2 {
 	return the method name via <code>name_ptr</code> and method signature via
 	<code>signature_ptr</code>.
         <p/>
-        Method signatures are defined in the JNI Specification and are referred to as
+        Method signatures are defined in the JNI Specification and are 
-        <vmspeclink id="ClassFile.doc.html#7035"
+        referred to as <code>method descriptors</code> in 
-                    name="method descriptors"
+        <vmspec chapter="4.3.3"/>.
-                    preposition="in"/>.
 	Note this is different
 	than method signatures as defined in the <i>Java Language Specification</i>.
       </description>
@@ -7902,9 +7881,7 @@ class C2 extends C1 implements I2 {
       <description>
 	For the method indicated by <code>method</code>,
 	return the access flags via <code>modifiers_ptr</code>.
-	Access flags are defined in the 
+	Access flags are defined in <vmspec chapter="4"/>.
-        <vmspeclink id="ClassFile.doc.html"
-                    name="Class File Format chapter"/>.
       </description>
       <origin>jvmdi</origin>
       <capabilities>
@@ -7941,9 +7918,7 @@ class C2 extends C1 implements I2 {
 	  including the local variables used to pass parameters to the
 	  method on its invocation. 
 	  <p/>
-	  See <code>max_locals</code> in the    
+	  See <code>max_locals</code> in <vmspec chapter="4.7.3"/>.
-          <vmspeclink id="ClassFile.doc.html#1546"
-                      name="Code Attribute section"/>.
       </description>
       <origin>jvmdi</origin>
       <capabilities>
@@ -8150,8 +8125,7 @@ class C2 extends C1 implements I2 {
 	    The local variable's type signature, encoded as a
 	    <internallink id="mUTF">modified UTF-8</internallink> string.
 	    The signature format is the same as that defined in
-            <vmspeclink id="ClassFile.doc.html#14152"
+	    <vmspec chapter="4.3.2"/>.
-                        name="Field Descriptors section"/>
 	  </description>
 	</field>
 	<field id="generic_signature">
@@ -10460,10 +10434,7 @@ myInit() {
       <synopsis>Add To Bootstrap Class Loader Search</synopsis>
       <description>
           This function can be used to cause instrumentation classes to be defined by the 
-          bootstrap class loader. See
+          bootstrap class loader. See <vmspec chapter="5.3.1"/>.
-          <vmspeclink id="ConstantPool.doc.html#79383"
-                      name="Loading Using the Bootstrap Class Loader"
-                      preposition="in"/>.
           After the bootstrap
 	  class loader unsuccessfully searches for a class, the specified platform-dependent 
 	  search path <paramlink id="segment"/> will be searched as well. Only one segment may be specified in 
@@ -10480,7 +10451,7 @@ myInit() {
           contain any classes or resources other than those to be defined by the bootstrap
           class loader for the purposes of instrumentation.
           <p/>
-          The <vmspeclink/> specifies that a subsequent attempt to resolve a symbolic
+          <vmspec/> specifies that a subsequent attempt to resolve a symbolic
           reference that the Java virtual machine has previously unsuccessfully attempted
           to resolve always fails with the same error that was thrown as a result of the
           initial resolution attempt. Consequently, if the JAR file contains an entry
@@ -10512,10 +10483,7 @@ myInit() {
       <synopsis>Add To System Class Loader Search</synopsis>
       <description>
 	  This function can be used to cause instrumentation classes to be
-	  defined by the system class loader. See
+	  defined by the system class loader. See <vmspec chapter="5.3.2"/>.
-          <vmspeclink id="ConstantPool.doc.html#79441"
-                      name="Loading Using a User-defined Class Loader"
-                      preposition="in"/>. 
 	  After the class loader unsuccessfully searches for a class, the specified platform-dependent search 
 	  path <paramlink id="segment"/> will be searched as well. Only one segment may be specified in the 
 	  <paramlink id="segment"/>. This function may be called multiple times to add multiple segments, the 
@@ -10536,7 +10504,7 @@ myInit() {
 	  which takes a single parameter of type <code>java.lang.String</code>. The method is not required 
 	  to have <code>public</code> access. 
 	  <p/>
-          The <vmspeclink/> specifies that a subsequent attempt to resolve a symbolic
+          <vmspec/> specifies that a subsequent attempt to resolve a symbolic
           reference that the Java virtual machine has previously unsuccessfully attempted
           to resolve always fails with the same error that was thrown as a result of the
           initial resolution attempt. Consequently, if the JAR file contains an entry
@@ -11438,7 +11406,7 @@ myInit() {
       at the finest granularity allowed by the VM. A single step event is
       generated whenever a thread reaches a new location. 
       Typically, single step events represent the completion of one VM 
-      instruction as defined in the <vmspeclink/>. However, some implementations 
+      instruction as defined in <vmspec/>. However, some implementations 
       may define locations differently. In any case the 
       <code>method</code> and <code>location</code>
       parameters  uniquely identify the current location and allow
@@ -13841,7 +13809,7 @@ typedef void (JNICALL *jvmtiEventVMInit)
       and can_get_source_debug_extension.
       PopFrame cannot have a native calling method.
       Removed incorrect statement in GetClassloaderClasses 
-      (see http://java.sun.com/docs/books/vmspec/2nd-edition/html/ConstantPool.doc.html#79383).
+      (see <vmspec chapter="4.4"/>).
   </change>
   <change date="24 July 2003" version="v79">
       XML and text fixes.

hotspot/src/share/vm/prims/jvmti.xsl

@@ -1039,34 +1039,14 @@ typedef struct {
   </a>
 </xsl:template>
 
-<xsl:template match="vmspeclink">
+<xsl:template match="vmspec">
-  <xsl:if test="count(@id)=1">
+  <cite>
-    <a>
+    <xsl:text>The Java&#8482; Virtual Machine Specification</xsl:text>
-      <xsl:attribute name="href">
+    <xsl:if test="count(@chapter)=1">
-        <xsl:text>http://java.sun.com/docs/books/vmspec/2nd-edition/html/</xsl:text>
+      <xsl:text>, Chapter </xsl:text> 
-        <xsl:value-of select="@id"/>
+      <xsl:value-of select="@chapter"/>
-      </xsl:attribute>
-      <xsl:value-of select="@name"/>
-    </a>
-    <xsl:text> </xsl:text>
-    <xsl:choose>
-      <xsl:when test="count(@preposition)=1">
-        <xsl:value-of select="@preposition"/>
-      </xsl:when>
-      <xsl:otherwise>
-        <xsl:text>of</xsl:text>
-      </xsl:otherwise>
-    </xsl:choose>
-    <xsl:text> the </xsl:text>
     </xsl:if>
-  <a>
+  </cite>
-    <xsl:attribute name="href">
-      <xsl:text>http://java.sun.com/docs/books/vmspec/</xsl:text>
-    </xsl:attribute>
-    <i>
-      <xsl:text>Java Virtual Machine Specification</xsl:text>
-    </i>
-  </a>
 </xsl:template>
 
 <xsl:template match="internallink">

hotspot/src/share/vm/prims/jvmtiExport.cpp

@@ -1804,6 +1804,8 @@ void JvmtiExport::post_compiled_method_load(JvmtiEnv* env, const jmethodID metho
 }
 
 void JvmtiExport::post_dynamic_code_generated_internal(const char *name, const void *code_begin, const void *code_end) {
+  assert(name != NULL && name[0] != '\0', "sanity check");
+
   JavaThread* thread = JavaThread::current();
   // In theory everyone coming thru here is in_vm but we need to be certain
   // because a callee will do a vm->native transition

hotspot/src/share/vm/prims/jvmtiImpl.cpp

@@ -38,6 +38,7 @@
 #include "runtime/handles.inline.hpp"
 #include "runtime/interfaceSupport.hpp"
 #include "runtime/javaCalls.hpp"
+#include "runtime/os.hpp"
 #include "runtime/serviceThread.hpp"
 #include "runtime/signature.hpp"
 #include "runtime/vframe.hpp"
@@ -939,10 +940,15 @@ JvmtiDeferredEvent JvmtiDeferredEvent::compiled_method_unload_event(
   nmethodLocker::lock_nmethod(nm, true /* zombie_ok */);
   return event;
 }
+
 JvmtiDeferredEvent JvmtiDeferredEvent::dynamic_code_generated_event(
       const char* name, const void* code_begin, const void* code_end) {
   JvmtiDeferredEvent event = JvmtiDeferredEvent(TYPE_DYNAMIC_CODE_GENERATED);
-  event._event_data.dynamic_code_generated.name = name;
+  // Need to make a copy of the name since we don't know how long
+  // the event poster will keep it around after we enqueue the
+  // deferred event and return. strdup() failure is handled in
+  // the post() routine below.
+  event._event_data.dynamic_code_generated.name = os::strdup(name);
   event._event_data.dynamic_code_generated.code_begin = code_begin;
   event._event_data.dynamic_code_generated.code_end = code_end;
   return event;
@@ -968,12 +974,19 @@ void JvmtiDeferredEvent::post() {
       nmethodLocker::unlock_nmethod(nm);
       break;
     }
-    case TYPE_DYNAMIC_CODE_GENERATED:
+    case TYPE_DYNAMIC_CODE_GENERATED: {
       JvmtiExport::post_dynamic_code_generated_internal(
-        _event_data.dynamic_code_generated.name,
+        // if strdup failed give the event a default name
+        (_event_data.dynamic_code_generated.name == NULL)
+          ? "unknown_code" : _event_data.dynamic_code_generated.name,
         _event_data.dynamic_code_generated.code_begin,
         _event_data.dynamic_code_generated.code_end);
+      if (_event_data.dynamic_code_generated.name != NULL) {
+        // release our copy
+        os::free((void *)_event_data.dynamic_code_generated.name);
+      }
       break;
+    }
     default:
       ShouldNotReachHere();
   }

hotspot/src/share/vm/runtime/arguments.cpp

@@ -960,7 +960,7 @@ void Arguments::set_mode_flags(Mode mode) {
   // Ensure Agent_OnLoad has the correct initial values.
   // This may not be the final mode; mode may change later in onload phase.
   PropertyList_unique_add(&_system_properties, "java.vm.info",
-                          (char*)Abstract_VM_Version::vm_info_string(), false);
+                          (char*)VM_Version::vm_info_string(), false);
 
   UseInterpreter             = true;
   UseCompiler                = true;
@@ -969,10 +969,10 @@ void Arguments::set_mode_flags(Mode mode) {
 #ifndef ZERO
   // Turn these off for mixed and comp.  Leave them on for Zero.
   if (FLAG_IS_DEFAULT(UseFastAccessorMethods)) {
-    UseFastAccessorMethods = mode == _int;
+    UseFastAccessorMethods = (mode == _int);
   }
   if (FLAG_IS_DEFAULT(UseFastEmptyMethods)) {
-    UseFastEmptyMethods = mode == _int;
+    UseFastEmptyMethods = (mode == _int);
   }
 #endif
 
@@ -1423,6 +1423,11 @@ void Arguments::set_parallel_gc_flags() {
       }
     }
   }
+  if (UseNUMA) {
+    if (FLAG_IS_DEFAULT(MinHeapDeltaBytes)) {
+      FLAG_SET_DEFAULT(MinHeapDeltaBytes, 64*M);
+    }
+  }
 }
 
 void Arguments::set_g1_gc_flags() {
@@ -1987,6 +1992,9 @@ jint Arguments::parse_vm_init_args(const JavaVMInitArgs* args) {
   Arguments::_ClipInlining             = ClipInlining;
   Arguments::_BackgroundCompilation    = BackgroundCompilation;
 
+  // Setup flags for mixed which is the default
+  set_mode_flags(_mixed);
+
   // Parse JAVA_TOOL_OPTIONS environment variable (if present)
   jint result = parse_java_tool_options_environment_variable(&scp, &scp_assembly_required);
   if (result != JNI_OK) {
@@ -2376,7 +2384,6 @@ jint Arguments::parse_each_vm_init_arg(const JavaVMInitArgs* args,
       _gc_log_filename = strdup(tail);
       FLAG_SET_CMDLINE(bool, PrintGC, true);
       FLAG_SET_CMDLINE(bool, PrintGCTimeStamps, true);
-      FLAG_SET_CMDLINE(bool, TraceClassUnloading, true);
 
     // JNI hooks
     } else if (match_option(option, "-Xcheck", &tail)) {

hotspot/src/share/vm/runtime/deoptimization.cpp

@@ -189,6 +189,10 @@ Deoptimization::UnrollBlock* Deoptimization::fetch_unroll_info_helper(JavaThread
   assert(thread->deopt_nmethod() == NULL, "Pending deopt!");
   thread->set_deopt_nmethod(deoptee.cb()->as_nmethod_or_null());
 
+  if (VerifyStack) {
+    thread->validate_frame_layout();
+  }
+
   // Create a growable array of VFrames where each VFrame represents an inlined
   // Java frame.  This storage is allocated with the usual system arena.
   assert(deoptee.is_compiled_frame(), "Wrong frame type");
@@ -421,6 +425,21 @@ Deoptimization::UnrollBlock* Deoptimization::fetch_unroll_info_helper(JavaThread
   frame deopt_sender = stub_frame.sender(&dummy_map); // First is the deoptee frame
   deopt_sender = deopt_sender.sender(&dummy_map);     // Now deoptee caller
 
+  // It's possible that the number of paramters at the call site is
+  // different than number of arguments in the callee when method
+  // handles are used.  If the caller is interpreted get the real
+  // value so that the proper amount of space can be added to it's
+  // frame.
+  int sender_callee_parameters = callee_parameters;
+  if (deopt_sender.is_interpreted_frame()) {
+    methodHandle method = deopt_sender.interpreter_frame_method();
+    Bytecode_invoke cur = Bytecode_invoke_check(method,
+                                                deopt_sender.interpreter_frame_bci());
+    Symbol* signature = method->constants()->signature_ref_at(cur.index());
+    ArgumentSizeComputer asc(signature);
+    sender_callee_parameters = asc.size() + (cur.has_receiver() ? 1 : 0);
+  }
+
   // Compute the amount the oldest interpreter frame will have to adjust
   // its caller's stack by. If the caller is a compiled frame then
   // we pretend that the callee has no parameters so that the
@@ -435,14 +454,13 @@ Deoptimization::UnrollBlock* Deoptimization::fetch_unroll_info_helper(JavaThread
 
   if (deopt_sender.is_compiled_frame()) {
     caller_adjustment = last_frame_adjust(0, callee_locals);
-  } else if (callee_locals > callee_parameters) {
+  } else if (callee_locals > sender_callee_parameters) {
     // The caller frame may need extending to accommodate
     // non-parameter locals of the first unpacked interpreted frame.
     // Compute that adjustment.
-    caller_adjustment = last_frame_adjust(callee_parameters, callee_locals);
+    caller_adjustment = last_frame_adjust(sender_callee_parameters, callee_locals);
   }
 
-
   // If the sender is deoptimized the we must retrieve the address of the handler
   // since the frame will "magically" show the original pc before the deopt
   // and we'd undo the deopt.
@@ -569,6 +587,8 @@ JRT_LEAF(BasicType, Deoptimization::unpack_frames(JavaThread* thread, int exec_m
   if (VerifyStack) {
     ResourceMark res_mark;
 
+    thread->validate_frame_layout();
+
     // Verify that the just-unpacked frames match the interpreter's
     // notions of expression stack and locals
     vframeArray* cur_array = thread->vframe_array_last();
@@ -1753,7 +1773,8 @@ const char* Deoptimization::_trap_reason_name[Reason_LIMIT] = {
   "constraint",
   "div0_check",
   "age",
-  "predicate"
+  "predicate",
+  "loop_limit_check"
 };
 const char* Deoptimization::_trap_action_name[Action_LIMIT] = {
   // Note:  Keep this in sync. with enum DeoptAction.

hotspot/src/share/vm/runtime/deoptimization.hpp

@@ -56,6 +56,7 @@ class Deoptimization : AllStatic {
     Reason_div0_check,            // a null_check due to division by zero
     Reason_age,                   // nmethod too old; tier threshold reached
     Reason_predicate,             // compiler generated predicate failed
+    Reason_loop_limit_check,      // compiler generated loop limits check failed
     Reason_LIMIT,
     // Note:  Keep this enum in sync. with _trap_reason_name.
     Reason_RECORDED_LIMIT = Reason_bimorphic  // some are not recorded per bc
@@ -78,7 +79,7 @@ class Deoptimization : AllStatic {
 
   enum {
     _action_bits = 3,
-    _reason_bits = 4,
+    _reason_bits = 5,
     _action_shift = 0,
     _reason_shift = _action_shift+_action_bits,
     BC_CASE_LIMIT = PRODUCT_ONLY(1) NOT_PRODUCT(4) // for _deoptimization_hist

hotspot/src/share/vm/runtime/frame.cpp

@@ -1308,6 +1308,72 @@ void frame::interpreter_frame_verify_monitor(BasicObjectLock* value) const {
   guarantee((current - low_mark) % monitor_size  ==  0         , "Misaligned bottom of BasicObjectLock*");
   guarantee( current >= low_mark                               , "Current BasicObjectLock* below than low_mark");
 }
+
+
+void frame::describe(FrameValues& values, int frame_no) {
+  if (is_entry_frame() || is_compiled_frame() || is_interpreted_frame() || is_native_frame()) {
+    // Label values common to most frames
+    values.describe(-1, unextended_sp(), err_msg("unextended_sp for #%d", frame_no));
+    values.describe(-1, sp(), err_msg("sp for #%d", frame_no));
+    values.describe(-1, fp(), err_msg("fp for #%d", frame_no));
+  }
+  if (is_interpreted_frame()) {
+    methodOop m = interpreter_frame_method();
+    int bci = interpreter_frame_bci();
+
+    // Label the method and current bci
+    values.describe(-1, MAX2(sp(), fp()),
+                    FormatBuffer<1024>("#%d method %s @ %d", frame_no, m->name_and_sig_as_C_string(), bci), 2);
+    values.describe(-1, MAX2(sp(), fp()),
+                    err_msg("- %d locals %d max stack", m->max_locals(), m->max_stack()), 1);
+    if (m->max_locals() > 0) {
+      intptr_t* l0 = interpreter_frame_local_at(0);
+      intptr_t* ln = interpreter_frame_local_at(m->max_locals() - 1);
+      values.describe(-1, MAX2(l0, ln), err_msg("locals for #%d", frame_no), 1);
+      // Report each local and mark as owned by this frame
+      for (int l = 0; l < m->max_locals(); l++) {
+        intptr_t* l0 = interpreter_frame_local_at(l);
+        values.describe(frame_no, l0, err_msg("local %d", l));
+      }
+    }
+
+    // Compute the actual expression stack size
+    InterpreterOopMap mask;
+    OopMapCache::compute_one_oop_map(m, bci, &mask);
+    intptr_t* tos = NULL;
+    // Report each stack element and mark as owned by this frame
+    for (int e = 0; e < mask.expression_stack_size(); e++) {
+      tos = MAX2(tos, interpreter_frame_expression_stack_at(e));
+      values.describe(frame_no, interpreter_frame_expression_stack_at(e),
+                      err_msg("stack %d", e));
+    }
+    if (tos != NULL) {
+      values.describe(-1, tos, err_msg("expression stack for #%d", frame_no), 1);
+    }
+    if (interpreter_frame_monitor_begin() != interpreter_frame_monitor_end()) {
+      values.describe(frame_no, (intptr_t*)interpreter_frame_monitor_begin(), "monitors begin");
+      values.describe(frame_no, (intptr_t*)interpreter_frame_monitor_end(), "monitors end");
+    }
+  } else if (is_entry_frame()) {
+    // For now just label the frame
+    values.describe(-1, MAX2(sp(), fp()), err_msg("#%d entry frame", frame_no), 2);
+  } else if (is_compiled_frame()) {
+    // For now just label the frame
+    nmethod* nm = cb()->as_nmethod_or_null();
+    values.describe(-1, MAX2(sp(), fp()),
+                    FormatBuffer<1024>("#%d nmethod " INTPTR_FORMAT " for method %s%s", frame_no,
+                                       nm, nm->method()->name_and_sig_as_C_string(),
+                                       is_deoptimized_frame() ? " (deoptimized" : ""), 2);
+  } else if (is_native_frame()) {
+    // For now just label the frame
+    nmethod* nm = cb()->as_nmethod_or_null();
+    values.describe(-1, MAX2(sp(), fp()),
+                    FormatBuffer<1024>("#%d nmethod " INTPTR_FORMAT " for native method %s", frame_no,
+                                       nm, nm->method()->name_and_sig_as_C_string()), 2);
+  }
+  describe_pd(values, frame_no);
+}
+
 #endif
 
 
@@ -1319,3 +1385,71 @@ StackFrameStream::StackFrameStream(JavaThread *thread, bool update) : _reg_map(t
   _fr = thread->last_frame();
   _is_done = false;
 }
+
+
+#ifdef ASSERT
+
+void FrameValues::describe(int owner, intptr_t* location, const char* description, int priority) {
+  FrameValue fv;
+  fv.location = location;
+  fv.owner = owner;
+  fv.priority = priority;
+  fv.description = NEW_RESOURCE_ARRAY(char, strlen(description) + 1);
+  strcpy(fv.description, description);
+  _values.append(fv);
+}
+
+
+bool FrameValues::validate() {
+  _values.sort(compare);
+  bool error = false;
+  FrameValue prev;
+  prev.owner = -1;
+  for (int i = _values.length() - 1; i >= 0; i--) {
+    FrameValue fv = _values.at(i);
+    if (fv.owner == -1) continue;
+    if (prev.owner == -1) {
+      prev = fv;
+      continue;
+    }
+    if (prev.location == fv.location) {
+      if (fv.owner != prev.owner) {
+        tty->print_cr("overlapping storage");
+        tty->print_cr(" " INTPTR_FORMAT ": " INTPTR_FORMAT " %s", prev.location, *prev.location, prev.description);
+        tty->print_cr(" " INTPTR_FORMAT ": " INTPTR_FORMAT " %s", fv.location, *fv.location, fv.description);
+        error = true;
+      }
+    } else {
+      prev = fv;
+    }
+  }
+  return error;
+}
+
+
+void FrameValues::print() {
+  _values.sort(compare);
+  intptr_t* v0 = _values.at(0).location;
+  intptr_t* v1 = _values.at(_values.length() - 1).location;
+  intptr_t* min = MIN2(v0, v1);
+  intptr_t* max = MAX2(v0, v1);
+  intptr_t* cur = max;
+  intptr_t* last = NULL;
+  for (int i = _values.length() - 1; i >= 0; i--) {
+    FrameValue fv = _values.at(i);
+    while (cur > fv.location) {
+      tty->print_cr(" " INTPTR_FORMAT ": " INTPTR_FORMAT, cur, *cur);
+      cur--;
+    }
+    if (last == fv.location) {
+      const char* spacer = "          " LP64_ONLY("        ");
+      tty->print_cr(" %s  %s %s", spacer, spacer, fv.description);
+    } else {
+      tty->print_cr(" " INTPTR_FORMAT ": " INTPTR_FORMAT " %s", fv.location, *fv.location, fv.description);
+      last = fv.location;
+      cur--;
+    }
+  }
+}
+
+#endif

hotspot/src/share/vm/runtime/frame.hpp

@@ -60,6 +60,7 @@
 typedef class BytecodeInterpreter* interpreterState;
 
 class CodeBlob;
+class FrameValues;
 class vframeArray;
 
 
@@ -381,6 +382,8 @@ class frame VALUE_OBJ_CLASS_SPEC {
  private:
   const char* print_name() const;
 
+  void describe_pd(FrameValues& values, int frame_no);
+
  public:
   void print_value() const { print_value_on(tty,NULL); }
   void print_value_on(outputStream* st, JavaThread *thread) const;
@@ -388,6 +391,9 @@ class frame VALUE_OBJ_CLASS_SPEC {
   void interpreter_frame_print_on(outputStream* st) const;
   void print_on_error(outputStream* st, char* buf, int buflen, bool verbose = false) const;
 
+  // Add annotated descriptions of memory locations belonging to this frame to values
+  void describe(FrameValues& values, int frame_no);
+
   // Conversion from an VMReg to physical stack location
   oop* oopmapreg_to_location(VMReg reg, const RegisterMap* regmap) const;
 
@@ -472,6 +478,41 @@ class frame VALUE_OBJ_CLASS_SPEC {
 
 };
 
+#ifdef ASSERT
+// A simple class to describe a location on the stack
+class FrameValue VALUE_OBJ_CLASS_SPEC {
+ public:
+  intptr_t* location;
+  char* description;
+  int owner;
+  int priority;
+};
+
+
+// A collection of described stack values that can print a symbolic
+// description of the stack memory.  Interpreter frame values can be
+// in the caller frames so all the values are collected first and then
+// sorted before being printed.
+class FrameValues {
+ private:
+  GrowableArray<FrameValue> _values;
+
+  static int compare(FrameValue* a, FrameValue* b) {
+    if (a->location == b->location) {
+      return a->priority - b->priority;
+    }
+    return a->location - b->location;
+  }
+
+ public:
+  // Used by frame functions to describe locations.
+  void describe(int owner, intptr_t* location, const char* description, int priority = 0);
+
+  bool validate();
+  void print();
+};
+
+#endif
 
 //
 // StackFrameStream iterates through the frames of a thread starting from

hotspot/src/share/vm/runtime/globals.hpp

@@ -1827,7 +1827,7 @@ class CommandLineFlags {
   develop(bool, VerifyBlockOffsetArray, false,                              \
           "Do (expensive!) block offset array verification")                \
                                                                             \
-  product(bool, BlockOffsetArrayUseUnallocatedBlock, false,                 \
+  diagnostic(bool, BlockOffsetArrayUseUnallocatedBlock, false,              \
           "Maintain _unallocated_block in BlockOffsetArray"                 \
           " (currently applicable only to CMS collector)")                  \
                                                                             \
@@ -2882,7 +2882,7 @@ class CommandLineFlags {
           "Max. no. of lines in the stack trace for Java exceptions "       \
           "(0 means all)")                                                  \
                                                                             \
-  NOT_EMBEDDED(develop(intx, GuaranteedSafepointInterval, 1000,             \
+  NOT_EMBEDDED(diagnostic(intx, GuaranteedSafepointInterval, 1000,          \
           "Guarantee a safepoint (at least) every so many milliseconds "    \
           "(0 means none)"))                                                \
                                                                             \

hotspot/src/share/vm/runtime/os.hpp

@@ -274,7 +274,7 @@ class os: AllStatic {
   static char*  reserve_memory_special(size_t size, char* addr = NULL,
                 bool executable = false);
   static bool   release_memory_special(char* addr, size_t bytes);
-  static bool   large_page_init();
+  static void   large_page_init();
   static size_t large_page_size();
   static bool   can_commit_large_page_memory();
   static bool   can_execute_large_page_memory();

hotspot/src/share/vm/runtime/thread.cpp

@@ -31,6 +31,7 @@
 #include "compiler/compileBroker.hpp"
 #include "interpreter/interpreter.hpp"
 #include "interpreter/linkResolver.hpp"
+#include "interpreter/oopMapCache.hpp"
 #include "jvmtifiles/jvmtiEnv.hpp"
 #include "memory/oopFactory.hpp"
 #include "memory/universe.inline.hpp"
@@ -2860,6 +2861,25 @@ void JavaThread::trace_frames() {
 }
 
 
+// Print or validate the layout of stack frames
+void JavaThread::print_frame_layout(int depth, bool validate_only) {
+  ResourceMark rm;
+  PRESERVE_EXCEPTION_MARK;
+  FrameValues values;
+  int frame_no = 0;
+  for(StackFrameStream fst(this, false); !fst.is_done(); fst.next()) {
+    fst.current()->describe(values, ++frame_no);
+    if (depth == frame_no) break;
+  }
+  if (validate_only) {
+    values.validate();
+  } else {
+    tty->print_cr("[Describe stack layout]");
+    values.print();
+  }
+}
+
+
 void JavaThread::trace_stack_from(vframe* start_vf) {
   ResourceMark rm;
   int vframe_no = 1;

hotspot/src/share/vm/runtime/thread.hpp

@@ -1380,6 +1380,12 @@ public:
   void trace_stack_from(vframe* start_vf)        PRODUCT_RETURN;
   void trace_frames()                            PRODUCT_RETURN;
 
+  // Print an annotated view of the stack frames
+  void print_frame_layout(int depth = 0, bool validate_only = false) PRODUCT_RETURN;
+  void validate_frame_layout() {
+    print_frame_layout(0, true);
+  }
+
   // Returns the number of stack frames on the stack
   int depth() const;
 

hotspot/src/share/vm/utilities/debug.cpp

@@ -469,6 +469,7 @@ extern "C" void verify() {
 extern "C" void pp(void* p) {
   Command c("pp");
   FlagSetting fl(PrintVMMessages, true);
+  FlagSetting f2(DisplayVMOutput, true);
   if (Universe::heap()->is_in(p)) {
     oop obj = oop(p);
     obj->print();
@@ -507,6 +508,17 @@ extern "C" void ps() { // print stack
 
 }
 
+extern "C" void pfl() {
+  // print frame layout
+  Command c("pfl");
+  JavaThread* p = JavaThread::active();
+  tty->print(" for thread: ");
+  p->print();
+  tty->cr();
+  if (p->has_last_Java_frame()) {
+    p->print_frame_layout();
+  }
+}
 
 extern "C" void psf() { // print stack frames
   {

hotspot/src/share/vm/utilities/elfFile.cpp

@@ -29,6 +29,7 @@
 #include <string.h>
 #include <stdio.h>
 #include <limits.h>
+#include <new>
 
 #include "memory/allocation.inline.hpp"
 #include "utilities/decoder.hpp"
@@ -46,7 +47,7 @@ ElfFile::ElfFile(const char* filepath) {
   m_status = Decoder::no_error;
 
   int len = strlen(filepath) + 1;
-  m_filepath = NEW_C_HEAP_ARRAY(char, len);
+  m_filepath = (const char*)os::malloc(len * sizeof(char));
   if (m_filepath != NULL) {
     strcpy((char*)m_filepath, filepath);
     m_file = fopen(filepath, "r");
@@ -74,7 +75,7 @@ ElfFile::~ElfFile() {
   }
 
   if (m_filepath != NULL) {
-    FREE_C_HEAP_ARRAY(char, m_filepath);
+    os::free((void*)m_filepath);
   }
 
   if (m_next != NULL) {
@@ -120,14 +121,14 @@ bool ElfFile::load_tables() {
       }
       // string table
       if (shdr.sh_type == SHT_STRTAB) {
-        ElfStringTable* table = new ElfStringTable(m_file, shdr, index);
+        ElfStringTable* table = new (std::nothrow) ElfStringTable(m_file, shdr, index);
         if (table == NULL) {
           m_status = Decoder::out_of_memory;
           return false;
         }
         add_string_table(table);
       } else if (shdr.sh_type == SHT_SYMTAB || shdr.sh_type == SHT_DYNSYM) {
-        ElfSymbolTable* table = new ElfSymbolTable(m_file, shdr);
+        ElfSymbolTable* table = new (std::nothrow) ElfSymbolTable(m_file, shdr);
         if (table == NULL) {
           m_status = Decoder::out_of_memory;
           return false;

hotspot/src/share/vm/utilities/elfStringTable.cpp

@@ -27,6 +27,7 @@
 #ifndef _WINDOWS
 
 #include "memory/allocation.inline.hpp"
+#include "runtime/os.hpp"
 #include "utilities/elfStringTable.hpp"
 
 // We will try to load whole string table into memory if we can.
@@ -41,14 +42,14 @@ ElfStringTable::ElfStringTable(FILE* file, Elf_Shdr shdr, int index) {
 
   // try to load the string table
   long cur_offset = ftell(file);
-  m_table = (char*)NEW_C_HEAP_ARRAY(char, shdr.sh_size);
+  m_table = (char*)os::malloc(sizeof(char) * shdr.sh_size);
   if (m_table != NULL) {
     // if there is an error, mark the error
     if (fseek(file, shdr.sh_offset, SEEK_SET) ||
       fread((void*)m_table, shdr.sh_size, 1, file) != 1 ||
       fseek(file, cur_offset, SEEK_SET)) {
       m_status = Decoder::file_invalid;
-      FREE_C_HEAP_ARRAY(char, m_table);
+      os::free((void*)m_table);
       m_table = NULL;
     }
   } else {
@@ -58,7 +59,7 @@ ElfStringTable::ElfStringTable(FILE* file, Elf_Shdr shdr, int index) {
 
 ElfStringTable::~ElfStringTable() {
   if (m_table != NULL) {
-    FREE_C_HEAP_ARRAY(char, m_table);
+    os::free((void*)m_table);
   }
 
   if (m_next != NULL) {

hotspot/test/compiler/5091921/Test5091921.java

@@ -0,0 +1,309 @@
+/*
+ * Copyright (c) 2011 Hewlett-Packard Company. 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.
+ *
+ */
+
+/**
+ * @test
+ * @bug 5091921
+ * @summary Sign flip issues in loop optimizer
+ *
+ * @run main/othervm -Xcomp -XX:CompileOnly=Test5091921 -XX:MaxInlineSize=1 Test5091921
+ */
+
+public class Test5091921 {
+  private static int result = 0;
+
+
+  /* Test for the bug of transforming indx >= MININT to indx > MININT-1 */
+  public static int test_ge1(int limit) {
+    int indx;
+    int sum = 0;
+    for (indx = 500; indx >= limit; indx -= 2) {
+      sum += 2000 / indx;
+      result = sum;
+    }
+    return sum;
+  }
+
+  /* Test for the bug of transforming indx <= MAXINT to indx < MAXINT+1 */
+  public static int test_le1(int limit) {
+    int indx;
+    int sum = 0;
+    for (indx = -500; indx <= limit; indx += 2)
+    {
+      sum += 3000 / indx;
+      result = sum;
+    }
+    return sum;
+  }
+
+  /* Run with -Xcomp -XX:CompileOnly=wrap1.test1 -XX:MaxInlineSize=1 */
+  /* limit reset to ((limit-init+stride-1)/stride)*stride+init */
+  /* Calculation may overflow */
+  public static volatile int c = 1;
+  public static int test_wrap1(int limit)
+  {
+    int indx;
+    int sum = 0;
+    for (indx = 0xffffffff; indx < limit; indx += 0x20000000)
+    {
+      sum += c;
+    }
+    return sum;
+  }
+
+  /* Test for range check elimination with bit flip issue for
+     scale*i+offset<limit where offset is not 0 */
+  static int[] box5 = {1,2,3,4,5,6,7,8,9};
+  public static int test_rce5(int[] b, int limit)
+  {
+    int indx;
+    int sum = b[1];
+    result = sum;
+    for (indx = 0x80000000; indx < limit; ++indx)
+    {
+      if (indx > 0x80000000)
+      {
+        // this test is not issued in pre-loop but issued in main loop
+        // trick rce into thinking expression is false when indx >= 0
+        // in fact it is false when indx==0x80000001
+        if (indx - 9 < -9)
+        {
+          sum += indx;
+          result = sum;
+          sum ^= b[indx & 7];
+          result = sum;
+        }
+        else
+          break;
+      }
+      else
+      {
+        sum += b[indx & 3];
+        result = sum;
+      }
+    }
+    return sum;
+  }
+
+  /* Test for range check elimination with bit flip issue for
+     scale*i<limit where scale > 1 */
+  static int[] box6 = {1,2,3,4,5,6,7,8,9};
+  public static int test_rce6(int[] b, int limit)
+  {
+    int indx;
+    int sum = b[1];
+    result = sum;
+    for (indx = 0x80000000; indx < limit; ++indx)
+    {
+      if (indx > 0x80000000)
+      {
+        // harmless rce target
+        if (indx < 0)
+        {
+          sum += result;
+          result = sum;
+        }
+        else
+          break;
+        // this test is not issued in pre-loop but issued in main loop
+        // trick rce into thinking expression is false when indx >= 0
+        // in fact it is false when indx==0x80000001
+        // In compilers that transform mulI to shiftI may mask this issue.
+        if (indx * 28 + 1 < 0)
+        {
+          sum += indx;
+          result = sum;
+          sum ^= b[indx & 7];
+          result = sum;
+        }
+        else
+          break;
+      }
+      else
+      {
+        sum += b[indx & 3];
+        result = sum;
+      }
+    }
+    return sum;
+  }
+
+  /* Test for range check elimination with i <= limit */
+  static int[] box7 = {1,2,3,4,5,6,7,8,9,0x7fffffff};
+  public static int test_rce7(int[] b)
+  {
+    int indx;
+    int max = b[9];
+    int sum = b[7];
+    result = sum;
+    for (indx = 0; indx < b.length; ++indx)
+    {
+      if (indx <= max)
+      {
+        sum += (indx ^ 15) + ((result != 0) ? 0 : sum);
+        result = sum;
+      }
+      else
+        throw new RuntimeException();
+    }
+    for (indx = -7; indx < b.length; ++indx)
+    {
+      if (indx <= 9)
+      {
+        sum += (sum ^ 15) + ((result != 0) ? 0 : sum);
+        result = sum;
+      }
+      else
+        throw new RuntimeException();
+    }
+    return sum;
+  }
+
+  /* Test for range check elimination with i >= limit */
+  static int[] box8 = {-1,0,1,2,3,4,5,6,7,8,0x80000000};
+  public static int test_rce8(int[] b)
+  {
+    int indx;
+    int sum = b[5];
+    int min = b[10];
+    result = sum;
+    for (indx = b.length-1; indx >= 0; --indx)
+    {
+      if (indx >= min)
+      {
+        sum += (sum ^ 9) + ((result != 0) ? 0 :sum);
+        result = sum;
+      }
+      else
+        throw new RuntimeException();
+    }
+    return sum;
+  }
+
+  public static void main(String[] args)
+  {
+    result=1;
+    int r = 0;
+    try {
+      r = test_ge1(0x80000000);
+      System.out.println(result);
+      System.out.println("test_ge1 FAILED");
+      System.exit(1);
+    }
+    catch (ArithmeticException e1) {
+      System.out.println("test_ge1: Expected exception caught");
+      if (result != 5986) {
+        System.out.println(result);
+        System.out.println("test_ge1 FAILED");
+        System.exit(97);
+      }
+    }
+    System.out.println("test_ge1 WORKED");
+
+    result=0;
+    try
+    {
+      r = test_le1(0x7fffffff);
+      System.out.println(result);
+      System.out.println("test_le1 FAILED");
+      System.exit(1);
+    }
+    catch (ArithmeticException e1)
+    {
+      System.out.println("test_le1: Expected exception caught");
+      if (result != -9039)
+      {
+        System.out.println(result);
+        System.out.println("test_le1 FAILED");
+        System.exit(97);
+      }
+    }
+    System.out.println("test_le1 WORKED");
+
+    result=0;
+    r = test_wrap1(0x7fffffff);
+    if (r != 4)
+    {
+      System.out.println(result);
+      System.out.println("test_wrap1 FAILED");
+      System.exit(97);
+    }
+    else
+    {
+      System.out.println("test_wrap1 WORKED");
+    }
+
+    result=0;
+    r = test_rce5(box5,0x80000100);
+    if (result != 3)
+    {
+      System.out.println(result);
+      System.out.println("test_rce5 FAILED");
+      System.exit(97);
+    }
+    else
+    {
+      System.out.println("test_rce5 WORKED");
+    }
+
+    result=0;
+    r = test_rce6(box6,0x80000100);
+    if (result != 6)
+    {
+      System.out.println(result);
+      System.out.println("test_rce6 FAILED");
+      System.exit(97);
+    }
+    else
+    {
+      System.out.println("test_rce6 WORKED");
+    }
+
+    result=0;
+    r = test_rce7(box7);
+    if (result != 14680079)
+    {
+      System.out.println(result);
+      System.out.println("test_rce7 FAILED");
+      System.exit(97);
+    }
+    else
+    {
+      System.out.println("test_rce7 WORKED");
+    }
+
+    result=0;
+    r = test_rce8(box8);
+    if (result != 16393)
+    {
+      System.out.println(result);
+      System.out.println("test_rce8 FAILED");
+      System.exit(97);
+    }
+    else
+    {
+      System.out.println("test_rce8 WORKED");
+    }
+  }
+}

hotspot/test/compiler/5091921/Test6186134.java

@@ -0,0 +1,71 @@
+/*
+ * 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.
+ *
+ */
+
+/**
+ * @test
+ * @bug 6186134
+ * @summary Server virtual machine produces/exeutes incorrect code.
+ *
+ * @run main Test6186134 100000
+ */
+import java.util.ArrayList;
+
+public class Test6186134 {
+
+  int num = 0;
+
+  public Test6186134(int n) {
+    num = n;
+  }
+
+  public boolean more() {
+    return num-- > 0;
+  }
+
+  public ArrayList test1() {
+    ArrayList res = new ArrayList();
+    int maxResults = Integer.MAX_VALUE;
+    int n = 0;
+    boolean more = more();
+    while ((n++ < maxResults) && more) {
+      res.add(new Object());
+      more = more();
+    }
+    return res;
+  }
+
+  public static void main(String[] pars) {
+    int n = Integer.parseInt(pars[0]);
+    for (int i=0; i<n; i++) {
+      Test6186134 t = new Test6186134(10);
+      int size = t.test1().size();
+      if (size != 10) {
+        System.out.println("wrong size: " + size +", should be 10");
+        System.exit(97);
+      }
+    }
+    System.out.println("Passed");
+  }
+}
+

hotspot/test/compiler/5091921/Test6196102.java

@@ -0,0 +1,47 @@
+/*
+ * 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.
+ *
+ */
+
+/**
+ * @test
+ * @bug 6196102
+ * @summary Integer seems to be greater than Integer.MAX_VALUE
+ *
+ * @run main Test6196102
+ */
+
+public class Test6196102 {
+    static public void main(String[] args) {
+        int i1 = 0;
+        int i2 = Integer.MAX_VALUE;
+
+        while (i1 >= 0) {
+            i1++;
+            if (i1 > i2) {
+                System.out.println("E R R O R: " + i1);
+                System.exit(97);
+            }
+        }
+    }
+}
+

hotspot/test/compiler/5091921/Test6357214.java

@@ -0,0 +1,179 @@
+/*
+ * 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.
+ *
+ */
+
+/**
+ * @test
+ * @bug 6357214
+ * @summary Hotspot server compiler gets integer comparison wrong
+ *
+ * @run main/othervm/timeout=60 -DshowAll=ffo -DeventID=444 Test6357214
+ */
+
+// The test hangs after few iterations before the fix. So it fails if timeout.
+class MyResult {
+        public boolean next() {
+                return true;
+        }
+
+        public String getString(String in) {
+                if (in.equals("id"))
+                        return "idFoo";
+                if (in.equals("contentKey"))
+                        return "ckFoo";
+                return "Foo";
+        }
+
+        public int getInt(String in) {
+                if (in.equals("processingComplete"))
+                        return 0;
+                return 1;
+        }
+
+        public byte[] getBytes(String in) {
+                byte[] arr = null;
+                if (in.equals("content")) {
+                        arr = new byte[65536];
+                        byte j = 32;
+                        for (int i=0; i<65536; i++) {
+                                arr[i] = j;
+                                if (++j == 127)
+                                        j=32;
+                        }
+                }
+                return arr;
+        }
+}
+
+public class Test6357214 {
+        public static volatile boolean bollocks = true;
+    public String create(String context) throws Exception {
+
+        //
+        // Extract HTTP parameters
+        //
+
+        boolean showAll = System.getProperty("showAll") != null;
+          String eventID = System.getProperty("eventID");
+          String eventContentKey = System.getProperty("cKey");
+        //
+        // Build ContentStaging query based on eventID or eventContentKey
+        //
+
+        String sql = "select id, processingComplete, contentKey, content "
+                   + "from   ContentStaging cs, ContentStagingKey csk "
+                   + "where  cs.eventContentKey = csk.eventContentKey ";
+
+        if (eventID != null) {
+            sql += "and id = " + eventID;
+        }
+        else if (eventContentKey != null) {
+            sql += "and cs.eventContentKey = '"
+                +  eventContentKey
+                +  "' having id = max(id)";
+        }
+        else {
+            throw new Exception("Need eventID or eventContentKey");
+        }
+
+        //
+        // This factory builds a static panel, there is no JSP
+        //
+
+        StringBuffer html = new StringBuffer();
+
+        try {
+
+                MyResult result = new MyResult();
+            if (result.next()) {
+
+                eventID = result.getString("id");
+                int processingComplete = result.getInt("processingComplete");
+                String contentKey = result.getString("contentKey");
+                byte[] bytes = result.getBytes("content");
+
+                //
+                // Print content status and associated controls
+                //
+
+                html.append("<br/><font class=\"small\">");
+                html.append("Status: ");
+                switch (processingComplete) {
+                    case  0 :
+                    case  1 : html.append("PENDING"); break;
+                    case  2 : html.append(contentKey); break;
+                    case  3 : html.append(eventID); break;
+                    default : html.append("UNKNONW");
+                }
+                html.append("</font><br/>");
+
+                //
+                // Print at most 20Kb of content unless "showAll" is set
+                //
+
+                int limit = showAll ? Integer.MAX_VALUE : 1024 * 20;
+                System.out.println(limit);
+                html.append("<pre>");
+                for (int i = 0; bytes != null && i < bytes.length; i++) {
+                    char c = (char) bytes[i];
+                    switch (c) {
+                        case '<' : html.append("&lt;");  break;
+                        case '>' : html.append("&gt;");  break;
+                        case '&' : html.append("&amp;"); break;
+                        default  : html.append(c);
+                    }
+
+                    if (i > limit) {
+                        while (bollocks);
+                        // System.out.println("i is " + i);
+                        // System.out.println("limit is " + limit);
+                        html.append("...\n</pre>");
+                        html.append(eventID);
+                        html.append("<pre>");
+                        break;
+                    }
+                }
+                html.append("</pre>");
+            }
+        }
+        catch (Exception exception) {
+            throw exception;
+        }
+        finally {
+            html.append("Oof!!");
+        }
+        String ret = html.toString();
+        System.out.println("Returning string length = "+ ret.length());
+        return ret;
+    }
+
+    public static void main(String[] args) throws Exception {
+                int length=0;
+
+                for (int i = 0; i < 100; i++) {
+                        length = new Test6357214().create("boo").length();
+                        System.out.println(length);
+                }
+    }
+}
+

hotspot/test/compiler/5091921/Test6559156.java

@@ -0,0 +1,81 @@
+/*
+ * 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.
+ *
+ */
+
+/**
+ * @test
+ * @bug 6559156
+ * @summary Server compiler generates bad code for "<= Integer.MAX_VALUE" expression
+ *
+ * @run main Test6559156
+ */
+
+public class Test6559156 {
+
+    static final int N_TESTS = 1000000;
+
+    public static void main(String[] args) throws Exception {
+
+        /*
+         * If MAX_VALUE is changed to MAX_VALUE - 1 below, the test passes
+         * because (apparently) bad code is only generated when comparing
+         * <= MAX_VALUE in the doTest method.
+         */
+        Test6559156 test = new Test6559156();
+        for (int i = 0; i < N_TESTS; i += 1) {
+            test.doTest1(10, Integer.MAX_VALUE, i);
+            test.doTest2(10, Integer.MAX_VALUE, i);
+        }
+        System.out.println("No failure");
+    }
+
+    void doTest1(int expected, int max, int i) {
+        int counted;
+        for (counted = 0;
+             (counted <= max) && (counted < expected);
+             counted += 1) {
+        }
+        if (counted != expected) {
+            throw new RuntimeException("Failed test1 iteration=" + i +
+                                       " max=" + max +
+                                       " counted=" + counted +
+                                       " expected=" + expected);
+        }
+    }
+
+    void doTest2(int expected, int max, int i) {
+        int counted;
+        for (counted = 0;
+             // change test sequence.
+             (counted < expected) && (counted <= max);
+             counted += 1) {
+        }
+        if (counted != expected) {
+            throw new RuntimeException("Failed test1 iteration=" + i +
+                                       " max=" + max +
+                                       " counted=" + counted +
+                                       " expected=" + expected);
+        }
+    }
+}
+

hotspot/test/compiler/5091921/Test6753639.java

@@ -0,0 +1,50 @@
+/*
+ * 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.
+ *
+ */
+
+/**
+ * @test
+ * @bug 6753639
+ * @summary Strange optimisation in for loop with cyclic integer condition
+ *
+ * @run main/othervm -Xbatch Test6753639
+ */
+
+public class Test6753639 {
+    public static void main(String[] args) throws InterruptedException {
+        int END = Integer.MAX_VALUE;
+        int count = 0;
+        for(int i = Integer.MAX_VALUE - 5; i <= END; i++) {
+            count++;
+            if (count > 100000) {
+                System.out.println("Passed");
+                System.exit(95);
+            }
+        }
+        System.out.println("broken " + count);
+        System.out.println("FAILED");
+        System.exit(97);
+    }
+}
+
+

hotspot/test/compiler/5091921/Test6850611.java

@@ -0,0 +1,53 @@
+/*
+ * 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.
+ *
+ */
+
+/**
+ * @test
+ * @bug 6850611
+ * @summary int / long arithmetic seems to be broken in 1.6.0_14 HotSpot Server VM (Win XP)
+ *
+ * @run main Test6850611
+ */
+
+public class Test6850611 {
+
+    public static void main(String[] args) {
+        test();
+    }
+
+    private static void test() {
+        for (int j = 0; j < 5; ++j) {
+            long x = 0;
+            for (int i = Integer.MIN_VALUE; i < Integer.MAX_VALUE; ++i) {
+                x += i;
+            }
+            System.out.println("sum: " + x);
+            if (x != -4294967295l) {
+                System.out.println("FAILED");
+                System.exit(97);
+            }
+        }
+    }
+}
+

hotspot/test/compiler/5091921/Test6890943.java

@@ -0,0 +1,189 @@
+/*
+ * 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.
+ *
+ */
+
+/**
+ * @test
+ * @bug 6890943
+ * @summary JVM mysteriously gives wrong result on 64-bit 1.6 VMs in hotspot mode.
+ *
+ * @run shell Test6890943.sh
+ */
+import java.util.*;
+import java.io.*;
+import java.util.regex.*;
+
+public class Test6890943 {
+  public static final boolean AIR = true, ROCK = false;
+  public static void main(String[] args) {
+    new Test6890943().go();
+  }
+
+  int r, c, f, t;
+  boolean[][] grid;
+
+  public void go() {
+    Scanner s = new Scanner(System.in);
+    s.useDelimiter("\\s+");
+    int T = s.nextInt();
+    for (t = 0 ; t < T ; t++) {
+      r = s.nextInt(); c = s.nextInt(); f = s.nextInt();
+      grid = new boolean[r][c];
+      for (int x = 0 ; x < r ; x++) {
+        String line = s.next();
+        for (int y = 0 ; y < c ; y++) grid[x][y] = line.charAt(y) == '.';
+      }
+      int digs = solve();
+      String res = digs == -1 ? "No" : "Yes " + digs;
+      System.out.printf("Case #%d: %s\n", t+1, res);
+    }
+  }
+
+  Map<Integer, Integer> M = new HashMap<Integer, Integer>();
+
+  private int solve() {
+    M = new HashMap<Integer, Integer>();
+    M.put(calcWalkingRange(0, 0), 0);
+    for (int digDown = 0 ; digDown < r ; digDown++) {
+      Map<Integer, Integer> tries = new HashMap<Integer, Integer>();
+      for (Map.Entry<Integer, Integer> m : M.entrySet()) {
+        int q = m.getKey();
+        if (depth(q) != (digDown)) continue;
+        if (stuck(q)) continue;
+        tries.put(q, m.getValue());
+      }
+
+      for (Map.Entry<Integer, Integer> m : tries.entrySet()) {
+        int q = m.getKey();
+        int fallLeftDelta = 0, fallRightDelta = 0;
+        //fall left
+        int fallLeft = fall(digDown, start(q));
+        if (fallLeft > 0) {
+          fallLeftDelta = 1;
+          if (fallLeft <= f) addToM(calcWalkingRange(digDown+fallLeft, start(q)), m.getValue());
+        }
+
+        //fall right
+        int fallRight = fall(digDown, end(q));
+        if (fallRight > 0) {
+          fallRightDelta = 1;
+
+          if (fallRight <= f) addToM(calcWalkingRange(digDown+fallRight, end(q)), m.getValue());
+        }
+
+        for (int p = start(q) + fallLeftDelta ; p <= end(q) - fallRightDelta ; p++) {
+          //goLeft
+          for (int digSpot = p ; digSpot > start(q) +fallLeftDelta ; digSpot--) {
+            int fallDown = 1+fall(digDown+1, digSpot);
+            if (fallDown <= f) {
+              if (fallDown == 1) {
+                addToM(calcWalkingRange(digDown + 1, digSpot, digSpot, p), m.getValue() + Math.abs(digSpot-p)+1);
+              } else {
+                addToM(calcWalkingRange(digDown + fallDown, digSpot), m.getValue() + Math.abs(digSpot-p)+1);
+              }
+            }
+          }
+
+          //goRight
+          for (int digSpot = p ; digSpot < end(q)-fallRightDelta ;digSpot++) {
+            int fallDown = 1+fall(digDown+1, digSpot);
+            if (fallDown <= f) {
+              if (fallDown == 1) {
+                addToM(calcWalkingRange(digDown + 1, digSpot, p, digSpot), m.getValue() + Math.abs(digSpot-p)+1);
+              } else {
+                addToM(calcWalkingRange(digDown + fallDown, digSpot), m.getValue() + Math.abs(digSpot-p)+1);
+              }
+            }
+          }
+        }
+      }
+    }
+
+    int result = Integer.MAX_VALUE;
+    for (Map.Entry<Integer, Integer> m : M.entrySet()) {
+      if (depth(m.getKey()) == r-1) result = Math.min(m.getValue(), result);
+    }
+
+    if (result == Integer.MAX_VALUE) return -1;
+    return result;
+  }
+
+  private void addToM(int q, int i) {
+    Integer original = M.get(q);
+    if ( original == null ) M.put(q, i);
+    else M.put(q, Math.min(original, i));
+  }
+
+  private int fall(int row, int column) {
+    int res = 0;
+    for ( int p = row+1 ; p < r ; p++) {
+      if (grid[p][column] == AIR) res++;
+      else break;
+    }
+    return res;
+  }
+
+  private boolean stuck(int q) {
+    return start(q) == end(q);
+  }
+
+  private int depth(int q) {
+    return q % 50;
+  }
+
+  private int start(int q) {
+    return q / (50*50);
+  }
+
+  private int end(int q) {
+    return (q / 50) % 50;
+  }
+
+  private int calcWalkingRange(int depth, int pos) {
+    return calcWalkingRange(depth, pos, Integer.MAX_VALUE, Integer.MIN_VALUE);
+  }
+
+  private int calcWalkingRange(int depth, int pos, int airOverrideStart, int airOverrideEnd) {
+    int left = pos, right = pos;
+    if (depth >= r) return (c-1)*50 + depth;
+
+    while (left > 0) {
+      if (grid[depth][left-1] == ROCK && (left-1 < airOverrideStart || left-1 > airOverrideEnd)) break;
+      if (depth < r-1 && grid[depth+1][left-1] == AIR) {
+        left--;
+        break;
+      }
+      left--;
+    }
+    while (right < c-1) {
+      if (grid[depth][right+1] == ROCK && (right+1 < airOverrideStart || right+1 > airOverrideEnd)) break;
+      if (depth < r-1 && grid[depth+1][right+1] == AIR) {
+        right++;
+        break;
+      }
+      right++;
+    }
+
+    return left *50*50 + right*50 + depth;
+  }
+}

hotspot/test/compiler/5091921/Test6890943.sh

@@ -0,0 +1,67 @@
+#!/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}/Test6890943.java .
+cp ${TESTSRC}/input6890943.txt .
+cp ${TESTSRC}/output6890943.txt .
+cp ${TESTSRC}/Test6890943.sh .
+
+${TESTJAVA}/bin/javac -d . Test6890943.java
+
+${TESTJAVA}/bin/java  ${TESTVMOPTS} Test6890943 < input6890943.txt > test.out 2>&1
+
+diff output6890943.txt test.out
+
+result=$?
+if [ $result -eq 0 ]
+then
+  echo "Passed"
+  exit 0
+else
+  echo "Failed"
+  exit 1
+fi

hotspot/test/compiler/5091921/Test6897150.java

@@ -0,0 +1,58 @@
+/*
+ * 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.
+ *
+ */
+
+/**
+ * @test
+ * @bug 6897150
+ * @summary Hotspot optimises away a valid loop
+ *
+ * @run main Test6897150
+ */
+
+// Should be compiled with javac from JDK1.3 to get bytecode which shows the problem.
+public class Test6897150 {
+    public static void main(String[] args) {
+        // This works
+        loopAndPrint(Integer.MAX_VALUE -1);
+        // This doesn't
+        loopAndPrint(Integer.MAX_VALUE);
+    }
+
+    static void verify(int max, int a) {
+        if ( a != (max - 1)) {
+            System.out.println("Expected: " + (max - 1));
+            System.out.println("Actual  : " + a);
+            System.exit(97);
+        }
+    }
+    static void loopAndPrint(int max) {
+        int a = -1;
+        int i = 1;
+        for (; i < max; i++) {
+            a = i;
+        }
+        verify(max, a);
+    }
+}
+

hotspot/test/compiler/5091921/Test6905845.java

@@ -0,0 +1,76 @@
+/*
+ * 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.
+ *
+ */
+
+/**
+ * @test
+ * @bug 6905845
+ * @summary Server VM improperly optimizing away loop.
+ *
+ * @run main Test6905845
+ */
+
+public class Test6905845 {
+
+   public static void main(String[] args){
+      for (int asdf = 0; asdf < 5; asdf++){
+         //test block
+         {
+            StringBuilder strBuf1 = new StringBuilder(65);
+            long          start   = System.currentTimeMillis();
+            int           count   = 0;
+
+            for (int i = Integer.MIN_VALUE; i < (Integer.MAX_VALUE - 80); i += 79){
+               strBuf1.append(i);
+               count++;
+               strBuf1.delete(0, 65);
+            }
+
+            System.out.println(count);
+            if (count != 54366674) {
+              System.out.println("wrong count: " + count +", should be 54366674");
+              System.exit(97);
+            }
+         }
+         //test block
+         {
+            StringBuilder strBuf1 = new StringBuilder(65);
+            long          start   = System.currentTimeMillis();
+            int           count   = 0;
+
+            for (int i = Integer.MIN_VALUE; i < (Integer.MAX_VALUE - 80); i += 79){
+               strBuf1.append(i);
+               count++;
+               strBuf1.delete(0, 65);
+            }
+
+            System.out.println(count);
+            if (count != 54366674) {
+              System.out.println("wrong count: " + count +", should be 54366674");
+              System.exit(97);
+            }
+         }
+      }
+   }
+}
+

hotspot/test/compiler/5091921/Test6931567.java

@@ -0,0 +1,62 @@
+/*
+ * 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.
+ *
+ */
+
+/**
+ * @test
+ * @bug 6931567
+ * @summary JIT Error (on class file compiled with eclipse) on JVM x64 (but not on x32!).
+ *
+ * @run main Test6931567
+ */
+
+// Should be compiled with javac from JDK1.3 to get bytecode which shows the problem.
+public class Test6931567 {
+
+    public static void main(final String[] args) {
+        booleanInvert(Integer.MAX_VALUE);
+        booleanInvert(Integer.MAX_VALUE - 1);
+    }
+
+    private static void booleanInvert(final int max) {
+        boolean test1 = false;
+        boolean test2 = false;
+
+        for (int i = 0; i < max; i++) {
+            test1 = !test1;
+        }
+
+        for (int i = 0; i < max; i++) {
+            test2 ^= true;
+        }
+
+        if (test1 != test2) {
+            System.out.println("ERROR: Boolean invert\n\ttest1=" + test1
+                    + "\n\ttest2=" + test2);
+            System.exit(97);
+        } else {
+            System.out.println("Passed!");
+        }
+    }
+}
+

hotspot/test/compiler/5091921/Test6935022.java

@@ -0,0 +1,74 @@
+/*
+ * 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.
+ *
+ */
+
+/**
+ * @test
+ * @bug 6935022
+ * @summary Server VM incorrectly breaks out of while loop
+ *
+ * @run main Test6935022
+ */
+
+public class Test6935022 {
+    public static final void main(String[] args) throws Exception {
+        Test6935022 test = new Test6935022();
+
+        int cnt = 0;
+
+        while (cnt < 10000) {
+            try {
+                ++cnt;
+                if ((cnt&1023) == 0)
+                  System.out.println("Thread="+Thread.currentThread().getName() + " iteration: " + cnt);
+                test.loop(2147483647, (cnt&1023));
+            }
+
+            catch (Exception e) {
+                System.out.println("Caught on iteration " + cnt);
+                e.printStackTrace();
+                System.exit(97);
+            }
+        }
+    }
+
+    private void loop(int endingRow, int mask) throws Exception {
+        int rows = 1;
+        boolean next = true;
+
+        while(rows <= endingRow && next) {
+            rows++;
+            if (rows == mask)
+              System.out.println("Rows="+rows+", end="+endingRow+", next="+next);
+            next = next(rows);
+        }
+
+        if (next)
+            throw new Exception("Ended on rows(no rs): " + rows);
+    }
+
+    private boolean next(int rows) {
+        return rows < 12;
+    }
+}
+

hotspot/test/compiler/5091921/Test6959129.java

@@ -0,0 +1,90 @@
+/*
+ * 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.
+ *
+ */
+
+/**
+ * @test
+ * @bug 6959129
+ * @summary COMPARISON WITH INTEGER.MAX_INT DOES NOT WORK CORRECTLY IN THE CLIENT VM.
+ *
+ * @run main/othervm -ea Test6959129
+ */
+
+public class Test6959129 {
+
+  public static void main(String[] args) {
+    long start  = System.currentTimeMillis();
+    int min = Integer.MAX_VALUE-30000;
+    int max = Integer.MAX_VALUE;
+    long maxmoves = 0;
+    try {
+      maxmoves = maxMoves(min, max);
+    } catch (AssertionError e) {
+      System.out.println("Passed");
+      System.exit(95);
+    }
+    System.out.println("maxMove:" + maxmoves);
+    System.out.println("FAILED");
+    System.exit(97);
+  }
+  /**
+   * Imperative implementation that returns the length hailstone moves
+   * for a given number.
+   */
+  public static long hailstoneLengthImp(long n) {
+    long moves = 0;
+    while (n != 1) {
+      assert n > 1;
+      if (isEven(n)) {
+        n = n / 2;
+      } else {
+        n = 3 * n + 1;
+      }
+      ++moves;
+    }
+    return moves;
+  }
+
+  private static boolean isEven(long n) {
+    return n % 2 == 0;
+  }
+
+  /**
+   * Returns the maximum length of the hailstone sequence for numbers
+   * between min to max.
+   *
+   * For rec1 - Assume that min is bigger than max.
+   */
+  public static long maxMoves(int min, int max) {
+    long maxmoves = 0;
+    for (int n = min; n <= max; n++) {
+      if ((n & 1023) == 0) System.out.println(n);
+      long moves = hailstoneLengthImp(n);
+      if (moves > maxmoves) {
+        maxmoves = moves;
+      }
+    }
+    return maxmoves;
+  }
+}
+

hotspot/test/compiler/5091921/Test6985295.java

@@ -0,0 +1,53 @@
+/*
+ * 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.
+ *
+ */
+
+/**
+ * @test
+ * @bug 6985295
+ * @summary JVM fails to evaluate condition randomly
+ *
+ * @run main/othervm -Xbatch Test6985295
+ */
+
+public class Test6985295 {
+
+    public static void main(String[] args) {
+        int min = Integer.MAX_VALUE-50000;
+        int max = Integer.MAX_VALUE;
+        System.out.println("max = " + max);
+        long counter = 0;
+        int i;
+        for(i = min; i <= max; i++) {
+            counter++;
+            if (counter > 1000000) {
+              System.out.println("Passed");
+              System.exit(95);
+            }
+        }
+        System.out.println("iteration went " + counter + " times (" + i + ")");
+        System.out.println("FAILED");
+        System.exit(97);
+    }
+}
+

hotspot/test/compiler/5091921/Test6992759.java

@@ -0,0 +1,66 @@
+/*
+ * 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.
+ *
+ */
+
+/**
+ * @test
+ * @bug 6992759
+ * @summary Bad code generated for integer <= comparison, fails for Integer.MAX_VALUE
+ *
+ * @run main Test6992759
+ */
+
+public class Test6992759 {
+
+    static final int N_TESTS = 1000000000;
+
+    public static void main(String[] args) throws Exception {
+
+        /*
+         * If MAX_VALUE is changed to MAX_VALUE - 1 below, the test passes
+         * because (apparently) bad code is only generated when comparing
+         * <= MAX_VALUE in the doTest method.
+         */
+        Test6992759 test = new Test6992759();
+        for (int i = 0; i < N_TESTS; i += 1) {
+            test.doTest(10, Integer.MAX_VALUE, i);
+            //test.doTest(10, Integer.MAX_VALUE - 1, i);
+        }
+        System.out.println("No failure");
+    }
+
+    void doTest(int expected, int max, int i) {
+        int counted;
+        for (counted = 0;
+             (counted <= max) && (counted < expected);
+             counted += 1) {
+        }
+        if (counted != expected) {
+            throw new RuntimeException("Failed test iteration=" + i +
+                                       " max=" + max +
+                                       " counted=" + counted +
+                                       " expected=" + expected);
+        }
+    }
+}
+

hotspot/test/compiler/5091921/Test7005594.java

@@ -0,0 +1,59 @@
+/*
+ * 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.
+ *
+ */
+
+/**
+ * @test
+ * @bug 7005594
+ * @summary Array overflow not handled correctly with loop optimzations
+ *
+ * @run main/othervm -Xms2048m -Xcomp -XX:CompileOnly=Test7005594.test Test7005594
+ */
+
+public class Test7005594 {
+
+      static int test(byte a[]){
+          int result=0;
+          for( int i=0; i<a.length; i+=((0x7fffffff>>1)+1) ){
+              result += a[i];
+          }
+          return result;
+      }
+
+      public static void main(String [] args){
+          byte a[]=new byte[(0x7fffffff>>1)+2];
+          int result = 0;
+          try {
+              result = test(a);
+          } catch (ArrayIndexOutOfBoundsException e) {
+              e.printStackTrace(System.out);
+              System.out.println("Passed");
+              System.exit(95);
+          }
+          System.out.println(result);
+          System.out.println("FAILED");
+          System.exit(97);
+      }
+
+}
+

hotspot/test/compiler/5091921/Test7020614.java

@@ -0,0 +1,58 @@
+/*
+ * 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.
+ *
+ */
+
+/**
+ * @test
+ * @bug 7020614
+ * @summary "-server" mode optimizer makes code hang
+ *
+ * @run main/othervm/timeout=30 -Xbatch Test7020614
+ */
+
+public class Test7020614 {
+
+    private static final int ITERATIONS = 1000;
+    private static int doNotOptimizeOut = 0;
+
+    public static long bitCountShort() {
+        long t0 = System.currentTimeMillis();
+        int sum = 0;
+        for (int it = 0; it < ITERATIONS; ++it) {
+            short value = 0;
+            do {
+                sum += Integer.bitCount(value);
+            } while (++value != 0);
+        }
+        doNotOptimizeOut += sum;
+        return System.currentTimeMillis() - t0;
+    }
+
+    public static void main(String[] args) {
+        for (int i = 0; i < 4; ++i) {
+            System.out.println((i + 1) + ": " + bitCountShort());
+        }
+        System.out.println("doNotOptimizeOut value: " + doNotOptimizeOut);
+    }
+}
+