# HG changeset patch # User dcubed # Date 1365203778 25200 # Node ID 46d24f112c273e51c0dd4f9af4aa682a53c67eb8 # Parent 4c8bb5e4f68f471f2ea56d29b6b8f6ea9643c96a# Parent 8be1318fbe77ebf8518ce1c7b590b4067edc5285 Merge diff -r 4c8bb5e4f68f -r 46d24f112c27 .hgtags --- a/.hgtags Fri Apr 05 12:19:19 2013 -0400 +++ b/.hgtags Fri Apr 05 16:16:18 2013 -0700 @@ -328,3 +328,5 @@ e3a41fc0234895eba4f272b984f7dacff495f8eb hs25-b24 1c8db54ee9f315e20d6d5d9bf0b5c10349e9d301 jdk8-b83 8d0f263a370c5f3e61791bb06054560804117288 hs25-b25 +af788b85010ebabbc1e8f52c6766e08c7a95cf99 jdk8-b84 +a947f40fb536e5b9e0aa210cf26abb430f80887a hs25-b26 diff -r 4c8bb5e4f68f -r 46d24f112c27 make/bsd/build.sh --- a/make/bsd/build.sh Fri Apr 05 12:19:19 2013 -0400 +++ /dev/null Thu Jan 01 00:00:00 1970 +0000 @@ -1,95 +0,0 @@ -#! /bin/sh -# -# Copyright (c) 1999, 2008, 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. -# -# - -# Make sure the variable JAVA_HOME is set before running this script. - -set -u - - -if [ $# != 2 ]; then - echo "Usage : $0 Build_Options Location" - echo "Build Options : debug or optimized or basicdebug or basic or clean" - echo "Location : specify any workspace which has gamma sources" - exit 1 -fi - -# Just in case: -case ${JAVA_HOME} in -/*) true;; -?*) JAVA_HOME=`( cd $JAVA_HOME; pwd )`;; -esac - -case `uname -m` in - i386|i486|i586|i686) - mach=i386 - ;; - *) - echo "Unsupported machine: " `uname -m` - exit 1 - ;; -esac - -if [ "${JAVA_HOME}" = "" -o ! -d "${JAVA_HOME}" -o ! -d ${JAVA_HOME}/jre/lib/${mach} ]; then - echo "JAVA_HOME needs to be set to a valid JDK path" - echo "ksh : export JAVA_HOME=/net/tetrasparc/export/gobi/JDK1.2_fcs_V/bsd" - echo "csh : setenv JAVA_HOME /net/tetrasparc/export/gobi/JDK1.2_fcs_V/bsd" - exit 1 -fi - - -LD_LIBRARY_PATH=${JAVA_HOME}/jre/lib/`uname -p`:\ -${JAVA_HOME}/jre/lib/`uname -p`/native_threads:${LD_LIBRARY_PATH-.} - -# This is necessary as long as we are using the old launcher -# with the new distribution format: -CLASSPATH=${JAVA_HOME}/jre/lib/rt.jar:${CLASSPATH-.} - - -for gm in gmake gnumake -do - if [ "${GNUMAKE-}" != "" ]; then break; fi - ($gm --version >/dev/null) 2>/dev/null && GNUMAKE=$gm -done -: ${GNUMAKE:?'Cannot locate the gnumake program. Stop.'} - - -echo "### ENVIRONMENT SETTINGS:" -export JAVA_HOME ; echo "JAVA_HOME=$JAVA_HOME" -export LD_LIBRARY_PATH ; echo "LD_LIBRARY_PATH=$LD_LIBRARY_PATH" -export CLASSPATH ; echo "CLASSPATH=$CLASSPATH" -export GNUMAKE ; echo "GNUMAKE=$GNUMAKE" -echo "###" - -Build_Options=$1 -Location=$2 - -case ${Location} in -/*) true;; -?*) Location=`(cd ${Location}; pwd)`;; -esac - -echo \ -${GNUMAKE} -f ${Location}/make/bsd/Makefile $Build_Options GAMMADIR=${Location} -${GNUMAKE} -f ${Location}/make/bsd/Makefile $Build_Options GAMMADIR=${Location} diff -r 4c8bb5e4f68f -r 46d24f112c27 make/bsd/makefiles/buildtree.make --- a/make/bsd/makefiles/buildtree.make Fri Apr 05 12:19:19 2013 -0400 +++ b/make/bsd/makefiles/buildtree.make Fri Apr 05 16:16:18 2013 -0700 @@ -1,5 +1,5 @@ # -# Copyright (c) 2005, 2012, Oracle and/or its affiliates. All rights reserved. +# Copyright (c) 2005, 2013, 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 @@ -190,6 +190,17 @@ # literal "$(GAMMADIR)/" suitable for inclusion in a Makefile. gamma-path=$(subst $(GAMMADIR),\$$(GAMMADIR),$(call $(1),$(HS_COMMON_SRC)/$(2))) +# This bit is needed to enable local rebuilds. +# Unless the makefile itself sets LP64, any environmental +# setting of LP64 will interfere with the build. +LP64_SETTING/32 = LP64 = \#empty +LP64_SETTING/64 = LP64 = 1 + +DATA_MODE/i486 = 32 +DATA_MODE/amd64 = 64 + +DATA_MODE = $(DATA_MODE/$(BUILDARCH)) + flags.make: $(BUILDTREE_MAKE) ../shared_dirs.lst @echo Creating $@ ... $(QUIETLY) ( \ @@ -212,6 +223,7 @@ echo "HOTSPOT_BUILD_USER = $(HOTSPOT_BUILD_USER)"; \ echo "HOTSPOT_VM_DISTRO = $(HOTSPOT_VM_DISTRO)"; \ echo "OPENJDK = $(OPENJDK)"; \ + echo "$(LP64_SETTING/$(DATA_MODE))"; \ echo; \ echo "# Used for platform dispatching"; \ echo "TARGET_DEFINES = -DTARGET_OS_FAMILY_\$$(Platform_os_family)"; \ diff -r 4c8bb5e4f68f -r 46d24f112c27 make/build.sh --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/make/build.sh Fri Apr 05 16:16:18 2013 -0700 @@ -0,0 +1,92 @@ +#! /bin/sh +# +# Copyright (c) 1998, 2013, 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. +# +# + +# Make sure the variable JAVA_HOME is set before running this script. + +set -u + + +if [ $# -lt 1 ]; then + echo "Usage : $0 BuildTarget [LP64=1] [BuildOptions]" + echo " Server VM | Client VM" + echo "BuildTarget : debug | debug1" + echo " fastdebug | fastdebug1" + echo " jvmg | jvmg1" + echo " optimized | optimized1" + echo " profiled | profiled1" + echo " product | product1" + exit 1 +fi + +if [ "${JAVA_HOME-}" = "" -o ! -d "${JAVA_HOME-}" -o ! -d ${JAVA_HOME-}/jre/lib/ ]; then + echo "JAVA_HOME needs to be set to a valid JDK path" + echo "JAVA_HOME: ${JAVA_HOME-}" + exit 1 +fi + +# Just in case: +JAVA_HOME=`( cd $JAVA_HOME; pwd )` + +if [ "${ALT_BOOTDIR-}" = "" -o ! -d "${ALT_BOOTDIR-}" -o ! -d ${ALT_BOOTDIR-}/jre/lib/ ]; then + ALT_BOOTDIR=${JAVA_HOME} +fi + +# build in current directory by default +if [ "${ALT_OUTPUTDIR-}" = "" -o ! -d "${ALT_OUTPUTDIR-}" ]; then + ALT_OUTPUTDIR=`(pwd)` +fi + +HOTSPOT_SRC=`(dirname $0)`/.. +HOTSPOT_SRC=`(cd ${HOTSPOT_SRC}; pwd)` + +for gm in gmake gnumake +do + if [ "${GNUMAKE-}" != "" ]; then break; fi + ($gm --version >/dev/null) 2>/dev/null && GNUMAKE=$gm +done +: ${GNUMAKE:?'Cannot locate the gnumake program. Stop.'} + +# quiet build by default +Quiet="MAKE_VERBOSE=" + +# no debug info by default +NoDebugInfo="ENABLE_FULL_DEBUG_SYMBOLS=" + +LANG=C + +echo "### ENVIRONMENT SETTINGS:" +export HOTSPOT_SRC ; echo "HOTSPOT_SRC=$HOTSPOT_SRC" +export JAVA_HOME ; echo "JAVA_HOME=$JAVA_HOME" +export ALT_BOOTDIR ; echo "ALT_BOOTDIR=$ALT_BOOTDIR" +export ALT_OUTPUTDIR ; echo "ALT_OUTPUTDIR=$ALT_OUTPUTDIR" +export GNUMAKE ; echo "GNUMAKE=$GNUMAKE" +export LANG ; echo "LANG=$LANG" +echo "###" + +BuildOptions="$Quiet $NoDebugInfo $*" + +echo \ +${GNUMAKE} -f ${HOTSPOT_SRC}/make/Makefile $BuildOptions GAMMADIR=${HOTSPOT_SRC} +${GNUMAKE} -f ${HOTSPOT_SRC}/make/Makefile $BuildOptions GAMMADIR=${HOTSPOT_SRC} diff -r 4c8bb5e4f68f -r 46d24f112c27 make/hotspot_version --- a/make/hotspot_version Fri Apr 05 12:19:19 2013 -0400 +++ b/make/hotspot_version Fri Apr 05 16:16:18 2013 -0700 @@ -35,7 +35,7 @@ HS_MAJOR_VER=25 HS_MINOR_VER=0 -HS_BUILD_NUMBER=26 +HS_BUILD_NUMBER=27 JDK_MAJOR_VER=1 JDK_MINOR_VER=8 diff -r 4c8bb5e4f68f -r 46d24f112c27 make/linux/build.sh --- a/make/linux/build.sh Fri Apr 05 12:19:19 2013 -0400 +++ /dev/null Thu Jan 01 00:00:00 1970 +0000 @@ -1,98 +0,0 @@ -#! /bin/sh -# -# Copyright (c) 1999, 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. -# -# - -# Make sure the variable JAVA_HOME is set before running this script. - -set -u - - -if [ $# != 2 ]; then - echo "Usage : $0 Build_Options Location" - echo "Build Options : debug or optimized or basicdebug or basic or clean" - echo "Location : specify any workspace which has gamma sources" - exit 1 -fi - -# Just in case: -case ${JAVA_HOME} in -/*) true;; -?*) JAVA_HOME=`( cd $JAVA_HOME; pwd )`;; -esac - -case `uname -m` in - i386|i486|i586|i686) - mach=i386 - ;; - x86_64) - mach=amd64 - ;; - *) - echo "Unsupported machine: " `uname -m` - exit 1 - ;; -esac - -if [ "${JAVA_HOME}" = "" -o ! -d "${JAVA_HOME}" -o ! -d ${JAVA_HOME}/jre/lib/${mach} ]; then - echo "JAVA_HOME needs to be set to a valid JDK path" - echo "ksh : export JAVA_HOME=/net/tetrasparc/export/gobi/JDK1.2_fcs_V/linux" - echo "csh : setenv JAVA_HOME /net/tetrasparc/export/gobi/JDK1.2_fcs_V/linux" - exit 1 -fi - - -LD_LIBRARY_PATH=${JAVA_HOME}/jre/lib/`uname -p`:\ -${JAVA_HOME}/jre/lib/`uname -p`/native_threads:${LD_LIBRARY_PATH-.} - -# This is necessary as long as we are using the old launcher -# with the new distribution format: -CLASSPATH=${JAVA_HOME}/jre/lib/rt.jar:${CLASSPATH-.} - - -for gm in gmake gnumake -do - if [ "${GNUMAKE-}" != "" ]; then break; fi - ($gm --version >/dev/null) 2>/dev/null && GNUMAKE=$gm -done -: ${GNUMAKE:?'Cannot locate the gnumake program. Stop.'} - - -echo "### ENVIRONMENT SETTINGS:" -export JAVA_HOME ; echo "JAVA_HOME=$JAVA_HOME" -export LD_LIBRARY_PATH ; echo "LD_LIBRARY_PATH=$LD_LIBRARY_PATH" -export CLASSPATH ; echo "CLASSPATH=$CLASSPATH" -export GNUMAKE ; echo "GNUMAKE=$GNUMAKE" -echo "###" - -Build_Options=$1 -Location=$2 - -case ${Location} in -/*) true;; -?*) Location=`(cd ${Location}; pwd)`;; -esac - -echo \ -${GNUMAKE} -f ${Location}/make/linux/Makefile $Build_Options GAMMADIR=${Location} -${GNUMAKE} -f ${Location}/make/linux/Makefile $Build_Options GAMMADIR=${Location} diff -r 4c8bb5e4f68f -r 46d24f112c27 make/linux/makefiles/buildtree.make --- a/make/linux/makefiles/buildtree.make Fri Apr 05 12:19:19 2013 -0400 +++ b/make/linux/makefiles/buildtree.make Fri Apr 05 16:16:18 2013 -0700 @@ -1,5 +1,5 @@ # -# Copyright (c) 2005, 2012, Oracle and/or its affiliates. All rights reserved. +# Copyright (c) 2005, 2013, 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 @@ -183,6 +183,19 @@ # literal "$(GAMMADIR)/" suitable for inclusion in a Makefile. gamma-path=$(subst $(GAMMADIR),\$$(GAMMADIR),$(call $(1),$(HS_COMMON_SRC)/$(2))) +# This bit is needed to enable local rebuilds. +# Unless the makefile itself sets LP64, any environmental +# setting of LP64 will interfere with the build. +LP64_SETTING/32 = LP64 = \#empty +LP64_SETTING/64 = LP64 = 1 + +DATA_MODE/i486 = 32 +DATA_MODE/sparc = 32 +DATA_MODE/sparcv9 = 64 +DATA_MODE/amd64 = 64 + +DATA_MODE = $(DATA_MODE/$(BUILDARCH)) + flags.make: $(BUILDTREE_MAKE) ../shared_dirs.lst @echo Creating $@ ... $(QUIETLY) ( \ @@ -205,6 +218,7 @@ echo "HOTSPOT_BUILD_USER = $(HOTSPOT_BUILD_USER)"; \ echo "HOTSPOT_VM_DISTRO = $(HOTSPOT_VM_DISTRO)"; \ echo "OPENJDK = $(OPENJDK)"; \ + echo "$(LP64_SETTING/$(DATA_MODE))"; \ echo; \ echo "# Used for platform dispatching"; \ echo "TARGET_DEFINES = -DTARGET_OS_FAMILY_\$$(Platform_os_family)"; \ diff -r 4c8bb5e4f68f -r 46d24f112c27 make/solaris/build.sh --- a/make/solaris/build.sh Fri Apr 05 12:19:19 2013 -0400 +++ /dev/null Thu Jan 01 00:00:00 1970 +0000 @@ -1,127 +0,0 @@ -#! /bin/sh -# -# Copyright (c) 1998, 2008, 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. -# -# - -# Make sure the variable JAVA_HOME is set before running this script. - -set -u - - -usage() { - ( - echo "Usage : $0 [-sb | -sbfast] config ws_path" - echo "" - echo "Where:" - echo " -sb ::= enable source browser info generation for" - echo " all configs during compilation" - echo "" - echo " -sbfast ::= enable source browser info generation for" - echo " all configs without compilation" - echo "" - echo " config ::= debug | debug1 | debugcore" - echo " fastdebug | fastdebug1 | fastdebugcore" - echo " jvmg | jvmg1 | jvmgcore" - echo " optimized | optimized1 | optimizedcore" - echo " profiled | profiled1 | profiledcore" - echo " product | product1 | productcore" - echo "" - echo " ws_path ::= path to HotSpot workspace" - ) >&2 - exit 1 -} - -# extract possible options -options="" -if [ $# -gt 2 ]; then - case "$1" in - -sb) - options="CFLAGS_BROWSE=-xsb" - shift - ;; - -sbfast) - options="CFLAGS_BROWSE=-xsbfast" - shift - ;; - *) - echo "Unknown option: '$1'" >&2 - usage - ;; - esac -fi - -# should be just two args left at this point -if [ $# != 2 ]; then - usage -fi - -# Just in case: -case ${JAVA_HOME} in -/*) true;; -?*) JAVA_HOME=`( cd $JAVA_HOME; pwd )`;; -esac - -if [ "${JAVA_HOME}" = "" -o ! -d "${JAVA_HOME}" -o ! -d ${JAVA_HOME}/jre/lib/`uname -p` ]; then - echo "JAVA_HOME needs to be set to a valid JDK path" - echo "ksh : export JAVA_HOME=/net/tetrasparc/export/gobi/JDK1.2_fcs_V/solaris" - echo "csh : setenv JAVA_HOME /net/tetrasparc/export/gobi/JDK1.2_fcs_V/solaris" - exit 1 -fi - - -LD_LIBRARY_PATH=${JAVA_HOME}/jre/lib/`uname -p`:\ -${JAVA_HOME}/jre/lib/`uname -p`/native_threads:${LD_LIBRARY_PATH-.} - -# This is necessary as long as we are using the old launcher -# with the new distribution format: -CLASSPATH=${JAVA_HOME}/jre/lib/rt.jar:${CLASSPATH-.} - - -for gm in gmake gnumake -do - if [ "${GNUMAKE-}" != "" ]; then break; fi - ($gm --version >/dev/null) 2>/dev/null && GNUMAKE=$gm -done -: ${GNUMAKE:?'Cannot locate the gnumake program. Stop.'} - - -echo "### ENVIRONMENT SETTINGS:" -export JAVA_HOME ; echo "JAVA_HOME=$JAVA_HOME" -export LD_LIBRARY_PATH ; echo "LD_LIBRARY_PATH=$LD_LIBRARY_PATH" -export CLASSPATH ; echo "CLASSPATH=$CLASSPATH" -export GNUMAKE ; echo "GNUMAKE=$GNUMAKE" -echo "###" - -config=$1 -ws_path=$2 - -case ${ws_path} in -/*) true;; -?*) ws_path=`(cd ${ws_path}; pwd)`;; -esac - -echo \ -${GNUMAKE} -f ${ws_path}/make/solaris/Makefile \ - $config GAMMADIR=${ws_path} $options -${GNUMAKE} -f ${ws_path}/make/solaris/Makefile \ - $config GAMMADIR=${ws_path} $options diff -r 4c8bb5e4f68f -r 46d24f112c27 make/solaris/makefiles/buildtree.make --- a/make/solaris/makefiles/buildtree.make Fri Apr 05 12:19:19 2013 -0400 +++ b/make/solaris/makefiles/buildtree.make Fri Apr 05 16:16:18 2013 -0700 @@ -1,5 +1,5 @@ # -# Copyright (c) 2000, 2012, Oracle and/or its affiliates. All rights reserved. +# Copyright (c) 2000, 2013, Oracle and/or its affiliates. All rights reserved. # DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. # # This code is free software; you can redistribute it and/or modify it @@ -176,6 +176,19 @@ # literal "$(GAMMADIR)/" suitable for inclusion in a Makefile. gamma-path=$(subst $(GAMMADIR),\$$(GAMMADIR),$(call $(1),$(HS_COMMON_SRC)/$(2))) +# This bit is needed to enable local rebuilds. +# Unless the makefile itself sets LP64, any environmental +# setting of LP64 will interfere with the build. +LP64_SETTING/32 = LP64 = \#empty +LP64_SETTING/64 = LP64 = 1 + +DATA_MODE/i486 = 32 +DATA_MODE/sparc = 32 +DATA_MODE/sparcv9 = 64 +DATA_MODE/amd64 = 64 + +DATA_MODE = $(DATA_MODE/$(BUILDARCH)) + flags.make: $(BUILDTREE_MAKE) ../shared_dirs.lst @echo Creating $@ ... $(QUIETLY) ( \ diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/sparc/vm/c1_CodeStubs_sparc.cpp --- a/src/cpu/sparc/vm/c1_CodeStubs_sparc.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/sparc/vm/c1_CodeStubs_sparc.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -51,6 +51,16 @@ void RangeCheckStub::emit_code(LIR_Assembler* ce) { __ bind(_entry); + if (_info->deoptimize_on_exception()) { + address a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id); + __ call(a, relocInfo::runtime_call_type); + __ delayed()->nop(); + ce->add_call_info_here(_info); + ce->verify_oop_map(_info); + debug_only(__ should_not_reach_here()); + return; + } + if (_index->is_register()) { __ mov(_index->as_register(), G4); } else { @@ -64,11 +74,22 @@ __ delayed()->nop(); ce->add_call_info_here(_info); ce->verify_oop_map(_info); -#ifdef ASSERT - __ should_not_reach_here(); -#endif + debug_only(__ should_not_reach_here()); +} + +PredicateFailedStub::PredicateFailedStub(CodeEmitInfo* info) { + _info = new CodeEmitInfo(info); } +void PredicateFailedStub::emit_code(LIR_Assembler* ce) { + __ bind(_entry); + address a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id); + __ call(a, relocInfo::runtime_call_type); + __ delayed()->nop(); + ce->add_call_info_here(_info); + ce->verify_oop_map(_info); + debug_only(__ should_not_reach_here()); +} void CounterOverflowStub::emit_code(LIR_Assembler* ce) { __ bind(_entry); @@ -99,10 +120,17 @@ void ImplicitNullCheckStub::emit_code(LIR_Assembler* ce) { + address a; + if (_info->deoptimize_on_exception()) { + // Deoptimize, do not throw the exception, because it is probably wrong to do it here. + a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id); + } else { + a = Runtime1::entry_for(Runtime1::throw_null_pointer_exception_id); + } + ce->compilation()->implicit_exception_table()->append(_offset, __ offset()); __ bind(_entry); - __ call(Runtime1::entry_for(Runtime1::throw_null_pointer_exception_id), - relocInfo::runtime_call_type); + __ call(a, relocInfo::runtime_call_type); __ delayed()->nop(); ce->add_call_info_here(_info); ce->verify_oop_map(_info); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/sparc/vm/c1_LIRAssembler_sparc.cpp --- a/src/cpu/sparc/vm/c1_LIRAssembler_sparc.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/sparc/vm/c1_LIRAssembler_sparc.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -3361,6 +3361,45 @@ __ mov(G2_thread, result_reg->as_register()); } +#ifdef ASSERT +// emit run-time assertion +void LIR_Assembler::emit_assert(LIR_OpAssert* op) { + assert(op->code() == lir_assert, "must be"); + + if (op->in_opr1()->is_valid()) { + assert(op->in_opr2()->is_valid(), "both operands must be valid"); + comp_op(op->condition(), op->in_opr1(), op->in_opr2(), op); + } else { + assert(op->in_opr2()->is_illegal(), "both operands must be illegal"); + assert(op->condition() == lir_cond_always, "no other conditions allowed"); + } + + Label ok; + if (op->condition() != lir_cond_always) { + Assembler::Condition acond; + switch (op->condition()) { + case lir_cond_equal: acond = Assembler::equal; break; + case lir_cond_notEqual: acond = Assembler::notEqual; break; + case lir_cond_less: acond = Assembler::less; break; + case lir_cond_lessEqual: acond = Assembler::lessEqual; break; + case lir_cond_greaterEqual: acond = Assembler::greaterEqual; break; + case lir_cond_greater: acond = Assembler::greater; break; + case lir_cond_aboveEqual: acond = Assembler::greaterEqualUnsigned; break; + case lir_cond_belowEqual: acond = Assembler::lessEqualUnsigned; break; + default: ShouldNotReachHere(); + }; + __ br(acond, false, Assembler::pt, ok); + __ delayed()->nop(); + } + if (op->halt()) { + const char* str = __ code_string(op->msg()); + __ stop(str); + } else { + breakpoint(); + } + __ bind(ok); +} +#endif void LIR_Assembler::peephole(LIR_List* lir) { LIR_OpList* inst = lir->instructions_list(); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/sparc/vm/c1_LIRGenerator_sparc.cpp --- a/src/cpu/sparc/vm/c1_LIRGenerator_sparc.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/sparc/vm/c1_LIRGenerator_sparc.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -324,7 +324,7 @@ void LIRGenerator::do_StoreIndexed(StoreIndexed* x) { assert(x->is_pinned(),""); - bool needs_range_check = true; + bool needs_range_check = x->compute_needs_range_check(); bool use_length = x->length() != NULL; bool obj_store = x->elt_type() == T_ARRAY || x->elt_type() == T_OBJECT; bool needs_store_check = obj_store && (x->value()->as_Constant() == NULL || @@ -339,12 +339,9 @@ array.load_item(); index.load_nonconstant(); - if (use_length) { - needs_range_check = x->compute_needs_range_check(); - if (needs_range_check) { - length.set_instruction(x->length()); - length.load_item(); - } + if (use_length && needs_range_check) { + length.set_instruction(x->length()); + length.load_item(); } if (needs_store_check) { value.load_item(); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/sparc/vm/c1_Runtime1_sparc.cpp --- a/src/cpu/sparc/vm/c1_Runtime1_sparc.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/sparc/vm/c1_Runtime1_sparc.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -987,6 +987,25 @@ break; #endif // INCLUDE_ALL_GCS + case predicate_failed_trap_id: + { + __ set_info("predicate_failed_trap", dont_gc_arguments); + OopMap* oop_map = save_live_registers(sasm); + + int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, predicate_failed_trap)); + + oop_maps = new OopMapSet(); + oop_maps->add_gc_map(call_offset, oop_map); + + DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob(); + assert(deopt_blob != NULL, "deoptimization blob must have been created"); + restore_live_registers(sasm); + __ restore(); + __ br(Assembler::always, false, Assembler::pt, deopt_blob->unpack_with_reexecution(), relocInfo::runtime_call_type); + __ delayed()->nop(); + } + break; + default: { __ set_info("unimplemented entry", dont_gc_arguments); __ save_frame(0); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/x86/vm/c1_CodeStubs_x86.cpp --- a/src/cpu/x86/vm/c1_CodeStubs_x86.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/x86/vm/c1_CodeStubs_x86.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -101,6 +101,15 @@ void RangeCheckStub::emit_code(LIR_Assembler* ce) { __ bind(_entry); + if (_info->deoptimize_on_exception()) { + address a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id); + __ call(RuntimeAddress(a)); + ce->add_call_info_here(_info); + ce->verify_oop_map(_info); + debug_only(__ should_not_reach_here()); + return; + } + // pass the array index on stack because all registers must be preserved if (_index->is_cpu_register()) { ce->store_parameter(_index->as_register(), 0); @@ -115,9 +124,22 @@ } __ call(RuntimeAddress(Runtime1::entry_for(stub_id))); ce->add_call_info_here(_info); + ce->verify_oop_map(_info); debug_only(__ should_not_reach_here()); } +PredicateFailedStub::PredicateFailedStub(CodeEmitInfo* info) { + _info = new CodeEmitInfo(info); +} + +void PredicateFailedStub::emit_code(LIR_Assembler* ce) { + __ bind(_entry); + address a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id); + __ call(RuntimeAddress(a)); + ce->add_call_info_here(_info); + ce->verify_oop_map(_info); + debug_only(__ should_not_reach_here()); +} void DivByZeroStub::emit_code(LIR_Assembler* ce) { if (_offset != -1) { @@ -414,10 +436,19 @@ void ImplicitNullCheckStub::emit_code(LIR_Assembler* ce) { + address a; + if (_info->deoptimize_on_exception()) { + // Deoptimize, do not throw the exception, because it is probably wrong to do it here. + a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id); + } else { + a = Runtime1::entry_for(Runtime1::throw_null_pointer_exception_id); + } + ce->compilation()->implicit_exception_table()->append(_offset, __ offset()); __ bind(_entry); - __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::throw_null_pointer_exception_id))); + __ call(RuntimeAddress(a)); ce->add_call_info_here(_info); + ce->verify_oop_map(_info); debug_only(__ should_not_reach_here()); } diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/x86/vm/c1_LIRAssembler_x86.cpp --- a/src/cpu/x86/vm/c1_LIRAssembler_x86.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/x86/vm/c1_LIRAssembler_x86.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -3755,6 +3755,44 @@ } } +#ifdef ASSERT +// emit run-time assertion +void LIR_Assembler::emit_assert(LIR_OpAssert* op) { + assert(op->code() == lir_assert, "must be"); + + if (op->in_opr1()->is_valid()) { + assert(op->in_opr2()->is_valid(), "both operands must be valid"); + comp_op(op->condition(), op->in_opr1(), op->in_opr2(), op); + } else { + assert(op->in_opr2()->is_illegal(), "both operands must be illegal"); + assert(op->condition() == lir_cond_always, "no other conditions allowed"); + } + + Label ok; + if (op->condition() != lir_cond_always) { + Assembler::Condition acond = Assembler::zero; + switch (op->condition()) { + case lir_cond_equal: acond = Assembler::equal; break; + case lir_cond_notEqual: acond = Assembler::notEqual; break; + case lir_cond_less: acond = Assembler::less; break; + case lir_cond_lessEqual: acond = Assembler::lessEqual; break; + case lir_cond_greaterEqual: acond = Assembler::greaterEqual;break; + case lir_cond_greater: acond = Assembler::greater; break; + case lir_cond_belowEqual: acond = Assembler::belowEqual; break; + case lir_cond_aboveEqual: acond = Assembler::aboveEqual; break; + default: ShouldNotReachHere(); + } + __ jcc(acond, ok); + } + if (op->halt()) { + const char* str = __ code_string(op->msg()); + __ stop(str); + } else { + breakpoint(); + } + __ bind(ok); +} +#endif void LIR_Assembler::membar() { // QQQ sparc TSO uses this, diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/x86/vm/c1_LIRGenerator_x86.cpp --- a/src/cpu/x86/vm/c1_LIRGenerator_x86.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/x86/vm/c1_LIRGenerator_x86.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -263,7 +263,7 @@ void LIRGenerator::do_StoreIndexed(StoreIndexed* x) { assert(x->is_pinned(),""); - bool needs_range_check = true; + bool needs_range_check = x->compute_needs_range_check(); bool use_length = x->length() != NULL; bool obj_store = x->elt_type() == T_ARRAY || x->elt_type() == T_OBJECT; bool needs_store_check = obj_store && (x->value()->as_Constant() == NULL || @@ -278,12 +278,10 @@ array.load_item(); index.load_nonconstant(); - if (use_length) { - needs_range_check = x->compute_needs_range_check(); - if (needs_range_check) { - length.set_instruction(x->length()); - length.load_item(); - } + if (use_length && needs_range_check) { + length.set_instruction(x->length()); + length.load_item(); + } if (needs_store_check) { value.load_item(); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/x86/vm/c1_LinearScan_x86.cpp --- a/src/cpu/x86/vm/c1_LinearScan_x86.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/x86/vm/c1_LinearScan_x86.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -675,7 +675,8 @@ switch (op2->code()) { case lir_cmp: case lir_cmp_fd2i: - case lir_ucmp_fd2i: { + case lir_ucmp_fd2i: + case lir_assert: { assert(left->is_fpu_register(), "invalid LIR"); assert(right->is_fpu_register(), "invalid LIR"); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/x86/vm/c1_Runtime1_x86.cpp --- a/src/cpu/x86/vm/c1_Runtime1_x86.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/x86/vm/c1_Runtime1_x86.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -1807,6 +1807,24 @@ break; #endif // INCLUDE_ALL_GCS + case predicate_failed_trap_id: + { + StubFrame f(sasm, "predicate_failed_trap", dont_gc_arguments); + + OopMap* map = save_live_registers(sasm, 1); + + int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, predicate_failed_trap)); + oop_maps = new OopMapSet(); + oop_maps->add_gc_map(call_offset, map); + restore_live_registers(sasm); + __ leave(); + DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob(); + assert(deopt_blob != NULL, "deoptimization blob must have been created"); + + __ jump(RuntimeAddress(deopt_blob->unpack_with_reexecution())); + } + break; + default: { StubFrame f(sasm, "unimplemented entry", dont_gc_arguments); __ movptr(rax, (int)id); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/x86/vm/cppInterpreter_x86.cpp --- a/src/cpu/x86/vm/cppInterpreter_x86.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/x86/vm/cppInterpreter_x86.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -1299,25 +1299,8 @@ __ push(rdx); #endif // _LP64 - // Either restore the MXCSR register after returning from the JNI Call - // or verify that it wasn't changed. - if (VM_Version::supports_sse()) { - if (RestoreMXCSROnJNICalls) { - __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std())); - } - else if (CheckJNICalls ) { - __ call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry())); - } - } - -#ifndef _LP64 - // Either restore the x87 floating pointer control word after returning - // from the JNI call or verify that it wasn't changed. - if (CheckJNICalls) { - __ call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry())); - } -#endif // _LP64 - + // Verify or restore cpu control state after JNI call + __ restore_cpu_control_state_after_jni(); // change thread state __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/x86/vm/macroAssembler_x86.cpp --- a/src/cpu/x86/vm/macroAssembler_x86.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/x86/vm/macroAssembler_x86.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -4765,6 +4765,31 @@ pop_CPU_state(); } +void MacroAssembler::restore_cpu_control_state_after_jni() { + // Either restore the MXCSR register after returning from the JNI Call + // or verify that it wasn't changed (with -Xcheck:jni flag). + if (VM_Version::supports_sse()) { + if (RestoreMXCSROnJNICalls) { + ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std())); + } else if (CheckJNICalls) { + call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry())); + } + } + if (VM_Version::supports_avx()) { + // Clear upper bits of YMM registers to avoid SSE <-> AVX transition penalty. + vzeroupper(); + } + +#ifndef _LP64 + // Either restore the x87 floating pointer control word after returning + // from the JNI call or verify that it wasn't changed. + if (CheckJNICalls) { + call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry())); + } +#endif // _LP64 +} + + void MacroAssembler::load_klass(Register dst, Register src) { #ifdef _LP64 if (UseCompressedKlassPointers) { @@ -5759,6 +5784,8 @@ addptr(result, stride2); subl(cnt2, stride2); jccb(Assembler::notZero, COMPARE_WIDE_VECTORS_LOOP); + // clean upper bits of YMM registers + vzeroupper(); // compare wide vectors tail bind(COMPARE_WIDE_TAIL); @@ -5772,6 +5799,8 @@ // Identifies the mismatching (higher or lower)16-bytes in the 32-byte vectors. bind(VECTOR_NOT_EQUAL); + // clean upper bits of YMM registers + vzeroupper(); lea(str1, Address(str1, result, scale)); lea(str2, Address(str2, result, scale)); jmp(COMPARE_16_CHARS); @@ -6028,6 +6057,10 @@ // That's it bind(DONE); + if (UseAVX >= 2) { + // clean upper bits of YMM registers + vzeroupper(); + } } void MacroAssembler::generate_fill(BasicType t, bool aligned, @@ -6157,6 +6190,10 @@ vmovdqu(Address(to, 0), xtmp); addptr(to, 32); subl(count, 8 << shift); + + BIND(L_check_fill_8_bytes); + // clean upper bits of YMM registers + vzeroupper(); } else { // Fill 32-byte chunks pshufd(xtmp, xtmp, 0); @@ -6180,8 +6217,9 @@ addptr(to, 32); subl(count, 8 << shift); jcc(Assembler::greaterEqual, L_fill_32_bytes_loop); + + BIND(L_check_fill_8_bytes); } - BIND(L_check_fill_8_bytes); addl(count, 8 << shift); jccb(Assembler::zero, L_exit); jmpb(L_fill_8_bytes); @@ -6316,6 +6354,10 @@ jccb(Assembler::lessEqual, L_copy_16_chars); bind(L_copy_16_chars_exit); + if (UseAVX >= 2) { + // clean upper bits of YMM registers + vzeroupper(); + } subptr(len, 8); jccb(Assembler::greater, L_copy_8_chars_exit); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/x86/vm/macroAssembler_x86.hpp --- a/src/cpu/x86/vm/macroAssembler_x86.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/x86/vm/macroAssembler_x86.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -582,6 +582,9 @@ // only if +VerifyFPU void verify_FPU(int stack_depth, const char* s = "illegal FPU state"); + // Verify or restore cpu control state after JNI call + void restore_cpu_control_state_after_jni(); + // prints msg, dumps registers and stops execution void stop(const char* msg); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/x86/vm/sharedRuntime_x86_32.cpp --- a/src/cpu/x86/vm/sharedRuntime_x86_32.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/x86/vm/sharedRuntime_x86_32.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -2065,6 +2065,9 @@ __ call(RuntimeAddress(native_func)); + // Verify or restore cpu control state after JNI call + __ restore_cpu_control_state_after_jni(); + // WARNING - on Windows Java Natives use pascal calling convention and pop the // arguments off of the stack. We could just re-adjust the stack pointer here // and continue to do SP relative addressing but we instead switch to FP diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/x86/vm/sharedRuntime_x86_64.cpp --- a/src/cpu/x86/vm/sharedRuntime_x86_64.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/x86/vm/sharedRuntime_x86_64.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -2315,16 +2315,8 @@ __ call(RuntimeAddress(native_func)); - // Either restore the MXCSR register after returning from the JNI Call - // or verify that it wasn't changed. - if (RestoreMXCSROnJNICalls) { - __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std())); - - } - else if (CheckJNICalls ) { - __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::verify_mxcsr_entry()))); - } - + // Verify or restore cpu control state after JNI call + __ restore_cpu_control_state_after_jni(); // Unpack native results. switch (ret_type) { diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/x86/vm/stubGenerator_x86_32.cpp --- a/src/cpu/x86/vm/stubGenerator_x86_32.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/x86/vm/stubGenerator_x86_32.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -835,6 +835,11 @@ __ BIND(L_copy_64_bytes); __ subl(qword_count, 8); __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop); + + if (UseUnalignedLoadStores && (UseAVX >= 2)) { + // clean upper bits of YMM registers + __ vzeroupper(); + } __ addl(qword_count, 8); __ jccb(Assembler::zero, L_exit); // diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/x86/vm/stubGenerator_x86_64.cpp --- a/src/cpu/x86/vm/stubGenerator_x86_64.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/x86/vm/stubGenerator_x86_64.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -1331,6 +1331,10 @@ } __ addptr(qword_count, 4); __ BIND(L_end); + if (UseAVX >= 2) { + // clean upper bits of YMM registers + __ vzeroupper(); + } } else { // Copy 32-bytes per iteration __ BIND(L_loop); @@ -1404,6 +1408,10 @@ } __ subptr(qword_count, 4); __ BIND(L_end); + if (UseAVX >= 2) { + // clean upper bits of YMM registers + __ vzeroupper(); + } } else { // Copy 32-bytes per iteration __ BIND(L_loop); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/x86/vm/templateInterpreter_x86_32.cpp --- a/src/cpu/x86/vm/templateInterpreter_x86_32.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/x86/vm/templateInterpreter_x86_32.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -1080,22 +1080,8 @@ // result potentially in rdx:rax or ST0 - // Either restore the MXCSR register after returning from the JNI Call - // or verify that it wasn't changed. - if (VM_Version::supports_sse()) { - if (RestoreMXCSROnJNICalls) { - __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std())); - } - else if (CheckJNICalls ) { - __ call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry())); - } - } - - // Either restore the x87 floating pointer control word after returning - // from the JNI call or verify that it wasn't changed. - if (CheckJNICalls) { - __ call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry())); - } + // Verify or restore cpu control state after JNI call + __ restore_cpu_control_state_after_jni(); // save potential result in ST(0) & rdx:rax // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 - diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/x86/vm/templateInterpreter_x86_64.cpp --- a/src/cpu/x86/vm/templateInterpreter_x86_64.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/x86/vm/templateInterpreter_x86_64.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -1079,15 +1079,8 @@ __ call(rax); // result potentially in rax or xmm0 - // Depending on runtime options, either restore the MXCSR - // register after returning from the JNI Call or verify that - // it wasn't changed during -Xcheck:jni. - if (RestoreMXCSROnJNICalls) { - __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std())); - } - else if (CheckJNICalls) { - __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::verify_mxcsr_entry()))); - } + // Verify or restore cpu control state after JNI call + __ restore_cpu_control_state_after_jni(); // NOTE: The order of these pushes is known to frame::interpreter_frame_result // in order to extract the result of a method call. If the order of these diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/x86/vm/x86_32.ad --- a/src/cpu/x86/vm/x86_32.ad Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/x86/vm/x86_32.ad Fri Apr 05 16:16:18 2013 -0700 @@ -228,10 +228,16 @@ static jlong *double_signflip_pool = double_quadword(&fp_signmask_pool[4*2], CONST64(0x8000000000000000), CONST64(0x8000000000000000)); // Offset hacking within calls. -static int pre_call_FPU_size() { - if (Compile::current()->in_24_bit_fp_mode()) - return 6; // fldcw - return 0; +static int pre_call_resets_size() { + int size = 0; + Compile* C = Compile::current(); + if (C->in_24_bit_fp_mode()) { + size += 6; // fldcw + } + if (C->max_vector_size() > 16) { + size += 3; // vzeroupper + } + return size; } static int preserve_SP_size() { @@ -242,21 +248,21 @@ // from the start of the call to the point where the return address // will point. int MachCallStaticJavaNode::ret_addr_offset() { - int offset = 5 + pre_call_FPU_size(); // 5 bytes from start of call to where return address points + int offset = 5 + pre_call_resets_size(); // 5 bytes from start of call to where return address points if (_method_handle_invoke) offset += preserve_SP_size(); return offset; } int MachCallDynamicJavaNode::ret_addr_offset() { - return 10 + pre_call_FPU_size(); // 10 bytes from start of call to where return address points + return 10 + pre_call_resets_size(); // 10 bytes from start of call to where return address points } static int sizeof_FFree_Float_Stack_All = -1; int MachCallRuntimeNode::ret_addr_offset() { assert(sizeof_FFree_Float_Stack_All != -1, "must have been emitted already"); - return sizeof_FFree_Float_Stack_All + 5 + pre_call_FPU_size(); + return sizeof_FFree_Float_Stack_All + 5 + pre_call_resets_size(); } // Indicate if the safepoint node needs the polling page as an input. @@ -272,7 +278,7 @@ // The address of the call instruction needs to be 4-byte aligned to // ensure that it does not span a cache line so that it can be patched. int CallStaticJavaDirectNode::compute_padding(int current_offset) const { - current_offset += pre_call_FPU_size(); // skip fldcw, if any + current_offset += pre_call_resets_size(); // skip fldcw, if any current_offset += 1; // skip call opcode byte return round_to(current_offset, alignment_required()) - current_offset; } @@ -280,7 +286,7 @@ // The address of the call instruction needs to be 4-byte aligned to // ensure that it does not span a cache line so that it can be patched. int CallStaticJavaHandleNode::compute_padding(int current_offset) const { - current_offset += pre_call_FPU_size(); // skip fldcw, if any + current_offset += pre_call_resets_size(); // skip fldcw, if any current_offset += preserve_SP_size(); // skip mov rbp, rsp current_offset += 1; // skip call opcode byte return round_to(current_offset, alignment_required()) - current_offset; @@ -289,7 +295,7 @@ // The address of the call instruction needs to be 4-byte aligned to // ensure that it does not span a cache line so that it can be patched. int CallDynamicJavaDirectNode::compute_padding(int current_offset) const { - current_offset += pre_call_FPU_size(); // skip fldcw, if any + current_offset += pre_call_resets_size(); // skip fldcw, if any current_offset += 5; // skip MOV instruction current_offset += 1; // skip call opcode byte return round_to(current_offset, alignment_required()) - current_offset; @@ -583,16 +589,20 @@ // Remove two words for return addr and rbp, framesize -= 2*wordSize; - if( C->in_24_bit_fp_mode() ) { + if (C->max_vector_size() > 16) { + st->print("VZEROUPPER"); + st->cr(); st->print("\t"); + } + if (C->in_24_bit_fp_mode()) { st->print("FLDCW standard control word"); st->cr(); st->print("\t"); } - if( framesize ) { + if (framesize) { st->print("ADD ESP,%d\t# Destroy frame",framesize); st->cr(); st->print("\t"); } st->print_cr("POPL EBP"); st->print("\t"); - if( do_polling() && C->is_method_compilation() ) { + if (do_polling() && C->is_method_compilation()) { st->print("TEST PollPage,EAX\t! Poll Safepoint"); st->cr(); st->print("\t"); } @@ -602,8 +612,14 @@ void MachEpilogNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const { Compile *C = ra_->C; + if (C->max_vector_size() > 16) { + // Clear upper bits of YMM registers when current compiled code uses + // wide vectors to avoid AVX <-> SSE transition penalty during call. + MacroAssembler masm(&cbuf); + masm.vzeroupper(); + } // If method set FPU control word, restore to standard control word - if( C->in_24_bit_fp_mode() ) { + if (C->in_24_bit_fp_mode()) { MacroAssembler masm(&cbuf); masm.fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); } @@ -615,12 +631,11 @@ // Note that VerifyStackAtCalls' Majik cookie does not change the frame size popped here - if( framesize >= 128 ) { + if (framesize >= 128) { emit_opcode(cbuf, 0x81); // add SP, #framesize emit_rm(cbuf, 0x3, 0x00, ESP_enc); emit_d32(cbuf, framesize); - } - else if( framesize ) { + } else if (framesize) { emit_opcode(cbuf, 0x83); // add SP, #framesize emit_rm(cbuf, 0x3, 0x00, ESP_enc); emit_d8(cbuf, framesize); @@ -628,7 +643,7 @@ emit_opcode(cbuf, 0x58 | EBP_enc); - if( do_polling() && C->is_method_compilation() ) { + if (do_polling() && C->is_method_compilation()) { cbuf.relocate(cbuf.insts_end(), relocInfo::poll_return_type, 0); emit_opcode(cbuf,0x85); emit_rm(cbuf, 0x0, EAX_enc, 0x5); // EAX @@ -640,7 +655,8 @@ Compile *C = ra_->C; // If method set FPU control word, restore to standard control word int size = C->in_24_bit_fp_mode() ? 6 : 0; - if( do_polling() && C->is_method_compilation() ) size += 6; + if (C->max_vector_size() > 16) size += 3; // vzeroupper + if (do_polling() && C->is_method_compilation()) size += 6; int framesize = C->frame_slots() << LogBytesPerInt; assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned"); @@ -649,7 +665,7 @@ size++; // popl rbp, - if( framesize >= 128 ) { + if (framesize >= 128) { size += 6; } else { size += framesize ? 3 : 0; @@ -1853,20 +1869,26 @@ %} - enc_class pre_call_FPU %{ + enc_class pre_call_resets %{ // If method sets FPU control word restore it here debug_only(int off0 = cbuf.insts_size()); - if( Compile::current()->in_24_bit_fp_mode() ) { - MacroAssembler masm(&cbuf); - masm.fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); + if (ra_->C->in_24_bit_fp_mode()) { + MacroAssembler _masm(&cbuf); + __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); + } + if (ra_->C->max_vector_size() > 16) { + // Clear upper bits of YMM registers when current compiled code uses + // wide vectors to avoid AVX <-> SSE transition penalty during call. + MacroAssembler _masm(&cbuf); + __ vzeroupper(); } debug_only(int off1 = cbuf.insts_size()); - assert(off1 - off0 == pre_call_FPU_size(), "correct size prediction"); + assert(off1 - off0 == pre_call_resets_size(), "correct size prediction"); %} enc_class post_call_FPU %{ // If method sets FPU control word do it here also - if( Compile::current()->in_24_bit_fp_mode() ) { + if (Compile::current()->in_24_bit_fp_mode()) { MacroAssembler masm(&cbuf); masm.fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_24())); } @@ -1877,17 +1899,17 @@ // who we intended to call. cbuf.set_insts_mark(); $$$emit8$primary; - if ( !_method ) { + if (!_method) { emit_d32_reloc(cbuf, ($meth$$method - (int)(cbuf.insts_end()) - 4), runtime_call_Relocation::spec(), RELOC_IMM32 ); - } else if(_optimized_virtual) { + } else if (_optimized_virtual) { emit_d32_reloc(cbuf, ($meth$$method - (int)(cbuf.insts_end()) - 4), opt_virtual_call_Relocation::spec(), RELOC_IMM32 ); } else { emit_d32_reloc(cbuf, ($meth$$method - (int)(cbuf.insts_end()) - 4), static_call_Relocation::spec(), RELOC_IMM32 ); } - if( _method ) { // Emit stub for static call + if (_method) { // Emit stub for static call emit_java_to_interp(cbuf); } %} @@ -12828,7 +12850,7 @@ ins_cost(300); format %{ "CALL,static " %} opcode(0xE8); /* E8 cd */ - ins_encode( pre_call_FPU, + ins_encode( pre_call_resets, Java_Static_Call( meth ), call_epilog, post_call_FPU ); @@ -12849,7 +12871,7 @@ ins_cost(300); format %{ "CALL,static/MethodHandle " %} opcode(0xE8); /* E8 cd */ - ins_encode( pre_call_FPU, + ins_encode( pre_call_resets, preserve_SP, Java_Static_Call( meth ), restore_SP, @@ -12870,7 +12892,7 @@ format %{ "MOV EAX,(oop)-1\n\t" "CALL,dynamic" %} opcode(0xE8); /* E8 cd */ - ins_encode( pre_call_FPU, + ins_encode( pre_call_resets, Java_Dynamic_Call( meth ), call_epilog, post_call_FPU ); @@ -12887,7 +12909,7 @@ format %{ "CALL,runtime " %} opcode(0xE8); /* E8 cd */ // Use FFREEs to clear entries in float stack - ins_encode( pre_call_FPU, + ins_encode( pre_call_resets, FFree_Float_Stack_All, Java_To_Runtime( meth ), post_call_FPU ); @@ -12902,7 +12924,7 @@ ins_cost(300); format %{ "CALL_LEAF,runtime " %} opcode(0xE8); /* E8 cd */ - ins_encode( pre_call_FPU, + ins_encode( pre_call_resets, FFree_Float_Stack_All, Java_To_Runtime( meth ), Verify_FPU_For_Leaf, post_call_FPU ); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/cpu/x86/vm/x86_64.ad --- a/src/cpu/x86/vm/x86_64.ad Fri Apr 05 12:19:19 2013 -0400 +++ b/src/cpu/x86/vm/x86_64.ad Fri Apr 05 16:16:18 2013 -0700 @@ -399,6 +399,9 @@ static int preserve_SP_size() { return 3; // rex.w, op, rm(reg/reg) } +static int clear_avx_size() { + return (Compile::current()->max_vector_size() > 16) ? 3 : 0; // vzeroupper +} // !!!!! Special hack to get all types of calls to specify the byte offset // from the start of the call to the point where the return address @@ -406,6 +409,7 @@ int MachCallStaticJavaNode::ret_addr_offset() { int offset = 5; // 5 bytes from start of call to where return address points + offset += clear_avx_size(); if (_method_handle_invoke) offset += preserve_SP_size(); return offset; @@ -413,11 +417,16 @@ int MachCallDynamicJavaNode::ret_addr_offset() { - return 15; // 15 bytes from start of call to where return address points + int offset = 15; // 15 bytes from start of call to where return address points + offset += clear_avx_size(); + return offset; } -// In os_cpu .ad file -// int MachCallRuntimeNode::ret_addr_offset() +int MachCallRuntimeNode::ret_addr_offset() { + int offset = 13; // movq r10,#addr; callq (r10) + offset += clear_avx_size(); + return offset; +} // Indicate if the safepoint node needs the polling page as an input, // it does if the polling page is more than disp32 away. @@ -434,6 +443,7 @@ // ensure that it does not span a cache line so that it can be patched. int CallStaticJavaDirectNode::compute_padding(int current_offset) const { + current_offset += clear_avx_size(); // skip vzeroupper current_offset += 1; // skip call opcode byte return round_to(current_offset, alignment_required()) - current_offset; } @@ -443,6 +453,7 @@ int CallStaticJavaHandleNode::compute_padding(int current_offset) const { current_offset += preserve_SP_size(); // skip mov rbp, rsp + current_offset += clear_avx_size(); // skip vzeroupper current_offset += 1; // skip call opcode byte return round_to(current_offset, alignment_required()) - current_offset; } @@ -451,6 +462,7 @@ // ensure that it does not span a cache line so that it can be patched. int CallDynamicJavaDirectNode::compute_padding(int current_offset) const { + current_offset += clear_avx_size(); // skip vzeroupper current_offset += 11; // skip movq instruction + call opcode byte return round_to(current_offset, alignment_required()) - current_offset; } @@ -764,6 +776,11 @@ void MachEpilogNode::format(PhaseRegAlloc* ra_, outputStream* st) const { Compile* C = ra_->C; + if (C->max_vector_size() > 16) { + st->print("vzeroupper"); + st->cr(); st->print("\t"); + } + int framesize = C->frame_slots() << LogBytesPerInt; assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned"); // Remove word for return adr already pushed @@ -793,6 +810,13 @@ void MachEpilogNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const { Compile* C = ra_->C; + if (C->max_vector_size() > 16) { + // Clear upper bits of YMM registers when current compiled code uses + // wide vectors to avoid AVX <-> SSE transition penalty during call. + MacroAssembler _masm(&cbuf); + __ vzeroupper(); + } + int framesize = C->frame_slots() << LogBytesPerInt; assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned"); // Remove word for return adr already pushed @@ -1997,6 +2021,25 @@ __ bind(miss); %} + enc_class clear_avx %{ + debug_only(int off0 = cbuf.insts_size()); + if (ra_->C->max_vector_size() > 16) { + // Clear upper bits of YMM registers when current compiled code uses + // wide vectors to avoid AVX <-> SSE transition penalty during call. + MacroAssembler _masm(&cbuf); + __ vzeroupper(); + } + debug_only(int off1 = cbuf.insts_size()); + assert(off1 - off0 == clear_avx_size(), "correct size prediction"); + %} + + enc_class Java_To_Runtime(method meth) %{ + // No relocation needed + MacroAssembler _masm(&cbuf); + __ mov64(r10, (int64_t) $meth$$method); + __ call(r10); + %} + enc_class Java_To_Interpreter(method meth) %{ // CALL Java_To_Interpreter @@ -11355,7 +11398,7 @@ ins_cost(300); format %{ "call,static " %} opcode(0xE8); /* E8 cd */ - ins_encode(Java_Static_Call(meth), call_epilog); + ins_encode(clear_avx, Java_Static_Call(meth), call_epilog); ins_pipe(pipe_slow); ins_alignment(4); %} @@ -11373,7 +11416,7 @@ ins_cost(300); format %{ "call,static/MethodHandle " %} opcode(0xE8); /* E8 cd */ - ins_encode(preserve_SP, + ins_encode(clear_avx, preserve_SP, Java_Static_Call(meth), restore_SP, call_epilog); @@ -11392,7 +11435,7 @@ ins_cost(300); format %{ "movq rax, #Universe::non_oop_word()\n\t" "call,dynamic " %} - ins_encode(Java_Dynamic_Call(meth), call_epilog); + ins_encode(clear_avx, Java_Dynamic_Call(meth), call_epilog); ins_pipe(pipe_slow); ins_alignment(4); %} @@ -11405,8 +11448,7 @@ ins_cost(300); format %{ "call,runtime " %} - opcode(0xE8); /* E8 cd */ - ins_encode(Java_To_Runtime(meth)); + ins_encode(clear_avx, Java_To_Runtime(meth)); ins_pipe(pipe_slow); %} @@ -11418,8 +11460,7 @@ ins_cost(300); format %{ "call_leaf,runtime " %} - opcode(0xE8); /* E8 cd */ - ins_encode(Java_To_Runtime(meth)); + ins_encode(clear_avx, Java_To_Runtime(meth)); ins_pipe(pipe_slow); %} @@ -11431,7 +11472,6 @@ ins_cost(300); format %{ "call_leaf_nofp,runtime " %} - opcode(0xE8); /* E8 cd */ ins_encode(Java_To_Runtime(meth)); ins_pipe(pipe_slow); %} diff -r 4c8bb5e4f68f -r 46d24f112c27 src/os/bsd/vm/os_bsd.cpp --- a/src/os/bsd/vm/os_bsd.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/os/bsd/vm/os_bsd.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -166,20 +166,6 @@ return Bsd::physical_memory(); } -julong os::allocatable_physical_memory(julong size) { -#ifdef _LP64 - return size; -#else - julong result = MIN2(size, (julong)3800*M); - if (!is_allocatable(result)) { - // See comments under solaris for alignment considerations - julong reasonable_size = (julong)2*G - 2 * os::vm_page_size(); - result = MIN2(size, reasonable_size); - } - return result; -#endif // _LP64 -} - //////////////////////////////////////////////////////////////////////////////// // environment support @@ -1214,7 +1200,7 @@ int n; char** pelements = split_path(pname, &n); if (pelements == NULL) { - return false; + return false; } for (int i = 0 ; i < n ; i++) { // Really shouldn't be NULL, but check can't hurt diff -r 4c8bb5e4f68f -r 46d24f112c27 src/os/linux/vm/os_linux.cpp --- a/src/os/linux/vm/os_linux.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/os/linux/vm/os_linux.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -193,20 +193,6 @@ return Linux::physical_memory(); } -julong os::allocatable_physical_memory(julong size) { -#ifdef _LP64 - return size; -#else - julong result = MIN2(size, (julong)3800*M); - if (!is_allocatable(result)) { - // See comments under solaris for alignment considerations - julong reasonable_size = (julong)2*G - 2 * os::vm_page_size(); - result = MIN2(size, reasonable_size); - } - return result; -#endif // _LP64 -} - //////////////////////////////////////////////////////////////////////////////// // environment support @@ -1647,7 +1633,7 @@ int n; char** pelements = split_path(pname, &n); if (pelements == NULL) { - return false; + return false; } for (int i = 0 ; i < n ; i++) { // Really shouldn't be NULL, but check can't hurt diff -r 4c8bb5e4f68f -r 46d24f112c27 src/os/posix/launcher/launcher.script --- a/src/os/posix/launcher/launcher.script Fri Apr 05 12:19:19 2013 -0400 +++ b/src/os/posix/launcher/launcher.script Fri Apr 05 16:16:18 2013 -0700 @@ -199,7 +199,7 @@ rm -f $GDBSCR ;; dbx) - $DBX -s $MYDIR/.dbxrc $LAUNCHER $JPARAMS + $DBX -s $HOME/.dbxrc $LAUNCHER $JPARMS ;; valgrind) echo Warning: Defaulting to 16Mb heap to make Valgrind run faster, use -Xmx for larger heap diff -r 4c8bb5e4f68f -r 46d24f112c27 src/os/posix/vm/os_posix.cpp --- a/src/os/posix/vm/os_posix.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/os/posix/vm/os_posix.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -188,4 +188,66 @@ st->cr(); } +bool os::has_allocatable_memory_limit(julong* limit) { + struct rlimit rlim; + int getrlimit_res = getrlimit(RLIMIT_AS, &rlim); + // if there was an error when calling getrlimit, assume that there is no limitation + // on virtual memory. + bool result; + if ((getrlimit_res != 0) || (rlim.rlim_cur == RLIM_INFINITY)) { + result = false; + } else { + *limit = (julong)rlim.rlim_cur; + result = true; + } +#ifdef _LP64 + return result; +#else + // arbitrary virtual space limit for 32 bit Unices found by testing. If + // getrlimit above returned a limit, bound it with this limit. Otherwise + // directly use it. + const julong max_virtual_limit = (julong)3800*M; + if (result) { + *limit = MIN2(*limit, max_virtual_limit); + } else { + *limit = max_virtual_limit; + } + // bound by actually allocatable memory. The algorithm uses two bounds, an + // upper and a lower limit. The upper limit is the current highest amount of + // memory that could not be allocated, the lower limit is the current highest + // amount of memory that could be allocated. + // The algorithm iteratively refines the result by halving the difference + // between these limits, updating either the upper limit (if that value could + // not be allocated) or the lower limit (if the that value could be allocated) + // until the difference between these limits is "small". + + // the minimum amount of memory we care about allocating. + const julong min_allocation_size = M; + + julong upper_limit = *limit; + + // first check a few trivial cases + if (is_allocatable(upper_limit) || (upper_limit <= min_allocation_size)) { + *limit = upper_limit; + } else if (!is_allocatable(min_allocation_size)) { + // we found that not even min_allocation_size is allocatable. Return it + // anyway. There is no point to search for a better value any more. + *limit = min_allocation_size; + } else { + // perform the binary search. + julong lower_limit = min_allocation_size; + while ((upper_limit - lower_limit) > min_allocation_size) { + julong temp_limit = ((upper_limit - lower_limit) / 2) + lower_limit; + temp_limit = align_size_down_(temp_limit, min_allocation_size); + if (is_allocatable(temp_limit)) { + lower_limit = temp_limit; + } else { + upper_limit = temp_limit; + } + } + *limit = lower_limit; + } + return true; +#endif +} diff -r 4c8bb5e4f68f -r 46d24f112c27 src/os/solaris/vm/os_solaris.cpp --- a/src/os/solaris/vm/os_solaris.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/os/solaris/vm/os_solaris.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -476,24 +476,6 @@ return Solaris::physical_memory(); } -julong os::allocatable_physical_memory(julong size) { -#ifdef _LP64 - return size; -#else - julong result = MIN2(size, (julong)3835*M); - if (!is_allocatable(result)) { - // Memory allocations will be aligned but the alignment - // is not known at this point. Alignments will - // be at most to LargePageSizeInBytes. Protect - // allocations from alignments up to illegal - // values. If at this point 2G is illegal. - julong reasonable_size = (julong)2*G - 2 * LargePageSizeInBytes; - result = MIN2(size, reasonable_size); - } - return result; -#endif -} - static hrtime_t first_hrtime = 0; static const hrtime_t hrtime_hz = 1000*1000*1000; const int LOCK_BUSY = 1; @@ -1904,7 +1886,7 @@ int n; char** pelements = split_path(pname, &n); if (pelements == NULL) { - return false; + return false; } for (int i = 0 ; i < n ; i++) { // really shouldn't be NULL but what the heck, check can't hurt diff -r 4c8bb5e4f68f -r 46d24f112c27 src/os/windows/vm/os_windows.cpp --- a/src/os/windows/vm/os_windows.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/os/windows/vm/os_windows.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -686,12 +686,17 @@ return win32::physical_memory(); } -julong os::allocatable_physical_memory(julong size) { +bool os::has_allocatable_memory_limit(julong* limit) { + MEMORYSTATUSEX ms; + ms.dwLength = sizeof(ms); + GlobalMemoryStatusEx(&ms); #ifdef _LP64 - return size; + *limit = (julong)ms.ullAvailVirtual; + return true; #else // Limit to 1400m because of the 2gb address space wall - return MIN2(size, (julong)1400*M); + *limit = MIN2((julong)1400*M, (julong)ms.ullAvailVirtual); + return true; #endif } @@ -1178,7 +1183,7 @@ int n; char** pelements = split_path(pname, &n); if (pelements == NULL) { - return false; + return false; } for (int i = 0 ; i < n ; i++) { char* path = pelements[i]; @@ -3771,6 +3776,8 @@ } } +static jint initSock(); + // this is called _after_ the global arguments have been parsed jint os::init_2(void) { // Allocate a single page and mark it as readable for safepoint polling @@ -3901,6 +3908,10 @@ if (!success) UseNUMAInterleaving = false; } + if (initSock() != JNI_OK) { + return JNI_ERR; + } + return JNI_OK; } @@ -4897,42 +4908,24 @@ // We don't build a headless jre for Windows bool os::is_headless_jre() { return false; } - -typedef CRITICAL_SECTION mutex_t; -#define mutexInit(m) InitializeCriticalSection(m) -#define mutexDestroy(m) DeleteCriticalSection(m) -#define mutexLock(m) EnterCriticalSection(m) -#define mutexUnlock(m) LeaveCriticalSection(m) - -static bool sock_initialized = FALSE; -static mutex_t sockFnTableMutex; - -static void initSock() { +static jint initSock() { WSADATA wsadata; if (!os::WinSock2Dll::WinSock2Available()) { - jio_fprintf(stderr, "Could not load Winsock 2 (error: %d)\n", + jio_fprintf(stderr, "Could not load Winsock (error: %d)\n", ::GetLastError()); - return; - } - if (sock_initialized == TRUE) return; - - ::mutexInit(&sockFnTableMutex); - ::mutexLock(&sockFnTableMutex); - if (os::WinSock2Dll::WSAStartup(MAKEWORD(1,1), &wsadata) != 0) { - jio_fprintf(stderr, "Could not initialize Winsock\n"); - } - sock_initialized = TRUE; - ::mutexUnlock(&sockFnTableMutex); + return JNI_ERR; + } + + if (os::WinSock2Dll::WSAStartup(MAKEWORD(2,2), &wsadata) != 0) { + jio_fprintf(stderr, "Could not initialize Winsock (error: %d)\n", + ::GetLastError()); + return JNI_ERR; + } + return JNI_OK; } struct hostent* os::get_host_by_name(char* name) { - if (!sock_initialized) { - initSock(); - } - if (!os::WinSock2Dll::WinSock2Available()) { - return NULL; - } return (struct hostent*)os::WinSock2Dll::gethostbyname(name); } diff -r 4c8bb5e4f68f -r 46d24f112c27 src/os_cpu/bsd_x86/vm/bsd_x86_64.ad --- a/src/os_cpu/bsd_x86/vm/bsd_x86_64.ad Fri Apr 05 12:19:19 2013 -0400 +++ b/src/os_cpu/bsd_x86/vm/bsd_x86_64.ad Fri Apr 05 16:16:18 2013 -0700 @@ -55,20 +55,6 @@ // adding a syntax that specifies the sizes of fields in an order, // so that the adlc can build the emit functions automagically - enc_class Java_To_Runtime(method meth) %{ - // No relocation needed - - // movq r10, - emit_opcode(cbuf, Assembler::REX_WB); - emit_opcode(cbuf, 0xB8 | (R10_enc - 8)); - emit_d64(cbuf, (int64_t) $meth$$method); - - // call (r10) - emit_opcode(cbuf, Assembler::REX_B); - emit_opcode(cbuf, 0xFF); - emit_opcode(cbuf, 0xD0 | (R10_enc - 8)); - %} - %} @@ -76,8 +62,4 @@ source %{ -int MachCallRuntimeNode::ret_addr_offset() { - return 13; // movq r10,#addr; callq (r10) -} - %} diff -r 4c8bb5e4f68f -r 46d24f112c27 src/os_cpu/linux_x86/vm/linux_x86_64.ad --- a/src/os_cpu/linux_x86/vm/linux_x86_64.ad Fri Apr 05 12:19:19 2013 -0400 +++ b/src/os_cpu/linux_x86/vm/linux_x86_64.ad Fri Apr 05 16:16:18 2013 -0700 @@ -55,20 +55,6 @@ // adding a syntax that specifies the sizes of fields in an order, // so that the adlc can build the emit functions automagically - enc_class Java_To_Runtime(method meth) %{ - // No relocation needed - - // movq r10, - emit_opcode(cbuf, Assembler::REX_WB); - emit_opcode(cbuf, 0xB8 | (R10_enc - 8)); - emit_d64(cbuf, (int64_t) $meth$$method); - - // call (r10) - emit_opcode(cbuf, Assembler::REX_B); - emit_opcode(cbuf, 0xFF); - emit_opcode(cbuf, 0xD0 | (R10_enc - 8)); - %} - %} @@ -76,8 +62,4 @@ source %{ -int MachCallRuntimeNode::ret_addr_offset() { - return 13; // movq r10,#addr; callq (r10) -} - %} diff -r 4c8bb5e4f68f -r 46d24f112c27 src/os_cpu/solaris_x86/vm/solaris_x86_64.ad --- a/src/os_cpu/solaris_x86/vm/solaris_x86_64.ad Fri Apr 05 12:19:19 2013 -0400 +++ b/src/os_cpu/solaris_x86/vm/solaris_x86_64.ad Fri Apr 05 16:16:18 2013 -0700 @@ -54,39 +54,10 @@ // main source block for now. In future, we can generalize this by // adding a syntax that specifies the sizes of fields in an order, // so that the adlc can build the emit functions automagically - - enc_class Java_To_Runtime(method meth) %{ - // No relocation needed - - // movq r10, - emit_opcode(cbuf, Assembler::REX_WB); - emit_opcode(cbuf, 0xB8 | (R10_enc - 8)); - emit_d64(cbuf, (int64_t) $meth$$method); - - // call (r10) - emit_opcode(cbuf, Assembler::REX_B); - emit_opcode(cbuf, 0xFF); - emit_opcode(cbuf, 0xD0 | (R10_enc - 8)); - %} - - enc_class post_call_verify_mxcsr %{ - MacroAssembler _masm(&cbuf); - if (RestoreMXCSROnJNICalls) { - __ ldmxcsr(ExternalAddress(StubRoutines::amd64::mxcsr_std())); - } - else if (CheckJNICalls) { - __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::amd64::verify_mxcsr_entry()))); - } - %} %} // Platform dependent source source %{ - -int MachCallRuntimeNode::ret_addr_offset() { - return 13; // movq r10,#addr; callq (r10) -} - %} diff -r 4c8bb5e4f68f -r 46d24f112c27 src/os_cpu/windows_x86/vm/windows_x86_64.ad --- a/src/os_cpu/windows_x86/vm/windows_x86_64.ad Fri Apr 05 12:19:19 2013 -0400 +++ b/src/os_cpu/windows_x86/vm/windows_x86_64.ad Fri Apr 05 16:16:18 2013 -0700 @@ -53,30 +53,11 @@ // adding a syntax that specifies the sizes of fields in an order, // so that the adlc can build the emit functions automagically - enc_class Java_To_Runtime (method meth) %{ // CALL Java_To_Runtime - // No relocation needed +%} + - // movq r10, - emit_opcode(cbuf, Assembler::REX_WB); - emit_opcode(cbuf, 0xB8 | (R10_enc - 8)); - emit_d64(cbuf, (int64_t) $meth$$method); +// Platform dependent source - // call (r10) - emit_opcode(cbuf, Assembler::REX_B); - emit_opcode(cbuf, 0xFF); - emit_opcode(cbuf, 0xD0 | (R10_enc - 8)); - %} +source %{ %} - -// -// Platform dependent source -// -source %{ - -int MachCallRuntimeNode::ret_addr_offset() -{ - return 13; // movq r10,#addr; callq (r10) -} - -%} diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/adlc/archDesc.cpp --- a/src/share/vm/adlc/archDesc.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/adlc/archDesc.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -832,6 +832,7 @@ int length = (int)strlen(rc_name) + (int)strlen(mask) + 5; char *regMask = new char[length]; sprintf(regMask,"%s%s()", rc_name, mask); + delete[] rc_name; return regMask; } } diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/adlc/dfa.cpp --- a/src/share/vm/adlc/dfa.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/adlc/dfa.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -191,12 +191,19 @@ // Macro equivalent to: _kids[0]->valid(FOO) && _kids[1]->valid(BAR) // static void child_test(FILE *fp, MatchList &mList) { - if( mList._lchild ) // If left child, check it - fprintf(fp, "STATE__VALID_CHILD(_kids[0], %s)", ArchDesc::getMachOperEnum(mList._lchild)); - if( mList._lchild && mList._rchild ) // If both, add the "&&" - fprintf(fp, " && " ); - if( mList._rchild ) // If right child, check it - fprintf(fp, "STATE__VALID_CHILD(_kids[1], %s)", ArchDesc::getMachOperEnum(mList._rchild)); + if (mList._lchild) { // If left child, check it + const char* lchild_to_upper = ArchDesc::getMachOperEnum(mList._lchild); + fprintf(fp, "STATE__VALID_CHILD(_kids[0], %s)", lchild_to_upper); + delete[] lchild_to_upper; + } + if (mList._lchild && mList._rchild) { // If both, add the "&&" + fprintf(fp, " && "); + } + if (mList._rchild) { // If right child, check it + const char* rchild_to_upper = ArchDesc::getMachOperEnum(mList._rchild); + fprintf(fp, "STATE__VALID_CHILD(_kids[1], %s)", rchild_to_upper); + delete[] rchild_to_upper; + } } //---------------------------calc_cost----------------------------------------- @@ -206,13 +213,17 @@ Expr *ArchDesc::calc_cost(FILE *fp, const char *spaces, MatchList &mList, ProductionState &status) { fprintf(fp, "%sunsigned int c = ", spaces); Expr *c = new Expr("0"); - if (mList._lchild ) { // If left child, add it in - sprintf(Expr::buffer(), "_kids[0]->_cost[%s]", ArchDesc::getMachOperEnum(mList._lchild)); + if (mList._lchild) { // If left child, add it in + const char* lchild_to_upper = ArchDesc::getMachOperEnum(mList._lchild); + sprintf(Expr::buffer(), "_kids[0]->_cost[%s]", lchild_to_upper); c->add(Expr::buffer()); + delete[] lchild_to_upper; } - if (mList._rchild) { // If right child, add it in - sprintf(Expr::buffer(), "_kids[1]->_cost[%s]", ArchDesc::getMachOperEnum(mList._rchild)); + if (mList._rchild) { // If right child, add it in + const char* rchild_to_upper = ArchDesc::getMachOperEnum(mList._rchild); + sprintf(Expr::buffer(), "_kids[1]->_cost[%s]", rchild_to_upper); c->add(Expr::buffer()); + delete[] rchild_to_upper; } // Add in cost of this rule const char *mList_cost = mList.get_cost(); @@ -232,15 +243,17 @@ fprintf(fp, "%s", spaces4); // Only generate child tests if this is not a leaf node bool has_child_constraints = mList._lchild || mList._rchild; - const char *predicate_test = mList.get_pred(); - if( has_child_constraints || predicate_test ) { + const char *predicate_test = mList.get_pred(); + if (has_child_constraints || predicate_test) { // Open the child-and-predicate-test braces fprintf(fp, "if( "); status.set_constraint(hasConstraint); child_test(fp, mList); // Only generate predicate test if one exists for this match - if( predicate_test ) { - if( has_child_constraints ) { fprintf(fp," &&\n"); } + if (predicate_test) { + if (has_child_constraints) { + fprintf(fp," &&\n"); + } fprintf(fp, "%s %s", spaces6, predicate_test); } // End of outer tests diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_Canonicalizer.cpp --- a/src/share/vm/c1/c1_Canonicalizer.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_Canonicalizer.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -937,4 +937,6 @@ void Canonicalizer::do_ProfileCall(ProfileCall* x) {} void Canonicalizer::do_ProfileInvoke(ProfileInvoke* x) {} void Canonicalizer::do_RuntimeCall(RuntimeCall* x) {} +void Canonicalizer::do_RangeCheckPredicate(RangeCheckPredicate* x) {} +void Canonicalizer::do_Assert(Assert* x) {} void Canonicalizer::do_MemBar(MemBar* x) {} diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_Canonicalizer.hpp --- a/src/share/vm/c1/c1_Canonicalizer.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_Canonicalizer.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -107,6 +107,8 @@ virtual void do_ProfileInvoke (ProfileInvoke* x); virtual void do_RuntimeCall (RuntimeCall* x); virtual void do_MemBar (MemBar* x); + virtual void do_RangeCheckPredicate(RangeCheckPredicate* x); + virtual void do_Assert (Assert* x); }; #endif // SHARE_VM_C1_C1_CANONICALIZER_HPP diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_CodeStubs.hpp --- a/src/share/vm/c1/c1_CodeStubs.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_CodeStubs.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -166,6 +166,22 @@ #endif // PRODUCT }; +// stub used when predicate fails and deoptimization is needed +class PredicateFailedStub: public CodeStub { + private: + CodeEmitInfo* _info; + + public: + PredicateFailedStub(CodeEmitInfo* info); + virtual void emit_code(LIR_Assembler* e); + virtual CodeEmitInfo* info() const { return _info; } + virtual void visit(LIR_OpVisitState* visitor) { + visitor->do_slow_case(_info); + } +#ifndef PRODUCT + virtual void print_name(outputStream* out) const { out->print("PredicateFailedStub"); } +#endif // PRODUCT +}; class DivByZeroStub: public CodeStub { private: diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_Compilation.cpp --- a/src/share/vm/c1/c1_Compilation.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_Compilation.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -33,13 +33,16 @@ #include "c1/c1_ValueStack.hpp" #include "code/debugInfoRec.hpp" #include "compiler/compileLog.hpp" +#include "c1/c1_RangeCheckElimination.hpp" typedef enum { _t_compile, _t_setup, - _t_optimizeIR, _t_buildIR, + _t_optimize_blocks, + _t_optimize_null_checks, + _t_rangeCheckElimination, _t_emit_lir, _t_linearScan, _t_lirGeneration, @@ -52,8 +55,10 @@ static const char * timer_name[] = { "compile", "setup", - "optimizeIR", "buildIR", + "optimize_blocks", + "optimize_null_checks", + "rangeCheckElimination", "emit_lir", "linearScan", "lirGeneration", @@ -159,9 +164,9 @@ if (UseC1Optimizations) { NEEDS_CLEANUP // optimization - PhaseTraceTime timeit(_t_optimizeIR); + PhaseTraceTime timeit(_t_optimize_blocks); - _hir->optimize(); + _hir->optimize_blocks(); } _hir->verify(); @@ -180,13 +185,47 @@ _hir->compute_code(); if (UseGlobalValueNumbering) { - ResourceMark rm; + // No resource mark here! LoopInvariantCodeMotion can allocate ValueStack objects. int instructions = Instruction::number_of_instructions(); GlobalValueNumbering gvn(_hir); assert(instructions == Instruction::number_of_instructions(), "shouldn't have created an instructions"); } + _hir->verify(); + +#ifndef PRODUCT + if (PrintCFGToFile) { + CFGPrinter::print_cfg(_hir, "Before RangeCheckElimination", true, false); + } +#endif + + if (RangeCheckElimination) { + if (_hir->osr_entry() == NULL) { + PhaseTraceTime timeit(_t_rangeCheckElimination); + RangeCheckElimination::eliminate(_hir); + } + } + +#ifndef PRODUCT + if (PrintCFGToFile) { + CFGPrinter::print_cfg(_hir, "After RangeCheckElimination", true, false); + } +#endif + + if (UseC1Optimizations) { + // loop invariant code motion reorders instructions and range + // check elimination adds new instructions so do null check + // elimination after. + NEEDS_CLEANUP + // optimization + PhaseTraceTime timeit(_t_optimize_null_checks); + + _hir->eliminate_null_checks(); + } + + _hir->verify(); + // compute use counts after global value numbering _hir->compute_use_counts(); @@ -502,6 +541,7 @@ , _next_id(0) , _next_block_id(0) , _code(buffer_blob) +, _has_access_indexed(false) , _current_instruction(NULL) #ifndef PRODUCT , _last_instruction_printed(NULL) @@ -567,7 +607,9 @@ tty->print_cr(" Detailed C1 Timings"); tty->print_cr(" Setup time: %6.3f s (%4.1f%%)", timers[_t_setup].seconds(), (timers[_t_setup].seconds() / total) * 100.0); tty->print_cr(" Build IR: %6.3f s (%4.1f%%)", timers[_t_buildIR].seconds(), (timers[_t_buildIR].seconds() / total) * 100.0); - tty->print_cr(" Optimize: %6.3f s (%4.1f%%)", timers[_t_optimizeIR].seconds(), (timers[_t_optimizeIR].seconds() / total) * 100.0); + float t_optimizeIR = timers[_t_optimize_blocks].seconds() + timers[_t_optimize_null_checks].seconds(); + tty->print_cr(" Optimize: %6.3f s (%4.1f%%)", t_optimizeIR, (t_optimizeIR / total) * 100.0); + tty->print_cr(" RCE: %6.3f s (%4.1f%%)", timers[_t_rangeCheckElimination].seconds(), (timers[_t_rangeCheckElimination].seconds() / total) * 100.0); tty->print_cr(" Emit LIR: %6.3f s (%4.1f%%)", timers[_t_emit_lir].seconds(), (timers[_t_emit_lir].seconds() / total) * 100.0); tty->print_cr(" LIR Gen: %6.3f s (%4.1f%%)", timers[_t_lirGeneration].seconds(), (timers[_t_lirGeneration].seconds() / total) * 100.0); tty->print_cr(" Linear Scan: %6.3f s (%4.1f%%)", timers[_t_linearScan].seconds(), (timers[_t_linearScan].seconds() / total) * 100.0); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_Compilation.hpp --- a/src/share/vm/c1/c1_Compilation.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_Compilation.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -26,8 +26,10 @@ #define SHARE_VM_C1_C1_COMPILATION_HPP #include "ci/ciEnv.hpp" +#include "ci/ciMethodData.hpp" #include "code/exceptionHandlerTable.hpp" #include "memory/resourceArea.hpp" +#include "runtime/deoptimization.hpp" class CompilationResourceObj; class XHandlers; @@ -85,6 +87,7 @@ LinearScan* _allocator; CodeOffsets _offsets; CodeBuffer _code; + bool _has_access_indexed; // compilation helpers void initialize(); @@ -140,6 +143,7 @@ C1_MacroAssembler* masm() const { return _masm; } CodeOffsets* offsets() { return &_offsets; } Arena* arena() { return _arena; } + bool has_access_indexed() { return _has_access_indexed; } // Instruction ids int get_next_id() { return _next_id++; } @@ -154,6 +158,7 @@ void set_has_fpu_code(bool f) { _has_fpu_code = f; } void set_has_unsafe_access(bool f) { _has_unsafe_access = f; } void set_would_profile(bool f) { _would_profile = f; } + void set_has_access_indexed(bool f) { _has_access_indexed = f; } // Add a set of exception handlers covering the given PC offset void add_exception_handlers_for_pco(int pco, XHandlers* exception_handlers); // Statistics gathering @@ -233,6 +238,14 @@ return env()->comp_level() == CompLevel_full_profile && C1UpdateMethodData && C1ProfileCheckcasts; } + + // will compilation make optimistic assumptions that might lead to + // deoptimization and that the runtime will account for? + bool is_optimistic() const { + return !TieredCompilation && + (RangeCheckElimination || UseLoopInvariantCodeMotion) && + method()->method_data()->trap_count(Deoptimization::Reason_none) == 0; + } }; diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_GraphBuilder.cpp --- a/src/share/vm/c1/c1_GraphBuilder.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_GraphBuilder.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -947,7 +947,9 @@ void GraphBuilder::load_indexed(BasicType type) { - ValueStack* state_before = copy_state_for_exception(); + // In case of in block code motion in range check elimination + ValueStack* state_before = copy_state_indexed_access(); + compilation()->set_has_access_indexed(true); Value index = ipop(); Value array = apop(); Value length = NULL; @@ -961,7 +963,9 @@ void GraphBuilder::store_indexed(BasicType type) { - ValueStack* state_before = copy_state_for_exception(); + // In case of in block code motion in range check elimination + ValueStack* state_before = copy_state_indexed_access(); + compilation()->set_has_access_indexed(true); Value value = pop(as_ValueType(type)); Value index = ipop(); Value array = apop(); @@ -1179,7 +1183,9 @@ BlockBegin* tsux = block_at(stream()->get_dest()); BlockBegin* fsux = block_at(stream()->next_bci()); bool is_bb = tsux->bci() < stream()->cur_bci() || fsux->bci() < stream()->cur_bci(); - Instruction *i = append(new If(x, cond, false, y, tsux, fsux, is_bb ? state_before : NULL, is_bb)); + // In case of loop invariant code motion or predicate insertion + // before the body of a loop the state is needed + Instruction *i = append(new If(x, cond, false, y, tsux, fsux, (is_bb || compilation()->is_optimistic()) ? state_before : NULL, is_bb)); assert(i->as_Goto() == NULL || (i->as_Goto()->sux_at(0) == tsux && i->as_Goto()->is_safepoint() == tsux->bci() < stream()->cur_bci()) || @@ -1294,7 +1300,9 @@ BlockBegin* tsux = block_at(bci() + sw.dest_offset_at(0)); BlockBegin* fsux = block_at(bci() + sw.default_offset()); bool is_bb = tsux->bci() < bci() || fsux->bci() < bci(); - ValueStack* state_before = is_bb ? copy_state_before() : NULL; + // In case of loop invariant code motion or predicate insertion + // before the body of a loop the state is needed + ValueStack* state_before = copy_state_if_bb(is_bb); append(new If(ipop(), If::eql, true, key, tsux, fsux, state_before, is_bb)); } else { // collect successors @@ -1308,7 +1316,9 @@ // add default successor if (sw.default_offset() < 0) has_bb = true; sux->at_put(i, block_at(bci() + sw.default_offset())); - ValueStack* state_before = has_bb ? copy_state_before() : NULL; + // In case of loop invariant code motion or predicate insertion + // before the body of a loop the state is needed + ValueStack* state_before = copy_state_if_bb(has_bb); Instruction* res = append(new TableSwitch(ipop(), sux, sw.low_key(), state_before, has_bb)); #ifdef ASSERT if (res->as_Goto()) { @@ -1336,7 +1346,9 @@ BlockBegin* tsux = block_at(bci() + pair.offset()); BlockBegin* fsux = block_at(bci() + sw.default_offset()); bool is_bb = tsux->bci() < bci() || fsux->bci() < bci(); - ValueStack* state_before = is_bb ? copy_state_before() : NULL; + // In case of loop invariant code motion or predicate insertion + // before the body of a loop the state is needed + ValueStack* state_before = copy_state_if_bb(is_bb);; append(new If(ipop(), If::eql, true, key, tsux, fsux, state_before, is_bb)); } else { // collect successors & keys @@ -1353,7 +1365,9 @@ // add default successor if (sw.default_offset() < 0) has_bb = true; sux->at_put(i, block_at(bci() + sw.default_offset())); - ValueStack* state_before = has_bb ? copy_state_before() : NULL; + // In case of loop invariant code motion or predicate insertion + // before the body of a loop the state is needed + ValueStack* state_before = copy_state_if_bb(has_bb); Instruction* res = append(new LookupSwitch(ipop(), sux, keys, state_before, has_bb)); #ifdef ASSERT if (res->as_Goto()) { diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_GraphBuilder.hpp --- a/src/share/vm/c1/c1_GraphBuilder.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_GraphBuilder.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -301,6 +301,8 @@ ValueStack* copy_state_exhandling(); ValueStack* copy_state_for_exception_with_bci(int bci); ValueStack* copy_state_for_exception(); + ValueStack* copy_state_if_bb(bool is_bb) { return (is_bb || compilation()->is_optimistic()) ? copy_state_before() : NULL; } + ValueStack* copy_state_indexed_access() { return compilation()->is_optimistic() ? copy_state_before() : copy_state_for_exception(); } // // Inlining support diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_IR.cpp --- a/src/share/vm/c1/c1_IR.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_IR.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -182,13 +182,14 @@ // Implementation of CodeEmitInfo // Stack must be NON-null -CodeEmitInfo::CodeEmitInfo(ValueStack* stack, XHandlers* exception_handlers) +CodeEmitInfo::CodeEmitInfo(ValueStack* stack, XHandlers* exception_handlers, bool deoptimize_on_exception) : _scope(stack->scope()) , _scope_debug_info(NULL) , _oop_map(NULL) , _stack(stack) , _exception_handlers(exception_handlers) - , _is_method_handle_invoke(false) { + , _is_method_handle_invoke(false) + , _deoptimize_on_exception(deoptimize_on_exception) { assert(_stack != NULL, "must be non null"); } @@ -199,7 +200,8 @@ , _scope_debug_info(NULL) , _oop_map(NULL) , _stack(stack == NULL ? info->_stack : stack) - , _is_method_handle_invoke(info->_is_method_handle_invoke) { + , _is_method_handle_invoke(info->_is_method_handle_invoke) + , _deoptimize_on_exception(info->_deoptimize_on_exception) { // deep copy of exception handlers if (info->_exception_handlers != NULL) { @@ -239,7 +241,7 @@ } -void IR::optimize() { +void IR::optimize_blocks() { Optimizer opt(this); if (!compilation()->profile_branches()) { if (DoCEE) { @@ -257,6 +259,10 @@ #endif } } +} + +void IR::eliminate_null_checks() { + Optimizer opt(this); if (EliminateNullChecks) { opt.eliminate_null_checks(); #ifndef PRODUCT @@ -429,6 +435,7 @@ BlockList _loop_end_blocks; // list of all loop end blocks collected during count_edges BitMap2D _loop_map; // two-dimensional bit set: a bit is set if a block is contained in a loop BlockList _work_list; // temporary list (used in mark_loops and compute_order) + BlockList _loop_headers; Compilation* _compilation; @@ -594,6 +601,7 @@ TRACE_LINEAR_SCAN(3, tty->print_cr("Block B%d is loop header of loop %d", cur->block_id(), _num_loops)); cur->set_loop_index(_num_loops); + _loop_headers.append(cur); _num_loops++; } @@ -656,6 +664,16 @@ // -> this is not a natural loop, so ignore it TRACE_LINEAR_SCAN(2, tty->print_cr("Loop %d is non-natural, so it is ignored", i)); + BlockBegin *loop_header = _loop_headers.at(i); + assert(loop_header->is_set(BlockBegin::linear_scan_loop_header_flag), "Must be loop header"); + + for (int j = 0; j < loop_header->number_of_preds(); j++) { + BlockBegin *pred = loop_header->pred_at(j); + pred->clear(BlockBegin::linear_scan_loop_end_flag); + } + + loop_header->clear(BlockBegin::linear_scan_loop_header_flag); + for (int block_id = _max_block_id - 1; block_id >= 0; block_id--) { clear_block_in_loop(i, block_id); } @@ -729,9 +747,20 @@ } else if (!(cur->is_set(BlockBegin::linear_scan_loop_header_flag) && parent->is_set(BlockBegin::linear_scan_loop_end_flag))) { TRACE_LINEAR_SCAN(4, tty->print_cr("DOM: computing dominator of B%d: common dominator of B%d and B%d is B%d", cur->block_id(), parent->block_id(), cur->dominator()->block_id(), common_dominator(cur->dominator(), parent)->block_id())); - assert(cur->number_of_preds() > 1, ""); + // Does not hold for exception blocks + assert(cur->number_of_preds() > 1 || cur->is_set(BlockBegin::exception_entry_flag), ""); cur->set_dominator(common_dominator(cur->dominator(), parent)); } + + // Additional edge to xhandler of all our successors + // range check elimination needs that the state at the end of a + // block be valid in every block it dominates so cur must dominate + // the exception handlers of its successors. + int num_cur_xhandler = cur->number_of_exception_handlers(); + for (int j = 0; j < num_cur_xhandler; j++) { + BlockBegin* xhandler = cur->exception_handler_at(j); + compute_dominator(xhandler, parent); + } } @@ -898,7 +927,6 @@ num_sux = cur->number_of_exception_handlers(); for (i = 0; i < num_sux; i++) { BlockBegin* sux = cur->exception_handler_at(i); - compute_dominator(sux, cur); if (ready_for_processing(sux)) { sort_into_work_list(sux); } @@ -918,8 +946,23 @@ BlockBegin* dominator = block->pred_at(0); int num_preds = block->number_of_preds(); - for (int i = 1; i < num_preds; i++) { - dominator = common_dominator(dominator, block->pred_at(i)); + + TRACE_LINEAR_SCAN(4, tty->print_cr("DOM: Processing B%d", block->block_id())); + + for (int j = 0; j < num_preds; j++) { + + BlockBegin *pred = block->pred_at(j); + TRACE_LINEAR_SCAN(4, tty->print_cr(" DOM: Subrocessing B%d", pred->block_id())); + + if (block->is_set(BlockBegin::exception_entry_flag)) { + dominator = common_dominator(dominator, pred); + int num_pred_preds = pred->number_of_preds(); + for (int k = 0; k < num_pred_preds; k++) { + dominator = common_dominator(dominator, pred->pred_at(k)); + } + } else { + dominator = common_dominator(dominator, pred); + } } if (dominator != block->dominator()) { @@ -946,6 +989,21 @@ // check that dominators are correct assert(!compute_dominators_iter(), "fix point not reached"); + + // Add Blocks to dominates-Array + int num_blocks = _linear_scan_order->length(); + for (int i = 0; i < num_blocks; i++) { + BlockBegin* block = _linear_scan_order->at(i); + + BlockBegin *dom = block->dominator(); + if (dom) { + assert(dom->dominator_depth() != -1, "Dominator must have been visited before"); + dom->dominates()->append(block); + block->set_dominator_depth(dom->dominator_depth() + 1); + } else { + block->set_dominator_depth(0); + } + } } @@ -1032,7 +1090,7 @@ BlockBegin* sux = cur->sux_at(j); assert(sux->linear_scan_number() >= 0 && sux->linear_scan_number() == _linear_scan_order->index_of(sux), "incorrect linear_scan_number"); - if (!cur->is_set(BlockBegin::linear_scan_loop_end_flag)) { + if (!sux->is_set(BlockBegin::backward_branch_target_flag)) { assert(cur->linear_scan_number() < sux->linear_scan_number(), "invalid order"); } if (cur->loop_depth() == sux->loop_depth()) { @@ -1044,7 +1102,7 @@ BlockBegin* pred = cur->pred_at(j); assert(pred->linear_scan_number() >= 0 && pred->linear_scan_number() == _linear_scan_order->index_of(pred), "incorrect linear_scan_number"); - if (!cur->is_set(BlockBegin::linear_scan_loop_header_flag)) { + if (!cur->is_set(BlockBegin::backward_branch_target_flag)) { assert(cur->linear_scan_number() > pred->linear_scan_number(), "invalid order"); } if (cur->loop_depth() == pred->loop_depth()) { @@ -1060,7 +1118,8 @@ } else { assert(cur->dominator() != NULL, "all but first block must have dominator"); } - assert(cur->number_of_preds() != 1 || cur->dominator() == cur->pred_at(0), "Single predecessor must also be dominator"); + // Assertion does not hold for exception handlers + assert(cur->number_of_preds() != 1 || cur->dominator() == cur->pred_at(0) || cur->is_set(BlockBegin::exception_entry_flag), "Single predecessor must also be dominator"); } // check that all loops are continuous @@ -1249,9 +1308,22 @@ } }; +class VerifyBlockBeginField : public BlockClosure { + +public: + + virtual void block_do(BlockBegin *block) { + for ( Instruction *cur = block; cur != NULL; cur = cur->next()) { + assert(cur->block() == block, "Block begin is not correct"); + } + } +}; + void IR::verify() { #ifdef ASSERT PredecessorValidator pv(this); + VerifyBlockBeginField verifier; + this->iterate_postorder(&verifier); #endif } diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_IR.hpp --- a/src/share/vm/c1/c1_IR.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_IR.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -254,6 +254,7 @@ OopMap* _oop_map; ValueStack* _stack; // used by deoptimization (contains also monitors bool _is_method_handle_invoke; // true if the associated call site is a MethodHandle call site. + bool _deoptimize_on_exception; FrameMap* frame_map() const { return scope()->compilation()->frame_map(); } Compilation* compilation() const { return scope()->compilation(); } @@ -261,7 +262,7 @@ public: // use scope from ValueStack - CodeEmitInfo(ValueStack* stack, XHandlers* exception_handlers); + CodeEmitInfo(ValueStack* stack, XHandlers* exception_handlers, bool deoptimize_on_exception = false); // make a copy CodeEmitInfo(CodeEmitInfo* info, ValueStack* stack = NULL); @@ -272,6 +273,7 @@ IRScope* scope() const { return _scope; } XHandlers* exception_handlers() const { return _exception_handlers; } ValueStack* stack() const { return _stack; } + bool deoptimize_on_exception() const { return _deoptimize_on_exception; } void add_register_oop(LIR_Opr opr); void record_debug_info(DebugInformationRecorder* recorder, int pc_offset); @@ -309,7 +311,8 @@ int max_stack() const { return top_scope()->max_stack(); } // expensive // ir manipulation - void optimize(); + void optimize_blocks(); + void eliminate_null_checks(); void compute_predecessors(); void split_critical_edges(); void compute_code(); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_Instruction.cpp --- a/src/share/vm/c1/c1_Instruction.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_Instruction.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -34,6 +34,15 @@ // Implementation of Instruction +int Instruction::dominator_depth() { + int result = -1; + if (block()) { + result = block()->dominator_depth(); + } + assert(result != -1 || this->as_Local(), "Only locals have dominator depth -1"); + return result; +} + Instruction::Condition Instruction::mirror(Condition cond) { switch (cond) { case eql: return eql; @@ -42,6 +51,8 @@ case leq: return geq; case gtr: return lss; case geq: return leq; + case aeq: return beq; + case beq: return aeq; } ShouldNotReachHere(); return eql; @@ -56,6 +67,8 @@ case leq: return gtr; case gtr: return leq; case geq: return lss; + case aeq: assert(false, "Above equal cannot be negated"); + case beq: assert(false, "Below equal cannot be negated"); } ShouldNotReachHere(); return eql; @@ -70,10 +83,10 @@ } } - -Instruction* Instruction::prev(BlockBegin* block) { +// Prev without need to have BlockBegin +Instruction* Instruction::prev() { Instruction* p = NULL; - Instruction* q = block; + Instruction* q = block(); while (q != this) { assert(q != NULL, "this is not in the block's instruction list"); p = q; q = q->next(); @@ -122,15 +135,24 @@ // perform constant and interval tests on index value bool AccessIndexed::compute_needs_range_check() { - Constant* clength = length()->as_Constant(); - Constant* cindex = index()->as_Constant(); - if (clength && cindex) { - IntConstant* l = clength->type()->as_IntConstant(); - IntConstant* i = cindex->type()->as_IntConstant(); - if (l && i && i->value() < l->value() && i->value() >= 0) { - return false; + + if (length()) { + + Constant* clength = length()->as_Constant(); + Constant* cindex = index()->as_Constant(); + if (clength && cindex) { + IntConstant* l = clength->type()->as_IntConstant(); + IntConstant* i = cindex->type()->as_IntConstant(); + if (l && i && i->value() < l->value() && i->value() >= 0) { + return false; + } } } + + if (!this->check_flag(NeedsRangeCheckFlag)) { + return false; + } + return true; } @@ -631,19 +653,25 @@ // of the inserted block, without recomputing the values of the other blocks // in the CFG. Therefore the value of "depth_first_number" in BlockBegin becomes meaningless. BlockBegin* BlockBegin::insert_block_between(BlockBegin* sux) { - BlockBegin* new_sux = new BlockBegin(end()->state()->bci()); + int bci = sux->bci(); + // critical edge splitting may introduce a goto after a if and array + // bound check elimination may insert a predicate between the if and + // goto. The bci of the goto can't be the one of the if otherwise + // the state and bci are inconsistent and a deoptimization triggered + // by the predicate would lead to incorrect execution/a crash. + BlockBegin* new_sux = new BlockBegin(bci); // mark this block (special treatment when block order is computed) new_sux->set(critical_edge_split_flag); // This goto is not a safepoint. Goto* e = new Goto(sux, false); - new_sux->set_next(e, end()->state()->bci()); + new_sux->set_next(e, bci); new_sux->set_end(e); // setup states ValueStack* s = end()->state(); - new_sux->set_state(s->copy()); - e->set_state(s->copy()); + new_sux->set_state(s->copy(s->kind(), bci)); + e->set_state(s->copy(s->kind(), bci)); assert(new_sux->state()->locals_size() == s->locals_size(), "local size mismatch!"); assert(new_sux->state()->stack_size() == s->stack_size(), "stack size mismatch!"); assert(new_sux->state()->locks_size() == s->locks_size(), "locks size mismatch!"); @@ -960,15 +988,14 @@ BlockList* sux = NULL; if (begin != NULL) { sux = begin->successors(); - } else if (_begin != NULL) { + } else if (this->begin() != NULL) { // copy our sux list - BlockList* sux = new BlockList(_begin->number_of_sux()); - for (int i = 0; i < _begin->number_of_sux(); i++) { - sux->append(_begin->sux_at(i)); + BlockList* sux = new BlockList(this->begin()->number_of_sux()); + for (int i = 0; i < this->begin()->number_of_sux(); i++) { + sux->append(this->begin()->sux_at(i)); } } _sux = sux; - _begin = begin; } @@ -1008,7 +1035,38 @@ } } +#ifdef ASSERT +// Constructor of Assert +Assert::Assert(Value x, Condition cond, bool unordered_is_true, Value y) : Instruction(illegalType) + , _x(x) + , _cond(cond) + , _y(y) +{ + set_flag(UnorderedIsTrueFlag, unordered_is_true); + assert(x->type()->tag() == y->type()->tag(), "types must match"); + pin(); + stringStream strStream; + Compilation::current()->method()->print_name(&strStream); + + stringStream strStream1; + InstructionPrinter ip1(1, &strStream1); + ip1.print_instr(x); + + stringStream strStream2; + InstructionPrinter ip2(1, &strStream2); + ip2.print_instr(y); + + stringStream ss; + ss.print("Assertion %s %s %s in method %s", strStream1.as_string(), ip2.cond_name(cond), strStream2.as_string(), strStream.as_string()); + + _message = ss.as_string(); +} +#endif + +void RangeCheckPredicate::check_state() { + assert(state()->kind() != ValueStack::EmptyExceptionState && state()->kind() != ValueStack::ExceptionState, "will deopt with empty state"); +} void ProfileInvoke::state_values_do(ValueVisitor* f) { if (state() != NULL) state()->values_do(f); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_Instruction.hpp --- a/src/share/vm/c1/c1_Instruction.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_Instruction.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -110,6 +110,8 @@ class ProfileInvoke; class RuntimeCall; class MemBar; +class RangeCheckPredicate; +class Assert; // A Value is a reference to the instruction creating the value typedef Instruction* Value; @@ -210,6 +212,10 @@ virtual void do_ProfileInvoke (ProfileInvoke* x) = 0; virtual void do_RuntimeCall (RuntimeCall* x) = 0; virtual void do_MemBar (MemBar* x) = 0; + virtual void do_RangeCheckPredicate(RangeCheckPredicate* x) = 0; +#ifdef ASSERT + virtual void do_Assert (Assert* x) = 0; +#endif }; @@ -306,8 +312,9 @@ void update_exception_state(ValueStack* state); - //protected: - public: + protected: + BlockBegin* _block; // Block that contains this instruction + void set_type(ValueType* type) { assert(type != NULL, "type must exist"); _type = type; @@ -342,6 +349,9 @@ ThrowIncompatibleClassChangeErrorFlag, ProfileMDOFlag, IsLinkedInBlockFlag, + NeedsRangeCheckFlag, + InWorkListFlag, + DeoptimizeOnException, InstructionLastFlag }; @@ -351,7 +361,7 @@ // 'globally' used condition values enum Condition { - eql, neq, lss, leq, gtr, geq + eql, neq, lss, leq, gtr, geq, aeq, beq }; // Instructions may be pinned for many reasons and under certain conditions @@ -381,6 +391,7 @@ , _pin_state(0) , _type(type) , _next(NULL) + , _block(NULL) , _subst(NULL) , _flags(0) , _operand(LIR_OprFact::illegalOpr) @@ -399,11 +410,13 @@ int printable_bci() const { assert(has_printable_bci(), "_printable_bci should have been set"); return _printable_bci; } void set_printable_bci(int bci) { _printable_bci = bci; } #endif + int dominator_depth(); int use_count() const { return _use_count; } int pin_state() const { return _pin_state; } bool is_pinned() const { return _pin_state != 0 || PinAllInstructions; } ValueType* type() const { return _type; } - Instruction* prev(BlockBegin* block); // use carefully, expensive operation + BlockBegin *block() const { return _block; } + Instruction* prev(); // use carefully, expensive operation Instruction* next() const { return _next; } bool has_subst() const { return _subst != NULL; } Instruction* subst() { return _subst == NULL ? this : _subst->subst(); } @@ -432,6 +445,9 @@ assert(as_BlockEnd() == NULL, "BlockEnd instructions must have no next"); assert(next->can_be_linked(), "shouldn't link these instructions into list"); + BlockBegin *block = this->block(); + next->_block = block; + next->set_flag(Instruction::IsLinkedInBlockFlag, true); _next = next; return next; @@ -444,6 +460,29 @@ return set_next(next); } + // when blocks are merged + void fixup_block_pointers() { + Instruction *cur = next()->next(); // next()'s block is set in set_next + while (cur && cur->_block != block()) { + cur->_block = block(); + cur = cur->next(); + } + } + + Instruction *insert_after(Instruction *i) { + Instruction* n = _next; + set_next(i); + i->set_next(n); + return _next; + } + + Instruction *insert_after_same_bci(Instruction *i) { +#ifndef PRODUCT + i->set_printable_bci(printable_bci()); +#endif + return insert_after(i); + } + void set_subst(Instruction* subst) { assert(subst == NULL || type()->base() == subst->type()->base() || @@ -452,6 +491,7 @@ } void set_exception_handlers(XHandlers *xhandlers) { _exception_handlers = xhandlers; } void set_exception_state(ValueStack* s) { check_state(s); _exception_state = s; } + void set_state_before(ValueStack* s) { check_state(s); _state_before = s; } // machine-specifics void set_operand(LIR_Opr operand) { assert(operand != LIR_OprFact::illegalOpr, "operand must exist"); _operand = operand; } @@ -509,6 +549,11 @@ virtual ExceptionObject* as_ExceptionObject() { return NULL; } virtual UnsafeOp* as_UnsafeOp() { return NULL; } virtual ProfileInvoke* as_ProfileInvoke() { return NULL; } + virtual RangeCheckPredicate* as_RangeCheckPredicate() { return NULL; } + +#ifdef ASSERT + virtual Assert* as_Assert() { return NULL; } +#endif virtual void visit(InstructionVisitor* v) = 0; @@ -570,7 +615,6 @@ LEAF(Phi, Instruction) private: - BlockBegin* _block; // the block to which the phi function belongs int _pf_flags; // the flags of the phi function int _index; // to value on operand stack (index < 0) or to local public: @@ -578,9 +622,9 @@ Phi(ValueType* type, BlockBegin* b, int index) : Instruction(type->base()) , _pf_flags(0) - , _block(b) , _index(index) { + _block = b; NOT_PRODUCT(set_printable_bci(Value(b)->printable_bci())); if (type->is_illegal()) { make_illegal(); @@ -603,8 +647,6 @@ Value operand_at(int i) const; int operand_count() const; - BlockBegin* block() const { return _block; } - void set(Flag f) { _pf_flags |= f; } void clear(Flag f) { _pf_flags &= ~f; } bool is_set(Flag f) const { return (_pf_flags & f) != 0; } @@ -670,6 +712,7 @@ pin(); } + // generic virtual bool can_trap() const { return state_before() != NULL; } virtual void input_values_do(ValueVisitor* f) { /* no values */ } @@ -852,6 +895,7 @@ , _length(length) , _elt_type(elt_type) { + set_flag(Instruction::NeedsRangeCheckFlag, true); ASSERT_VALUES } @@ -860,6 +904,7 @@ Value length() const { return _length; } BasicType elt_type() const { return _elt_type; } + void clear_length() { _length = NULL; } // perform elimination of range checks involving constants bool compute_needs_range_check(); @@ -1524,6 +1569,7 @@ int _bci; // start-bci of block int _depth_first_number; // number of this block in a depth-first ordering int _linear_scan_number; // number of this block in linear-scan ordering + int _dominator_depth; int _loop_depth; // the loop nesting level of this block int _loop_index; // number of the innermost loop of this block int _flags; // the flags associated with this block @@ -1535,6 +1581,7 @@ // SSA specific fields: (factor out later) BlockList _successors; // the successors of this block BlockList _predecessors; // the predecessors of this block + BlockList _dominates; // list of blocks that are dominated by this block BlockBegin* _dominator; // the dominator of this block // SSA specific ends BlockEnd* _end; // the last instruction of this block @@ -1583,10 +1630,12 @@ , _linear_scan_number(-1) , _loop_depth(0) , _flags(0) + , _dominator_depth(-1) , _dominator(NULL) , _end(NULL) , _predecessors(2) , _successors(2) + , _dominates(2) , _exception_handlers(1) , _exception_states(NULL) , _exception_handler_pco(-1) @@ -1603,6 +1652,7 @@ , _total_preds(0) , _stores_to_locals() { + _block = this; #ifndef PRODUCT set_printable_bci(bci); #endif @@ -1612,8 +1662,10 @@ int block_id() const { return _block_id; } int bci() const { return _bci; } BlockList* successors() { return &_successors; } + BlockList* dominates() { return &_dominates; } BlockBegin* dominator() const { return _dominator; } int loop_depth() const { return _loop_depth; } + int dominator_depth() const { return _dominator_depth; } int depth_first_number() const { return _depth_first_number; } int linear_scan_number() const { return _linear_scan_number; } BlockEnd* end() const { return _end; } @@ -1634,6 +1686,7 @@ // manipulation void set_dominator(BlockBegin* dom) { _dominator = dom; } void set_loop_depth(int d) { _loop_depth = d; } + void set_dominator_depth(int d) { _dominator_depth = d; } void set_depth_first_number(int dfn) { _depth_first_number = dfn; } void set_linear_scan_number(int lsn) { _linear_scan_number = lsn; } void set_end(BlockEnd* end); @@ -1695,7 +1748,8 @@ parser_loop_header_flag = 1 << 7, // set by parser to identify blocks where phi functions can not be created on demand critical_edge_split_flag = 1 << 8, // set for all blocks that are introduced when critical edges are split linear_scan_loop_header_flag = 1 << 9, // set during loop-detection for LinearScan - linear_scan_loop_end_flag = 1 << 10 // set during loop-detection for LinearScan + linear_scan_loop_end_flag = 1 << 10, // set during loop-detection for LinearScan + donot_eliminate_range_checks = 1 << 11 // Should be try to eliminate range checks in this block }; void set(Flag f) { _flags |= f; } @@ -1728,7 +1782,6 @@ BASE(BlockEnd, StateSplit) private: - BlockBegin* _begin; BlockList* _sux; protected: @@ -1746,7 +1799,6 @@ // creation BlockEnd(ValueType* type, ValueStack* state_before, bool is_safepoint) : StateSplit(type, state_before) - , _begin(NULL) , _sux(NULL) { set_flag(IsSafepointFlag, is_safepoint); @@ -1754,7 +1806,8 @@ // accessors bool is_safepoint() const { return check_flag(IsSafepointFlag); } - BlockBegin* begin() const { return _begin; } + // For compatibility with old code, for new code use block() + BlockBegin* begin() const { return _block; } // manipulation void set_begin(BlockBegin* begin); @@ -1811,6 +1864,74 @@ void set_direction(Direction d) { _direction = d; } }; +#ifdef ASSERT +LEAF(Assert, Instruction) + private: + Value _x; + Condition _cond; + Value _y; + char *_message; + + public: + // creation + // unordered_is_true is valid for float/double compares only + Assert(Value x, Condition cond, bool unordered_is_true, Value y); + + // accessors + Value x() const { return _x; } + Condition cond() const { return _cond; } + bool unordered_is_true() const { return check_flag(UnorderedIsTrueFlag); } + Value y() const { return _y; } + const char *message() const { return _message; } + + // generic + virtual void input_values_do(ValueVisitor* f) { f->visit(&_x); f->visit(&_y); } +}; +#endif + +LEAF(RangeCheckPredicate, StateSplit) + private: + Value _x; + Condition _cond; + Value _y; + + void check_state(); + + public: + // creation + // unordered_is_true is valid for float/double compares only + RangeCheckPredicate(Value x, Condition cond, bool unordered_is_true, Value y, ValueStack* state) : StateSplit(illegalType) + , _x(x) + , _cond(cond) + , _y(y) + { + ASSERT_VALUES + set_flag(UnorderedIsTrueFlag, unordered_is_true); + assert(x->type()->tag() == y->type()->tag(), "types must match"); + this->set_state(state); + check_state(); + } + + // Always deoptimize + RangeCheckPredicate(ValueStack* state) : StateSplit(illegalType) + { + this->set_state(state); + _x = _y = NULL; + check_state(); + } + + // accessors + Value x() const { return _x; } + Condition cond() const { return _cond; } + bool unordered_is_true() const { return check_flag(UnorderedIsTrueFlag); } + Value y() const { return _y; } + + void always_fail() { _x = _y = NULL; } + + // generic + virtual void input_values_do(ValueVisitor* f) { StateSplit::input_values_do(f); f->visit(&_x); f->visit(&_y); } + HASHING3(RangeCheckPredicate, true, x()->subst(), y()->subst(), cond()) +}; LEAF(If, BlockEnd) private: diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_InstructionPrinter.cpp --- a/src/share/vm/c1/c1_InstructionPrinter.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_InstructionPrinter.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -57,6 +57,8 @@ case If::leq: return "<="; case If::gtr: return ">"; case If::geq: return ">="; + case If::aeq: return "|>=|"; + case If::beq: return "|<=|"; } ShouldNotReachHere(); return NULL; @@ -181,6 +183,11 @@ output()->put('['); print_value(indexed->index()); output()->put(']'); + if (indexed->length() != NULL) { + output()->put('('); + print_value(indexed->length()); + output()->put(')'); + } } @@ -373,6 +380,7 @@ void InstructionPrinter::do_LoadField(LoadField* x) { print_field(x); output()->print(" (%c)", type2char(x->field()->type()->basic_type())); + output()->print(" %s", x->field()->name()->as_utf8()); } @@ -381,6 +389,7 @@ output()->print(" := "); print_value(x->value()); output()->print(" (%c)", type2char(x->field()->type()->basic_type())); + output()->print(" %s", x->field()->name()->as_utf8()); } @@ -393,6 +402,9 @@ void InstructionPrinter::do_LoadIndexed(LoadIndexed* x) { print_indexed(x); output()->print(" (%c)", type2char(x->elt_type())); + if (x->check_flag(Instruction::NeedsRangeCheckFlag)) { + output()->print(" [rc]"); + } } @@ -401,6 +413,9 @@ output()->print(" := "); print_value(x->value()); output()->print(" (%c)", type2char(x->elt_type())); + if (x->check_flag(Instruction::NeedsRangeCheckFlag)) { + output()->print(" [rc]"); + } } void InstructionPrinter::do_NegateOp(NegateOp* x) { @@ -843,6 +858,25 @@ output()->put(')'); } +void InstructionPrinter::do_RangeCheckPredicate(RangeCheckPredicate* x) { + + if (x->x() != NULL && x->y() != NULL) { + output()->print("if "); + print_value(x->x()); + output()->print(" %s ", cond_name(x->cond())); + print_value(x->y()); + output()->print(" then deoptimize!"); + } else { + output()->print("always deoptimize!"); + } +} + +void InstructionPrinter::do_Assert(Assert* x) { + output()->print("assert "); + print_value(x->x()); + output()->print(" %s ", cond_name(x->cond())); + print_value(x->y()); +} void InstructionPrinter::do_UnsafePrefetchWrite(UnsafePrefetchWrite* x) { print_unsafe_object_op(x, "UnsafePrefetchWrite"); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_InstructionPrinter.hpp --- a/src/share/vm/c1/c1_InstructionPrinter.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_InstructionPrinter.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -135,6 +135,8 @@ virtual void do_ProfileInvoke (ProfileInvoke* x); virtual void do_RuntimeCall (RuntimeCall* x); virtual void do_MemBar (MemBar* x); + virtual void do_RangeCheckPredicate(RangeCheckPredicate* x); + virtual void do_Assert (Assert* x); }; #endif // PRODUCT diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_LIR.cpp --- a/src/share/vm/c1/c1_LIR.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_LIR.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -633,6 +633,7 @@ case lir_ushr: case lir_xadd: case lir_xchg: + case lir_assert: { assert(op->as_Op2() != NULL, "must be"); LIR_Op2* op2 = (LIR_Op2*)op; @@ -1112,6 +1113,11 @@ } } +#ifdef ASSERT +void LIR_OpAssert::emit_code(LIR_Assembler* masm) { + masm->emit_assert(this); +} +#endif void LIR_OpDelay::emit_code(LIR_Assembler* masm) { masm->emit_delay(this); @@ -1771,6 +1777,8 @@ case lir_cas_int: s = "cas_int"; break; // LIR_OpProfileCall case lir_profile_call: s = "profile_call"; break; + // LIR_OpAssert + case lir_assert: s = "assert"; break; case lir_none: ShouldNotReachHere();break; default: s = "illegal_op"; break; } @@ -2017,6 +2025,13 @@ out->print("[lbl:0x%x]", stub()->entry()); } +void LIR_OpAssert::print_instr(outputStream* out) const { + print_condition(out, condition()); out->print(" "); + in_opr1()->print(out); out->print(" "); + in_opr2()->print(out); out->print(", \""); + out->print(msg()); out->print("\""); +} + void LIR_OpDelay::print_instr(outputStream* out) const { _op->print_on(out); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_LIR.hpp --- a/src/share/vm/c1/c1_LIR.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_LIR.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -881,6 +881,7 @@ class LIR_OpTypeCheck; class LIR_OpCompareAndSwap; class LIR_OpProfileCall; +class LIR_OpAssert; // LIR operation codes @@ -1000,6 +1001,9 @@ , begin_opMDOProfile , lir_profile_call , end_opMDOProfile + , begin_opAssert + , lir_assert + , end_opAssert }; @@ -1135,6 +1139,7 @@ virtual LIR_OpTypeCheck* as_OpTypeCheck() { return NULL; } virtual LIR_OpCompareAndSwap* as_OpCompareAndSwap() { return NULL; } virtual LIR_OpProfileCall* as_OpProfileCall() { return NULL; } + virtual LIR_OpAssert* as_OpAssert() { return NULL; } virtual void verify() const {} }; @@ -1623,7 +1628,7 @@ , _tmp3(LIR_OprFact::illegalOpr) , _tmp4(LIR_OprFact::illegalOpr) , _tmp5(LIR_OprFact::illegalOpr) { - assert(code == lir_cmp, "code check"); + assert(code == lir_cmp || code == lir_assert, "code check"); } LIR_Op2(LIR_Code code, LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type) @@ -1683,7 +1688,7 @@ LIR_Opr tmp4_opr() const { return _tmp4; } LIR_Opr tmp5_opr() const { return _tmp5; } LIR_Condition condition() const { - assert(code() == lir_cmp || code() == lir_cmove, "only valid for cmp and cmove"); return _condition; + assert(code() == lir_cmp || code() == lir_cmove || code() == lir_assert, "only valid for cmp and cmove and assert"); return _condition; } void set_condition(LIR_Condition condition) { assert(code() == lir_cmp || code() == lir_cmove, "only valid for cmp and cmove"); _condition = condition; @@ -1823,6 +1828,30 @@ CodeEmitInfo* call_info() const { return info(); } }; +#ifdef ASSERT +// LIR_OpAssert +class LIR_OpAssert : public LIR_Op2 { + friend class LIR_OpVisitState; + + private: + const char* _msg; + bool _halt; + + public: + LIR_OpAssert(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, const char* msg, bool halt) + : LIR_Op2(lir_assert, condition, opr1, opr2) + , _halt(halt) + , _msg(msg) { + } + + const char* msg() const { return _msg; } + bool halt() const { return _halt; } + + virtual void emit_code(LIR_Assembler* masm); + virtual LIR_OpAssert* as_OpAssert() { return this; } + virtual void print_instr(outputStream* out) const PRODUCT_RETURN; +}; +#endif // LIR_OpCompareAndSwap class LIR_OpCompareAndSwap : public LIR_Op { @@ -2196,6 +2225,9 @@ void xadd(LIR_Opr src, LIR_Opr add, LIR_Opr res, LIR_Opr tmp) { append(new LIR_Op2(lir_xadd, src, add, res, tmp)); } void xchg(LIR_Opr src, LIR_Opr set, LIR_Opr res, LIR_Opr tmp) { append(new LIR_Op2(lir_xchg, src, set, res, tmp)); } +#ifdef ASSERT + void lir_assert(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, const char* msg, bool halt) { append(new LIR_OpAssert(condition, opr1, opr2, msg, halt)); } +#endif }; void print_LIR(BlockList* blocks); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_LIRAssembler.hpp --- a/src/share/vm/c1/c1_LIRAssembler.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_LIRAssembler.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -210,6 +210,9 @@ void arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info, bool pop_fpu_stack); void arithmetic_idiv(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr temp, LIR_Opr result, CodeEmitInfo* info); void intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr unused, LIR_Opr dest, LIR_Op* op); +#ifdef ASSERT + void emit_assert(LIR_OpAssert* op); +#endif void logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_LIRGenerator.cpp --- a/src/share/vm/c1/c1_LIRGenerator.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_LIRGenerator.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -403,6 +403,10 @@ CodeEmitInfo* LIRGenerator::state_for(Instruction* x, ValueStack* state, bool ignore_xhandler) { assert(state != NULL, "state must be defined"); +#ifndef PRODUCT + state->verify(); +#endif + ValueStack* s = state; for_each_state(s) { if (s->kind() == ValueStack::EmptyExceptionState) { @@ -453,7 +457,7 @@ } } - return new CodeEmitInfo(state, ignore_xhandler ? NULL : x->exception_handlers()); + return new CodeEmitInfo(state, ignore_xhandler ? NULL : x->exception_handlers(), x->check_flag(Instruction::DeoptimizeOnException)); } @@ -1773,11 +1777,18 @@ } #endif + bool stress_deopt = StressLoopInvariantCodeMotion && info && info->deoptimize_on_exception(); if (x->needs_null_check() && (needs_patching || - MacroAssembler::needs_explicit_null_check(x->offset()))) { + MacroAssembler::needs_explicit_null_check(x->offset()) || + stress_deopt)) { + LIR_Opr obj = object.result(); + if (stress_deopt) { + obj = new_register(T_OBJECT); + __ move(LIR_OprFact::oopConst(NULL), obj); + } // emit an explicit null check because the offset is too large - __ null_check(object.result(), new CodeEmitInfo(info)); + __ null_check(obj, new CodeEmitInfo(info)); } LIR_Opr reg = rlock_result(x, field_type); @@ -1854,6 +1865,11 @@ } else { info = state_for(nc); } + if (StressLoopInvariantCodeMotion && info->deoptimize_on_exception()) { + LIR_Opr obj = new_register(T_OBJECT); + __ move(LIR_OprFact::oopConst(NULL), obj); + __ null_check(obj, new CodeEmitInfo(info)); + } } __ load(new LIR_Address(array.result(), arrayOopDesc::length_offset_in_bytes(), T_INT), reg, info, lir_patch_none); } @@ -1864,14 +1880,11 @@ LIRItem array(x->array(), this); LIRItem index(x->index(), this); LIRItem length(this); - bool needs_range_check = true; - - if (use_length) { - needs_range_check = x->compute_needs_range_check(); - if (needs_range_check) { - length.set_instruction(x->length()); - length.load_item(); - } + bool needs_range_check = x->compute_needs_range_check(); + + if (use_length && needs_range_check) { + length.set_instruction(x->length()); + length.load_item(); } array.load_item(); @@ -1891,13 +1904,20 @@ } else { null_check_info = range_check_info; } + if (StressLoopInvariantCodeMotion && null_check_info->deoptimize_on_exception()) { + LIR_Opr obj = new_register(T_OBJECT); + __ move(LIR_OprFact::oopConst(NULL), obj); + __ null_check(obj, new CodeEmitInfo(null_check_info)); + } } // emit array address setup early so it schedules better LIR_Address* array_addr = emit_array_address(array.result(), index.result(), x->elt_type(), false); if (GenerateRangeChecks && needs_range_check) { - if (use_length) { + if (StressLoopInvariantCodeMotion && range_check_info->deoptimize_on_exception()) { + __ branch(lir_cond_always, T_ILLEGAL, new RangeCheckStub(range_check_info, index.result())); + } else if (use_length) { // TODO: use a (modified) version of array_range_check that does not require a // constant length to be loaded to a register __ cmp(lir_cond_belowEqual, length.result(), index.result()); @@ -2615,7 +2635,7 @@ LIR_Opr lock = new_register(T_INT); __ load_stack_address_monitor(0, lock); - CodeEmitInfo* info = new CodeEmitInfo(scope()->start()->state()->copy(ValueStack::StateBefore, SynchronizationEntryBCI), NULL); + CodeEmitInfo* info = new CodeEmitInfo(scope()->start()->state()->copy(ValueStack::StateBefore, SynchronizationEntryBCI), NULL, x->check_flag(Instruction::DeoptimizeOnException)); CodeStub* slow_path = new MonitorEnterStub(obj, lock, info); // receiver is guaranteed non-NULL so don't need CodeEmitInfo @@ -2625,7 +2645,7 @@ // increment invocation counters if needed if (!method()->is_accessor()) { // Accessors do not have MDOs, so no counting. - CodeEmitInfo* info = new CodeEmitInfo(scope()->start()->state()->copy(ValueStack::StateBefore, SynchronizationEntryBCI), NULL); + CodeEmitInfo* info = new CodeEmitInfo(scope()->start()->state()->copy(ValueStack::StateBefore, SynchronizationEntryBCI), NULL, false); increment_invocation_counter(info); } @@ -3083,6 +3103,95 @@ } } +void LIRGenerator::do_Assert(Assert *x) { +#ifdef ASSERT + ValueTag tag = x->x()->type()->tag(); + If::Condition cond = x->cond(); + + LIRItem xitem(x->x(), this); + LIRItem yitem(x->y(), this); + LIRItem* xin = &xitem; + LIRItem* yin = &yitem; + + assert(tag == intTag, "Only integer assertions are valid!"); + + xin->load_item(); + yin->dont_load_item(); + + set_no_result(x); + + LIR_Opr left = xin->result(); + LIR_Opr right = yin->result(); + + __ lir_assert(lir_cond(x->cond()), left, right, x->message(), true); +#endif +} + + +void LIRGenerator::do_RangeCheckPredicate(RangeCheckPredicate *x) { + + + Instruction *a = x->x(); + Instruction *b = x->y(); + if (!a || StressRangeCheckElimination) { + assert(!b || StressRangeCheckElimination, "B must also be null"); + + CodeEmitInfo *info = state_for(x, x->state()); + CodeStub* stub = new PredicateFailedStub(info); + + __ jump(stub); + } else if (a->type()->as_IntConstant() && b->type()->as_IntConstant()) { + int a_int = a->type()->as_IntConstant()->value(); + int b_int = b->type()->as_IntConstant()->value(); + + bool ok = false; + + switch(x->cond()) { + case Instruction::eql: ok = (a_int == b_int); break; + case Instruction::neq: ok = (a_int != b_int); break; + case Instruction::lss: ok = (a_int < b_int); break; + case Instruction::leq: ok = (a_int <= b_int); break; + case Instruction::gtr: ok = (a_int > b_int); break; + case Instruction::geq: ok = (a_int >= b_int); break; + case Instruction::aeq: ok = ((unsigned int)a_int >= (unsigned int)b_int); break; + case Instruction::beq: ok = ((unsigned int)a_int <= (unsigned int)b_int); break; + default: ShouldNotReachHere(); + } + + if (ok) { + + CodeEmitInfo *info = state_for(x, x->state()); + CodeStub* stub = new PredicateFailedStub(info); + + __ jump(stub); + } + } else { + + ValueTag tag = x->x()->type()->tag(); + If::Condition cond = x->cond(); + LIRItem xitem(x->x(), this); + LIRItem yitem(x->y(), this); + LIRItem* xin = &xitem; + LIRItem* yin = &yitem; + + assert(tag == intTag, "Only integer deoptimizations are valid!"); + + xin->load_item(); + yin->dont_load_item(); + set_no_result(x); + + LIR_Opr left = xin->result(); + LIR_Opr right = yin->result(); + + CodeEmitInfo *info = state_for(x, x->state()); + CodeStub* stub = new PredicateFailedStub(info); + + __ cmp(lir_cond(cond), left, right); + __ branch(lir_cond(cond), right->type(), stub); + } +} + + LIR_Opr LIRGenerator::call_runtime(Value arg1, address entry, ValueType* result_type, CodeEmitInfo* info) { LIRItemList args(1); LIRItem value(arg1, this); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_LIRGenerator.hpp --- a/src/share/vm/c1/c1_LIRGenerator.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_LIRGenerator.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -412,6 +412,8 @@ case If::leq: l = lir_cond_lessEqual; break; case If::geq: l = lir_cond_greaterEqual; break; case If::gtr: l = lir_cond_greater; break; + case If::aeq: l = lir_cond_aboveEqual; break; + case If::beq: l = lir_cond_belowEqual; break; }; return l; } @@ -534,6 +536,8 @@ virtual void do_ProfileInvoke (ProfileInvoke* x); virtual void do_RuntimeCall (RuntimeCall* x); virtual void do_MemBar (MemBar* x); + virtual void do_RangeCheckPredicate(RangeCheckPredicate* x); + virtual void do_Assert (Assert* x); }; diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_LinearScan.cpp --- a/src/share/vm/c1/c1_LinearScan.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_LinearScan.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -6231,26 +6231,29 @@ assert(prev_op->as_OpBranch() != NULL, "branch must be of type LIR_OpBranch"); LIR_OpBranch* prev_branch = (LIR_OpBranch*)prev_op; - LIR_Op2* prev_cmp = NULL; - - for(int j = instructions->length() - 3; j >= 0 && prev_cmp == NULL; j--) { - prev_op = instructions->at(j); - if(prev_op->code() == lir_cmp) { - assert(prev_op->as_Op2() != NULL, "branch must be of type LIR_Op2"); - prev_cmp = (LIR_Op2*)prev_op; - assert(prev_branch->cond() == prev_cmp->condition(), "should be the same"); + if (prev_branch->stub() == NULL) { + + LIR_Op2* prev_cmp = NULL; + + for(int j = instructions->length() - 3; j >= 0 && prev_cmp == NULL; j--) { + prev_op = instructions->at(j); + if (prev_op->code() == lir_cmp) { + assert(prev_op->as_Op2() != NULL, "branch must be of type LIR_Op2"); + prev_cmp = (LIR_Op2*)prev_op; + assert(prev_branch->cond() == prev_cmp->condition(), "should be the same"); + } } - } - assert(prev_cmp != NULL, "should have found comp instruction for branch"); - if (prev_branch->block() == code->at(i + 1) && prev_branch->info() == NULL) { - - TRACE_LINEAR_SCAN(3, tty->print_cr("Negating conditional branch and deleting unconditional branch at end of block B%d", block->block_id())); - - // eliminate a conditional branch to the immediate successor - prev_branch->change_block(last_branch->block()); - prev_branch->negate_cond(); - prev_cmp->set_condition(prev_branch->cond()); - instructions->truncate(instructions->length() - 1); + assert(prev_cmp != NULL, "should have found comp instruction for branch"); + if (prev_branch->block() == code->at(i + 1) && prev_branch->info() == NULL) { + + TRACE_LINEAR_SCAN(3, tty->print_cr("Negating conditional branch and deleting unconditional branch at end of block B%d", block->block_id())); + + // eliminate a conditional branch to the immediate successor + prev_branch->change_block(last_branch->block()); + prev_branch->negate_cond(); + prev_cmp->set_condition(prev_branch->cond()); + instructions->truncate(instructions->length() - 1); + } } } } diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_Optimizer.cpp --- a/src/share/vm/c1/c1_Optimizer.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_Optimizer.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -178,7 +178,7 @@ // 2) substitute conditional expression // with an IfOp followed by a Goto // cut if_ away and get node before - Instruction* cur_end = if_->prev(block); + Instruction* cur_end = if_->prev(); // append constants of true- and false-block if necessary // clone constants because original block must not be destroyed @@ -202,7 +202,7 @@ } // append Goto to successor - ValueStack* state_before = if_->is_safepoint() ? if_->state_before() : NULL; + ValueStack* state_before = if_->state_before(); Goto* goto_ = new Goto(sux, state_before, if_->is_safepoint() || t_goto->is_safepoint() || f_goto->is_safepoint()); // prepare state for Goto @@ -367,10 +367,11 @@ #endif // find instruction before end & append first instruction of sux block - Instruction* prev = end->prev(block); + Instruction* prev = end->prev(); Instruction* next = sux->next(); assert(prev->as_BlockEnd() == NULL, "must not be a BlockEnd"); prev->set_next(next); + prev->fixup_block_pointers(); sux->disconnect_from_graph(); block->set_end(sux->end()); // add exception handlers of deleted block, if any @@ -533,6 +534,8 @@ void do_ProfileInvoke (ProfileInvoke* x); void do_RuntimeCall (RuntimeCall* x); void do_MemBar (MemBar* x); + void do_RangeCheckPredicate(RangeCheckPredicate* x); + void do_Assert (Assert* x); }; @@ -714,6 +717,8 @@ void NullCheckVisitor::do_ProfileInvoke (ProfileInvoke* x) {} void NullCheckVisitor::do_RuntimeCall (RuntimeCall* x) {} void NullCheckVisitor::do_MemBar (MemBar* x) {} +void NullCheckVisitor::do_RangeCheckPredicate(RangeCheckPredicate* x) {} +void NullCheckVisitor::do_Assert (Assert* x) {} void NullCheckEliminator::visit(Value* p) { diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_RangeCheckElimination.cpp --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/src/share/vm/c1/c1_RangeCheckElimination.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -0,0 +1,1517 @@ +/* + * Copyright (c) 2012, 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. + * + */ + +#include "precompiled.hpp" +#include "c1/c1_ValueStack.hpp" +#include "c1/c1_RangeCheckElimination.hpp" +#include "c1/c1_IR.hpp" +#include "c1/c1_Canonicalizer.hpp" +#include "c1/c1_ValueMap.hpp" +#include "ci/ciMethodData.hpp" +#include "runtime/deoptimization.hpp" + +// Macros for the Trace and the Assertion flag +#ifdef ASSERT +#define TRACE_RANGE_CHECK_ELIMINATION(code) if (TraceRangeCheckElimination) { code; } +#define ASSERT_RANGE_CHECK_ELIMINATION(code) if (AssertRangeCheckElimination) { code; } +#define TRACE_OR_ASSERT_RANGE_CHECK_ELIMINATION(code) if (TraceRangeCheckElimination || AssertRangeCheckElimination) { code; } +#else +#define TRACE_RANGE_CHECK_ELIMINATION(code) +#define ASSERT_RANGE_CHECK_ELIMINATION(code) +#define TRACE_OR_ASSERT_RANGE_CHECK_ELIMINATION(code) +#endif + +// Entry point for the optimization +void RangeCheckElimination::eliminate(IR *ir) { + bool do_elimination = ir->compilation()->has_access_indexed(); + ASSERT_RANGE_CHECK_ELIMINATION(do_elimination = true); + if (do_elimination) { + RangeCheckEliminator rce(ir); + } +} + +// Constructor +RangeCheckEliminator::RangeCheckEliminator(IR *ir) : + _bounds(Instruction::number_of_instructions(), NULL), + _access_indexed_info(Instruction::number_of_instructions(), NULL) +{ + _visitor.set_range_check_eliminator(this); + _ir = ir; + _number_of_instructions = Instruction::number_of_instructions(); + _optimistic = ir->compilation()->is_optimistic(); + + TRACE_RANGE_CHECK_ELIMINATION( + tty->print_cr(""); + tty->print_cr("Range check elimination"); + ir->method()->print_name(tty); + tty->print_cr(""); + ); + + TRACE_RANGE_CHECK_ELIMINATION( + tty->print_cr("optimistic=%d", (int)_optimistic); + ); + +#ifdef ASSERT + // Verifies several conditions that must be true on the IR-input. Only used for debugging purposes. + TRACE_RANGE_CHECK_ELIMINATION( + tty->print_cr("Verification of IR . . ."); + ); + Verification verification(ir); +#endif + + // Set process block flags + // Optimization so a blocks is only processed if it contains an access indexed instruction or if + // one of its children in the dominator tree contains an access indexed instruction. + set_process_block_flags(ir->start()); + + // Pass over instructions in the dominator tree + TRACE_RANGE_CHECK_ELIMINATION( + tty->print_cr("Starting pass over dominator tree . . .") + ); + calc_bounds(ir->start(), NULL); + + TRACE_RANGE_CHECK_ELIMINATION( + tty->print_cr("Finished!") + ); +} + +// Instruction specific work for some instructions +// Constant +void RangeCheckEliminator::Visitor::do_Constant(Constant *c) { + IntConstant *ic = c->type()->as_IntConstant(); + if (ic != NULL) { + int value = ic->value(); + _bound = new Bound(value, NULL, value, NULL); + } +} + +// LogicOp +void RangeCheckEliminator::Visitor::do_LogicOp(LogicOp *lo) { + if (lo->type()->as_IntType() && lo->op() == Bytecodes::_iand && (lo->x()->as_Constant() || lo->y()->as_Constant())) { + int constant = 0; + Constant *c = lo->x()->as_Constant(); + if (c != NULL) { + constant = c->type()->as_IntConstant()->value(); + } else { + constant = lo->y()->as_Constant()->type()->as_IntConstant()->value(); + } + if (constant >= 0) { + _bound = new Bound(0, NULL, constant, NULL); + } + } +} + +// Phi +void RangeCheckEliminator::Visitor::do_Phi(Phi *phi) { + if (!phi->type()->as_IntType() && !phi->type()->as_ObjectType()) return; + + BlockBegin *block = phi->block(); + int op_count = phi->operand_count(); + bool has_upper = true; + bool has_lower = true; + assert(phi, "Phi must not be null"); + Bound *bound = NULL; + + // TODO: support more difficult phis + for (int i=0; ioperand_at(i); + + if (v == phi) continue; + + // Check if instruction is connected with phi itself + Op2 *op2 = v->as_Op2(); + if (op2 != NULL) { + Value x = op2->x(); + Value y = op2->y(); + if ((x == phi || y == phi)) { + Value other = x; + if (other == phi) { + other = y; + } + ArithmeticOp *ao = v->as_ArithmeticOp(); + if (ao != NULL && ao->op() == Bytecodes::_iadd) { + assert(ao->op() == Bytecodes::_iadd, "Has to be add!"); + if (ao->type()->as_IntType()) { + Constant *c = other->as_Constant(); + if (c != NULL) { + assert(c->type()->as_IntConstant(), "Constant has to be of type integer"); + int value = c->type()->as_IntConstant()->value(); + if (value == 1) { + has_upper = false; + } else if (value > 1) { + // Overflow not guaranteed + has_upper = false; + has_lower = false; + } else if (value < 0) { + has_lower = false; + } + continue; + } + } + } + } + } + + // No connection -> new bound + Bound *v_bound = _rce->get_bound(v); + Bound *cur_bound; + int cur_constant = 0; + Value cur_value = v; + + if (v->type()->as_IntConstant()) { + cur_constant = v->type()->as_IntConstant()->value(); + cur_value = NULL; + } + if (!v_bound->has_upper() || !v_bound->has_lower()) { + cur_bound = new Bound(cur_constant, cur_value, cur_constant, cur_value); + } else { + cur_bound = v_bound; + } + if (cur_bound) { + if (!bound) { + bound = cur_bound->copy(); + } else { + bound->or_op(cur_bound); + } + } else { + // No bound! + bound = NULL; + break; + } + } + + if (bound) { + if (!has_upper) { + bound->remove_upper(); + } + if (!has_lower) { + bound->remove_lower(); + } + _bound = bound; + } else { + _bound = new Bound(); + } +} + + +// ArithmeticOp +void RangeCheckEliminator::Visitor::do_ArithmeticOp(ArithmeticOp *ao) { + Value x = ao->x(); + Value y = ao->y(); + + if (ao->op() == Bytecodes::_irem) { + Bound* x_bound = _rce->get_bound(x); + Bound* y_bound = _rce->get_bound(y); + if (x_bound->lower() >= 0 && x_bound->lower_instr() == NULL && y->as_ArrayLength() != NULL) { + _bound = new Bound(0, NULL, -1, y); + } else { + _bound = new Bound(); + } + } else if (!x->as_Constant() || !y->as_Constant()) { + assert(!x->as_Constant() || !y->as_Constant(), "One of the operands must be non-constant!"); + if (((x->as_Constant() || y->as_Constant()) && (ao->op() == Bytecodes::_iadd)) || (y->as_Constant() && ao->op() == Bytecodes::_isub)) { + assert(ao->op() == Bytecodes::_iadd || ao->op() == Bytecodes::_isub, "Operand must be iadd or isub"); + + if (y->as_Constant()) { + Value tmp = x; + x = y; + y = tmp; + } + assert(x->as_Constant()->type()->as_IntConstant(), "Constant must be int constant!"); + + // Constant now in x + int const_value = x->as_Constant()->type()->as_IntConstant()->value(); + if (ao->op() == Bytecodes::_iadd || const_value != min_jint) { + if (ao->op() == Bytecodes::_isub) { + const_value = -const_value; + } + + Bound * bound = _rce->get_bound(y); + if (bound->has_upper() && bound->has_lower()) { + int new_lower = bound->lower() + const_value; + jlong new_lowerl = ((jlong)bound->lower()) + const_value; + int new_upper = bound->upper() + const_value; + jlong new_upperl = ((jlong)bound->upper()) + const_value; + + if (((jlong)new_lower) == new_lowerl && ((jlong)new_upper == new_upperl)) { + Bound *newBound = new Bound(new_lower, bound->lower_instr(), new_upper, bound->upper_instr()); + _bound = newBound; + } else { + // overflow + _bound = new Bound(); + } + } else { + _bound = new Bound(); + } + } else { + _bound = new Bound(); + } + } else { + Bound *bound = _rce->get_bound(x); + if (ao->op() == Bytecodes::_isub) { + if (bound->lower_instr() == y) { + _bound = new Bound(Instruction::geq, NULL, bound->lower()); + } else { + _bound = new Bound(); + } + } else { + _bound = new Bound(); + } + } + } +} + +// IfOp +void RangeCheckEliminator::Visitor::do_IfOp(IfOp *ifOp) +{ + if (ifOp->tval()->type()->as_IntConstant() && ifOp->fval()->type()->as_IntConstant()) { + int min = ifOp->tval()->type()->as_IntConstant()->value(); + int max = ifOp->fval()->type()->as_IntConstant()->value(); + if (min > max) { + // min ^= max ^= min ^= max; + int tmp = min; + min = max; + max = tmp; + } + _bound = new Bound(min, NULL, max, NULL); + } +} + +// Get bound. Returns the current bound on Value v. Normally this is the topmost element on the bound stack. +RangeCheckEliminator::Bound *RangeCheckEliminator::get_bound(Value v) { + // Wrong type or NULL -> No bound + if (!v || (!v->type()->as_IntType() && !v->type()->as_ObjectType())) return NULL; + + if (!_bounds[v->id()]) { + // First (default) bound is calculated + // Create BoundStack + _bounds[v->id()] = new BoundStack(); + _visitor.clear_bound(); + Value visit_value = v; + visit_value->visit(&_visitor); + Bound *bound = _visitor.bound(); + if (bound) { + _bounds[v->id()]->push(bound); + } + if (_bounds[v->id()]->length() == 0) { + assert(!(v->as_Constant() && v->type()->as_IntConstant()), "constants not handled here"); + _bounds[v->id()]->push(new Bound()); + } + } else if (_bounds[v->id()]->length() == 0) { + // To avoid endless loops, bound is currently in calculation -> nothing known about it + return new Bound(); + } + + // Return bound + return _bounds[v->id()]->top(); +} + +// Update bound +void RangeCheckEliminator::update_bound(IntegerStack &pushed, Value v, Instruction::Condition cond, Value value, int constant) { + if (cond == Instruction::gtr) { + cond = Instruction::geq; + constant++; + } else if (cond == Instruction::lss) { + cond = Instruction::leq; + constant--; + } + Bound *bound = new Bound(cond, value, constant); + update_bound(pushed, v, bound); +} + +// Checks for loop invariance. Returns true if the instruction is outside of the loop which is identified by loop_header. +bool RangeCheckEliminator::loop_invariant(BlockBegin *loop_header, Instruction *instruction) { + assert(loop_header, "Loop header must not be null!"); + if (!instruction) return true; + return instruction->dominator_depth() < loop_header->dominator_depth(); +} + +// Update bound. Pushes a new bound onto the stack. Tries to do a conjunction with the current bound. +void RangeCheckEliminator::update_bound(IntegerStack &pushed, Value v, Bound *bound) { + if (v->as_Constant()) { + // No bound update for constants + return; + } + if (!_bounds[v->id()]) { + get_bound(v); + assert(_bounds[v->id()], "Now Stack must exist"); + } + Bound *top = NULL; + if (_bounds[v->id()]->length() > 0) { + top = _bounds[v->id()]->top(); + } + if (top) { + bound->and_op(top); + } + _bounds[v->id()]->push(bound); + pushed.append(v->id()); +} + +// Add instruction + idx for in block motion +void RangeCheckEliminator::add_access_indexed_info(InstructionList &indices, int idx, Value instruction, AccessIndexed *ai) { + int id = instruction->id(); + AccessIndexedInfo *aii = _access_indexed_info[id]; + if (aii == NULL) { + aii = new AccessIndexedInfo(); + _access_indexed_info[id] = aii; + indices.append(instruction); + aii->_min = idx; + aii->_max = idx; + aii->_list = new AccessIndexedList(); + } else if (idx >= aii->_min && idx <= aii->_max) { + remove_range_check(ai); + return; + } + aii->_min = MIN2(aii->_min, idx); + aii->_max = MAX2(aii->_max, idx); + aii->_list->append(ai); +} + +// In block motion. Tries to reorder checks in order to reduce some of them. +// Example: +// a[i] = 0; +// a[i+2] = 0; +// a[i+1] = 0; +// In this example the check for a[i+1] would be considered as unnecessary during the first iteration. +// After this i is only checked once for i >= 0 and i+2 < a.length before the first array access. If this +// check fails, deoptimization is called. +void RangeCheckEliminator::in_block_motion(BlockBegin *block, AccessIndexedList &accessIndexed, InstructionList &arrays) { + InstructionList indices; + + // Now iterate over all arrays + for (int i=0; iarray() != array || !ai->check_flag(Instruction::NeedsRangeCheckFlag)) continue; + + Value index = ai->index(); + Constant *c = index->as_Constant(); + if (c != NULL) { + int constant_value = c->type()->as_IntConstant()->value(); + if (constant_value >= 0) { + if (constant_value <= max_constant) { + // No range check needed for this + remove_range_check(ai); + } else { + max_constant = constant_value; + list_constant.append(ai); + } + } + } else { + int last_integer = 0; + Instruction *last_instruction = index; + int base = 0; + ArithmeticOp *ao = index->as_ArithmeticOp(); + + while (ao != NULL && (ao->x()->as_Constant() || ao->y()->as_Constant()) && (ao->op() == Bytecodes::_iadd || ao->op() == Bytecodes::_isub)) { + c = ao->y()->as_Constant(); + Instruction *other = ao->x(); + if (!c && ao->op() == Bytecodes::_iadd) { + c = ao->x()->as_Constant(); + other = ao->y(); + } + + if (c) { + int value = c->type()->as_IntConstant()->value(); + if (value != min_jint) { + if (ao->op() == Bytecodes::_isub) { + value = -value; + } + base += value; + last_integer = base; + last_instruction = other; + } + index = other; + } else { + break; + } + ao = index->as_ArithmeticOp(); + } + add_access_indexed_info(indices, last_integer, last_instruction, ai); + } + } + + // Iterate over all different indices + if (_optimistic) { + for (int i=0; iid()]; + assert(info != NULL, "Info must not be null"); + + // if idx < 0, max > 0, max + idx may fall between 0 and + // length-1 and if min < 0, min + idx may overflow and be >= + // 0. The predicate wouldn't trigger but some accesses could + // be with a negative index. This test guarantees that for the + // min and max value that are kept the predicate can't let + // some incorrect accesses happen. + bool range_cond = (info->_max < 0 || info->_max + min_jint <= info->_min); + + // Generate code only if more than 2 range checks can be eliminated because of that. + // 2 because at least 2 comparisons are done + if (info->_list->length() > 2 && range_cond) { + AccessIndexed *first = info->_list->at(0); + Instruction *insert_position = first->prev(); + assert(insert_position->next() == first, "prev was calculated"); + ValueStack *state = first->state_before(); + + // Load min Constant + Constant *min_constant = NULL; + if (info->_min != 0) { + min_constant = new Constant(new IntConstant(info->_min)); + NOT_PRODUCT(min_constant->set_printable_bci(first->printable_bci())); + insert_position = insert_position->insert_after(min_constant); + } + + // Load max Constant + Constant *max_constant = NULL; + if (info->_max != 0) { + max_constant = new Constant(new IntConstant(info->_max)); + NOT_PRODUCT(max_constant->set_printable_bci(first->printable_bci())); + insert_position = insert_position->insert_after(max_constant); + } + + // Load array length + Value length_instr = first->length(); + if (!length_instr) { + ArrayLength *length = new ArrayLength(array, first->state_before()->copy()); + length->set_exception_state(length->state_before()); + length->set_flag(Instruction::DeoptimizeOnException, true); + insert_position = insert_position->insert_after_same_bci(length); + length_instr = length; + } + + // Calculate lower bound + Instruction *lower_compare = index_instruction; + if (min_constant) { + ArithmeticOp *ao = new ArithmeticOp(Bytecodes::_iadd, min_constant, lower_compare, false, NULL); + insert_position = insert_position->insert_after_same_bci(ao); + lower_compare = ao; + } + + // Calculate upper bound + Instruction *upper_compare = index_instruction; + if (max_constant) { + ArithmeticOp *ao = new ArithmeticOp(Bytecodes::_iadd, max_constant, upper_compare, false, NULL); + insert_position = insert_position->insert_after_same_bci(ao); + upper_compare = ao; + } + + // Trick with unsigned compare is done + int bci = NOT_PRODUCT(first->printable_bci()) PRODUCT_ONLY(-1); + insert_position = predicate(upper_compare, Instruction::aeq, length_instr, state, insert_position, bci); + insert_position = predicate_cmp_with_const(lower_compare, Instruction::leq, -1, state, insert_position); + for (int j = 0; j_list->length(); j++) { + AccessIndexed *ai = info->_list->at(j); + remove_range_check(ai); + } + } + _access_indexed_info[index_instruction->id()] = NULL; + } + indices.clear(); + + if (list_constant.length() > 1) { + AccessIndexed *first = list_constant.at(0); + Instruction *insert_position = first->prev(); + ValueStack *state = first->state_before(); + // Load max Constant + Constant *constant = new Constant(new IntConstant(max_constant)); + NOT_PRODUCT(constant->set_printable_bci(first->printable_bci())); + insert_position = insert_position->insert_after(constant); + Instruction *compare_instr = constant; + Value length_instr = first->length(); + if (!length_instr) { + ArrayLength *length = new ArrayLength(array, state->copy()); + length->set_exception_state(length->state_before()); + length->set_flag(Instruction::DeoptimizeOnException, true); + insert_position = insert_position->insert_after_same_bci(length); + length_instr = length; + } + // Compare for greater or equal to array length + insert_position = predicate(compare_instr, Instruction::geq, length_instr, state, insert_position); + for (int j = 0; jas_AccessIndexed() != NULL); + cur = cur->next(); + } + + BlockList *dominates = block->dominates(); + for (int i=0; ilength(); i++) { + BlockBegin *next = dominates->at(i); + process |= set_process_block_flags(next); + } + + if (!process) { + block->set(BlockBegin::donot_eliminate_range_checks); + } + return process; +} + +bool RangeCheckEliminator::is_ok_for_deoptimization(Instruction *insert_position, Instruction *array_instr, Instruction *length_instr, Instruction *lower_instr, int lower, Instruction *upper_instr, int upper) { + bool upper_check = true; + assert(lower_instr || lower >= 0, "If no lower_instr present, lower must be greater 0"); + assert(!lower_instr || lower_instr->dominator_depth() <= insert_position->dominator_depth(), "Dominator depth must be smaller"); + assert(!upper_instr || upper_instr->dominator_depth() <= insert_position->dominator_depth(), "Dominator depth must be smaller"); + assert(array_instr, "Array instruction must exist"); + assert(array_instr->dominator_depth() <= insert_position->dominator_depth(), "Dominator depth must be smaller"); + assert(!length_instr || length_instr->dominator_depth() <= insert_position->dominator_depth(), "Dominator depth must be smaller"); + + if (upper_instr && upper_instr->as_ArrayLength() && upper_instr->as_ArrayLength()->array() == array_instr) { + // static check + if (upper >= 0) return false; // would always trigger a deopt: + // array_length + x >= array_length, x >= 0 is always true + upper_check = false; + } + if (lower_instr && lower_instr->as_ArrayLength() && lower_instr->as_ArrayLength()->array() == array_instr) { + if (lower > 0) return false; + } + // No upper check required -> skip + if (upper_check && upper_instr && upper_instr->type()->as_ObjectType() && upper_instr == array_instr) { + // upper_instr is object means that the upper bound is the length + // of the upper_instr. + return false; + } + return true; +} + +Instruction* RangeCheckEliminator::insert_after(Instruction* insert_position, Instruction* instr, int bci) { + if (bci != -1) { + NOT_PRODUCT(instr->set_printable_bci(bci)); + return insert_position->insert_after(instr); + } else { + return insert_position->insert_after_same_bci(instr); + } +} + +Instruction* RangeCheckEliminator::predicate(Instruction* left, Instruction::Condition cond, Instruction* right, ValueStack* state, Instruction *insert_position, int bci) { + RangeCheckPredicate *deoptimize = new RangeCheckPredicate(left, cond, true, right, state->copy()); + return insert_after(insert_position, deoptimize, bci); +} + +Instruction* RangeCheckEliminator::predicate_cmp_with_const(Instruction* instr, Instruction::Condition cond, int constant, ValueStack* state, Instruction *insert_position, int bci) { + Constant *const_instr = new Constant(new IntConstant(constant)); + insert_position = insert_after(insert_position, const_instr, bci); + return predicate(instr, cond, const_instr, state, insert_position); +} + +Instruction* RangeCheckEliminator::predicate_add(Instruction* left, int left_const, Instruction::Condition cond, Instruction* right, ValueStack* state, Instruction *insert_position, int bci) { + Constant *constant = new Constant(new IntConstant(left_const)); + insert_position = insert_after(insert_position, constant, bci); + ArithmeticOp *ao = new ArithmeticOp(Bytecodes::_iadd, constant, left, false, NULL); + insert_position = insert_position->insert_after_same_bci(ao); + return predicate(ao, cond, right, state, insert_position); +} + +Instruction* RangeCheckEliminator::predicate_add_cmp_with_const(Instruction* left, int left_const, Instruction::Condition cond, int constant, ValueStack* state, Instruction *insert_position, int bci) { + Constant *const_instr = new Constant(new IntConstant(constant)); + insert_position = insert_after(insert_position, const_instr, bci); + return predicate_add(left, left_const, cond, const_instr, state, insert_position); +} + +// Insert deoptimization +void RangeCheckEliminator::insert_deoptimization(ValueStack *state, Instruction *insert_position, Instruction *array_instr, Instruction *length_instr, Instruction *lower_instr, int lower, Instruction *upper_instr, int upper, AccessIndexed *ai) { + assert(is_ok_for_deoptimization(insert_position, array_instr, length_instr, lower_instr, lower, upper_instr, upper), "should have been tested before"); + bool upper_check = !(upper_instr && upper_instr->as_ArrayLength() && upper_instr->as_ArrayLength()->array() == array_instr); + + int bci = NOT_PRODUCT(ai->printable_bci()) PRODUCT_ONLY(-1); + if (lower_instr) { + assert(!lower_instr->type()->as_ObjectType(), "Must not be object type"); + if (lower == 0) { + // Compare for less than 0 + insert_position = predicate_cmp_with_const(lower_instr, Instruction::lss, 0, state, insert_position, bci); + } else if (lower > 0) { + // Compare for smaller 0 + insert_position = predicate_add_cmp_with_const(lower_instr, lower, Instruction::lss, 0, state, insert_position, bci); + } else { + assert(lower < 0, ""); + // Add 1 + lower++; + lower = -lower; + // Compare for smaller or equal 0 + insert_position = predicate_cmp_with_const(lower_instr, Instruction::leq, lower, state, insert_position, bci); + } + } + + // No upper check required -> skip + if (!upper_check) return; + + // We need to know length of array + if (!length_instr) { + // Load length if necessary + ArrayLength *length = new ArrayLength(array_instr, state->copy()); + NOT_PRODUCT(length->set_printable_bci(ai->printable_bci())); + length->set_exception_state(length->state_before()); + length->set_flag(Instruction::DeoptimizeOnException, true); + insert_position = insert_position->insert_after(length); + length_instr = length; + } + + if (!upper_instr) { + // Compare for geq array.length + insert_position = predicate_cmp_with_const(length_instr, Instruction::leq, upper, state, insert_position, bci); + } else { + if (upper_instr->type()->as_ObjectType()) { + assert(state, "must not be null"); + assert(upper_instr != array_instr, "should be"); + ArrayLength *length = new ArrayLength(upper_instr, state->copy()); + NOT_PRODUCT(length->set_printable_bci(ai->printable_bci())); + length->set_flag(Instruction::DeoptimizeOnException, true); + length->set_exception_state(length->state_before()); + insert_position = insert_position->insert_after(length); + upper_instr = length; + } + assert(upper_instr->type()->as_IntType(), "Must not be object type!"); + + if (upper == 0) { + // Compare for geq array.length + insert_position = predicate(upper_instr, Instruction::geq, length_instr, state, insert_position, bci); + } else if (upper < 0) { + // Compare for geq array.length + insert_position = predicate_add(upper_instr, upper, Instruction::geq, length_instr, state, insert_position, bci); + } else { + assert(upper > 0, ""); + upper = -upper; + // Compare for geq array.length + insert_position = predicate_add(length_instr, upper, Instruction::leq, upper_instr, state, insert_position, bci); + } + } +} + +// Add if condition +void RangeCheckEliminator::add_if_condition(IntegerStack &pushed, Value x, Value y, Instruction::Condition condition) { + if (y->as_Constant()) return; + + int const_value = 0; + Value instr_value = x; + Constant *c = x->as_Constant(); + ArithmeticOp *ao = x->as_ArithmeticOp(); + + if (c != NULL) { + const_value = c->type()->as_IntConstant()->value(); + instr_value = NULL; + } else if (ao != NULL && (!ao->x()->as_Constant() || !ao->y()->as_Constant()) && ((ao->op() == Bytecodes::_isub && ao->y()->as_Constant()) || ao->op() == Bytecodes::_iadd)) { + assert(!ao->x()->as_Constant() || !ao->y()->as_Constant(), "At least one operator must be non-constant!"); + assert(ao->op() == Bytecodes::_isub || ao->op() == Bytecodes::_iadd, "Operation has to be add or sub!"); + c = ao->x()->as_Constant(); + if (c != NULL) { + const_value = c->type()->as_IntConstant()->value(); + instr_value = ao->y(); + } else { + c = ao->y()->as_Constant(); + if (c != NULL) { + const_value = c->type()->as_IntConstant()->value(); + instr_value = ao->x(); + } + } + if (ao->op() == Bytecodes::_isub) { + assert(ao->y()->as_Constant(), "1 - x not supported, only x - 1 is valid!"); + if (const_value > min_jint) { + const_value = -const_value; + } else { + const_value = 0; + instr_value = x; + } + } + } + + update_bound(pushed, y, condition, instr_value, const_value); +} + +// Process If +void RangeCheckEliminator::process_if(IntegerStack &pushed, BlockBegin *block, If *cond) { + // Only if we are direct true / false successor and NOT both ! (even this may occur) + if ((cond->tsux() == block || cond->fsux() == block) && cond->tsux() != cond->fsux()) { + Instruction::Condition condition = cond->cond(); + if (cond->fsux() == block) { + condition = Instruction::negate(condition); + } + Value x = cond->x(); + Value y = cond->y(); + if (x->type()->as_IntType() && y->type()->as_IntType()) { + add_if_condition(pushed, y, x, condition); + add_if_condition(pushed, x, y, Instruction::mirror(condition)); + } + } +} + +// Process access indexed +void RangeCheckEliminator::process_access_indexed(BlockBegin *loop_header, BlockBegin *block, AccessIndexed *ai) { + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2) + ); + TRACE_RANGE_CHECK_ELIMINATION( + tty->print_cr("Access indexed: index=%d length=%d", ai->index()->id(), (ai->length() != NULL ? ai->length()->id() :-1 )) + ); + + if (ai->check_flag(Instruction::NeedsRangeCheckFlag)) { + Bound *index_bound = get_bound(ai->index()); + if (!index_bound->has_lower() || !index_bound->has_upper()) { + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("Index instruction %d has no lower and/or no upper bound!", ai->index()->id()) + ); + return; + } + + Bound *array_bound; + if (ai->length()) { + array_bound = get_bound(ai->length()); + } else { + array_bound = get_bound(ai->array()); + } + + if (in_array_bound(index_bound, ai->array()) || + (index_bound && array_bound && index_bound->is_smaller(array_bound) && !index_bound->lower_instr() && index_bound->lower() >= 0)) { + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("Bounds check for instruction %d in block B%d can be fully eliminated!", ai->id(), ai->block()->block_id()) + ); + + remove_range_check(ai); + } else if (_optimistic && loop_header) { + assert(ai->array(), "Array must not be null!"); + assert(ai->index(), "Index must not be null!"); + + // Array instruction + Instruction *array_instr = ai->array(); + if (!loop_invariant(loop_header, array_instr)) { + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("Array %d is not loop invariant to header B%d", ai->array()->id(), loop_header->block_id()) + ); + return; + } + + // Lower instruction + Value index_instr = ai->index(); + Value lower_instr = index_bound->lower_instr(); + if (!loop_invariant(loop_header, lower_instr)) { + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("Lower instruction %d not loop invariant!", lower_instr->id()) + ); + return; + } + if (!lower_instr && index_bound->lower() < 0) { + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("Lower bound smaller than 0 (%d)!", index_bound->lower()) + ); + return; + } + + // Upper instruction + Value upper_instr = index_bound->upper_instr(); + if (!loop_invariant(loop_header, upper_instr)) { + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("Upper instruction %d not loop invariant!", upper_instr->id()) + ); + return; + } + + // Length instruction + Value length_instr = ai->length(); + if (!loop_invariant(loop_header, length_instr)) { + // Generate length instruction yourself! + length_instr = NULL; + } + + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("LOOP INVARIANT access indexed %d found in block B%d!", ai->id(), ai->block()->block_id()) + ); + + BlockBegin *pred_block = loop_header->dominator(); + assert(pred_block != NULL, "Every loop header has a dominator!"); + BlockEnd *pred_block_end = pred_block->end(); + Instruction *insert_position = pred_block_end->prev(); + ValueStack *state = pred_block_end->state_before(); + if (pred_block_end->as_Goto() && state == NULL) state = pred_block_end->state(); + assert(state, "State must not be null"); + + // Add deoptimization to dominator of loop header + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("Inserting deopt at bci %d in block B%d!", state->bci(), insert_position->block()->block_id()) + ); + + if (!is_ok_for_deoptimization(insert_position, array_instr, length_instr, lower_instr, index_bound->lower(), upper_instr, index_bound->upper())) { + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("Could not eliminate because of static analysis!") + ); + return; + } + + insert_deoptimization(state, insert_position, array_instr, length_instr, lower_instr, index_bound->lower(), upper_instr, index_bound->upper(), ai); + + // Finally remove the range check! + remove_range_check(ai); + } + } +} + +void RangeCheckEliminator::remove_range_check(AccessIndexed *ai) { + ai->set_flag(Instruction::NeedsRangeCheckFlag, false); + // no range check, no need for the length instruction anymore + ai->clear_length(); + + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(ai->dominator_depth()*2); + tty->print_cr("Range check for instruction %d eliminated!", ai->id()); + ); + + ASSERT_RANGE_CHECK_ELIMINATION( + Value array_length = ai->length(); + if (!array_length) { + array_length = ai->array(); + assert(array_length->type()->as_ObjectType(), "Has to be object type!"); + } + int cur_constant = -1; + Value cur_value = array_length; + if (cur_value->type()->as_IntConstant()) { + cur_constant += cur_value->type()->as_IntConstant()->value(); + cur_value = NULL; + } + Bound *new_index_bound = new Bound(0, NULL, cur_constant, cur_value); + add_assertions(new_index_bound, ai->index(), ai); + ); +} + +// Calculate bounds for instruction in this block and children blocks in the dominator tree +void RangeCheckEliminator::calc_bounds(BlockBegin *block, BlockBegin *loop_header) { + // Ensures a valid loop_header + assert(!loop_header || loop_header->is_set(BlockBegin::linear_scan_loop_header_flag), "Loop header has to be real !"); + + // Tracing output + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("Block B%d", block->block_id()); + ); + + // Pushed stack for conditions + IntegerStack pushed; + // Process If + BlockBegin *parent = block->dominator(); + if (parent != NULL) { + If *cond = parent->end()->as_If(); + if (cond != NULL) { + process_if(pushed, block, cond); + } + } + + // Interate over current block + InstructionList arrays; + AccessIndexedList accessIndexed; + Instruction *cur = block; + + while (cur) { + // Ensure cur wasn't inserted during the elimination + if (cur->id() < this->_bounds.length()) { + // Process only if it is an access indexed instruction + AccessIndexed *ai = cur->as_AccessIndexed(); + if (ai != NULL) { + process_access_indexed(loop_header, block, ai); + accessIndexed.append(ai); + if (!arrays.contains(ai->array())) { + arrays.append(ai->array()); + } + Bound *b = get_bound(ai->index()); + if (!b->lower_instr()) { + // Lower bound is constant + update_bound(pushed, ai->index(), Instruction::geq, NULL, 0); + } + if (!b->has_upper()) { + if (ai->length() && ai->length()->type()->as_IntConstant()) { + int value = ai->length()->type()->as_IntConstant()->value(); + update_bound(pushed, ai->index(), Instruction::lss, NULL, value); + } else { + // Has no upper bound + Instruction *instr = ai->length(); + if (instr != NULL) instr = ai->array(); + update_bound(pushed, ai->index(), Instruction::lss, instr, 0); + } + } + } + } + cur = cur->next(); + } + + // Output current condition stack + TRACE_RANGE_CHECK_ELIMINATION(dump_condition_stack(block)); + + // Do in block motion of range checks + in_block_motion(block, accessIndexed, arrays); + + // Call all dominated blocks + for (int i=0; idominates()->length(); i++) { + BlockBegin *next = block->dominates()->at(i); + if (!next->is_set(BlockBegin::donot_eliminate_range_checks)) { + // if current block is a loop header and: + // - next block belongs to the same loop + // or + // - next block belongs to an inner loop + // then current block is the loop header for next block + if (block->is_set(BlockBegin::linear_scan_loop_header_flag) && (block->loop_index() == next->loop_index() || next->loop_depth() > block->loop_depth())) { + calc_bounds(next, block); + } else { + calc_bounds(next, loop_header); + } + } + } + + // Reset stack + for (int i=0; ipop(); + } +} + +#ifndef PRODUCT +// Dump condition stack +void RangeCheckEliminator::dump_condition_stack(BlockBegin *block) { + for (int i=0; i<_ir->linear_scan_order()->length(); i++) { + BlockBegin *cur_block = _ir->linear_scan_order()->at(i); + Instruction *instr = cur_block; + for_each_phi_fun(cur_block, phi, + BoundStack *bound_stack = _bounds.at(phi->id()); + if (bound_stack && bound_stack->length() > 0) { + Bound *bound = bound_stack->top(); + if ((bound->has_lower() || bound->has_upper()) && (bound->lower_instr() != phi || bound->upper_instr() != phi || bound->lower() != 0 || bound->upper() != 0)) { + TRACE_RANGE_CHECK_ELIMINATION(tty->fill_to(2*block->dominator_depth()); + tty->print("i%d", phi->id()); + tty->print(": "); + bound->print(); + tty->print_cr(""); + ); + } + }); + + while (!instr->as_BlockEnd()) { + if (instr->id() < _bounds.length()) { + BoundStack *bound_stack = _bounds.at(instr->id()); + if (bound_stack && bound_stack->length() > 0) { + Bound *bound = bound_stack->top(); + if ((bound->has_lower() || bound->has_upper()) && (bound->lower_instr() != instr || bound->upper_instr() != instr || bound->lower() != 0 || bound->upper() != 0)) { + TRACE_RANGE_CHECK_ELIMINATION(tty->fill_to(2*block->dominator_depth()); + tty->print("i%d", instr->id()); + tty->print(": "); + bound->print(); + tty->print_cr(""); + ); + } + } + } + instr = instr->next(); + } + } +} +#endif + +// Verification or the IR +RangeCheckEliminator::Verification::Verification(IR *ir) : _used(BlockBegin::number_of_blocks(), false) { + this->_ir = ir; + ir->iterate_linear_scan_order(this); +} + +// Verify this block +void RangeCheckEliminator::Verification::block_do(BlockBegin *block) { + If *cond = block->end()->as_If(); + // Watch out: tsux and fsux can be the same! + if (block->number_of_sux() > 1) { + for (int i=0; inumber_of_sux(); i++) { + BlockBegin *sux = block->sux_at(i); + BlockBegin *pred = NULL; + for (int j=0; jnumber_of_preds(); j++) { + BlockBegin *cur = sux->pred_at(j); + assert(cur != NULL, "Predecessor must not be null"); + if (!pred) { + pred = cur; + } + assert(cur == pred, "Block must not have more than one predecessor if its predecessor has more than one successor"); + } + assert(sux->number_of_preds() >= 1, "Block must have at least one predecessor"); + assert(sux->pred_at(0) == block, "Wrong successor"); + } + } + + BlockBegin *dominator = block->dominator(); + if (dominator) { + assert(block != _ir->start(), "Start block must not have a dominator!"); + assert(can_reach(dominator, block), "Dominator can't reach his block !"); + assert(can_reach(_ir->start(), dominator), "Dominator is unreachable !"); + assert(!can_reach(_ir->start(), block, dominator), "Wrong dominator ! Block can be reached anyway !"); + BlockList *all_blocks = _ir->linear_scan_order(); + for (int i=0; ilength(); i++) { + BlockBegin *cur = all_blocks->at(i); + if (cur != dominator && cur != block) { + assert(can_reach(dominator, block, cur), "There has to be another dominator!"); + } + } + } else { + assert(block == _ir->start(), "Only start block must not have a dominator"); + } + + if (block->is_set(BlockBegin::linear_scan_loop_header_flag)) { + int loop_index = block->loop_index(); + BlockList *all_blocks = _ir->linear_scan_order(); + assert(block->number_of_preds() >= 1, "Block must have at least one predecessor"); + assert(!block->is_set(BlockBegin::exception_entry_flag), "Loop header must not be exception handler!"); + // Sometimes, the backbranch comes from an exception handler. In + // this case, loop indexes/loop depths may not appear correct. + bool loop_through_xhandler = false; + for (int i = 0; i < block->number_of_exception_handlers(); i++) { + BlockBegin *xhandler = block->exception_handler_at(i); + for (int j = 0; j < block->number_of_preds(); j++) { + if (dominates(xhandler, block->pred_at(j)) || xhandler == block->pred_at(j)) { + loop_through_xhandler = true; + } + } + } + + for (int i=0; inumber_of_sux(); i++) { + BlockBegin *sux = block->sux_at(i); + assert(sux->loop_depth() != block->loop_depth() || sux->loop_index() == block->loop_index() || loop_through_xhandler, "Loop index has to be same"); + assert(sux->loop_depth() == block->loop_depth() || sux->loop_index() != block->loop_index(), "Loop index has to be different"); + } + + for (int i=0; ilength(); i++) { + BlockBegin *cur = all_blocks->at(i); + if (cur->loop_index() == loop_index && cur != block) { + assert(dominates(block->dominator(), cur), "Dominator of loop header must dominate all loop blocks"); + } + } + } + + Instruction *cur = block; + while (cur) { + assert(cur->block() == block, "Block begin has to be set correctly!"); + cur = cur->next(); + } +} + +// Loop header must dominate all loop blocks +bool RangeCheckEliminator::Verification::dominates(BlockBegin *dominator, BlockBegin *block) { + BlockBegin *cur = block->dominator(); + while (cur && cur != dominator) { + cur = cur->dominator(); + } + return cur == dominator; +} + +// Try to reach Block end beginning in Block start and not using Block dont_use +bool RangeCheckEliminator::Verification::can_reach(BlockBegin *start, BlockBegin *end, BlockBegin *dont_use /* = NULL */) { + if (start == end) return start != dont_use; + // Simple BSF from start to end + // BlockBeginList _current; + for (int i=0; i<_used.length(); i++) { + _used[i] = false; + } + _current.truncate(0); + _successors.truncate(0); + if (start != dont_use) { + _current.push(start); + _used[start->block_id()] = true; + } + + // BlockBeginList _successors; + while (_current.length() > 0) { + BlockBegin *cur = _current.pop(); + // Add exception handlers to list + for (int i=0; inumber_of_exception_handlers(); i++) { + BlockBegin *xhandler = cur->exception_handler_at(i); + _successors.push(xhandler); + // Add exception handlers of _successors to list + for (int j=0; jnumber_of_exception_handlers(); j++) { + BlockBegin *sux_xhandler = xhandler->exception_handler_at(j); + _successors.push(sux_xhandler); + } + } + // Add normal _successors to list + for (int i=0; inumber_of_sux(); i++) { + BlockBegin *sux = cur->sux_at(i); + _successors.push(sux); + // Add exception handlers of _successors to list + for (int j=0; jnumber_of_exception_handlers(); j++) { + BlockBegin *xhandler = sux->exception_handler_at(j); + _successors.push(xhandler); + } + } + for (int i=0; i<_successors.length(); i++) { + BlockBegin *sux = _successors[i]; + assert(sux != NULL, "Successor must not be NULL!"); + if (sux == end) { + return true; + } + if (sux != dont_use && !_used[sux->block_id()]) { + _used[sux->block_id()] = true; + _current.push(sux); + } + } + _successors.truncate(0); + } + + return false; +} + +// Bound +RangeCheckEliminator::Bound::~Bound() { +} + +// Bound constructor +RangeCheckEliminator::Bound::Bound() { + init(); + this->_lower = min_jint; + this->_upper = max_jint; + this->_lower_instr = NULL; + this->_upper_instr = NULL; +} + +// Bound constructor +RangeCheckEliminator::Bound::Bound(int lower, Value lower_instr, int upper, Value upper_instr) { + init(); + assert(!lower_instr || !lower_instr->as_Constant() || !lower_instr->type()->as_IntConstant(), "Must not be constant!"); + assert(!upper_instr || !upper_instr->as_Constant() || !upper_instr->type()->as_IntConstant(), "Must not be constant!"); + this->_lower = lower; + this->_upper = upper; + this->_lower_instr = lower_instr; + this->_upper_instr = upper_instr; +} + +// Bound constructor +RangeCheckEliminator::Bound::Bound(Instruction::Condition cond, Value v, int constant) { + assert(!v || (v->type() && (v->type()->as_IntType() || v->type()->as_ObjectType())), "Type must be array or integer!"); + assert(!v || !v->as_Constant() || !v->type()->as_IntConstant(), "Must not be constant!"); + + init(); + if (cond == Instruction::eql) { + _lower = constant; + _lower_instr = v; + _upper = constant; + _upper_instr = v; + } else if (cond == Instruction::neq) { + _lower = min_jint; + _upper = max_jint; + _lower_instr = NULL; + _upper_instr = NULL; + if (v == NULL) { + if (constant == min_jint) { + _lower++; + } + if (constant == max_jint) { + _upper--; + } + } + } else if (cond == Instruction::geq) { + _lower = constant; + _lower_instr = v; + _upper = max_jint; + _upper_instr = NULL; + } else if (cond == Instruction::leq) { + _lower = min_jint; + _lower_instr = NULL; + _upper = constant; + _upper_instr = v; + } else { + ShouldNotReachHere(); + } +} + +// Set lower +void RangeCheckEliminator::Bound::set_lower(int value, Value v) { + assert(!v || !v->as_Constant() || !v->type()->as_IntConstant(), "Must not be constant!"); + this->_lower = value; + this->_lower_instr = v; +} + +// Set upper +void RangeCheckEliminator::Bound::set_upper(int value, Value v) { + assert(!v || !v->as_Constant() || !v->type()->as_IntConstant(), "Must not be constant!"); + this->_upper = value; + this->_upper_instr = v; +} + +// Add constant -> no overflow may occur +void RangeCheckEliminator::Bound::add_constant(int value) { + this->_lower += value; + this->_upper += value; +} + +// Init +void RangeCheckEliminator::Bound::init() { +} + +// or +void RangeCheckEliminator::Bound::or_op(Bound *b) { + // Watch out, bound is not guaranteed not to overflow! + // Update lower bound + if (_lower_instr != b->_lower_instr || (_lower_instr && _lower != b->_lower)) { + _lower_instr = NULL; + _lower = min_jint; + } else { + _lower = MIN2(_lower, b->_lower); + } + // Update upper bound + if (_upper_instr != b->_upper_instr || (_upper_instr && _upper != b->_upper)) { + _upper_instr = NULL; + _upper = max_jint; + } else { + _upper = MAX2(_upper, b->_upper); + } +} + +// and +void RangeCheckEliminator::Bound::and_op(Bound *b) { + // Update lower bound + if (_lower_instr == b->_lower_instr) { + _lower = MAX2(_lower, b->_lower); + } + if (b->has_lower()) { + bool set = true; + if (_lower_instr != NULL && b->_lower_instr != NULL) { + set = (_lower_instr->dominator_depth() > b->_lower_instr->dominator_depth()); + } + if (set) { + _lower = b->_lower; + _lower_instr = b->_lower_instr; + } + } + // Update upper bound + if (_upper_instr == b->_upper_instr) { + _upper = MIN2(_upper, b->_upper); + } + if (b->has_upper()) { + bool set = true; + if (_upper_instr != NULL && b->_upper_instr != NULL) { + set = (_upper_instr->dominator_depth() > b->_upper_instr->dominator_depth()); + } + if (set) { + _upper = b->_upper; + _upper_instr = b->_upper_instr; + } + } +} + +// has_upper +bool RangeCheckEliminator::Bound::has_upper() { + return _upper_instr != NULL || _upper < max_jint; +} + +// is_smaller +bool RangeCheckEliminator::Bound::is_smaller(Bound *b) { + if (b->_lower_instr != _upper_instr) { + return false; + } + return _upper < b->_lower; +} + +// has_lower +bool RangeCheckEliminator::Bound::has_lower() { + return _lower_instr != NULL || _lower > min_jint; +} + +// in_array_bound +bool RangeCheckEliminator::in_array_bound(Bound *bound, Value array){ + if (!bound) return false; + assert(array != NULL, "Must not be null!"); + assert(bound != NULL, "Must not be null!"); + if (bound->lower() >=0 && bound->lower_instr() == NULL && bound->upper() < 0 && bound->upper_instr() != NULL) { + ArrayLength *len = bound->upper_instr()->as_ArrayLength(); + if (bound->upper_instr() == array || (len != NULL && len->array() == array)) { + return true; + } + } + return false; +} + +// remove_lower +void RangeCheckEliminator::Bound::remove_lower() { + _lower = min_jint; + _lower_instr = NULL; +} + +// remove_upper +void RangeCheckEliminator::Bound::remove_upper() { + _upper = max_jint; + _upper_instr = NULL; +} + +// upper +int RangeCheckEliminator::Bound::upper() { + return _upper; +} + +// lower +int RangeCheckEliminator::Bound::lower() { + return _lower; +} + +// upper_instr +Value RangeCheckEliminator::Bound::upper_instr() { + return _upper_instr; +} + +// lower_instr +Value RangeCheckEliminator::Bound::lower_instr() { + return _lower_instr; +} + +// print +void RangeCheckEliminator::Bound::print() { + tty->print(""); + if (this->_lower_instr || this->_lower != min_jint) { + if (this->_lower_instr) { + tty->print("i%d", this->_lower_instr->id()); + if (this->_lower > 0) { + tty->print("+%d", _lower); + } + if (this->_lower < 0) { + tty->print("%d", _lower); + } + } else { + tty->print("%d", _lower); + } + tty->print(" <= "); + } + tty->print("x"); + if (this->_upper_instr || this->_upper != max_jint) { + tty->print(" <= "); + if (this->_upper_instr) { + tty->print("i%d", this->_upper_instr->id()); + if (this->_upper > 0) { + tty->print("+%d", _upper); + } + if (this->_upper < 0) { + tty->print("%d", _upper); + } + } else { + tty->print("%d", _upper); + } + } +} + +// Copy +RangeCheckEliminator::Bound *RangeCheckEliminator::Bound::copy() { + Bound *b = new Bound(); + b->_lower = _lower; + b->_lower_instr = _lower_instr; + b->_upper = _upper; + b->_upper_instr = _upper_instr; + return b; +} + +#ifdef ASSERT +// Add assertion +void RangeCheckEliminator::Bound::add_assertion(Instruction *instruction, Instruction *position, int i, Value instr, Instruction::Condition cond) { + Instruction *result = position; + Instruction *compare_with = NULL; + ValueStack *state = position->state_before(); + if (position->as_BlockEnd() && !position->as_Goto()) { + state = position->as_BlockEnd()->state_before(); + } + Instruction *instruction_before = position->prev(); + if (position->as_Return() && Compilation::current()->method()->is_synchronized() && instruction_before->as_MonitorExit()) { + instruction_before = instruction_before->prev(); + } + result = instruction_before; + // Load constant only if needed + Constant *constant = NULL; + if (i != 0 || !instr) { + constant = new Constant(new IntConstant(i)); + NOT_PRODUCT(constant->set_printable_bci(position->printable_bci())); + result = result->insert_after(constant); + compare_with = constant; + } + + if (instr) { + assert(instr->type()->as_ObjectType() || instr->type()->as_IntType(), "Type must be array or integer!"); + compare_with = instr; + // Load array length if necessary + Instruction *op = instr; + if (instr->type()->as_ObjectType()) { + assert(state, "must not be null"); + ArrayLength *length = new ArrayLength(instr, state->copy()); + NOT_PRODUCT(length->set_printable_bci(position->printable_bci())); + length->set_exception_state(length->state_before()); + result = result->insert_after(length); + op = length; + compare_with = length; + } + // Add operation only if necessary + if (constant) { + ArithmeticOp *ao = new ArithmeticOp(Bytecodes::_iadd, constant, op, false, NULL); + NOT_PRODUCT(ao->set_printable_bci(position->printable_bci())); + result = result->insert_after(ao); + compare_with = ao; + // TODO: Check that add operation does not overflow! + } + } + assert(compare_with != NULL, "You have to compare with something!"); + assert(instruction != NULL, "Instruction must not be null!"); + + if (instruction->type()->as_ObjectType()) { + // Load array length if necessary + Instruction *op = instruction; + assert(state, "must not be null"); + ArrayLength *length = new ArrayLength(instruction, state->copy()); + length->set_exception_state(length->state_before()); + NOT_PRODUCT(length->set_printable_bci(position->printable_bci())); + result = result->insert_after(length); + instruction = length; + } + + Assert *assert = new Assert(instruction, cond, false, compare_with); + NOT_PRODUCT(assert->set_printable_bci(position->printable_bci())); + result->insert_after(assert); +} + +// Add assertions +void RangeCheckEliminator::add_assertions(Bound *bound, Instruction *instruction, Instruction *position) { + // Add lower bound assertion + if (bound->has_lower()) { + bound->add_assertion(instruction, position, bound->lower(), bound->lower_instr(), Instruction::geq); + } + // Add upper bound assertion + if (bound->has_upper()) { + bound->add_assertion(instruction, position, bound->upper(), bound->upper_instr(), Instruction::leq); + } +} +#endif + diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_RangeCheckElimination.hpp --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/src/share/vm/c1/c1_RangeCheckElimination.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -0,0 +1,241 @@ +/* + * Copyright (c) 2012, 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. + * + */ + +#ifndef SHARE_VM_C1_C1_RANGECHECKELIMINATION_HPP +#define SHARE_VM_C1_C1_RANGECHECKELIMINATION_HPP + +#include "c1/c1_Instruction.hpp" + +// Base class for range check elimination +class RangeCheckElimination : AllStatic { +public: + static void eliminate(IR *ir); +}; + +// Implementation +class RangeCheckEliminator VALUE_OBJ_CLASS_SPEC { +private: + int _number_of_instructions; + bool _optimistic; // Insert predicates and deoptimize when they fail + IR *_ir; + + define_array(BlockBeginArray, BlockBegin*) + define_stack(BlockBeginList, BlockBeginArray) + define_stack(IntegerStack, intArray) + define_array(IntegerMap, IntegerStack*) + + class Verification : public _ValueObj /*VALUE_OBJ_CLASS_SPEC*/, public BlockClosure { + private: + IR *_ir; + boolArray _used; + BlockBeginList _current; + BlockBeginList _successors; + + public: + Verification(IR *ir); + virtual void block_do(BlockBegin *block); + bool can_reach(BlockBegin *start, BlockBegin *end, BlockBegin *dont_use = NULL); + bool dominates(BlockBegin *dominator, BlockBegin *block); + }; + +public: + // Bounds for an instruction in the form x + c which c integer + // constant and x another instruction + class Bound : public CompilationResourceObj { + private: + int _upper; + Value _upper_instr; + int _lower; + Value _lower_instr; + + public: + Bound(); + Bound(Value v); + Bound(Instruction::Condition cond, Value v, int constant = 0); + Bound(int lower, Value lower_instr, int upper, Value upper_instr); + ~Bound(); + +#ifdef ASSERT + void add_assertion(Instruction *instruction, Instruction *position, int i, Value instr, Instruction::Condition cond); +#endif + int upper(); + Value upper_instr(); + int lower(); + Value lower_instr(); + void print(); + bool check_no_overflow(int const_value); + void or_op(Bound *b); + void and_op(Bound *b); + bool has_upper(); + bool has_lower(); + void set_upper(int upper, Value upper_instr); + void set_lower(int lower, Value lower_instr); + bool is_smaller(Bound *b); + void remove_upper(); + void remove_lower(); + void add_constant(int value); + Bound *copy(); + + private: + void init(); + }; + + + class Visitor : public InstructionVisitor { + private: + Bound *_bound; + RangeCheckEliminator *_rce; + + public: + void set_range_check_eliminator(RangeCheckEliminator *rce) { _rce = rce; } + Bound *bound() const { return _bound; } + void clear_bound() { _bound = NULL; } + + protected: + // visitor functions + void do_Constant (Constant* x); + void do_IfOp (IfOp* x); + void do_LogicOp (LogicOp* x); + void do_ArithmeticOp (ArithmeticOp* x); + void do_Phi (Phi* x); + + void do_StoreField (StoreField* x) { /* nothing to do */ }; + void do_StoreIndexed (StoreIndexed* x) { /* nothing to do */ }; + void do_MonitorEnter (MonitorEnter* x) { /* nothing to do */ }; + void do_MonitorExit (MonitorExit* x) { /* nothing to do */ }; + void do_Invoke (Invoke* x) { /* nothing to do */ }; + void do_UnsafePutRaw (UnsafePutRaw* x) { /* nothing to do */ }; + void do_UnsafePutObject(UnsafePutObject* x) { /* nothing to do */ }; + void do_Intrinsic (Intrinsic* x) { /* nothing to do */ }; + void do_Local (Local* x) { /* nothing to do */ }; + void do_LoadField (LoadField* x) { /* nothing to do */ }; + void do_ArrayLength (ArrayLength* x) { /* nothing to do */ }; + void do_LoadIndexed (LoadIndexed* x) { /* nothing to do */ }; + void do_NegateOp (NegateOp* x) { /* nothing to do */ }; + void do_ShiftOp (ShiftOp* x) { /* nothing to do */ }; + void do_CompareOp (CompareOp* x) { /* nothing to do */ }; + void do_Convert (Convert* x) { /* nothing to do */ }; + void do_NullCheck (NullCheck* x) { /* nothing to do */ }; + void do_TypeCast (TypeCast* x) { /* nothing to do */ }; + void do_NewInstance (NewInstance* x) { /* nothing to do */ }; + void do_NewTypeArray (NewTypeArray* x) { /* nothing to do */ }; + void do_NewObjectArray (NewObjectArray* x) { /* nothing to do */ }; + void do_NewMultiArray (NewMultiArray* x) { /* nothing to do */ }; + void do_CheckCast (CheckCast* x) { /* nothing to do */ }; + void do_InstanceOf (InstanceOf* x) { /* nothing to do */ }; + void do_BlockBegin (BlockBegin* x) { /* nothing to do */ }; + void do_Goto (Goto* x) { /* nothing to do */ }; + void do_If (If* x) { /* nothing to do */ }; + void do_IfInstanceOf (IfInstanceOf* x) { /* nothing to do */ }; + void do_TableSwitch (TableSwitch* x) { /* nothing to do */ }; + void do_LookupSwitch (LookupSwitch* x) { /* nothing to do */ }; + void do_Return (Return* x) { /* nothing to do */ }; + void do_Throw (Throw* x) { /* nothing to do */ }; + void do_Base (Base* x) { /* nothing to do */ }; + void do_OsrEntry (OsrEntry* x) { /* nothing to do */ }; + void do_ExceptionObject(ExceptionObject* x) { /* nothing to do */ }; + void do_RoundFP (RoundFP* x) { /* nothing to do */ }; + void do_UnsafeGetRaw (UnsafeGetRaw* x) { /* nothing to do */ }; + void do_UnsafeGetObject(UnsafeGetObject* x) { /* nothing to do */ }; + void do_UnsafeGetAndSetObject(UnsafeGetAndSetObject* x) { /* nothing to do */ }; + void do_UnsafePrefetchRead (UnsafePrefetchRead* x) { /* nothing to do */ }; + void do_UnsafePrefetchWrite(UnsafePrefetchWrite* x) { /* nothing to do */ }; + void do_ProfileCall (ProfileCall* x) { /* nothing to do */ }; + void do_ProfileInvoke (ProfileInvoke* x) { /* nothing to do */ }; + void do_RuntimeCall (RuntimeCall* x) { /* nothing to do */ }; + void do_MemBar (MemBar* x) { /* nothing to do */ }; + void do_RangeCheckPredicate(RangeCheckPredicate* x) { /* nothing to do */ }; + void do_Assert (Assert* x) { /* nothing to do */ }; + }; + +#ifdef ASSERT + void add_assertions(Bound *bound, Instruction *instruction, Instruction *position); +#endif + + define_array(BoundArray, Bound *) + define_stack(BoundStack, BoundArray) + define_array(BoundMap, BoundStack *) + define_array(AccessIndexedArray, AccessIndexed *) + define_stack(AccessIndexedList, AccessIndexedArray) + define_array(InstructionArray, Instruction *) + define_stack(InstructionList, InstructionArray) + + class AccessIndexedInfo : public CompilationResourceObj { + public: + AccessIndexedList *_list; + int _min; + int _max; + }; + + define_array(AccessIndexedInfoArray, AccessIndexedInfo *) + BoundMap _bounds; // Mapping from Instruction's id to current bound + AccessIndexedInfoArray _access_indexed_info; // Mapping from Instruction's id to AccessIndexedInfo for in block motion + Visitor _visitor; + +public: + RangeCheckEliminator(IR *ir); + + IR *ir() const { return _ir; } + + // Pass over the dominator tree to identify blocks where there's an oppportunity for optimization + bool set_process_block_flags(BlockBegin *block); + // The core of the optimization work: pass over the dominator tree + // to propagate bound information, insert predicate out of loops, + // eliminate bound checks when possible and perform in block motion + void calc_bounds(BlockBegin *block, BlockBegin *loop_header); + // reorder bound checks within a block in order to eliminate some of them + void in_block_motion(BlockBegin *block, AccessIndexedList &accessIndexed, InstructionList &arrays); + + // update/access current bound + void update_bound(IntegerStack &pushed, Value v, Instruction::Condition cond, Value value, int constant); + void update_bound(IntegerStack &pushed, Value v, Bound *bound); + Bound *get_bound(Value v); + + bool loop_invariant(BlockBegin *loop_header, Instruction *instruction); // check for loop invariance + void add_access_indexed_info(InstructionList &indices, int i, Value instruction, AccessIndexed *ai); // record indexed access for in block motion + void remove_range_check(AccessIndexed *ai); // Mark this instructions as not needing a range check + void add_if_condition(IntegerStack &pushed, Value x, Value y, Instruction::Condition condition); // Update bound for an If + bool in_array_bound(Bound *bound, Value array); // Check whether bound is known to fall within array + + // helper functions to work with predicates + Instruction* insert_after(Instruction* insert_position, Instruction* instr, int bci); + Instruction* predicate(Instruction* left, Instruction::Condition cond, Instruction* right, ValueStack* state, Instruction *insert_position, int bci=-1); + Instruction* predicate_cmp_with_const(Instruction* instr, Instruction::Condition cond, int constant, ValueStack* state, Instruction *insert_position, int bci=1); + Instruction* predicate_add(Instruction* left, int left_const, Instruction::Condition cond, Instruction* right, ValueStack* state, Instruction *insert_position, int bci=-1); + Instruction* predicate_add_cmp_with_const(Instruction* left, int left_const, Instruction::Condition cond, int constant, ValueStack* state, Instruction *insert_position, int bci=-1); + + void insert_deoptimization(ValueStack *state, Instruction *insert_position, Instruction *array_instr, // Add predicate + Instruction *length_instruction, Instruction *lower_instr, int lower, + Instruction *upper_instr, int upper, AccessIndexed *ai); + bool is_ok_for_deoptimization(Instruction *insert_position, Instruction *array_instr, // Can we safely add a predicate? + Instruction *length_instr, Instruction *lower_instr, + int lower, Instruction *upper_instr, int upper); + void process_if(IntegerStack &pushed, BlockBegin *block, If *cond); // process If Instruction + void process_access_indexed(BlockBegin *loop_header, BlockBegin *block, AccessIndexed *ai); // process indexed access + + void dump_condition_stack(BlockBegin *cur_block); + static void print_statistics(); +}; + +#endif // SHARE_VM_C1_C1_RANGECHECKELIMINATION_HPP diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_Runtime1.cpp --- a/src/share/vm/c1/c1_Runtime1.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_Runtime1.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -1330,6 +1330,50 @@ return (k != NULL && obj != NULL && obj->is_a(k)) ? 1 : 0; JRT_END +JRT_ENTRY(void, Runtime1::predicate_failed_trap(JavaThread* thread)) + ResourceMark rm; + + assert(!TieredCompilation, "incompatible with tiered compilation"); + + RegisterMap reg_map(thread, false); + frame runtime_frame = thread->last_frame(); + frame caller_frame = runtime_frame.sender(®_map); + + nmethod* nm = CodeCache::find_nmethod(caller_frame.pc()); + assert (nm != NULL, "no more nmethod?"); + nm->make_not_entrant(); + + methodHandle m(nm->method()); + MethodData* mdo = m->method_data(); + + if (mdo == NULL && !HAS_PENDING_EXCEPTION) { + // Build an MDO. Ignore errors like OutOfMemory; + // that simply means we won't have an MDO to update. + Method::build_interpreter_method_data(m, THREAD); + if (HAS_PENDING_EXCEPTION) { + assert((PENDING_EXCEPTION->is_a(SystemDictionary::OutOfMemoryError_klass())), "we expect only an OOM error here"); + CLEAR_PENDING_EXCEPTION; + } + mdo = m->method_data(); + } + + if (mdo != NULL) { + mdo->inc_trap_count(Deoptimization::Reason_none); + } + + if (TracePredicateFailedTraps) { + stringStream ss1, ss2; + vframeStream vfst(thread); + methodHandle inlinee = methodHandle(vfst.method()); + inlinee->print_short_name(&ss1); + m->print_short_name(&ss2); + tty->print_cr("Predicate failed trap in method %s at bci %d inlined in %s at pc %x", ss1.as_string(), vfst.bci(), ss2.as_string(), caller_frame.pc()); + } + + + Deoptimization::deoptimize_frame(thread, caller_frame.id()); + +JRT_END #ifndef PRODUCT void Runtime1::print_statistics() { diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_Runtime1.hpp --- a/src/share/vm/c1/c1_Runtime1.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_Runtime1.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -71,6 +71,7 @@ stub(g1_post_barrier_slow) \ stub(fpu2long_stub) \ stub(counter_overflow) \ + stub(predicate_failed_trap) \ last_entry(number_of_ids) #define DECLARE_STUB_ID(x) x ## _id , @@ -190,6 +191,8 @@ static void oop_arraycopy(HeapWord* src, HeapWord* dst, int length); static int is_instance_of(oopDesc* mirror, oopDesc* obj); + static void predicate_failed_trap(JavaThread* thread); + static void print_statistics() PRODUCT_RETURN; }; diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_ValueMap.cpp --- a/src/share/vm/c1/c1_ValueMap.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_ValueMap.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -26,9 +26,9 @@ #include "c1/c1_Canonicalizer.hpp" #include "c1/c1_IR.hpp" #include "c1/c1_ValueMap.hpp" +#include "c1/c1_ValueStack.hpp" #include "utilities/bitMap.inline.hpp" - #ifndef PRODUCT int ValueMap::_number_of_finds = 0; @@ -192,10 +192,6 @@ && lf->field()->holder() == field->holder() \ && (all_offsets || lf->field()->offset() == field->offset()); -#define MUST_KILL_EXCEPTION(must_kill, entry, value) \ - assert(entry->nesting() < nesting(), "must not find bigger nesting than current"); \ - bool must_kill = (entry->nesting() == nesting() - 1); - void ValueMap::kill_memory() { GENERIC_KILL_VALUE(MUST_KILL_MEMORY); @@ -209,11 +205,6 @@ GENERIC_KILL_VALUE(MUST_KILL_FIELD); } -void ValueMap::kill_exception() { - GENERIC_KILL_VALUE(MUST_KILL_EXCEPTION); -} - - void ValueMap::kill_map(ValueMap* map) { assert(is_global_value_numbering(), "only for global value numbering"); _killed_values.set_union(&map->_killed_values); @@ -274,6 +265,8 @@ GlobalValueNumbering* _gvn; BlockList _loop_blocks; bool _too_complicated_loop; + bool _has_field_store[T_ARRAY + 1]; + bool _has_indexed_store[T_ARRAY + 1]; // simplified access to methods of GlobalValueNumbering ValueMap* current_map() { return _gvn->current_map(); } @@ -281,8 +274,16 @@ // implementation for abstract methods of ValueNumberingVisitor void kill_memory() { _too_complicated_loop = true; } - void kill_field(ciField* field, bool all_offsets) { current_map()->kill_field(field, all_offsets); }; - void kill_array(ValueType* type) { current_map()->kill_array(type); }; + void kill_field(ciField* field, bool all_offsets) { + current_map()->kill_field(field, all_offsets); + assert(field->type()->basic_type() >= 0 && field->type()->basic_type() <= T_ARRAY, "Invalid type"); + _has_field_store[field->type()->basic_type()] = true; + } + void kill_array(ValueType* type) { + current_map()->kill_array(type); + BasicType basic_type = as_BasicType(type); assert(basic_type >= 0 && basic_type <= T_ARRAY, "Invalid type"); + _has_indexed_store[basic_type] = true; + } public: ShortLoopOptimizer(GlobalValueNumbering* gvn) @@ -290,11 +291,141 @@ , _loop_blocks(ValueMapMaxLoopSize) , _too_complicated_loop(false) { + for (int i=0; i<= T_ARRAY; i++){ + _has_field_store[i] = false; + _has_indexed_store[i] = false; + } + } + + bool has_field_store(BasicType type) { + assert(type >= 0 && type <= T_ARRAY, "Invalid type"); + return _has_field_store[type]; + } + + bool has_indexed_store(BasicType type) { + assert(type >= 0 && type <= T_ARRAY, "Invalid type"); + return _has_indexed_store[type]; } bool process(BlockBegin* loop_header); }; +class LoopInvariantCodeMotion : public StackObj { + private: + GlobalValueNumbering* _gvn; + ShortLoopOptimizer* _short_loop_optimizer; + Instruction* _insertion_point; + ValueStack * _state; + + void set_invariant(Value v) const { _gvn->set_processed(v); } + bool is_invariant(Value v) const { return _gvn->is_processed(v); } + + void process_block(BlockBegin* block); + + public: + LoopInvariantCodeMotion(ShortLoopOptimizer *slo, GlobalValueNumbering* gvn, BlockBegin* loop_header, BlockList* loop_blocks); +}; + +LoopInvariantCodeMotion::LoopInvariantCodeMotion(ShortLoopOptimizer *slo, GlobalValueNumbering* gvn, BlockBegin* loop_header, BlockList* loop_blocks) + : _gvn(gvn), _short_loop_optimizer(slo) { + + TRACE_VALUE_NUMBERING(tty->print_cr("using loop invariant code motion loop_header = %d", loop_header->block_id())); + TRACE_VALUE_NUMBERING(tty->print_cr("** loop invariant code motion for short loop B%d", loop_header->block_id())); + + BlockBegin* insertion_block = loop_header->dominator(); + if (insertion_block->number_of_preds() == 0) { + return; // only the entry block does not have a predecessor + } + + assert(insertion_block->end()->as_Base() == NULL, "cannot insert into entry block"); + _insertion_point = insertion_block->end()->prev(); + + BlockEnd *block_end = insertion_block->end(); + _state = block_end->state_before(); + + if (!_state) { + // If, TableSwitch and LookupSwitch always have state_before when + // loop invariant code motion happens.. + assert(block_end->as_Goto(), "Block has to be goto"); + _state = block_end->state(); + } + + // the loop_blocks are filled by going backward from the loop header, so this processing order is best + assert(loop_blocks->at(0) == loop_header, "loop header must be first loop block"); + process_block(loop_header); + for (int i = loop_blocks->length() - 1; i >= 1; i--) { + process_block(loop_blocks->at(i)); + } +} + +void LoopInvariantCodeMotion::process_block(BlockBegin* block) { + TRACE_VALUE_NUMBERING(tty->print_cr("processing block B%d", block->block_id())); + + Instruction* prev = block; + Instruction* cur = block->next(); + + while (cur != NULL) { + + // determine if cur instruction is loop invariant + // only selected instruction types are processed here + bool cur_invariant = false; + + if (cur->as_Constant() != NULL) { + cur_invariant = !cur->can_trap(); + } else if (cur->as_ArithmeticOp() != NULL || cur->as_LogicOp() != NULL || cur->as_ShiftOp() != NULL) { + assert(cur->as_Op2() != NULL, "must be Op2"); + Op2* op2 = (Op2*)cur; + cur_invariant = !op2->can_trap() && is_invariant(op2->x()) && is_invariant(op2->y()); + } else if (cur->as_LoadField() != NULL) { + LoadField* lf = (LoadField*)cur; + // deoptimizes on NullPointerException + cur_invariant = !lf->needs_patching() && !lf->field()->is_volatile() && !_short_loop_optimizer->has_field_store(lf->field()->type()->basic_type()) && is_invariant(lf->obj()); + } else if (cur->as_ArrayLength() != NULL) { + ArrayLength *length = cur->as_ArrayLength(); + cur_invariant = is_invariant(length->array()); + } else if (cur->as_LoadIndexed() != NULL) { + LoadIndexed *li = (LoadIndexed *)cur->as_LoadIndexed(); + cur_invariant = !_short_loop_optimizer->has_indexed_store(as_BasicType(cur->type())) && is_invariant(li->array()) && is_invariant(li->index()); + } + + if (cur_invariant) { + // perform value numbering and mark instruction as loop-invariant + _gvn->substitute(cur); + + if (cur->as_Constant() == NULL) { + // ensure that code for non-constant instructions is always generated + cur->pin(); + } + + // remove cur instruction from loop block and append it to block before loop + Instruction* next = cur->next(); + Instruction* in = _insertion_point->next(); + _insertion_point = _insertion_point->set_next(cur); + cur->set_next(in); + + // Deoptimize on exception + cur->set_flag(Instruction::DeoptimizeOnException, true); + + // Clear exception handlers + cur->set_exception_handlers(NULL); + + TRACE_VALUE_NUMBERING(tty->print_cr("Instruction %c%d is loop invariant", cur->type()->tchar(), cur->id())); + + if (cur->state_before() != NULL) { + cur->set_state_before(_state->copy()); + } + if (cur->exception_state() != NULL) { + cur->set_exception_state(_state->copy()); + } + + cur = prev->set_next(next); + + } else { + prev = cur; + cur = cur->next(); + } + } +} bool ShortLoopOptimizer::process(BlockBegin* loop_header) { TRACE_VALUE_NUMBERING(tty->print_cr("** loop header block")); @@ -316,6 +447,10 @@ for (int j = block->number_of_preds() - 1; j >= 0; j--) { BlockBegin* pred = block->pred_at(j); + if (pred->is_set(BlockBegin::osr_entry_flag)) { + return false; + } + ValueMap* pred_map = value_map_of(pred); if (pred_map != NULL) { current_map()->kill_map(pred_map); @@ -336,6 +471,12 @@ } } + bool optimistic = this->_gvn->compilation()->is_optimistic(); + + if (UseLoopInvariantCodeMotion && optimistic) { + LoopInvariantCodeMotion code_motion(this, _gvn, loop_header, &_loop_blocks); + } + TRACE_VALUE_NUMBERING(tty->print_cr("** loop successfully optimized")); return true; } @@ -344,11 +485,11 @@ GlobalValueNumbering::GlobalValueNumbering(IR* ir) : _current_map(NULL) , _value_maps(ir->linear_scan_order()->length(), NULL) + , _compilation(ir->compilation()) { TRACE_VALUE_NUMBERING(tty->print_cr("****** start of global value numbering")); ShortLoopOptimizer short_loop_optimizer(this); - int subst_count = 0; BlockList* blocks = ir->linear_scan_order(); int num_blocks = blocks->length(); @@ -357,6 +498,12 @@ assert(start_block == ir->start() && start_block->number_of_preds() == 0 && start_block->dominator() == NULL, "must be start block"); assert(start_block->next()->as_Base() != NULL && start_block->next()->next() == NULL, "start block must not have instructions"); + // method parameters are not linked in instructions list, so process them separateley + for_each_state_value(start_block->state(), value, + assert(value->as_Local() != NULL, "only method parameters allowed"); + set_processed(value); + ); + // initial, empty value map with nesting 0 set_value_map_of(start_block, new ValueMap()); @@ -374,7 +521,7 @@ // create new value map with increased nesting _current_map = new ValueMap(value_map_of(dominator)); - if (num_preds == 1) { + if (num_preds == 1 && !block->is_set(BlockBegin::exception_entry_flag)) { assert(dominator == block->pred_at(0), "dominator must be equal to predecessor"); // nothing to do here @@ -403,36 +550,41 @@ } } - if (block->is_set(BlockBegin::exception_entry_flag)) { - current_map()->kill_exception(); - } + // phi functions are not linked in instructions list, so process them separateley + for_each_phi_fun(block, phi, + set_processed(phi); + ); TRACE_VALUE_NUMBERING(tty->print("value map before processing block: "); current_map()->print()); // visit all instructions of this block for (Value instr = block->next(); instr != NULL; instr = instr->next()) { - assert(!instr->has_subst(), "substitution already set"); - // check if instruction kills any values instr->visit(this); - - if (instr->hash() != 0) { - Value f = current_map()->find_insert(instr); - if (f != instr) { - assert(!f->has_subst(), "can't have a substitution"); - instr->set_subst(f); - subst_count++; - } - } + // perform actual value numbering + substitute(instr); } // remember value map for successors set_value_map_of(block, current_map()); } - if (subst_count != 0) { + if (_has_substitutions) { SubstitutionResolver resolver(ir); } TRACE_VALUE_NUMBERING(tty->print("****** end of global value numbering. "); ValueMap::print_statistics()); } + +void GlobalValueNumbering::substitute(Instruction* instr) { + assert(!instr->has_subst(), "substitution already set"); + Value subst = current_map()->find_insert(instr); + if (subst != instr) { + assert(!subst->has_subst(), "can't have a substitution"); + + TRACE_VALUE_NUMBERING(tty->print_cr("substitution for %d set to %d", instr->id(), subst->id())); + instr->set_subst(subst); + _has_substitutions = true; + } + set_processed(instr); +} diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_ValueMap.hpp --- a/src/share/vm/c1/c1_ValueMap.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_ValueMap.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -206,6 +206,8 @@ void do_ProfileInvoke (ProfileInvoke* x) { /* nothing to do */ }; void do_RuntimeCall (RuntimeCall* x) { /* nothing to do */ }; void do_MemBar (MemBar* x) { /* nothing to do */ }; + void do_RangeCheckPredicate(RangeCheckPredicate* x) { /* nothing to do */ }; + void do_Assert (Assert* x) { /* nothing to do */ }; }; @@ -225,15 +227,22 @@ class GlobalValueNumbering: public ValueNumberingVisitor { private: + Compilation* _compilation; // compilation data ValueMap* _current_map; // value map of current block ValueMapArray _value_maps; // list of value maps for all blocks + ValueSet _processed_values; // marker for instructions that were already processed + bool _has_substitutions; // set to true when substitutions must be resolved public: // accessors + Compilation* compilation() const { return _compilation; } ValueMap* current_map() { return _current_map; } ValueMap* value_map_of(BlockBegin* block) { return _value_maps.at(block->linear_scan_number()); } void set_value_map_of(BlockBegin* block, ValueMap* map) { assert(value_map_of(block) == NULL, ""); _value_maps.at_put(block->linear_scan_number(), map); } + bool is_processed(Value v) { return _processed_values.contains(v); } + void set_processed(Value v) { _processed_values.put(v); } + // implementation for abstract methods of ValueNumberingVisitor void kill_memory() { current_map()->kill_memory(); } void kill_field(ciField* field, bool all_offsets) { current_map()->kill_field(field, all_offsets); } @@ -241,6 +250,7 @@ // main entry point that performs global value numbering GlobalValueNumbering(IR* ir); + void substitute(Instruction* instr); // substitute instruction if it is contained in current value map }; #endif // SHARE_VM_C1_C1_VALUEMAP_HPP diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/c1/c1_globals.hpp --- a/src/share/vm/c1/c1_globals.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/c1/c1_globals.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -119,6 +119,24 @@ develop(bool, UseGlobalValueNumbering, true, \ "Use Global Value Numbering (separate phase)") \ \ + product(bool, UseLoopInvariantCodeMotion, true, \ + "Simple loop invariant code motion for short loops during GVN") \ + \ + develop(bool, TracePredicateFailedTraps, false, \ + "trace runtime traps caused by predicate failure") \ + \ + develop(bool, StressLoopInvariantCodeMotion, false, \ + "stress loop invariant code motion") \ + \ + develop(bool, TraceRangeCheckElimination, false, \ + "Trace Range Check Elimination") \ + \ + develop(bool, AssertRangeCheckElimination, false, \ + "Assert Range Check Elimination") \ + \ + develop(bool, StressRangeCheckElimination, false, \ + "stress Range Check Elimination") \ + \ develop(bool, PrintValueNumbering, false, \ "Print Value Numbering") \ \ diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/ci/ciMethod.cpp --- a/src/share/vm/ci/ciMethod.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/ci/ciMethod.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -790,6 +790,17 @@ return count; } + +// ------------------------------------------------------------------ +// ciMethod::is_special_get_caller_class_method +// +bool ciMethod::is_ignored_by_security_stack_walk() const { + check_is_loaded(); + VM_ENTRY_MARK; + return get_Method()->is_ignored_by_security_stack_walk(); +} + + // ------------------------------------------------------------------ // invokedynamic support diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/ci/ciMethod.hpp --- a/src/share/vm/ci/ciMethod.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/ci/ciMethod.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -166,8 +166,9 @@ // Code size for inlining decisions. int code_size_for_inlining(); - bool force_inline() { return get_Method()->force_inline(); } - bool dont_inline() { return get_Method()->dont_inline(); } + bool caller_sensitive() { return get_Method()->caller_sensitive(); } + bool force_inline() { return get_Method()->force_inline(); } + bool dont_inline() { return get_Method()->dont_inline(); } int comp_level(); int highest_osr_comp_level(); @@ -264,6 +265,9 @@ int instructions_size(); int scale_count(int count, float prof_factor = 1.); // make MDO count commensurate with IIC + // Stack walking support + bool is_ignored_by_security_stack_walk() const; + // JSR 292 support bool is_method_handle_intrinsic() const; bool is_compiled_lambda_form() const; diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/classfile/classFileParser.cpp --- a/src/share/vm/classfile/classFileParser.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/classfile/classFileParser.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -1735,9 +1735,14 @@ Symbol* name) { vmSymbols::SID sid = vmSymbols::find_sid(name); // Privileged code can use all annotations. Other code silently drops some. - bool privileged = loader_data->is_the_null_class_loader_data() || - loader_data->is_anonymous(); + const bool privileged = loader_data->is_the_null_class_loader_data() || + loader_data->is_ext_class_loader_data() || + loader_data->is_anonymous(); switch (sid) { + case vmSymbols::VM_SYMBOL_ENUM_NAME(sun_reflect_CallerSensitive_signature): + if (_location != _in_method) break; // only allow for methods + if (!privileged) break; // only allow in privileged code + return _method_CallerSensitive; case vmSymbols::VM_SYMBOL_ENUM_NAME(java_lang_invoke_ForceInline_signature): if (_location != _in_method) break; // only allow for methods if (!privileged) break; // only allow in privileged code @@ -1775,6 +1780,8 @@ } void ClassFileParser::MethodAnnotationCollector::apply_to(methodHandle m) { + if (has_annotation(_method_CallerSensitive)) + m->set_caller_sensitive(true); if (has_annotation(_method_ForceInline)) m->set_force_inline(true); if (has_annotation(_method_DontInline)) diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/classfile/classFileParser.hpp --- a/src/share/vm/classfile/classFileParser.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/classfile/classFileParser.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -119,6 +119,7 @@ enum Location { _in_field, _in_method, _in_class }; enum ID { _unknown = 0, + _method_CallerSensitive, _method_ForceInline, _method_DontInline, _method_LambdaForm_Compiled, diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/classfile/classLoaderData.cpp --- a/src/share/vm/classfile/classLoaderData.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/classfile/classLoaderData.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -321,6 +321,13 @@ } } +/** + * Returns true if this class loader data is for the extension class loader. + */ +bool ClassLoaderData::is_ext_class_loader_data() const { + return SystemDictionary::is_ext_class_loader(class_loader()); +} + Metaspace* ClassLoaderData::metaspace_non_null() { assert(!DumpSharedSpaces, "wrong metaspace!"); // If the metaspace has not been allocated, create a new one. Might want diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/classfile/classLoaderData.hpp --- a/src/share/vm/classfile/classLoaderData.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/classfile/classLoaderData.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -191,6 +191,7 @@ bool is_the_null_class_loader_data() const { return this == _the_null_class_loader_data; } + bool is_ext_class_loader_data() const; // The Metaspace is created lazily so may be NULL. This // method will allocate a Metaspace if needed. diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/classfile/javaClasses.hpp --- a/src/share/vm/classfile/javaClasses.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/classfile/javaClasses.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -1050,15 +1050,16 @@ // Relevant integer codes (keep these in synch. with MethodHandleNatives.Constants): enum { - MN_IS_METHOD = 0x00010000, // method (not constructor) - MN_IS_CONSTRUCTOR = 0x00020000, // constructor - MN_IS_FIELD = 0x00040000, // field - MN_IS_TYPE = 0x00080000, // nested type + MN_IS_METHOD = 0x00010000, // method (not constructor) + MN_IS_CONSTRUCTOR = 0x00020000, // constructor + MN_IS_FIELD = 0x00040000, // field + MN_IS_TYPE = 0x00080000, // nested type + MN_CALLER_SENSITIVE = 0x00100000, // @CallerSensitive annotation detected MN_REFERENCE_KIND_SHIFT = 24, // refKind - MN_REFERENCE_KIND_MASK = 0x0F000000 >> MN_REFERENCE_KIND_SHIFT, + MN_REFERENCE_KIND_MASK = 0x0F000000 >> MN_REFERENCE_KIND_SHIFT, // The SEARCH_* bits are not for MN.flags but for the matchFlags argument of MHN.getMembers: - MN_SEARCH_SUPERCLASSES = 0x00100000, // walk super classes - MN_SEARCH_INTERFACES = 0x00200000 // walk implemented interfaces + MN_SEARCH_SUPERCLASSES = 0x00100000, // walk super classes + MN_SEARCH_INTERFACES = 0x00200000 // walk implemented interfaces }; // Accessors for code generation: diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/classfile/systemDictionary.cpp --- a/src/share/vm/classfile/systemDictionary.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/classfile/systemDictionary.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -146,6 +146,17 @@ } return false; } + +/** + * Returns true if the passed class loader is the extension class loader. + */ +bool SystemDictionary::is_ext_class_loader(Handle class_loader) { + if (class_loader.is_null()) { + return false; + } + return (class_loader->klass()->name() == vmSymbols::sun_misc_Launcher_ExtClassLoader()); +} + // ---------------------------------------------------------------------------- // Resolving of classes diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/classfile/systemDictionary.hpp --- a/src/share/vm/classfile/systemDictionary.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/classfile/systemDictionary.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -106,6 +106,7 @@ do_klass(ThreadDeath_klass, java_lang_ThreadDeath, Pre ) \ do_klass(Exception_klass, java_lang_Exception, Pre ) \ do_klass(RuntimeException_klass, java_lang_RuntimeException, Pre ) \ + do_klass(SecurityManager_klass, java_lang_SecurityManager, Pre ) \ do_klass(ProtectionDomain_klass, java_security_ProtectionDomain, Pre ) \ do_klass(AccessControlContext_klass, java_security_AccessControlContext, Pre ) \ do_klass(ClassNotFoundException_klass, java_lang_ClassNotFoundException, Pre ) \ @@ -138,13 +139,14 @@ /* NOTE: needed too early in bootstrapping process to have checks based on JDK version */ \ /* Universe::is_gte_jdk14x_version() is not set up by this point. */ \ /* It's okay if this turns out to be NULL in non-1.4 JDKs. */ \ - do_klass(lambda_MagicLambdaImpl_klass, java_lang_invoke_MagicLambdaImpl, Opt ) \ + do_klass(lambda_MagicLambdaImpl_klass, java_lang_invoke_MagicLambdaImpl, Opt ) \ do_klass(reflect_MagicAccessorImpl_klass, sun_reflect_MagicAccessorImpl, Opt ) \ do_klass(reflect_MethodAccessorImpl_klass, sun_reflect_MethodAccessorImpl, Opt_Only_JDK14NewRef) \ do_klass(reflect_ConstructorAccessorImpl_klass, sun_reflect_ConstructorAccessorImpl, Opt_Only_JDK14NewRef) \ do_klass(reflect_DelegatingClassLoader_klass, sun_reflect_DelegatingClassLoader, Opt ) \ do_klass(reflect_ConstantPool_klass, sun_reflect_ConstantPool, Opt_Only_JDK15 ) \ do_klass(reflect_UnsafeStaticFieldAccessorImpl_klass, sun_reflect_UnsafeStaticFieldAccessorImpl, Opt_Only_JDK15 ) \ + do_klass(reflect_CallerSensitive_klass, sun_reflect_CallerSensitive, Opt ) \ \ /* support for dynamic typing; it's OK if these are NULL in earlier JDKs */ \ do_klass(MethodHandle_klass, java_lang_invoke_MethodHandle, Pre_JSR292 ) \ @@ -628,12 +630,15 @@ static bool is_parallelCapable(Handle class_loader); static bool is_parallelDefine(Handle class_loader); +public: + static bool is_ext_class_loader(Handle class_loader); + +private: static Klass* find_shared_class(Symbol* class_name); // Setup link to hierarchy static void add_to_hierarchy(instanceKlassHandle k, TRAPS); -private: // We pass in the hashtable index so we can calculate it outside of // the SystemDictionary_lock. diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/classfile/vmSymbols.hpp --- a/src/share/vm/classfile/vmSymbols.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/classfile/vmSymbols.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -91,6 +91,7 @@ template(java_lang_StringBuffer, "java/lang/StringBuffer") \ template(java_lang_StringBuilder, "java/lang/StringBuilder") \ template(java_lang_CharSequence, "java/lang/CharSequence") \ + template(java_lang_SecurityManager, "java/lang/SecurityManager") \ template(java_security_AccessControlContext, "java/security/AccessControlContext") \ template(java_security_ProtectionDomain, "java/security/ProtectionDomain") \ template(java_io_OutputStream, "java/io/OutputStream") \ @@ -211,6 +212,8 @@ template(sun_reflect_SerializationConstructorAccessorImpl, "sun/reflect/SerializationConstructorAccessorImpl") \ template(sun_reflect_DelegatingClassLoader, "sun/reflect/DelegatingClassLoader") \ template(sun_reflect_Reflection, "sun/reflect/Reflection") \ + template(sun_reflect_CallerSensitive, "sun/reflect/CallerSensitive") \ + template(sun_reflect_CallerSensitive_signature, "Lsun/reflect/CallerSensitive;") \ template(checkedExceptions_name, "checkedExceptions") \ template(clazz_name, "clazz") \ template(exceptionTypes_name, "exceptionTypes") \ @@ -343,6 +346,7 @@ template(contextClassLoader_name, "contextClassLoader") \ template(inheritedAccessControlContext_name, "inheritedAccessControlContext") \ template(isPrivileged_name, "isPrivileged") \ + template(getClassContext_name, "getClassContext") \ template(wait_name, "wait") \ template(checkPackageAccess_name, "checkPackageAccess") \ template(stackSize_name, "stackSize") \ @@ -463,6 +467,7 @@ template(void_classloader_signature, "()Ljava/lang/ClassLoader;") \ template(void_object_signature, "()Ljava/lang/Object;") \ template(void_class_signature, "()Ljava/lang/Class;") \ + template(void_class_array_signature, "()[Ljava/lang/Class;") \ template(void_string_signature, "()Ljava/lang/String;") \ template(object_array_object_signature, "([Ljava/lang/Object;)Ljava/lang/Object;") \ template(object_object_array_object_signature, "(Ljava/lang/Object;[Ljava/lang/Object;)Ljava/lang/Object;")\ @@ -705,9 +710,8 @@ do_intrinsic(_getLength, java_lang_reflect_Array, getLength_name, object_int_signature, F_SN) \ do_name( getLength_name, "getLength") \ \ - do_intrinsic(_getCallerClass, sun_reflect_Reflection, getCallerClass_name, getCallerClass_signature, F_SN) \ + do_intrinsic(_getCallerClass, sun_reflect_Reflection, getCallerClass_name, void_class_signature, F_SN) \ do_name( getCallerClass_name, "getCallerClass") \ - do_signature(getCallerClass_signature, "(I)Ljava/lang/Class;") \ \ do_intrinsic(_newArray, java_lang_reflect_Array, newArray_name, newArray_signature, F_SN) \ do_name( newArray_name, "newArray") \ diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/code/codeCache.hpp --- a/src/share/vm/code/codeCache.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/code/codeCache.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -156,6 +156,11 @@ static address low_bound() { return (address) _heap->low_boundary(); } static address high_bound() { return (address) _heap->high_boundary(); } + static bool has_space(int size) { + // Always leave some room in the CodeCache for I2C/C2I adapters + return largest_free_block() > (CodeCacheMinimumFreeSpace + size); + } + // Profiling static address first_address(); // first address used for CodeBlobs static address last_address(); // last address used for CodeBlobs diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/code/nmethod.cpp --- a/src/share/vm/code/nmethod.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/code/nmethod.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -486,7 +486,6 @@ #endif // def HAVE_DTRACE_H } - nmethod* nmethod::new_native_nmethod(methodHandle method, int compile_id, CodeBuffer *code_buffer, @@ -502,17 +501,19 @@ { MutexLockerEx mu(CodeCache_lock, Mutex::_no_safepoint_check_flag); int native_nmethod_size = allocation_size(code_buffer, sizeof(nmethod)); - CodeOffsets offsets; - offsets.set_value(CodeOffsets::Verified_Entry, vep_offset); - offsets.set_value(CodeOffsets::Frame_Complete, frame_complete); - nm = new (native_nmethod_size) - nmethod(method(), native_nmethod_size, compile_id, &offsets, - code_buffer, frame_size, - basic_lock_owner_sp_offset, basic_lock_sp_offset, - oop_maps); - NOT_PRODUCT(if (nm != NULL) nmethod_stats.note_native_nmethod(nm)); - if (PrintAssembly && nm != NULL) - Disassembler::decode(nm); + if (CodeCache::has_space(native_nmethod_size)) { + CodeOffsets offsets; + offsets.set_value(CodeOffsets::Verified_Entry, vep_offset); + offsets.set_value(CodeOffsets::Frame_Complete, frame_complete); + nm = new (native_nmethod_size) nmethod(method(), native_nmethod_size, + compile_id, &offsets, + code_buffer, frame_size, + basic_lock_owner_sp_offset, + basic_lock_sp_offset, oop_maps); + NOT_PRODUCT(if (nm != NULL) nmethod_stats.note_native_nmethod(nm)); + if (PrintAssembly && nm != NULL) + Disassembler::decode(nm); + } } // verify nmethod debug_only(if (nm) nm->verify();) // might block @@ -537,16 +538,19 @@ { MutexLockerEx mu(CodeCache_lock, Mutex::_no_safepoint_check_flag); int nmethod_size = allocation_size(code_buffer, sizeof(nmethod)); - CodeOffsets offsets; - offsets.set_value(CodeOffsets::Verified_Entry, vep_offset); - offsets.set_value(CodeOffsets::Dtrace_trap, trap_offset); - offsets.set_value(CodeOffsets::Frame_Complete, frame_complete); + if (CodeCache::has_space(nmethod_size)) { + CodeOffsets offsets; + offsets.set_value(CodeOffsets::Verified_Entry, vep_offset); + offsets.set_value(CodeOffsets::Dtrace_trap, trap_offset); + offsets.set_value(CodeOffsets::Frame_Complete, frame_complete); - nm = new (nmethod_size) nmethod(method(), nmethod_size, &offsets, code_buffer, frame_size); + nm = new (nmethod_size) nmethod(method(), nmethod_size, + &offsets, code_buffer, frame_size); - NOT_PRODUCT(if (nm != NULL) nmethod_stats.note_nmethod(nm)); - if (PrintAssembly && nm != NULL) - Disassembler::decode(nm); + NOT_PRODUCT(if (nm != NULL) nmethod_stats.note_nmethod(nm)); + if (PrintAssembly && nm != NULL) + Disassembler::decode(nm); + } } // verify nmethod debug_only(if (nm) nm->verify();) // might block @@ -587,7 +591,8 @@ + round_to(handler_table->size_in_bytes(), oopSize) + round_to(nul_chk_table->size_in_bytes(), oopSize) + round_to(debug_info->data_size() , oopSize); - nm = new (nmethod_size) + if (CodeCache::has_space(nmethod_size)) { + nm = new (nmethod_size) nmethod(method(), nmethod_size, compile_id, entry_bci, offsets, orig_pc_offset, debug_info, dependencies, code_buffer, frame_size, oop_maps, @@ -595,6 +600,7 @@ nul_chk_table, compiler, comp_level); + } if (nm != NULL) { // To make dependency checking during class loading fast, record // the nmethod dependencies in the classes it is dependent on. @@ -793,9 +799,9 @@ #endif // def HAVE_DTRACE_H void* nmethod::operator new(size_t size, int nmethod_size) { - // Always leave some room in the CodeCache for I2C/C2I adapters - if (CodeCache::largest_free_block() < CodeCacheMinimumFreeSpace) return NULL; - return CodeCache::allocate(nmethod_size); + void* alloc = CodeCache::allocate(nmethod_size); + guarantee(alloc != NULL, "CodeCache should have enough space"); + return alloc; } diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/compiler/compileBroker.cpp --- a/src/share/vm/compiler/compileBroker.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/compiler/compileBroker.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -2166,6 +2166,9 @@ comp->print_timers(); } tty->cr(); + tty->print_cr(" Total compiled methods : %6d methods", CompileBroker::_total_compile_count); + tty->print_cr(" Standard compilation : %6d methods", CompileBroker::_total_standard_compile_count); + tty->print_cr(" On stack replacement : %6d methods", CompileBroker::_total_osr_compile_count); int tcb = CompileBroker::_sum_osr_bytes_compiled + CompileBroker::_sum_standard_bytes_compiled; tty->print_cr(" Total compiled bytecodes : %6d bytes", tcb); tty->print_cr(" Standard compilation : %6d bytes", CompileBroker::_sum_standard_bytes_compiled); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp --- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -854,7 +854,8 @@ assert(!isHumongous(word_size), "we do not allow humongous TLABs"); unsigned int dummy_gc_count_before; - return attempt_allocation(word_size, &dummy_gc_count_before); + int dummy_gclocker_retry_count = 0; + return attempt_allocation(word_size, &dummy_gc_count_before, &dummy_gclocker_retry_count); } HeapWord* @@ -863,14 +864,14 @@ assert_heap_not_locked_and_not_at_safepoint(); // Loop until the allocation is satisified, or unsatisfied after GC. - for (int try_count = 1; /* we'll return */; try_count += 1) { + for (int try_count = 1, gclocker_retry_count = 0; /* we'll return */; try_count += 1) { unsigned int gc_count_before; HeapWord* result = NULL; if (!isHumongous(word_size)) { - result = attempt_allocation(word_size, &gc_count_before); + result = attempt_allocation(word_size, &gc_count_before, &gclocker_retry_count); } else { - result = attempt_allocation_humongous(word_size, &gc_count_before); + result = attempt_allocation_humongous(word_size, &gc_count_before, &gclocker_retry_count); } if (result != NULL) { return result; @@ -894,6 +895,9 @@ } return result; } else { + if (gclocker_retry_count > GCLockerRetryAllocationCount) { + return NULL; + } assert(op.result() == NULL, "the result should be NULL if the VM op did not succeed"); } @@ -910,7 +914,8 @@ } HeapWord* G1CollectedHeap::attempt_allocation_slow(size_t word_size, - unsigned int *gc_count_before_ret) { + unsigned int *gc_count_before_ret, + int* gclocker_retry_count_ret) { // Make sure you read the note in attempt_allocation_humongous(). assert_heap_not_locked_and_not_at_safepoint(); @@ -986,10 +991,16 @@ return NULL; } } else { + if (*gclocker_retry_count_ret > GCLockerRetryAllocationCount) { + MutexLockerEx x(Heap_lock); + *gc_count_before_ret = total_collections(); + return NULL; + } // The GCLocker is either active or the GCLocker initiated // GC has not yet been performed. Stall until it is and // then retry the allocation. GC_locker::stall_until_clear(); + (*gclocker_retry_count_ret) += 1; } // We can reach here if we were unsuccessul in scheduling a @@ -1019,7 +1030,8 @@ } HeapWord* G1CollectedHeap::attempt_allocation_humongous(size_t word_size, - unsigned int * gc_count_before_ret) { + unsigned int * gc_count_before_ret, + int* gclocker_retry_count_ret) { // The structure of this method has a lot of similarities to // attempt_allocation_slow(). The reason these two were not merged // into a single one is that such a method would require several "if @@ -1104,10 +1116,16 @@ return NULL; } } else { + if (*gclocker_retry_count_ret > GCLockerRetryAllocationCount) { + MutexLockerEx x(Heap_lock); + *gc_count_before_ret = total_collections(); + return NULL; + } // The GCLocker is either active or the GCLocker initiated // GC has not yet been performed. Stall until it is and // then retry the allocation. GC_locker::stall_until_clear(); + (*gclocker_retry_count_ret) += 1; } // We can reach here if we were unsuccessul in scheduling a @@ -3270,12 +3288,12 @@ void G1CollectedHeap::verify(bool silent, VerifyOption vo) { - if (SafepointSynchronize::is_at_safepoint() || ! UseTLAB) { + if (SafepointSynchronize::is_at_safepoint()) { if (!silent) { gclog_or_tty->print("Roots "); } VerifyRootsClosure rootsCl(vo); assert(Thread::current()->is_VM_thread(), - "Expected to be executed serially by the VM thread at this point"); + "Expected to be executed serially by the VM thread at this point"); CodeBlobToOopClosure blobsCl(&rootsCl, /*do_marking=*/ false); VerifyKlassClosure klassCl(this, &rootsCl); @@ -3360,7 +3378,8 @@ } guarantee(!failures, "there should not have been any failures"); } else { - if (!silent) gclog_or_tty->print("(SKIPPING roots, heapRegions, remset) "); + if (!silent) + gclog_or_tty->print("(SKIPPING roots, heapRegionSets, heapRegions, remset) "); } } diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp --- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -559,18 +559,21 @@ // the mutator alloc region without taking the Heap_lock. This // should only be used for non-humongous allocations. inline HeapWord* attempt_allocation(size_t word_size, - unsigned int* gc_count_before_ret); + unsigned int* gc_count_before_ret, + int* gclocker_retry_count_ret); // Second-level mutator allocation attempt: take the Heap_lock and // retry the allocation attempt, potentially scheduling a GC // pause. This should only be used for non-humongous allocations. HeapWord* attempt_allocation_slow(size_t word_size, - unsigned int* gc_count_before_ret); + unsigned int* gc_count_before_ret, + int* gclocker_retry_count_ret); // Takes the Heap_lock and attempts a humongous allocation. It can // potentially schedule a GC pause. HeapWord* attempt_allocation_humongous(size_t word_size, - unsigned int* gc_count_before_ret); + unsigned int* gc_count_before_ret, + int* gclocker_retry_count_ret); // Allocation attempt that should be called during safepoints (e.g., // at the end of a successful GC). expect_null_mutator_alloc_region diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/gc_implementation/g1/g1CollectedHeap.inline.hpp --- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.inline.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.inline.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -60,7 +60,8 @@ inline HeapWord* G1CollectedHeap::attempt_allocation(size_t word_size, - unsigned int* gc_count_before_ret) { + unsigned int* gc_count_before_ret, + int* gclocker_retry_count_ret) { assert_heap_not_locked_and_not_at_safepoint(); assert(!isHumongous(word_size), "attempt_allocation() should not " "be called for humongous allocation requests"); @@ -68,7 +69,9 @@ HeapWord* result = _mutator_alloc_region.attempt_allocation(word_size, false /* bot_updates */); if (result == NULL) { - result = attempt_allocation_slow(word_size, gc_count_before_ret); + result = attempt_allocation_slow(word_size, + gc_count_before_ret, + gclocker_retry_count_ret); } assert_heap_not_locked(); if (result != NULL) { diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.cpp --- a/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -326,6 +326,7 @@ uint loop_count = 0; uint gc_count = 0; + int gclocker_stalled_count = 0; while (result == NULL) { // We don't want to have multiple collections for a single filled generation. @@ -354,6 +355,10 @@ return result; } + if (gclocker_stalled_count > GCLockerRetryAllocationCount) { + return NULL; + } + // Failed to allocate without a gc. if (GC_locker::is_active_and_needs_gc()) { // If this thread is not in a jni critical section, we stall @@ -366,6 +371,7 @@ if (!jthr->in_critical()) { MutexUnlocker mul(Heap_lock); GC_locker::stall_until_clear(); + gclocker_stalled_count += 1; continue; } else { if (CheckJNICalls) { diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/gc_implementation/parallelScavenge/psOldGen.cpp --- a/src/share/vm/gc_implementation/parallelScavenge/psOldGen.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/gc_implementation/parallelScavenge/psOldGen.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -256,7 +256,7 @@ } if (PrintGC && Verbose) { - if (success && GC_locker::is_active()) { + if (success && GC_locker::is_active_and_needs_gc()) { gclog_or_tty->print_cr("Garbage collection disabled, expanded heap instead"); } } diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/memory/collectorPolicy.cpp --- a/src/share/vm/memory/collectorPolicy.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/memory/collectorPolicy.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -532,7 +532,7 @@ // Loop until the allocation is satisified, // or unsatisfied after GC. - for (int try_count = 1; /* return or throw */; try_count += 1) { + for (int try_count = 1, gclocker_stalled_count = 0; /* return or throw */; try_count += 1) { HandleMark hm; // discard any handles allocated in each iteration // First allocation attempt is lock-free. @@ -576,6 +576,10 @@ } } + if (gclocker_stalled_count > GCLockerRetryAllocationCount) { + return NULL; // we didn't get to do a GC and we didn't get any memory + } + // If this thread is not in a jni critical section, we stall // the requestor until the critical section has cleared and // GC allowed. When the critical section clears, a GC is @@ -587,6 +591,7 @@ MutexUnlocker mul(Heap_lock); // Wait for JNI critical section to be exited GC_locker::stall_until_clear(); + gclocker_stalled_count += 1; continue; } else { if (CheckJNICalls) { diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/memory/universe.cpp --- a/src/share/vm/memory/universe.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/memory/universe.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -953,15 +953,6 @@ void universe2_init() { EXCEPTION_MARK; Universe::genesis(CATCH); - // Although we'd like to verify here that the state of the heap - // is good, we can't because the main thread has not yet added - // itself to the threads list (so, using current interfaces - // we can't "fill" its TLAB), unless TLABs are disabled. - if (VerifyBeforeGC && !UseTLAB && - Universe::heap()->total_collections() >= VerifyGCStartAt) { - Universe::heap()->prepare_for_verify(); - Universe::verify(); // make sure we're starting with a clean slate - } } diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/oops/instanceKlass.cpp --- a/src/share/vm/oops/instanceKlass.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/oops/instanceKlass.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -2228,8 +2228,6 @@ } void InstanceKlass::clean_method_data(BoolObjectClosure* is_alive) { -#ifdef COMPILER2 - // Currently only used by C2. for (int m = 0; m < methods()->length(); m++) { MethodData* mdo = methods()->at(m)->method_data(); if (mdo != NULL) { @@ -2240,15 +2238,6 @@ } } } -#else -#ifdef ASSERT - // Verify that we haven't started to use MDOs for C1. - for (int m = 0; m < methods()->length(); m++) { - MethodData* mdo = methods()->at(m)->method_data(); - assert(mdo == NULL, "Didn't expect C1 to use MDOs"); - } -#endif // ASSERT -#endif // !COMPILER2 } diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/oops/method.cpp --- a/src/share/vm/oops/method.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/oops/method.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -967,6 +967,32 @@ return false; } + +/** + * Returns true if this is one of the specially treated methods for + * security related stack walks (like Reflection.getCallerClass). + */ +bool Method::is_ignored_by_security_stack_walk() const { + const bool use_new_reflection = JDK_Version::is_gte_jdk14x_version() && UseNewReflection; + + assert(intrinsic_id() != vmIntrinsics::_invoke || Universe::reflect_invoke_cache()->is_same_method((Method*)this), "sanity"); + if (intrinsic_id() == vmIntrinsics::_invoke) { + // This is Method.invoke() -- ignore it + return true; + } + if (use_new_reflection && + method_holder()->is_subclass_of(SystemDictionary::reflect_MethodAccessorImpl_klass())) { + // This is an auxilary frame -- ignore it + return true; + } + if (is_method_handle_intrinsic() || is_compiled_lambda_form()) { + // This is an internal adapter frame for method handles -- ignore it + return true; + } + return false; +} + + // Constant pool structure for invoke methods: enum { _imcp_invoke_name = 1, // utf8: 'invokeExact', etc. @@ -1180,13 +1206,13 @@ // because we are not loading from core libraries // exception: the AES intrinsics come from lib/ext/sunjce_provider.jar // which does not use the class default class loader so we check for its loader here - if ((InstanceKlass::cast(holder)->class_loader() != NULL) && - InstanceKlass::cast(holder)->class_loader()->klass()->name() != vmSymbols::sun_misc_Launcher_ExtClassLoader()) { + InstanceKlass* ik = InstanceKlass::cast(holder); + if ((ik->class_loader() != NULL) && !SystemDictionary::is_ext_class_loader(ik->class_loader())) { return vmSymbols::NO_SID; // regardless of name, no intrinsics here } // see if the klass name is well-known: - Symbol* klass_name = InstanceKlass::cast(holder)->name(); + Symbol* klass_name = ik->name(); return vmSymbols::find_sid(klass_name); } diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/oops/method.hpp --- a/src/share/vm/oops/method.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/oops/method.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -118,11 +118,12 @@ #endif u2 _method_size; // size of this object u1 _intrinsic_id; // vmSymbols::intrinsic_id (0 == _none) - u1 _jfr_towrite : 1, // Flags - _force_inline : 1, - _hidden : 1, - _dont_inline : 1, - : 4; + u1 _jfr_towrite : 1, // Flags + _caller_sensitive : 1, + _force_inline : 1, + _hidden : 1, + _dont_inline : 1, + : 3; u2 _interpreter_throwout_count; // Count of times method was exited via exception while interpreting u2 _number_of_breakpoints; // fullspeed debugging support InvocationCounter _invocation_counter; // Incremented before each activation of the method - used to trigger frequency-based optimizations @@ -618,6 +619,9 @@ // Reflection support bool is_overridden_in(Klass* k) const; + // Stack walking support + bool is_ignored_by_security_stack_walk() const; + // JSR 292 support bool is_method_handle_intrinsic() const; // MethodHandles::is_signature_polymorphic_intrinsic(intrinsic_id) bool is_compiled_lambda_form() const; // intrinsic_id() == vmIntrinsics::_compiledLambdaForm @@ -705,15 +709,16 @@ void init_intrinsic_id(); // updates from _none if a match static vmSymbols::SID klass_id_for_intrinsics(Klass* holder); - bool jfr_towrite() { return _jfr_towrite; } - void set_jfr_towrite(bool towrite) { _jfr_towrite = towrite; } - - bool force_inline() { return _force_inline; } - void set_force_inline(bool x) { _force_inline = x; } - bool dont_inline() { return _dont_inline; } - void set_dont_inline(bool x) { _dont_inline = x; } - bool is_hidden() { return _hidden; } - void set_hidden(bool x) { _hidden = x; } + bool jfr_towrite() { return _jfr_towrite; } + void set_jfr_towrite(bool x) { _jfr_towrite = x; } + bool caller_sensitive() { return _caller_sensitive; } + void set_caller_sensitive(bool x) { _caller_sensitive = x; } + bool force_inline() { return _force_inline; } + void set_force_inline(bool x) { _force_inline = x; } + bool dont_inline() { return _dont_inline; } + void set_dont_inline(bool x) { _dont_inline = x; } + bool is_hidden() { return _hidden; } + void set_hidden(bool x) { _hidden = x; } ConstMethod::MethodType method_type() const { return _constMethod->method_type(); } diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/oops/methodData.cpp --- a/src/share/vm/oops/methodData.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/oops/methodData.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -392,6 +392,9 @@ } int MethodData::bytecode_cell_count(Bytecodes::Code code) { +#if defined(COMPILER1) && !defined(COMPILER2) + return no_profile_data; +#else switch (code) { case Bytecodes::_checkcast: case Bytecodes::_instanceof: @@ -438,6 +441,7 @@ return variable_cell_count; } return no_profile_data; +#endif } // Compute the size of the profiling information corresponding to @@ -509,6 +513,9 @@ // the segment in bytes. int MethodData::initialize_data(BytecodeStream* stream, int data_index) { +#if defined(COMPILER1) && !defined(COMPILER2) + return 0; +#else int cell_count = -1; int tag = DataLayout::no_tag; DataLayout* data_layout = data_layout_at(data_index); @@ -587,6 +594,7 @@ assert(!bytecode_has_profile(c), "agree w/ !BHP"); return 0; } +#endif } // Get the data at an arbitrary (sort of) data index. diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/opto/graphKit.cpp --- a/src/share/vm/opto/graphKit.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/opto/graphKit.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -3445,7 +3445,6 @@ void GraphKit::final_sync(IdealKit& ideal) { // Final sync IdealKit and graphKit. - __ drain_delay_transform(); sync_kit(ideal); } diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/opto/idealKit.cpp --- a/src/share/vm/opto/idealKit.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/opto/idealKit.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -48,9 +48,9 @@ _cvstate = NULL; // We can go memory state free or else we need the entire memory state assert(_initial_memory == NULL || _initial_memory->Opcode() == Op_MergeMem, "memory must be pre-split"); + assert(!_gvn.is_IterGVN(), "IdealKit can't be used during Optimize phase"); int init_size = 5; _pending_cvstates = new (C->node_arena()) GrowableArray(C->node_arena(), init_size, 0, 0); - _delay_transform = new (C->node_arena()) GrowableArray(C->node_arena(), init_size, 0, 0); DEBUG_ONLY(_state = new (C->node_arena()) GrowableArray(C->node_arena(), init_size, 0, 0)); if (!has_declarations) { declarations_done(); @@ -296,19 +296,16 @@ return delay_transform(n); } else { n = gvn().transform(n); - if (!gvn().is_IterGVN()) { - C->record_for_igvn(n); - } + C->record_for_igvn(n); return n; } } //-----------------------------delay_transform----------------------------------- Node* IdealKit::delay_transform(Node* n) { - if (!gvn().is_IterGVN() || !gvn().is_IterGVN()->delay_transform()) { - gvn().set_type(n, n->bottom_type()); - } - _delay_transform->push(n); + // Delay transform until IterativeGVN + gvn().set_type(n, n->bottom_type()); + C->record_for_igvn(n); return n; } @@ -332,17 +329,6 @@ for (uint i = 0; i < m->req(); i++) m->set_req(i, NULL); } -//-----------------------------drain_delay_transform---------------------------- -void IdealKit::drain_delay_transform() { - while (_delay_transform->length() > 0) { - Node* n = _delay_transform->pop(); - gvn().transform(n); - if (!gvn().is_IterGVN()) { - C->record_for_igvn(n); - } - } -} - //-----------------------------IdealVariable---------------------------- IdealVariable::IdealVariable(IdealKit &k) { k.declare(this); @@ -351,9 +337,7 @@ Node* IdealKit::memory(uint alias_idx) { MergeMemNode* mem = merged_memory(); Node* p = mem->memory_at(alias_idx); - if (!gvn().is_IterGVN() || !gvn().is_IterGVN()->delay_transform()) { - _gvn.set_type(p, Type::MEMORY); // must be mapped - } + _gvn.set_type(p, Type::MEMORY); // must be mapped return p; } diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/opto/idealKit.hpp --- a/src/share/vm/opto/idealKit.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/opto/idealKit.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -102,7 +102,6 @@ Compile * const C; PhaseGVN &_gvn; GrowableArray* _pending_cvstates; // stack of cvstates - GrowableArray* _delay_transform; // delay invoking gvn.transform until drain Node* _cvstate; // current cvstate (control, memory and variables) uint _var_ct; // number of variables bool _delay_all_transforms; // flag forcing all transforms to be delayed @@ -121,7 +120,7 @@ void clear(Node* m); // clear a cvstate void stop() { clear(_cvstate); } // clear current cvstate Node* delay_transform(Node* n); - Node* transform(Node* n); // gvn.transform or push node on delay list + Node* transform(Node* n); // gvn.transform or skip it Node* promote_to_phi(Node* n, Node* reg);// Promote "n" to a phi on region "reg" bool was_promoted_to_phi(Node* n, Node* reg) { return (n->is_Phi() && n->in(0) == reg); @@ -146,7 +145,6 @@ IdealKit(GraphKit* gkit, bool delay_all_transforms = false, bool has_declarations = false); ~IdealKit() { stop(); - drain_delay_transform(); } void sync_kit(GraphKit* gkit); @@ -173,7 +171,6 @@ void bind(Node* lab); void goto_(Node* lab, bool bind = false); void declarations_done(); - void drain_delay_transform(); Node* IfTrue(IfNode* iff) { return transform(new (C) IfTrueNode(iff)); } Node* IfFalse(IfNode* iff) { return transform(new (C) IfFalseNode(iff)); } @@ -198,7 +195,11 @@ Node* thread() { return gvn().transform(new (C) ThreadLocalNode()); } // Pointers - Node* AddP(Node *base, Node *ptr, Node *off) { return transform(new (C) AddPNode(base, ptr, off)); } + + // Raw address should be transformed regardless 'delay_transform' flag + // to produce canonical form CastX2P(offset). + Node* AddP(Node *base, Node *ptr, Node *off) { return _gvn.transform(new (C) AddPNode(base, ptr, off)); } + Node* CmpP(Node* l, Node* r) { return transform(new (C) CmpPNode(l, r)); } #ifdef _LP64 Node* XorX(Node* l, Node* r) { return transform(new (C) XorLNode(l, r)); } @@ -208,8 +209,6 @@ Node* URShiftX(Node* l, Node* r) { return transform(new (C) URShiftXNode(l, r)); } Node* ConX(jint k) { return (Node*)gvn().MakeConX(k); } Node* CastPX(Node* ctl, Node* p) { return transform(new (C) CastP2XNode(ctl, p)); } - // Add a fixed offset to a pointer - Node* basic_plus_adr(Node* base, Node* ptr, intptr_t offset); // Memory operations diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/opto/ifg.cpp --- a/src/share/vm/opto/ifg.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/opto/ifg.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -37,8 +37,6 @@ #include "opto/memnode.hpp" #include "opto/opcodes.hpp" -#define EXACT_PRESSURE 1 - //============================================================================= //------------------------------IFG-------------------------------------------- PhaseIFG::PhaseIFG( Arena *arena ) : Phase(Interference_Graph), _arena(arena) { @@ -445,23 +443,15 @@ pressure[1] -= lrg->reg_pressure(); if( pressure[1] == (uint)FLOATPRESSURE ) { hrp_index[1] = where; -#ifdef EXACT_PRESSURE - if( pressure[1] > b->_freg_pressure ) - b->_freg_pressure = pressure[1]+1; -#else - b->_freg_pressure = (uint)FLOATPRESSURE+1; -#endif + if( pressure[1] > b->_freg_pressure ) + b->_freg_pressure = pressure[1]+1; } } else if( lrg->mask().overlap(*Matcher::idealreg2regmask[Op_RegI]) ) { pressure[0] -= lrg->reg_pressure(); if( pressure[0] == (uint)INTPRESSURE ) { hrp_index[0] = where; -#ifdef EXACT_PRESSURE - if( pressure[0] > b->_reg_pressure ) - b->_reg_pressure = pressure[0]+1; -#else - b->_reg_pressure = (uint)INTPRESSURE+1; -#endif + if( pressure[0] > b->_reg_pressure ) + b->_reg_pressure = pressure[0]+1; } } } @@ -526,17 +516,13 @@ if (lrg.mask().is_UP() && lrg.mask_size()) { if (lrg._is_float || lrg._is_vector) { // Count float pressure pressure[1] += lrg.reg_pressure(); -#ifdef EXACT_PRESSURE if( pressure[1] > b->_freg_pressure ) b->_freg_pressure = pressure[1]; -#endif // Count int pressure, but do not count the SP, flags } else if( lrgs(lidx).mask().overlap(*Matcher::idealreg2regmask[Op_RegI]) ) { pressure[0] += lrg.reg_pressure(); -#ifdef EXACT_PRESSURE if( pressure[0] > b->_reg_pressure ) b->_reg_pressure = pressure[0]; -#endif } } } @@ -589,30 +575,20 @@ RegMask itmp = lrgs(r).mask(); itmp.AND(*Matcher::idealreg2regmask[Op_RegI]); int iregs = itmp.Size(); -#ifdef EXACT_PRESSURE if( pressure[0]+iregs > b->_reg_pressure ) b->_reg_pressure = pressure[0]+iregs; -#endif if( pressure[0] <= (uint)INTPRESSURE && pressure[0]+iregs > (uint)INTPRESSURE ) { -#ifndef EXACT_PRESSURE - b->_reg_pressure = (uint)INTPRESSURE+1; -#endif hrp_index[0] = j-1; } // Count the float-only registers RegMask ftmp = lrgs(r).mask(); ftmp.AND(*Matcher::idealreg2regmask[Op_RegD]); int fregs = ftmp.Size(); -#ifdef EXACT_PRESSURE if( pressure[1]+fregs > b->_freg_pressure ) b->_freg_pressure = pressure[1]+fregs; -#endif if( pressure[1] <= (uint)FLOATPRESSURE && pressure[1]+fregs > (uint)FLOATPRESSURE ) { -#ifndef EXACT_PRESSURE - b->_freg_pressure = (uint)FLOATPRESSURE+1; -#endif hrp_index[1] = j-1; } } @@ -769,16 +745,12 @@ if (lrg.mask().is_UP() && lrg.mask_size()) { if (lrg._is_float || lrg._is_vector) { pressure[1] += lrg.reg_pressure(); -#ifdef EXACT_PRESSURE if( pressure[1] > b->_freg_pressure ) b->_freg_pressure = pressure[1]; -#endif } else if( lrg.mask().overlap(*Matcher::idealreg2regmask[Op_RegI]) ) { pressure[0] += lrg.reg_pressure(); -#ifdef EXACT_PRESSURE if( pressure[0] > b->_reg_pressure ) b->_reg_pressure = pressure[0]; -#endif } } assert( pressure[0] == count_int_pressure (&liveout), "" ); @@ -794,21 +766,13 @@ // the whole block is high pressure. if( pressure[0] > (uint)INTPRESSURE ) { hrp_index[0] = 0; -#ifdef EXACT_PRESSURE if( pressure[0] > b->_reg_pressure ) b->_reg_pressure = pressure[0]; -#else - b->_reg_pressure = (uint)INTPRESSURE+1; -#endif } if( pressure[1] > (uint)FLOATPRESSURE ) { hrp_index[1] = 0; -#ifdef EXACT_PRESSURE if( pressure[1] > b->_freg_pressure ) b->_freg_pressure = pressure[1]; -#else - b->_freg_pressure = (uint)FLOATPRESSURE+1; -#endif } // Compute high pressure indice; avoid landing in the middle of projnodes diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/opto/library_call.cpp --- a/src/share/vm/opto/library_call.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/opto/library_call.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -231,7 +231,6 @@ void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark); bool inline_native_clone(bool is_virtual); bool inline_native_Reflection_getCallerClass(); - bool is_method_invoke_or_aux_frame(JVMState* jvms); // Helper function for inlining native object hash method bool inline_native_hashcode(bool is_virtual, bool is_static); bool inline_native_getClass(); @@ -393,7 +392,7 @@ case vmIntrinsics::_getCallerClass: if (!UseNewReflection) return NULL; if (!InlineReflectionGetCallerClass) return NULL; - if (!JDK_Version::is_gte_jdk14x_version()) return NULL; + if (SystemDictionary::reflect_CallerSensitive_klass() == NULL) return NULL; break; case vmIntrinsics::_bitCount_i: @@ -3872,13 +3871,13 @@ } //-----------------inline_native_Reflection_getCallerClass--------------------- -// public static native Class sun.reflect.Reflection.getCallerClass(int realFramesToSkip); +// public static native Class sun.reflect.Reflection.getCallerClass(); // // In the presence of deep enough inlining, getCallerClass() becomes a no-op. // -// NOTE that this code must perform the same logic as -// vframeStream::security_get_caller_frame in that it must skip -// Method.invoke() and auxiliary frames. +// NOTE: This code must perform the same logic as JVM_GetCallerClass +// in that it must skip particular security frames and checks for +// caller sensitive methods. bool LibraryCallKit::inline_native_Reflection_getCallerClass() { #ifndef PRODUCT if ((PrintIntrinsics || PrintInlining || PrintOptoInlining) && Verbose) { @@ -3886,35 +3885,6 @@ } #endif - Node* caller_depth_node = argument(0); - - // The depth value must be a constant in order for the runtime call - // to be eliminated. - const TypeInt* caller_depth_type = _gvn.type(caller_depth_node)->isa_int(); - if (caller_depth_type == NULL || !caller_depth_type->is_con()) { -#ifndef PRODUCT - if ((PrintIntrinsics || PrintInlining || PrintOptoInlining) && Verbose) { - tty->print_cr(" Bailing out because caller depth was not a constant"); - } -#endif - return false; - } - // Note that the JVM state at this point does not include the - // getCallerClass() frame which we are trying to inline. The - // semantics of getCallerClass(), however, are that the "first" - // frame is the getCallerClass() frame, so we subtract one from the - // requested depth before continuing. We don't inline requests of - // getCallerClass(0). - int caller_depth = caller_depth_type->get_con() - 1; - if (caller_depth < 0) { -#ifndef PRODUCT - if ((PrintIntrinsics || PrintInlining || PrintOptoInlining) && Verbose) { - tty->print_cr(" Bailing out because caller depth was %d", caller_depth); - } -#endif - return false; - } - if (!jvms()->has_method()) { #ifndef PRODUCT if ((PrintIntrinsics || PrintInlining || PrintOptoInlining) && Verbose) { @@ -3923,95 +3893,67 @@ #endif return false; } - int _depth = jvms()->depth(); // cache call chain depth // Walk back up the JVM state to find the caller at the required - // depth. NOTE that this code must perform the same logic as - // vframeStream::security_get_caller_frame in that it must skip - // Method.invoke() and auxiliary frames. Note also that depth is - // 1-based (1 is the bottom of the inlining). - int inlining_depth = _depth; - JVMState* caller_jvms = NULL; - - if (inlining_depth > 0) { - caller_jvms = jvms(); - assert(caller_jvms = jvms()->of_depth(inlining_depth), "inlining_depth == our depth"); - do { - // The following if-tests should be performed in this order - if (is_method_invoke_or_aux_frame(caller_jvms)) { - // Skip a Method.invoke() or auxiliary frame - } else if (caller_depth > 0) { - // Skip real frame - --caller_depth; - } else { - // We're done: reached desired caller after skipping. - break; + // depth. + JVMState* caller_jvms = jvms(); + + // Cf. JVM_GetCallerClass + // NOTE: Start the loop at depth 1 because the current JVM state does + // not include the Reflection.getCallerClass() frame. + for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) { + ciMethod* m = caller_jvms->method(); + switch (n) { + case 0: + fatal("current JVM state does not include the Reflection.getCallerClass frame"); + break; + case 1: + // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass). + if (!m->caller_sensitive()) { +#ifndef PRODUCT + if ((PrintIntrinsics || PrintInlining || PrintOptoInlining) && Verbose) { + tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n); + } +#endif + return false; // bail-out; let JVM_GetCallerClass do the work } - caller_jvms = caller_jvms->caller(); - --inlining_depth; - } while (inlining_depth > 0); - } - - if (inlining_depth == 0) { + break; + default: + if (!m->is_ignored_by_security_stack_walk()) { + // We have reached the desired frame; return the holder class. + // Acquire method holder as java.lang.Class and push as constant. + ciInstanceKlass* caller_klass = caller_jvms->method()->holder(); + ciInstance* caller_mirror = caller_klass->java_mirror(); + set_result(makecon(TypeInstPtr::make(caller_mirror))); + #ifndef PRODUCT - if ((PrintIntrinsics || PrintInlining || PrintOptoInlining) && Verbose) { - tty->print_cr(" Bailing out because caller depth (%d) exceeded inlining depth (%d)", caller_depth_type->get_con(), _depth); - tty->print_cr(" JVM state at this point:"); - for (int i = _depth; i >= 1; i--) { - ciMethod* m = jvms()->of_depth(i)->method(); - tty->print_cr(" %d) %s.%s", i, m->holder()->name()->as_utf8(), m->name()->as_utf8()); - } - } + if ((PrintIntrinsics || PrintInlining || PrintOptoInlining) && Verbose) { + tty->print_cr(" Succeeded: caller = %d) %s.%s, JVMS depth = %d", n, caller_klass->name()->as_utf8(), caller_jvms->method()->name()->as_utf8(), jvms()->depth()); + tty->print_cr(" JVM state at this point:"); + for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) { + ciMethod* m = jvms()->of_depth(i)->method(); + tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8()); + } + } #endif - return false; // Reached end of inlining + return true; + } + break; + } } - // Acquire method holder as java.lang.Class - ciInstanceKlass* caller_klass = caller_jvms->method()->holder(); - ciInstance* caller_mirror = caller_klass->java_mirror(); - - // Push this as a constant - set_result(makecon(TypeInstPtr::make(caller_mirror))); - #ifndef PRODUCT if ((PrintIntrinsics || PrintInlining || PrintOptoInlining) && Verbose) { - tty->print_cr(" Succeeded: caller = %s.%s, caller depth = %d, depth = %d", caller_klass->name()->as_utf8(), caller_jvms->method()->name()->as_utf8(), caller_depth_type->get_con(), _depth); + tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth()); tty->print_cr(" JVM state at this point:"); - for (int i = _depth; i >= 1; i--) { + for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) { ciMethod* m = jvms()->of_depth(i)->method(); - tty->print_cr(" %d) %s.%s", i, m->holder()->name()->as_utf8(), m->name()->as_utf8()); + tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8()); } } #endif - return true; -} - -// Helper routine for above -bool LibraryCallKit::is_method_invoke_or_aux_frame(JVMState* jvms) { - ciMethod* method = jvms->method(); - - // Is this the Method.invoke method itself? - if (method->intrinsic_id() == vmIntrinsics::_invoke) - return true; - - // Is this a helper, defined somewhere underneath MethodAccessorImpl. - ciKlass* k = method->holder(); - if (k->is_instance_klass()) { - ciInstanceKlass* ik = k->as_instance_klass(); - for (; ik != NULL; ik = ik->super()) { - if (ik->name() == ciSymbol::sun_reflect_MethodAccessorImpl() && - ik == env()->find_system_klass(ik->name())) { - return true; - } - } - } - else if (method->is_method_handle_intrinsic() || - method->is_compiled_lambda_form()) { - // This is an internal adapter frame from the MethodHandleCompiler -- skip it - return true; - } - - return false; + + return false; // bail-out; let JVM_GetCallerClass do the work } bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) { diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/opto/loopnode.cpp --- a/src/share/vm/opto/loopnode.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/opto/loopnode.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -2251,6 +2251,11 @@ return; } + // clear out the dead code after build_loop_late + while (_deadlist.size()) { + _igvn.remove_globally_dead_node(_deadlist.pop()); + } + if (stop_early) { assert(do_expensive_nodes, "why are we here?"); if (process_expensive_nodes()) { @@ -2260,9 +2265,7 @@ // nodes again. C->set_major_progress(); } - _igvn.optimize(); - return; } @@ -2273,11 +2276,6 @@ eliminate_useless_predicates(); } - // clear out the dead code - while(_deadlist.size()) { - _igvn.remove_globally_dead_node(_deadlist.pop()); - } - #ifndef PRODUCT C->verify_graph_edges(); if (_verify_me) { // Nested verify pass? diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/opto/output.cpp --- a/src/share/vm/opto/output.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/opto/output.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -449,6 +449,17 @@ int max_loop_pad = nb->code_alignment()-relocInfo::addr_unit(); if (max_loop_pad > 0) { assert(is_power_of_2(max_loop_pad+relocInfo::addr_unit()), ""); + // Adjust last_call_adr and/or last_avoid_back_to_back_adr. + // If either is the last instruction in this block, bump by + // max_loop_pad in lock-step with blk_size, so sizing + // calculations in subsequent blocks still can conservatively + // detect that it may the last instruction in this block. + if (last_call_adr == blk_starts[i]+blk_size) { + last_call_adr += max_loop_pad; + } + if (last_avoid_back_to_back_adr == blk_starts[i]+blk_size) { + last_avoid_back_to_back_adr += max_loop_pad; + } blk_size += max_loop_pad; } } @@ -1193,8 +1204,6 @@ int last_call_offset = -1; int last_avoid_back_to_back_offset = -1; #ifdef ASSERT - int block_alignment_padding = 0; - uint* jmp_target = NEW_RESOURCE_ARRAY(uint,nblocks); uint* jmp_offset = NEW_RESOURCE_ARRAY(uint,nblocks); uint* jmp_size = NEW_RESOURCE_ARRAY(uint,nblocks); @@ -1228,8 +1237,6 @@ Node *delay_slot = NULL; for (uint i=0; i < nblocks; i++) { - guarantee(blk_starts[i] >= (uint)cb->insts_size(),"should not increase size"); - Block *b = _cfg->_blocks[i]; Node *head = b->head(); @@ -1250,14 +1257,6 @@ jmp_offset[i] = 0; jmp_size[i] = 0; jmp_rule[i] = 0; - - // Maximum alignment padding for loop block was used - // during first round of branches shortening, as result - // padding for nodes (sfpt after call) was not added. - // Take this into account for block's size change check - // and allow increase block's size by the difference - // of maximum and actual alignment paddings. - int orig_blk_size = blk_starts[i+1] - blk_starts[i] + block_alignment_padding; #endif int blk_offset = current_offset; @@ -1557,8 +1556,6 @@ } } // End for all instructions in block - assert((uint)blk_offset <= blk_starts[i], "shouldn't increase distance"); - blk_starts[i] = blk_offset; // If the next block is the top of a loop, pad this block out to align // the loop top a little. Helps prevent pipe stalls at loop back branches. @@ -1572,16 +1569,13 @@ nop->emit(*cb, _regalloc); current_offset = cb->insts_size(); } -#ifdef ASSERT - int max_loop_pad = nb->code_alignment()-relocInfo::addr_unit(); - block_alignment_padding = (max_loop_pad - padding); - assert(block_alignment_padding >= 0, "sanity"); -#endif } // Verify that the distance for generated before forward // short branches is still valid. - assert(orig_blk_size >= (current_offset - blk_offset), "shouldn't increase block size"); - + guarantee((int)(blk_starts[i+1] - blk_starts[i]) >= (current_offset - blk_offset), "shouldn't increase block size"); + + // Save new block start offset + blk_starts[i] = blk_offset; } // End of for all blocks blk_starts[nblocks] = current_offset; diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/opto/parse2.cpp --- a/src/share/vm/opto/parse2.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/opto/parse2.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -104,7 +104,8 @@ if (C->log() != NULL) C->log()->elem("observe that='!need_range_check'"); } - if (!arytype->klass()->is_loaded()) { + ciKlass * arytype_klass = arytype->klass(); + if ((arytype_klass != NULL) && (!arytype_klass->is_loaded())) { // Only fails for some -Xcomp runs // The class is unloaded. We have to run this bytecode in the interpreter. uncommon_trap(Deoptimization::Reason_unloaded, @@ -1385,6 +1386,7 @@ if (TraceOptoParse) { tty->print(" @"); dump_bci(bci()); + tty->cr(); } #endif diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/opto/phaseX.cpp --- a/src/share/vm/opto/phaseX.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/opto/phaseX.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -1166,31 +1166,30 @@ if (progress_state == PROCESS_INPUTS) { // After following inputs, continue to outputs _stack.set_index(PROCESS_OUTPUTS); - // Remove from iterative worklist - _worklist.remove(dead); if (!dead->is_Con()) { // Don't kill cons but uses bool recurse = false; // Remove from hash table _table.hash_delete( dead ); // Smash all inputs to 'dead', isolating him completely - for( uint i = 0; i < dead->req(); i++ ) { + for (uint i = 0; i < dead->req(); i++) { Node *in = dead->in(i); - if( in ) { // Points to something? - dead->set_req(i,NULL); // Kill the edge - if (in->outcnt() == 0 && in != C->top()) {// Made input go dead? + if (in != NULL && in != C->top()) { // Points to something? + int nrep = dead->replace_edge(in, NULL); // Kill edges + assert((nrep > 0), "sanity"); + if (in->outcnt() == 0) { // Made input go dead? _stack.push(in, PROCESS_INPUTS); // Recursively remove recurse = true; } else if (in->outcnt() == 1 && in->has_special_unique_user()) { _worklist.push(in->unique_out()); } else if (in->outcnt() <= 2 && dead->is_Phi()) { - if( in->Opcode() == Op_Region ) + if (in->Opcode() == Op_Region) { _worklist.push(in); - else if( in->is_Store() ) { + } else if (in->is_Store()) { DUIterator_Fast imax, i = in->fast_outs(imax); _worklist.push(in->fast_out(i)); i++; - if(in->outcnt() == 2) { + if (in->outcnt() == 2) { _worklist.push(in->fast_out(i)); i++; } @@ -1209,38 +1208,42 @@ } } } - } - } - C->record_dead_node(dead->_idx); - if (dead->is_macro()) { - C->remove_macro_node(dead); - } - if (dead->is_expensive()) { - C->remove_expensive_node(dead); - } - + } // if (in != NULL && in != C->top()) + } // for (uint i = 0; i < dead->req(); i++) if (recurse) { continue; } - } - // Constant node that has no out-edges and has only one in-edge from - // root is usually dead. However, sometimes reshaping walk makes - // it reachable by adding use edges. So, we will NOT count Con nodes - // as dead to be conservative about the dead node count at any - // given time. - } + } // if (!dead->is_Con()) + } // if (progress_state == PROCESS_INPUTS) // Aggressively kill globally dead uses // (Rather than pushing all the outs at once, we push one at a time, // plus the parent to resume later, because of the indefinite number // of edge deletions per loop trip.) if (dead->outcnt() > 0) { - // Recursively remove + // Recursively remove output edges _stack.push(dead->raw_out(0), PROCESS_INPUTS); } else { + // Finished disconnecting all input and output edges. _stack.pop(); + // Remove dead node from iterative worklist + _worklist.remove(dead); + // Constant node that has no out-edges and has only one in-edge from + // root is usually dead. However, sometimes reshaping walk makes + // it reachable by adding use edges. So, we will NOT count Con nodes + // as dead to be conservative about the dead node count at any + // given time. + if (!dead->is_Con()) { + C->record_dead_node(dead->_idx); + } + if (dead->is_macro()) { + C->remove_macro_node(dead); + } + if (dead->is_expensive()) { + C->remove_expensive_node(dead); + } } - } + } // while (_stack.is_nonempty()) } //------------------------------subsume_node----------------------------------- diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/prims/jvm.cpp --- a/src/share/vm/prims/jvm.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/prims/jvm.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -30,6 +30,7 @@ #include "classfile/systemDictionary.hpp" #include "classfile/vmSymbols.hpp" #include "gc_interface/collectedHeap.inline.hpp" +#include "interpreter/bytecode.hpp" #include "memory/oopFactory.hpp" #include "memory/universe.inline.hpp" #include "oops/fieldStreams.hpp" @@ -665,8 +666,51 @@ JVM_ENTRY(jclass, JVM_GetCallerClass(JNIEnv* env, int depth)) JVMWrapper("JVM_GetCallerClass"); - Klass* k = thread->security_get_caller_class(depth); - return (k == NULL) ? NULL : (jclass) JNIHandles::make_local(env, k->java_mirror()); + + // Pre-JDK 8 and early builds of JDK 8 don't have a CallerSensitive annotation. + if (SystemDictionary::reflect_CallerSensitive_klass() == NULL) { + Klass* k = thread->security_get_caller_class(depth); + return (k == NULL) ? NULL : (jclass) JNIHandles::make_local(env, k->java_mirror()); + } else { + // Basic handshaking with Java_sun_reflect_Reflection_getCallerClass + assert(depth == -1, "wrong handshake depth"); + } + + // Getting the class of the caller frame. + // + // The call stack at this point looks something like this: + // + // [0] [ @CallerSensitive public sun.reflect.Reflection.getCallerClass ] + // [1] [ @CallerSensitive API.method ] + // [.] [ (skipped intermediate frames) ] + // [n] [ caller ] + vframeStream vfst(thread); + // Cf. LibraryCallKit::inline_native_Reflection_getCallerClass + for (int n = 0; !vfst.at_end(); vfst.security_next(), n++) { + Method* m = vfst.method(); + assert(m != NULL, "sanity"); + switch (n) { + case 0: + // This must only be called from Reflection.getCallerClass + if (m->intrinsic_id() != vmIntrinsics::_getCallerClass) { + THROW_MSG_NULL(vmSymbols::java_lang_InternalError(), "JVM_GetCallerClass must only be called from Reflection.getCallerClass"); + } + // fall-through + case 1: + // Frame 0 and 1 must be caller sensitive. + if (!m->caller_sensitive()) { + THROW_MSG_NULL(vmSymbols::java_lang_InternalError(), err_msg("CallerSensitive annotation expected at frame %d", n)); + } + break; + default: + if (!m->is_ignored_by_security_stack_walk()) { + // We have reached the desired frame; return the holder class. + return (jclass) JNIHandles::make_local(env, m->method_holder()->java_mirror()); + } + break; + } + } + return NULL; JVM_END @@ -3208,11 +3252,24 @@ KlassLink* first = NULL; KlassLink* last = NULL; int depth = 0; - - for(vframeStream vfst(thread); !vfst.at_end(); vfst.security_get_caller_frame(1)) { + vframeStream vfst(thread); + + if (SystemDictionary::reflect_CallerSensitive_klass() != NULL) { + // This must only be called from SecurityManager.getClassContext + Method* m = vfst.method(); + if (!(m->method_holder() == SystemDictionary::SecurityManager_klass() && + m->name() == vmSymbols::getClassContext_name() && + m->signature() == vmSymbols::void_class_array_signature())) { + THROW_MSG_NULL(vmSymbols::java_lang_InternalError(), "JVM_GetClassContext must only be called from SecurityManager.getClassContext"); + } + } + + // Collect method holders + for (; !vfst.at_end(); vfst.security_next()) { + Method* m = vfst.method(); // Native frames are not returned - if (!vfst.method()->is_native()) { - Klass* holder = vfst.method()->method_holder(); + if (!m->is_ignored_by_security_stack_walk() && !m->is_native()) { + Klass* holder = m->method_holder(); assert(holder->is_klass(), "just checking"); depth++; KlassLink* l = new KlassLink(KlassHandle(thread, holder)); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/prims/jvmtiEventController.cpp --- a/src/share/vm/prims/jvmtiEventController.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/prims/jvmtiEventController.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -39,7 +39,12 @@ #include "runtime/vm_operations.hpp" #ifdef JVMTI_TRACE -#define EC_TRACE(out) if (JvmtiTrace::trace_event_controller()) { SafeResourceMark rm; tty->print_cr out; } while (0) +#define EC_TRACE(out) do { \ + if (JvmtiTrace::trace_event_controller()) { \ + SafeResourceMark rm; \ + tty->print_cr out; \ + } \ +} while (0) #else #define EC_TRACE(out) #endif /*JVMTI_TRACE */ diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/prims/jvmtiRedefineClassesTrace.hpp --- a/src/share/vm/prims/jvmtiRedefineClassesTrace.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/prims/jvmtiRedefineClassesTrace.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -72,36 +72,6 @@ // 0x20000000 | 536870912 - unused // 0x40000000 | 1073741824 - unused // 0x80000000 | 2147483648 - unused -// -// Note: The ResourceMark is to cleanup resource allocated args. -// The "while (0)" is so we can use semi-colon at end of RC_TRACE(). -#define RC_TRACE(level, args) \ - if ((TraceRedefineClasses & level) != 0) { \ - ResourceMark rm; \ - tty->print("RedefineClasses-0x%x: ", level); \ - tty->print_cr args; \ - } while (0) - -#define RC_TRACE_NO_CR(level, args) \ - if ((TraceRedefineClasses & level) != 0) { \ - ResourceMark rm; \ - tty->print("RedefineClasses-0x%x: ", level); \ - tty->print args; \ - } while (0) - -#define RC_TRACE_WITH_THREAD(level, thread, args) \ - if ((TraceRedefineClasses & level) != 0) { \ - ResourceMark rm(thread); \ - tty->print("RedefineClasses-0x%x: ", level); \ - tty->print_cr args; \ - } while (0) - -#define RC_TRACE_MESG(args) \ - { \ - ResourceMark rm; \ - tty->print("RedefineClasses: "); \ - tty->print_cr args; \ - } while (0) // Macro for checking if TraceRedefineClasses has a specific bit // enabled. Returns true if the bit specified by level is set. @@ -120,16 +90,49 @@ #define RC_TRACE_IN_RANGE(low, high) \ (((TraceRedefineClasses & ((high << 1) - 1)) & ~(low - 1)) != 0) -// Timer support macros. Only do timer operations if timer tracing -// is enabled. The "while (0)" is so we can use semi-colon at end of -// the macro. -#define RC_TIMER_START(t) \ +// Note: The ResourceMark is to cleanup resource allocated args. +// The "do {...} while (0)" is so we can use semi-colon at end of RC_TRACE(). +#define RC_TRACE(level, args) do { \ + if (RC_TRACE_ENABLED(level)) { \ + ResourceMark rm; \ + tty->print("RedefineClasses-0x%x: ", level); \ + tty->print_cr args; \ + } \ +} while (0) + +#define RC_TRACE_NO_CR(level, args) do { \ + if (RC_TRACE_ENABLED(level)) { \ + ResourceMark rm; \ + tty->print("RedefineClasses-0x%x: ", level); \ + tty->print args; \ + } \ +} while (0) + +#define RC_TRACE_WITH_THREAD(level, thread, args) do { \ + if (RC_TRACE_ENABLED(level)) { \ + ResourceMark rm(thread); \ + tty->print("RedefineClasses-0x%x: ", level); \ + tty->print_cr args; \ + } \ +} while (0) + +#define RC_TRACE_MESG(args) do { \ + ResourceMark rm; \ + tty->print("RedefineClasses: "); \ + tty->print_cr args; \ +} while (0) + +// Timer support macros. Only do timer operations if timer tracing is enabled. +// The "do {...} while (0)" is so we can use semi-colon at end of the macro. +#define RC_TIMER_START(t) do { \ if (RC_TRACE_ENABLED(0x00000004)) { \ t.start(); \ - } while (0) -#define RC_TIMER_STOP(t) \ + } \ +} while (0) +#define RC_TIMER_STOP(t) do { \ if (RC_TRACE_ENABLED(0x00000004)) { \ t.stop(); \ - } while (0) + } \ +} while (0) #endif // SHARE_VM_PRIMS_JVMTIREDEFINECLASSESTRACE_HPP diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/prims/methodHandles.cpp --- a/src/share/vm/prims/methodHandles.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/prims/methodHandles.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -105,14 +105,15 @@ // import java_lang_invoke_MemberName.* enum { - IS_METHOD = java_lang_invoke_MemberName::MN_IS_METHOD, - IS_CONSTRUCTOR = java_lang_invoke_MemberName::MN_IS_CONSTRUCTOR, - IS_FIELD = java_lang_invoke_MemberName::MN_IS_FIELD, - IS_TYPE = java_lang_invoke_MemberName::MN_IS_TYPE, + IS_METHOD = java_lang_invoke_MemberName::MN_IS_METHOD, + IS_CONSTRUCTOR = java_lang_invoke_MemberName::MN_IS_CONSTRUCTOR, + IS_FIELD = java_lang_invoke_MemberName::MN_IS_FIELD, + IS_TYPE = java_lang_invoke_MemberName::MN_IS_TYPE, + CALLER_SENSITIVE = java_lang_invoke_MemberName::MN_CALLER_SENSITIVE, REFERENCE_KIND_SHIFT = java_lang_invoke_MemberName::MN_REFERENCE_KIND_SHIFT, REFERENCE_KIND_MASK = java_lang_invoke_MemberName::MN_REFERENCE_KIND_MASK, - SEARCH_SUPERCLASSES = java_lang_invoke_MemberName::MN_SEARCH_SUPERCLASSES, - SEARCH_INTERFACES = java_lang_invoke_MemberName::MN_SEARCH_INTERFACES, + SEARCH_SUPERCLASSES = java_lang_invoke_MemberName::MN_SEARCH_SUPERCLASSES, + SEARCH_INTERFACES = java_lang_invoke_MemberName::MN_SEARCH_INTERFACES, ALL_KINDS = IS_METHOD | IS_CONSTRUCTOR | IS_FIELD | IS_TYPE }; @@ -207,10 +208,15 @@ vmindex = m->vtable_index(); } - java_lang_invoke_MemberName::set_flags(mname_oop, flags); + // @CallerSensitive annotation detected + if (m->caller_sensitive()) { + flags |= CALLER_SENSITIVE; + } + + java_lang_invoke_MemberName::set_flags( mname_oop, flags); java_lang_invoke_MemberName::set_vmtarget(mname_oop, m); - java_lang_invoke_MemberName::set_vmindex(mname_oop, vmindex); // vtable/itable index - java_lang_invoke_MemberName::set_clazz(mname_oop, receiver_limit->java_mirror()); + java_lang_invoke_MemberName::set_vmindex( mname_oop, vmindex); // vtable/itable index + java_lang_invoke_MemberName::set_clazz( mname_oop, receiver_limit->java_mirror()); // Note: name and type can be lazily computed by resolve_MemberName, // if Java code needs them as resolved String and MethodType objects. // The clazz must be eagerly stored, because it provides a GC @@ -940,6 +946,7 @@ template(java_lang_invoke_MemberName,MN_IS_CONSTRUCTOR) \ template(java_lang_invoke_MemberName,MN_IS_FIELD) \ template(java_lang_invoke_MemberName,MN_IS_TYPE) \ + template(java_lang_invoke_MemberName,MN_CALLER_SENSITIVE) \ template(java_lang_invoke_MemberName,MN_SEARCH_SUPERCLASSES) \ template(java_lang_invoke_MemberName,MN_SEARCH_INTERFACES) \ template(java_lang_invoke_MemberName,MN_REFERENCE_KIND_SHIFT) \ diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/prims/unsafe.cpp --- a/src/share/vm/prims/unsafe.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/prims/unsafe.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -868,7 +868,7 @@ env->ThrowNew(cls, msg); } -static jclass Unsafe_DefineClass(JNIEnv *env, jstring name, jbyteArray data, int offset, int length, jobject loader, jobject pd) { +static jclass Unsafe_DefineClass_impl(JNIEnv *env, jstring name, jbyteArray data, int offset, int length, jobject loader, jobject pd) { { // Code lifted from JDK 1.3 ClassLoader.c @@ -939,6 +939,15 @@ } +UNSAFE_ENTRY(jclass, Unsafe_DefineClass(JNIEnv *env, jobject unsafe, jstring name, jbyteArray data, int offset, int length, jobject loader, jobject pd)) + UnsafeWrapper("Unsafe_DefineClass"); + { + ThreadToNativeFromVM ttnfv(thread); + return Unsafe_DefineClass_impl(env, name, data, offset, length, loader, pd); + } +UNSAFE_END + + UNSAFE_ENTRY(jclass, Unsafe_DefineClass0(JNIEnv *env, jobject unsafe, jstring name, jbyteArray data, int offset, int length)) UnsafeWrapper("Unsafe_DefineClass"); { @@ -949,20 +958,11 @@ jobject loader = (caller == NULL) ? NULL : JVM_GetClassLoader(env, caller); jobject pd = (caller == NULL) ? NULL : JVM_GetProtectionDomain(env, caller); - return Unsafe_DefineClass(env, name, data, offset, length, loader, pd); + return Unsafe_DefineClass_impl(env, name, data, offset, length, loader, pd); } UNSAFE_END -UNSAFE_ENTRY(jclass, Unsafe_DefineClass1(JNIEnv *env, jobject unsafe, jstring name, jbyteArray data, int offset, int length, jobject loader, jobject pd)) - UnsafeWrapper("Unsafe_DefineClass"); - { - ThreadToNativeFromVM ttnfv(thread); - - return Unsafe_DefineClass(env, name, data, offset, length, loader, pd); - } -UNSAFE_END - #define DAC_Args CLS"[B["OBJ // define a class but do not make it known to the class loader or system dictionary // - host_class: supplies context for linkage, access control, protection domain, and class loader @@ -1323,7 +1323,7 @@ #define THR LANG"Throwable;" #define DC0_Args LANG"String;[BII" -#define DC1_Args DC0_Args LANG"ClassLoader;" "Ljava/security/ProtectionDomain;" +#define DC_Args DC0_Args LANG"ClassLoader;" "Ljava/security/ProtectionDomain;" #define CC (char*) /*cast a literal from (const char*)*/ #define FN_PTR(f) CAST_FROM_FN_PTR(void*, &f) @@ -1352,10 +1352,8 @@ -// %%% These are temporarily supported until the SDK sources -// contain the necessarily updated Unsafe.java. +// These are the methods for 1.4.0 static JNINativeMethod methods_140[] = { - {CC"getObject", CC"("OBJ"I)"OBJ"", FN_PTR(Unsafe_GetObject140)}, {CC"putObject", CC"("OBJ"I"OBJ")V", FN_PTR(Unsafe_SetObject140)}, @@ -1381,12 +1379,10 @@ {CC"allocateMemory", CC"(J)"ADR, FN_PTR(Unsafe_AllocateMemory)}, {CC"reallocateMemory", CC"("ADR"J)"ADR, FN_PTR(Unsafe_ReallocateMemory)}, -// {CC"setMemory", CC"("ADR"JB)V", FN_PTR(Unsafe_SetMemory)}, -// {CC"copyMemory", CC"("ADR ADR"J)V", FN_PTR(Unsafe_CopyMemory)}, {CC"freeMemory", CC"("ADR")V", FN_PTR(Unsafe_FreeMemory)}, - {CC"fieldOffset", CC"("FLD")I", FN_PTR(Unsafe_FieldOffset)}, //deprecated - {CC"staticFieldBase", CC"("CLS")"OBJ, FN_PTR(Unsafe_StaticFieldBaseFromClass)}, //deprecated + {CC"fieldOffset", CC"("FLD")I", FN_PTR(Unsafe_FieldOffset)}, + {CC"staticFieldBase", CC"("CLS")"OBJ, FN_PTR(Unsafe_StaticFieldBaseFromClass)}, {CC"ensureClassInitialized",CC"("CLS")V", FN_PTR(Unsafe_EnsureClassInitialized)}, {CC"arrayBaseOffset", CC"("CLS")I", FN_PTR(Unsafe_ArrayBaseOffset)}, {CC"arrayIndexScale", CC"("CLS")I", FN_PTR(Unsafe_ArrayIndexScale)}, @@ -1394,16 +1390,15 @@ {CC"pageSize", CC"()I", FN_PTR(Unsafe_PageSize)}, {CC"defineClass", CC"("DC0_Args")"CLS, FN_PTR(Unsafe_DefineClass0)}, - {CC"defineClass", CC"("DC1_Args")"CLS, FN_PTR(Unsafe_DefineClass1)}, + {CC"defineClass", CC"("DC_Args")"CLS, FN_PTR(Unsafe_DefineClass)}, {CC"allocateInstance", CC"("CLS")"OBJ, FN_PTR(Unsafe_AllocateInstance)}, {CC"monitorEnter", CC"("OBJ")V", FN_PTR(Unsafe_MonitorEnter)}, {CC"monitorExit", CC"("OBJ")V", FN_PTR(Unsafe_MonitorExit)}, {CC"throwException", CC"("THR")V", FN_PTR(Unsafe_ThrowException)} }; -// These are the old methods prior to the JSR 166 changes in 1.5.0 +// These are the methods prior to the JSR 166 changes in 1.5.0 static JNINativeMethod methods_141[] = { - {CC"getObject", CC"("OBJ"J)"OBJ"", FN_PTR(Unsafe_GetObject)}, {CC"putObject", CC"("OBJ"J"OBJ")V", FN_PTR(Unsafe_SetObject)}, @@ -1429,8 +1424,6 @@ {CC"allocateMemory", CC"(J)"ADR, FN_PTR(Unsafe_AllocateMemory)}, {CC"reallocateMemory", CC"("ADR"J)"ADR, FN_PTR(Unsafe_ReallocateMemory)}, -// {CC"setMemory", CC"("ADR"JB)V", FN_PTR(Unsafe_SetMemory)}, -// {CC"copyMemory", CC"("ADR ADR"J)V", FN_PTR(Unsafe_CopyMemory)}, {CC"freeMemory", CC"("ADR")V", FN_PTR(Unsafe_FreeMemory)}, {CC"objectFieldOffset", CC"("FLD")J", FN_PTR(Unsafe_ObjectFieldOffset)}, @@ -1443,7 +1436,7 @@ {CC"pageSize", CC"()I", FN_PTR(Unsafe_PageSize)}, {CC"defineClass", CC"("DC0_Args")"CLS, FN_PTR(Unsafe_DefineClass0)}, - {CC"defineClass", CC"("DC1_Args")"CLS, FN_PTR(Unsafe_DefineClass1)}, + {CC"defineClass", CC"("DC_Args")"CLS, FN_PTR(Unsafe_DefineClass)}, {CC"allocateInstance", CC"("CLS")"OBJ, FN_PTR(Unsafe_AllocateInstance)}, {CC"monitorEnter", CC"("OBJ")V", FN_PTR(Unsafe_MonitorEnter)}, {CC"monitorExit", CC"("OBJ")V", FN_PTR(Unsafe_MonitorExit)}, @@ -1451,9 +1444,8 @@ }; -// These are the old methods prior to the JSR 166 changes in 1.6.0 +// These are the methods prior to the JSR 166 changes in 1.6.0 static JNINativeMethod methods_15[] = { - {CC"getObject", CC"("OBJ"J)"OBJ"", FN_PTR(Unsafe_GetObject)}, {CC"putObject", CC"("OBJ"J"OBJ")V", FN_PTR(Unsafe_SetObject)}, {CC"getObjectVolatile",CC"("OBJ"J)"OBJ"", FN_PTR(Unsafe_GetObjectVolatile)}, @@ -1482,8 +1474,6 @@ {CC"allocateMemory", CC"(J)"ADR, FN_PTR(Unsafe_AllocateMemory)}, {CC"reallocateMemory", CC"("ADR"J)"ADR, FN_PTR(Unsafe_ReallocateMemory)}, -// {CC"setMemory", CC"("ADR"JB)V", FN_PTR(Unsafe_SetMemory)}, -// {CC"copyMemory", CC"("ADR ADR"J)V", FN_PTR(Unsafe_CopyMemory)}, {CC"freeMemory", CC"("ADR")V", FN_PTR(Unsafe_FreeMemory)}, {CC"objectFieldOffset", CC"("FLD")J", FN_PTR(Unsafe_ObjectFieldOffset)}, @@ -1496,7 +1486,7 @@ {CC"pageSize", CC"()I", FN_PTR(Unsafe_PageSize)}, {CC"defineClass", CC"("DC0_Args")"CLS, FN_PTR(Unsafe_DefineClass0)}, - {CC"defineClass", CC"("DC1_Args")"CLS, FN_PTR(Unsafe_DefineClass1)}, + {CC"defineClass", CC"("DC_Args")"CLS, FN_PTR(Unsafe_DefineClass)}, {CC"allocateInstance", CC"("CLS")"OBJ, FN_PTR(Unsafe_AllocateInstance)}, {CC"monitorEnter", CC"("OBJ")V", FN_PTR(Unsafe_MonitorEnter)}, {CC"monitorExit", CC"("OBJ")V", FN_PTR(Unsafe_MonitorExit)}, @@ -1509,14 +1499,69 @@ }; -// These are the correct methods, moving forward: -static JNINativeMethod methods[] = { - +// These are the methods for 1.6.0 and 1.7.0 +static JNINativeMethod methods_16[] = { {CC"getObject", CC"("OBJ"J)"OBJ"", FN_PTR(Unsafe_GetObject)}, {CC"putObject", CC"("OBJ"J"OBJ")V", FN_PTR(Unsafe_SetObject)}, {CC"getObjectVolatile",CC"("OBJ"J)"OBJ"", FN_PTR(Unsafe_GetObjectVolatile)}, {CC"putObjectVolatile",CC"("OBJ"J"OBJ")V", FN_PTR(Unsafe_SetObjectVolatile)}, + DECLARE_GETSETOOP(Boolean, Z), + DECLARE_GETSETOOP(Byte, B), + DECLARE_GETSETOOP(Short, S), + DECLARE_GETSETOOP(Char, C), + DECLARE_GETSETOOP(Int, I), + DECLARE_GETSETOOP(Long, J), + DECLARE_GETSETOOP(Float, F), + DECLARE_GETSETOOP(Double, D), + + DECLARE_GETSETNATIVE(Byte, B), + DECLARE_GETSETNATIVE(Short, S), + DECLARE_GETSETNATIVE(Char, C), + DECLARE_GETSETNATIVE(Int, I), + DECLARE_GETSETNATIVE(Long, J), + DECLARE_GETSETNATIVE(Float, F), + DECLARE_GETSETNATIVE(Double, D), + + {CC"getAddress", CC"("ADR")"ADR, FN_PTR(Unsafe_GetNativeAddress)}, + {CC"putAddress", CC"("ADR""ADR")V", FN_PTR(Unsafe_SetNativeAddress)}, + + {CC"allocateMemory", CC"(J)"ADR, FN_PTR(Unsafe_AllocateMemory)}, + {CC"reallocateMemory", CC"("ADR"J)"ADR, FN_PTR(Unsafe_ReallocateMemory)}, + {CC"freeMemory", CC"("ADR")V", FN_PTR(Unsafe_FreeMemory)}, + + {CC"objectFieldOffset", CC"("FLD")J", FN_PTR(Unsafe_ObjectFieldOffset)}, + {CC"staticFieldOffset", CC"("FLD")J", FN_PTR(Unsafe_StaticFieldOffset)}, + {CC"staticFieldBase", CC"("FLD")"OBJ, FN_PTR(Unsafe_StaticFieldBaseFromField)}, + {CC"ensureClassInitialized",CC"("CLS")V", FN_PTR(Unsafe_EnsureClassInitialized)}, + {CC"arrayBaseOffset", CC"("CLS")I", FN_PTR(Unsafe_ArrayBaseOffset)}, + {CC"arrayIndexScale", CC"("CLS")I", FN_PTR(Unsafe_ArrayIndexScale)}, + {CC"addressSize", CC"()I", FN_PTR(Unsafe_AddressSize)}, + {CC"pageSize", CC"()I", FN_PTR(Unsafe_PageSize)}, + + {CC"defineClass", CC"("DC0_Args")"CLS, FN_PTR(Unsafe_DefineClass0)}, + {CC"defineClass", CC"("DC_Args")"CLS, FN_PTR(Unsafe_DefineClass)}, + {CC"allocateInstance", CC"("CLS")"OBJ, FN_PTR(Unsafe_AllocateInstance)}, + {CC"monitorEnter", CC"("OBJ")V", FN_PTR(Unsafe_MonitorEnter)}, + {CC"monitorExit", CC"("OBJ")V", FN_PTR(Unsafe_MonitorExit)}, + {CC"tryMonitorEnter", CC"("OBJ")Z", FN_PTR(Unsafe_TryMonitorEnter)}, + {CC"throwException", CC"("THR")V", FN_PTR(Unsafe_ThrowException)}, + {CC"compareAndSwapObject", CC"("OBJ"J"OBJ""OBJ")Z", FN_PTR(Unsafe_CompareAndSwapObject)}, + {CC"compareAndSwapInt", CC"("OBJ"J""I""I"")Z", FN_PTR(Unsafe_CompareAndSwapInt)}, + {CC"compareAndSwapLong", CC"("OBJ"J""J""J"")Z", FN_PTR(Unsafe_CompareAndSwapLong)}, + {CC"putOrderedObject", CC"("OBJ"J"OBJ")V", FN_PTR(Unsafe_SetOrderedObject)}, + {CC"putOrderedInt", CC"("OBJ"JI)V", FN_PTR(Unsafe_SetOrderedInt)}, + {CC"putOrderedLong", CC"("OBJ"JJ)V", FN_PTR(Unsafe_SetOrderedLong)}, + {CC"park", CC"(ZJ)V", FN_PTR(Unsafe_Park)}, + {CC"unpark", CC"("OBJ")V", FN_PTR(Unsafe_Unpark)} +}; + +// These are the methods for 1.8.0 +static JNINativeMethod methods_18[] = { + {CC"getObject", CC"("OBJ"J)"OBJ"", FN_PTR(Unsafe_GetObject)}, + {CC"putObject", CC"("OBJ"J"OBJ")V", FN_PTR(Unsafe_SetObject)}, + {CC"getObjectVolatile",CC"("OBJ"J)"OBJ"", FN_PTR(Unsafe_GetObjectVolatile)}, + {CC"putObjectVolatile",CC"("OBJ"J"OBJ")V", FN_PTR(Unsafe_SetObjectVolatile)}, DECLARE_GETSETOOP(Boolean, Z), DECLARE_GETSETOOP(Byte, B), @@ -1540,8 +1585,6 @@ {CC"allocateMemory", CC"(J)"ADR, FN_PTR(Unsafe_AllocateMemory)}, {CC"reallocateMemory", CC"("ADR"J)"ADR, FN_PTR(Unsafe_ReallocateMemory)}, -// {CC"setMemory", CC"("ADR"JB)V", FN_PTR(Unsafe_SetMemory)}, -// {CC"copyMemory", CC"("ADR ADR"J)V", FN_PTR(Unsafe_CopyMemory)}, {CC"freeMemory", CC"("ADR")V", FN_PTR(Unsafe_FreeMemory)}, {CC"objectFieldOffset", CC"("FLD")J", FN_PTR(Unsafe_ObjectFieldOffset)}, @@ -1553,8 +1596,7 @@ {CC"addressSize", CC"()I", FN_PTR(Unsafe_AddressSize)}, {CC"pageSize", CC"()I", FN_PTR(Unsafe_PageSize)}, - {CC"defineClass", CC"("DC0_Args")"CLS, FN_PTR(Unsafe_DefineClass0)}, - {CC"defineClass", CC"("DC1_Args")"CLS, FN_PTR(Unsafe_DefineClass1)}, + {CC"defineClass", CC"("DC_Args")"CLS, FN_PTR(Unsafe_DefineClass)}, {CC"allocateInstance", CC"("CLS")"OBJ, FN_PTR(Unsafe_AllocateInstance)}, {CC"monitorEnter", CC"("OBJ")V", FN_PTR(Unsafe_MonitorEnter)}, {CC"monitorExit", CC"("OBJ")V", FN_PTR(Unsafe_MonitorExit)}, @@ -1566,23 +1608,12 @@ {CC"putOrderedObject", CC"("OBJ"J"OBJ")V", FN_PTR(Unsafe_SetOrderedObject)}, {CC"putOrderedInt", CC"("OBJ"JI)V", FN_PTR(Unsafe_SetOrderedInt)}, {CC"putOrderedLong", CC"("OBJ"JJ)V", FN_PTR(Unsafe_SetOrderedLong)}, - {CC"loadFence", CC"()V", FN_PTR(Unsafe_LoadFence)}, - {CC"storeFence", CC"()V", FN_PTR(Unsafe_StoreFence)}, - {CC"fullFence", CC"()V", FN_PTR(Unsafe_FullFence)}, {CC"park", CC"(ZJ)V", FN_PTR(Unsafe_Park)}, {CC"unpark", CC"("OBJ")V", FN_PTR(Unsafe_Unpark)} - -// {CC"getLoadAverage", CC"([DI)I", FN_PTR(Unsafe_Loadavg)}, - -// {CC"prefetchRead", CC"("OBJ"J)V", FN_PTR(Unsafe_PrefetchRead)}, -// {CC"prefetchWrite", CC"("OBJ"J)V", FN_PTR(Unsafe_PrefetchWrite)} -// {CC"prefetchReadStatic", CC"("OBJ"J)V", FN_PTR(Unsafe_PrefetchRead)}, -// {CC"prefetchWriteStatic",CC"("OBJ"J)V", FN_PTR(Unsafe_PrefetchWrite)} - }; JNINativeMethod loadavg_method[] = { - {CC"getLoadAverage", CC"([DI)I", FN_PTR(Unsafe_Loadavg)} + {CC"getLoadAverage", CC"([DI)I", FN_PTR(Unsafe_Loadavg)} }; JNINativeMethod prefetch_methods[] = { @@ -1592,7 +1623,7 @@ {CC"prefetchWriteStatic",CC"("OBJ"J)V", FN_PTR(Unsafe_PrefetchWrite)} }; -JNINativeMethod memcopy_methods[] = { +JNINativeMethod memcopy_methods_17[] = { {CC"copyMemory", CC"("OBJ"J"OBJ"JJ)V", FN_PTR(Unsafe_CopyMemory2)}, {CC"setMemory", CC"("OBJ"JJB)V", FN_PTR(Unsafe_SetMemory2)} }; @@ -1610,6 +1641,12 @@ {CC"shouldBeInitialized",CC"("CLS")Z", FN_PTR(Unsafe_ShouldBeInitialized)}, }; +JNINativeMethod fence_methods[] = { + {CC"loadFence", CC"()V", FN_PTR(Unsafe_LoadFence)}, + {CC"storeFence", CC"()V", FN_PTR(Unsafe_StoreFence)}, + {CC"fullFence", CC"()V", FN_PTR(Unsafe_FullFence)}, +}; + #undef CC #undef FN_PTR @@ -1622,12 +1659,32 @@ #undef MTH #undef THR #undef DC0_Args -#undef DC1_Args +#undef DC_Args #undef DECLARE_GETSETOOP #undef DECLARE_GETSETNATIVE +/** + * Helper method to register native methods. + */ +static bool register_natives(const char* message, JNIEnv* env, jclass clazz, const JNINativeMethod* methods, jint nMethods) { + int status = env->RegisterNatives(clazz, methods, nMethods); + if (status < 0 || env->ExceptionOccurred()) { + if (PrintMiscellaneous && (Verbose || WizardMode)) { + tty->print_cr("Unsafe: failed registering %s", message); + } + env->ExceptionClear(); + return false; + } else { + if (PrintMiscellaneous && (Verbose || WizardMode)) { + tty->print_cr("Unsafe: successfully registered %s", message); + } + return true; + } +} + + // This one function is exported, used by NativeLookup. // The Unsafe_xxx functions above are called only from the interpreter. // The optimizer looks at names and signatures to recognize @@ -1637,83 +1694,57 @@ UnsafeWrapper("JVM_RegisterUnsafeMethods"); { ThreadToNativeFromVM ttnfv(thread); - { - env->RegisterNatives(unsafecls, loadavg_method, sizeof(loadavg_method)/sizeof(JNINativeMethod)); - if (env->ExceptionOccurred()) { - if (PrintMiscellaneous && (Verbose || WizardMode)) { - tty->print_cr("Warning: SDK 1.6 Unsafe.loadavg not found."); - } - env->ExceptionClear(); - } - } + + // Unsafe methods { - env->RegisterNatives(unsafecls, prefetch_methods, sizeof(prefetch_methods)/sizeof(JNINativeMethod)); - if (env->ExceptionOccurred()) { - if (PrintMiscellaneous && (Verbose || WizardMode)) { - tty->print_cr("Warning: SDK 1.6 Unsafe.prefetchRead/Write not found."); - } - env->ExceptionClear(); + bool success = false; + // We need to register the 1.6 methods first because the 1.8 methods would register fine on 1.7 and 1.6 + if (!success) { + success = register_natives("1.6 methods", env, unsafecls, methods_16, sizeof(methods_16)/sizeof(JNINativeMethod)); + } + if (!success) { + success = register_natives("1.8 methods", env, unsafecls, methods_18, sizeof(methods_18)/sizeof(JNINativeMethod)); + } + if (!success) { + success = register_natives("1.5 methods", env, unsafecls, methods_15, sizeof(methods_15)/sizeof(JNINativeMethod)); + } + if (!success) { + success = register_natives("1.4.1 methods", env, unsafecls, methods_141, sizeof(methods_141)/sizeof(JNINativeMethod)); } + if (!success) { + success = register_natives("1.4.0 methods", env, unsafecls, methods_140, sizeof(methods_140)/sizeof(JNINativeMethod)); + } + guarantee(success, "register unsafe natives"); } + + // Unsafe.getLoadAverage + register_natives("1.6 loadavg method", env, unsafecls, loadavg_method, sizeof(loadavg_method)/sizeof(JNINativeMethod)); + + // Prefetch methods + register_natives("1.6 prefetch methods", env, unsafecls, prefetch_methods, sizeof(prefetch_methods)/sizeof(JNINativeMethod)); + + // Memory copy methods { - env->RegisterNatives(unsafecls, memcopy_methods, sizeof(memcopy_methods)/sizeof(JNINativeMethod)); - if (env->ExceptionOccurred()) { - if (PrintMiscellaneous && (Verbose || WizardMode)) { - tty->print_cr("Warning: SDK 1.7 Unsafe.copyMemory not found."); - } - env->ExceptionClear(); - env->RegisterNatives(unsafecls, memcopy_methods_15, sizeof(memcopy_methods_15)/sizeof(JNINativeMethod)); - if (env->ExceptionOccurred()) { - if (PrintMiscellaneous && (Verbose || WizardMode)) { - tty->print_cr("Warning: SDK 1.5 Unsafe.copyMemory not found."); - } - env->ExceptionClear(); - } + bool success = false; + if (!success) { + success = register_natives("1.7 memory copy methods", env, unsafecls, memcopy_methods_17, sizeof(memcopy_methods_17)/sizeof(JNINativeMethod)); + } + if (!success) { + success = register_natives("1.5 memory copy methods", env, unsafecls, memcopy_methods_15, sizeof(memcopy_methods_15)/sizeof(JNINativeMethod)); } } + + // Unsafe.defineAnonymousClass if (EnableInvokeDynamic) { - env->RegisterNatives(unsafecls, anonk_methods, sizeof(anonk_methods)/sizeof(JNINativeMethod)); - if (env->ExceptionOccurred()) { - if (PrintMiscellaneous && (Verbose || WizardMode)) { - tty->print_cr("Warning: SDK 1.7 Unsafe.defineClass (anonymous version) not found."); - } - env->ExceptionClear(); - } - } - if (EnableInvokeDynamic) { - env->RegisterNatives(unsafecls, lform_methods, sizeof(lform_methods)/sizeof(JNINativeMethod)); - if (env->ExceptionOccurred()) { - if (PrintMiscellaneous && (Verbose || WizardMode)) { - tty->print_cr("Warning: SDK 1.7 LambdaForm support in Unsafe not found."); - } - env->ExceptionClear(); - } + register_natives("1.7 define anonymous class method", env, unsafecls, anonk_methods, sizeof(anonk_methods)/sizeof(JNINativeMethod)); } - int status = env->RegisterNatives(unsafecls, methods, sizeof(methods)/sizeof(JNINativeMethod)); - if (env->ExceptionOccurred()) { - if (PrintMiscellaneous && (Verbose || WizardMode)) { - tty->print_cr("Warning: SDK 1.6 version of Unsafe not found."); - } - env->ExceptionClear(); - // %%% For now, be backward compatible with an older class: - status = env->RegisterNatives(unsafecls, methods_15, sizeof(methods_15)/sizeof(JNINativeMethod)); + + // Unsafe.shouldBeInitialized + if (EnableInvokeDynamic) { + register_natives("1.7 LambdaForm support", env, unsafecls, lform_methods, sizeof(lform_methods)/sizeof(JNINativeMethod)); } - if (env->ExceptionOccurred()) { - if (PrintMiscellaneous && (Verbose || WizardMode)) { - tty->print_cr("Warning: SDK 1.5 version of Unsafe not found."); - } - env->ExceptionClear(); - // %%% For now, be backward compatible with an older class: - status = env->RegisterNatives(unsafecls, methods_141, sizeof(methods_141)/sizeof(JNINativeMethod)); - } - if (env->ExceptionOccurred()) { - if (PrintMiscellaneous && (Verbose || WizardMode)) { - tty->print_cr("Warning: SDK 1.4.1 version of Unsafe not found."); - } - env->ExceptionClear(); - // %%% For now, be backward compatible with an older class: - status = env->RegisterNatives(unsafecls, methods_140, sizeof(methods_140)/sizeof(JNINativeMethod)); - } - guarantee(status == 0, "register unsafe natives"); + + // Fence methods + register_natives("1.8 fence methods", env, unsafecls, fence_methods, sizeof(fence_methods)/sizeof(JNINativeMethod)); } JVM_END diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/runtime/arguments.cpp --- a/src/share/vm/runtime/arguments.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/runtime/arguments.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -1553,6 +1553,15 @@ } } +julong Arguments::limit_by_allocatable_memory(julong limit) { + julong max_allocatable; + julong result = limit; + if (os::has_allocatable_memory_limit(&max_allocatable)) { + result = MIN2(result, max_allocatable / MaxVirtMemFraction); + } + return result; +} + void Arguments::set_heap_size() { if (!FLAG_IS_DEFAULT(DefaultMaxRAMFraction)) { // Deprecated flag @@ -1591,12 +1600,12 @@ } reasonable_max = MIN2(reasonable_max, max_coop_heap); } - reasonable_max = os::allocatable_physical_memory(reasonable_max); + reasonable_max = limit_by_allocatable_memory(reasonable_max); if (!FLAG_IS_DEFAULT(InitialHeapSize)) { // An initial heap size was specified on the command line, // so be sure that the maximum size is consistent. Done - // after call to allocatable_physical_memory because that + // after call to limit_by_allocatable_memory because that // method might reduce the allocation size. reasonable_max = MAX2(reasonable_max, (julong)InitialHeapSize); } @@ -1616,14 +1625,14 @@ reasonable_minimum = MIN2(reasonable_minimum, (julong)MaxHeapSize); - reasonable_minimum = os::allocatable_physical_memory(reasonable_minimum); + reasonable_minimum = limit_by_allocatable_memory(reasonable_minimum); julong reasonable_initial = phys_mem / InitialRAMFraction; reasonable_initial = MAX2(reasonable_initial, reasonable_minimum); reasonable_initial = MIN2(reasonable_initial, (julong)MaxHeapSize); - reasonable_initial = os::allocatable_physical_memory(reasonable_initial); + reasonable_initial = limit_by_allocatable_memory(reasonable_initial); if (PrintGCDetails && Verbose) { // Cannot use gclog_or_tty yet. @@ -2613,9 +2622,7 @@ initHeapSize = MIN2(total_memory / (julong)2, total_memory - (julong)160*M); - // Make sure that if we have a lot of memory we cap the 32 bit - // process space. The 64bit VM version of this function is a nop. - initHeapSize = os::allocatable_physical_memory(initHeapSize); + initHeapSize = limit_by_allocatable_memory(initHeapSize); if (FLAG_IS_DEFAULT(MaxHeapSize)) { FLAG_SET_CMDLINE(uintx, MaxHeapSize, initHeapSize); diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/runtime/arguments.hpp --- a/src/share/vm/runtime/arguments.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/runtime/arguments.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -312,6 +312,9 @@ static void set_use_compressed_oops(); static void set_ergonomics_flags(); static void set_shared_spaces_flags(); + // limits the given memory size by the maximum amount of memory this process is + // currently allowed to allocate or reserve. + static julong limit_by_allocatable_memory(julong size); // Setup heap size static void set_heap_size(); // Based on automatic selection criteria, should the diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/runtime/globals.hpp --- a/src/share/vm/runtime/globals.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/runtime/globals.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -1407,6 +1407,10 @@ "How much the GC can expand the eden by while the GC locker " \ "is active (as a percentage)") \ \ + diagnostic(intx, GCLockerRetryAllocationCount, 2, \ + "Number of times to retry allocations when" \ + " blocked by the GC locker") \ + \ develop(bool, UseCMSAdaptiveFreeLists, true, \ "Use Adaptive Free Lists in the CMS generation") \ \ @@ -1963,6 +1967,10 @@ product(uintx, InitialRAMFraction, 64, \ "Fraction (1/n) of real memory used for initial heap size") \ \ + develop(uintx, MaxVirtMemFraction, 2, \ + "Maximum fraction (1/n) of virtual memory used for ergonomically" \ + "determining maximum heap size") \ + \ product(bool, UseAutoGCSelectPolicy, false, \ "Use automatic collection selection policy") \ \ @@ -2520,7 +2528,7 @@ "disable locking assertions (for speed)") \ \ product(bool, RangeCheckElimination, true, \ - "Split loop iterations to eliminate range checks") \ + "Eliminate range checks") \ \ develop_pd(bool, UncommonNullCast, \ "track occurrences of null in casts; adjust compiler tactics") \ diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/runtime/init.cpp --- a/src/share/vm/runtime/init.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/runtime/init.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -132,15 +132,6 @@ javaClasses_init(); // must happen after vtable initialization stubRoutines_init2(); // note: StubRoutines need 2-phase init - // Although we'd like to, we can't easily do a heap verify - // here because the main thread isn't yet a JavaThread, so - // its TLAB may not be made parseable from the usual interfaces. - if (VerifyBeforeGC && !UseTLAB && - Universe::heap()->total_collections() >= VerifyGCStartAt) { - Universe::heap()->prepare_for_verify(); - Universe::verify(); // make sure we're starting with a clean slate - } - // All the flags that get adjusted by VM_Version_init and os::init_2 // have been set so dump the flags now. if (PrintFlagsFinal) { diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/runtime/os.hpp --- a/src/share/vm/runtime/os.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/runtime/os.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -184,7 +184,7 @@ } static julong available_memory(); static julong physical_memory(); - static julong allocatable_physical_memory(julong size); + static bool has_allocatable_memory_limit(julong* limit); static bool is_server_class_machine(); // number of CPUs diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/runtime/thread.cpp --- a/src/share/vm/runtime/thread.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/runtime/thread.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -3423,12 +3423,6 @@ // real raw monitor. VM is setup enough here for raw monitor enter. JvmtiExport::transition_pending_onload_raw_monitors(); - if (VerifyBeforeGC && - Universe::heap()->total_collections() >= VerifyGCStartAt) { - Universe::heap()->prepare_for_verify(); - Universe::verify(); // make sure we're starting with a clean slate - } - // Fully start NMT MemTracker::start(); @@ -3452,6 +3446,11 @@ } assert (Universe::is_fully_initialized(), "not initialized"); + if (VerifyBeforeGC && VerifyGCStartAt == 0) { + Universe::heap()->prepare_for_verify(); + Universe::verify(); // make sure we're starting with a clean slate + } + EXCEPTION_MARK; // At this point, the Universe is initialized, but we have not executed diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/runtime/vframe.cpp --- a/src/share/vm/runtime/vframe.cpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/runtime/vframe.cpp Fri Apr 05 16:16:18 2013 -0700 @@ -391,40 +391,27 @@ // Step back n frames, skip any pseudo frames in between. // This function is used in Class.forName, Class.newInstance, Method.Invoke, // AccessController.doPrivileged. -// -// NOTE that in JDK 1.4 this has been exposed to Java as -// sun.reflect.Reflection.getCallerClass(), which can be inlined. -// Inlined versions must match this routine's logic. -// Native method prefixing logic does not need to match since -// the method names don't match and inlining will not occur. -// See, for example, -// Parse::inline_native_Reflection_getCallerClass in -// opto/library_call.cpp. void vframeStreamCommon::security_get_caller_frame(int depth) { - bool use_new_reflection = JDK_Version::is_gte_jdk14x_version() && UseNewReflection; + assert(depth >= 0, err_msg("invalid depth: %d", depth)); + for (int n = 0; !at_end(); security_next()) { + if (!method()->is_ignored_by_security_stack_walk()) { + if (n == depth) { + // We have reached the desired depth; return. + return; + } + n++; // this is a non-skipped frame; count it against the depth + } + } + // NOTE: At this point there were not enough frames on the stack + // to walk to depth. Callers of this method have to check for at_end. +} - while (!at_end()) { - if (Universe::reflect_invoke_cache()->is_same_method(method())) { - // This is Method.invoke() -- skip it - } else if (use_new_reflection && - method()->method_holder() - ->is_subclass_of(SystemDictionary::reflect_MethodAccessorImpl_klass())) { - // This is an auxilary frame -- skip it - } else if (method()->is_method_handle_intrinsic() || - method()->is_compiled_lambda_form()) { - // This is an internal adapter frame for method handles -- skip it - } else { - // This is non-excluded frame, we need to count it against the depth - if (depth-- <= 0) { - // we have reached the desired depth, we are done - break; - } - } - if (method()->is_prefixed_native()) { - skip_prefixed_method_and_wrappers(); - } else { - next(); - } + +void vframeStreamCommon::security_next() { + if (method()->is_prefixed_native()) { + skip_prefixed_method_and_wrappers(); // calls next() + } else { + next(); } } diff -r 4c8bb5e4f68f -r 46d24f112c27 src/share/vm/runtime/vframe.hpp --- a/src/share/vm/runtime/vframe.hpp Fri Apr 05 12:19:19 2013 -0400 +++ b/src/share/vm/runtime/vframe.hpp Fri Apr 05 16:16:18 2013 -0700 @@ -336,6 +336,7 @@ _frame = _frame.sender(&_reg_map); } while (!fill_from_frame()); } + void security_next(); bool at_end() const { return _mode == at_end_mode; } diff -r 4c8bb5e4f68f -r 46d24f112c27 test/compiler/8009761/Test8009761.java --- a/test/compiler/8009761/Test8009761.java Fri Apr 05 12:19:19 2013 -0400 +++ b/test/compiler/8009761/Test8009761.java Fri Apr 05 16:16:18 2013 -0700 @@ -25,7 +25,7 @@ * @test * @bug 8009761 * @summary Deoptimization on sparc doesn't set Llast_SP correctly in the interpreter frames it creates - * @run main/othervm -XX:-UseOnStackReplacement -XX:-BackgroundCompilation Test8009761 + * @run main/othervm -Xmixed -XX:-UseOnStackReplacement -XX:-BackgroundCompilation Test8009761 * */ diff -r 4c8bb5e4f68f -r 46d24f112c27 test/gc/TestVerifyBeforeGCDuringStartup.java --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/gc/TestVerifyBeforeGCDuringStartup.java Fri Apr 05 16:16:18 2013 -0700 @@ -0,0 +1,45 @@ +/* + * Copyright (c) 2013, 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 TestVerifyBeforeGCDuringStartup.java + * @key gc + * @bug 8010463 + * @summary Simple test run with -XX:+VerifyBeforeGC -XX:-UseTLAB to verify 8010463 + * @library /testlibrary + */ + +import com.oracle.java.testlibrary.OutputAnalyzer; +import com.oracle.java.testlibrary.ProcessTools; + +public class TestVerifyBeforeGCDuringStartup { + public static void main(String args[]) throws Exception { + ProcessBuilder pb = + ProcessTools.createJavaProcessBuilder(System.getProperty("test.vm.opts"), + "-XX:-UseTLAB", + "-XX:+UnlockDiagnosticVMOptions", + "-XX:+VerifyBeforeGC", "-version"); + OutputAnalyzer output = new OutputAnalyzer(pb.start()); + output.shouldContain("[Verifying"); + output.shouldHaveExitValue(0); + } +} diff -r 4c8bb5e4f68f -r 46d24f112c27 test/gc/metaspace/ClassMetaspaceSizeInJmapHeap.java --- a/test/gc/metaspace/ClassMetaspaceSizeInJmapHeap.java Fri Apr 05 12:19:19 2013 -0400 +++ b/test/gc/metaspace/ClassMetaspaceSizeInJmapHeap.java Fri Apr 05 16:16:18 2013 -0700 @@ -39,8 +39,10 @@ public static void main(String[] args) throws Exception { String pid = Integer.toString(ProcessTools.getProcessId()); - ProcessBuilder pb = new ProcessBuilder(); - pb.command(JDKToolFinder.getJDKTool("jmap"), "-heap", pid); + JDKToolLauncher jmap = JDKToolLauncher.create("jmap") + .addToolArg("-heap") + .addToolArg(pid); + ProcessBuilder pb = new ProcessBuilder(jmap.getCommand()); File out = new File("ClassMetaspaceSizeInJmapHeap.stdout.txt"); pb.redirectOutput(out); diff -r 4c8bb5e4f68f -r 46d24f112c27 test/testlibrary/com/oracle/java/testlibrary/JDKToolLauncher.java --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/testlibrary/com/oracle/java/testlibrary/JDKToolLauncher.java Fri Apr 05 16:16:18 2013 -0700 @@ -0,0 +1,114 @@ +/* + * Copyright (c) 2013, 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. + */ + +package com.oracle.java.testlibrary; + +import java.util.List; +import java.util.ArrayList; +import java.util.Arrays; + +import com.oracle.java.testlibrary.JDKToolFinder; +import com.oracle.java.testlibrary.ProcessTools; + +/** + * A utility for constructing command lines for starting JDK tool processes. + * + * The JDKToolLauncher can in particular be combined with a + * java.lang.ProcessBuilder to easily run a JDK tool. For example, the + * following code run {@code jmap -heap} against a process with GC logging + * turned on for the {@code jmap} process: + * + *
+ * {@code
+ * JDKToolLauncher jmap = JDKToolLauncher.create("jmap")
+ *                                       .addVMArg("-XX:+PrintGC");
+ *                                       .addVMArg("-XX:+PrintGCDetails")
+ *                                       .addToolArg("-heap")
+ *                                       .addToolArg(pid);
+ * ProcessBuilder pb = new ProcessBuilder(jmap.getCommand());
+ * Process p = pb.start();
+ * }
+ * 
+ */ +public class JDKToolLauncher { + private final String executable; + private final List vmArgs = new ArrayList(); + private final List toolArgs = new ArrayList(); + + private JDKToolLauncher(String tool) { + executable = JDKToolFinder.getJDKTool(tool); + vmArgs.addAll(Arrays.asList(ProcessTools.getPlatformSpecificVMArgs())); + } + + /** + * Creates a new JDKToolLauncher for the specified tool. + * + * @param tool The name of the tool + * @return A new JDKToolLauncher + */ + public static JDKToolLauncher create(String tool) { + return new JDKToolLauncher(tool); + } + + /** + * Adds an argument to the JVM running the tool. + * + * The JVM arguments are passed to the underlying JVM running the tool. + * Arguments will automatically be prepended with "-J". + * + * Any platform specific arguments required for running the tool are + * automatically added. + * + * + * @param arg The argument to VM running the tool + * @return The JDKToolLauncher instance + */ + public JDKToolLauncher addVMArg(String arg) { + vmArgs.add("-J" + arg); + return this; + } + + /** + * Adds an argument to the tool. + * + * @param arg The argument to the tool + * @return The JDKToolLauncher instance + */ + public JDKToolLauncher addToolArg(String arg) { + toolArgs.add(arg); + return this; + } + + /** + * Returns the command that can be used for running the tool. + * + * @return An array whose elements are the arguments of the command. + */ + public String[] getCommand() { + List command = new ArrayList(); + command.add(executable); + command.addAll(vmArgs); + command.addAll(toolArgs); + return command.toArray(new String[command.size()]); + } +}