|
|
|
|
Quellcode-Bibliothek library_call.cpp
Interaktion und
PortierbarkeitC
|
|
/*
* Copyright (c) 1999, 2022, 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 "asm/macroAssembler.hpp"
#include "ci/ciUtilities.inline.hpp"
#include "classfile/vmIntrinsics.hpp"
#include "compiler/compileBroker.hpp"
#include "compiler/compileLog.hpp"
#include "gc/shared/barrierSet.hpp"
#include "jfr/support/jfrIntrinsics.hpp"
#include "memory/resourceArea.hpp"
#include "oops/klass.inline.hpp"
#include "oops/objArrayKlass.hpp"
#include "opto/addnode.hpp"
#include "opto/arraycopynode.hpp"
#include "opto/c2compiler.hpp"
#include "opto/castnode.hpp"
#include "opto/cfgnode.hpp"
#include "opto/convertnode.hpp"
#include "opto/countbitsnode.hpp"
#include "opto/idealKit.hpp"
#include "opto/library_call.hpp"
#include "opto/mathexactnode.hpp"
#include "opto/mulnode.hpp"
#include "opto/narrowptrnode.hpp"
#include "opto/opaquenode.hpp"
#include "opto/parse.hpp"
#include "opto/runtime.hpp"
#include "opto/rootnode.hpp"
#include "opto/subnode.hpp"
#include "prims/unsafe.hpp"
#include "runtime/objectMonitor.hpp"
#include "runtime/sharedRuntime.hpp"
#include "runtime/stubRoutines.hpp"
#include "utilities/macros.hpp"
#include "utilities/powerOfTwo.hpp"
//---------------------------make_vm_intrinsic----------------------------
CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
vmIntrinsicID id = m->intrinsic_id();
assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
if (!m->is_loaded()) {
// Do not attempt to inline unloaded methods.
return NULL;
}
C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
bool is_available = false;
{
// For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
// the compiler must transition to '_thread_in_vm' state because both
// methods access VM-internal data.
VM_ENTRY_MARK;
methodHandle mh(THREAD, m->get_Method());
is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
!C->directive()->is_intrinsic_disabled(mh) &&
!vmIntrinsics::is_disabled_by_flags(mh);
}
if (is_available) {
assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
return new LibraryIntrinsic(m, is_virtual,
vmIntrinsics::predicates_needed(id),
vmIntrinsics::does_virtual_dispatch(id),
id);
} else {
return NULL;
}
}
JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
LibraryCallKit kit(jvms, this);
Compile* C = kit.C;
int nodes = C->unique();
#ifndef PRODUCT
if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
char buf[1000];
const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
tty->print_cr("Intrinsic %s", str);
}
#endif
ciMethod* callee = kit.callee();
const int bci = kit.bci();
#ifdef ASSERT
Node* ctrl = kit.control();
#endif
// Try to inline the intrinsic.
if (callee->check_intrinsic_candidate() &&
kit.try_to_inline(_last_predicate)) {
const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
: "(intrinsic)";
CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
if (C->print_intrinsics() || C->print_inlining()) {
C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
}
C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
if (C->log()) {
C->log()->elem("intrinsic id='%s'%s nodes='%d'",
vmIntrinsics::name_at(intrinsic_id()),
(is_virtual() ? " virtual='1'" : ""),
C->unique() - nodes);
}
// Push the result from the inlined method onto the stack.
kit.push_result();
C->print_inlining_update(this);
return kit.transfer_exceptions_into_jvms();
}
// The intrinsic bailed out
assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
if (jvms->has_method()) {
// Not a root compile.
const char* msg;
if (callee->intrinsic_candidate()) {
msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
} else {
msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
: "failed to inline (intrinsic), method not annotated";
}
CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
if (C->print_intrinsics() || C->print_inlining()) {
C->print_inlining(callee, jvms->depth() - 1, bci, msg);
}
} else {
// Root compile
ResourceMark rm;
stringStream msg_stream;
msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
vmIntrinsics::name_at(intrinsic_id()),
is_virtual() ? " (virtual)" : "", bci);
const char *msg = msg_stream.freeze();
log_debug(jit, inlining)("%s", msg);
if (C->print_intrinsics() || C->print_inlining()) {
tty->print("%s", msg);
}
}
C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
C->print_inlining_update(this);
return NULL;
}
Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
LibraryCallKit kit(jvms, this);
Compile* C = kit.C;
int nodes = C->unique();
_last_predicate = predicate;
#ifndef PRODUCT
assert(is_predicated() && predicate < predicates_count(), "sanity");
if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
char buf[1000];
const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
tty->print_cr("Predicate for intrinsic %s", str);
}
#endif
ciMethod* callee = kit.callee();
const int bci = kit.bci();
Node* slow_ctl = kit.try_to_predicate(predicate);
if (!kit.failing()) {
const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
: "(intrinsic, predicate)";
CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
if (C->print_intrinsics() || C->print_inlining()) {
C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
}
C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
if (C->log()) {
C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
vmIntrinsics::name_at(intrinsic_id()),
(is_virtual() ? " virtual='1'" : ""),
C->unique() - nodes);
}
return slow_ctl; // Could be NULL if the check folds.
}
// The intrinsic bailed out
if (jvms->has_method()) {
// Not a root compile.
const char* msg = "failed to generate predicate for intrinsic";
CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
if (C->print_intrinsics() || C->print_inlining()) {
C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
}
} else {
// Root compile
ResourceMark rm;
stringStream msg_stream;
msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
vmIntrinsics::name_at(intrinsic_id()),
is_virtual() ? " (virtual)" : "", bci);
const char *msg = msg_stream.freeze();
log_debug(jit, inlining)("%s", msg);
if (C->print_intrinsics() || C->print_inlining()) {
C->print_inlining_stream()->print("%s", msg);
}
}
C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
return NULL;
}
bool LibraryCallKit::try_to_inline(int predicate) {
// Handle symbolic names for otherwise undistinguished boolean switches:
const bool is_store = true;
const bool is_compress = true;
const bool is_static = true;
const bool is_volatile = true;
if (!jvms()->has_method()) {
// Root JVMState has a null method.
assert(map()->memory()->Opcode() == Op_Parm, "");
// Insert the memory aliasing node
set_all_memory(reset_memory());
}
assert(merged_memory(), "");
switch (intrinsic_id()) {
case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
case vmIntrinsics::_getClass: return inline_native_getClass();
case vmIntrinsics::_ceil:
case vmIntrinsics::_floor:
case vmIntrinsics::_rint:
case vmIntrinsics::_dsin:
case vmIntrinsics::_dcos:
case vmIntrinsics::_dtan:
case vmIntrinsics::_dabs:
case vmIntrinsics::_fabs:
case vmIntrinsics::_iabs:
case vmIntrinsics::_labs:
case vmIntrinsics::_datan2:
case vmIntrinsics::_dsqrt:
case vmIntrinsics::_dsqrt_strict:
case vmIntrinsics::_dexp:
case vmIntrinsics::_dlog:
case vmIntrinsics::_dlog10:
case vmIntrinsics::_dpow:
case vmIntrinsics::_dcopySign:
case vmIntrinsics::_fcopySign:
case vmIntrinsics::_dsignum:
case vmIntrinsics::_roundF:
case vmIntrinsics::_roundD:
case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
case vmIntrinsics::_notify:
case vmIntrinsics::_notifyAll:
return inline_notify(intrinsic_id());
case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
case vmIntrinsics::_arraycopy: return inline_arraycopy();
case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
case vmIntrinsics::_compressStringC:
case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
case vmIntrinsics::_inflateStringC:
case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
case vmIntrinsics::_loadFence:
case vmIntrinsics::_storeFence:
case vmIntrinsics::_storeStoreFence:
case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
case vmIntrinsics::_onSpinWait: return inline_onspinwait();
case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
case vmIntrinsics::_currentThread: return inline_native_currentThread();
case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
#ifdef JFR_HAVE_INTRINSICS
case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
#endif
case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
case vmIntrinsics::_getLength: return inline_native_getLength();
case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
case vmIntrinsics::_isInstance:
case vmIntrinsics::_getModifiers:
case vmIntrinsics::_isInterface:
case vmIntrinsics::_isArray:
case vmIntrinsics::_isPrimitive:
case vmIntrinsics::_isHidden:
case vmIntrinsics::_getSuperclass:
case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
case vmIntrinsics::_floatToRawIntBits:
case vmIntrinsics::_floatToIntBits:
case vmIntrinsics::_intBitsToFloat:
case vmIntrinsics::_doubleToRawLongBits:
case vmIntrinsics::_doubleToLongBits:
case vmIntrinsics::_longBitsToDouble:
case vmIntrinsics::_floatToFloat16:
case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
case vmIntrinsics::_floatIsFinite:
case vmIntrinsics::_floatIsInfinite:
case vmIntrinsics::_doubleIsFinite:
case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
case vmIntrinsics::_numberOfLeadingZeros_i:
case vmIntrinsics::_numberOfLeadingZeros_l:
case vmIntrinsics::_numberOfTrailingZeros_i:
case vmIntrinsics::_numberOfTrailingZeros_l:
case vmIntrinsics::_bitCount_i:
case vmIntrinsics::_bitCount_l:
case vmIntrinsics::_reverse_i:
case vmIntrinsics::_reverse_l:
case vmIntrinsics::_reverseBytes_i:
case vmIntrinsics::_reverseBytes_l:
case vmIntrinsics::_reverseBytes_s:
case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
case vmIntrinsics::_compress_i:
case vmIntrinsics::_compress_l:
case vmIntrinsics::_expand_i:
case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
case vmIntrinsics::_compareUnsigned_i:
case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
case vmIntrinsics::_divideUnsigned_i:
case vmIntrinsics::_divideUnsigned_l:
case vmIntrinsics::_remainderUnsigned_i:
case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
case vmIntrinsics::_Reference_get: return inline_reference_get();
case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
case vmIntrinsics::_Class_cast: return inline_Class_cast();
case vmIntrinsics::_aescrypt_encryptBlock:
case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
return inline_electronicCodeBook_AESCrypt(intrinsic_id());
case vmIntrinsics::_counterMode_AESCrypt:
return inline_counterMode_AESCrypt(intrinsic_id());
case vmIntrinsics::_galoisCounterMode_AESCrypt:
return inline_galoisCounterMode_AESCrypt();
case vmIntrinsics::_md5_implCompress:
case vmIntrinsics::_sha_implCompress:
case vmIntrinsics::_sha2_implCompress:
case vmIntrinsics::_sha5_implCompress:
case vmIntrinsics::_sha3_implCompress:
return inline_digestBase_implCompress(intrinsic_id());
case vmIntrinsics::_digestBase_implCompressMB:
return inline_digestBase_implCompressMB(predicate);
case vmIntrinsics::_multiplyToLen:
return inline_multiplyToLen();
case vmIntrinsics::_squareToLen:
return inline_squareToLen();
case vmIntrinsics::_mulAdd:
return inline_mulAdd();
case vmIntrinsics::_montgomeryMultiply:
return inline_montgomeryMultiply();
case vmIntrinsics::_montgomerySquare:
return inline_montgomerySquare();
case vmIntrinsics::_bigIntegerRightShiftWorker:
return inline_bigIntegerShift(true);
case vmIntrinsics::_bigIntegerLeftShiftWorker:
return inline_bigIntegerShift(false);
case vmIntrinsics::_vectorizedMismatch:
return inline_vectorizedMismatch();
case vmIntrinsics::_ghash_processBlocks:
return inline_ghash_processBlocks();
case vmIntrinsics::_chacha20Block:
return inline_chacha20Block();
case vmIntrinsics::_base64_encodeBlock:
return inline_base64_encodeBlock();
case vmIntrinsics::_base64_decodeBlock:
return inline_base64_decodeBlock();
case vmIntrinsics::_poly1305_processBlocks:
return inline_poly1305_processBlocks();
case vmIntrinsics::_encodeISOArray:
case vmIntrinsics::_encodeByteISOArray:
return inline_encodeISOArray(false);
case vmIntrinsics::_encodeAsciiArray:
return inline_encodeISOArray(true);
case vmIntrinsics::_updateCRC32:
return inline_updateCRC32();
case vmIntrinsics::_updateBytesCRC32:
return inline_updateBytesCRC32();
case vmIntrinsics::_updateByteBufferCRC32:
return inline_updateByteBufferCRC32();
case vmIntrinsics::_updateBytesCRC32C:
return inline_updateBytesCRC32C();
case vmIntrinsics::_updateDirectByteBufferCRC32C:
return inline_updateDirectByteBufferCRC32C();
case vmIntrinsics::_updateBytesAdler32:
return inline_updateBytesAdler32();
case vmIntrinsics::_updateByteBufferAdler32:
return inline_updateByteBufferAdler32();
case vmIntrinsics::_profileBoolean:
return inline_profileBoolean();
case vmIntrinsics::_isCompileConstant:
return inline_isCompileConstant();
case vmIntrinsics::_countPositives:
return inline_countPositives();
case vmIntrinsics::_fmaD:
case vmIntrinsics::_fmaF:
return inline_fma(intrinsic_id());
case vmIntrinsics::_isDigit:
case vmIntrinsics::_isLowerCase:
case vmIntrinsics::_isUpperCase:
case vmIntrinsics::_isWhitespace:
return inline_character_compare(intrinsic_id());
case vmIntrinsics::_min:
case vmIntrinsics::_max:
case vmIntrinsics::_min_strict:
case vmIntrinsics::_max_strict:
return inline_min_max(intrinsic_id());
case vmIntrinsics::_maxF:
case vmIntrinsics::_minF:
case vmIntrinsics::_maxD:
case vmIntrinsics::_minD:
case vmIntrinsics::_maxF_strict:
case vmIntrinsics::_minF_strict:
case vmIntrinsics::_maxD_strict:
case vmIntrinsics::_minD_strict:
return inline_fp_min_max(intrinsic_id());
case vmIntrinsics::_VectorUnaryOp:
return inline_vector_nary_operation(1);
case vmIntrinsics::_VectorBinaryOp:
return inline_vector_nary_operation(2);
case vmIntrinsics::_VectorTernaryOp:
return inline_vector_nary_operation(3);
case vmIntrinsics::_VectorFromBitsCoerced:
return inline_vector_frombits_coerced();
case vmIntrinsics::_VectorShuffleIota:
return inline_vector_shuffle_iota();
case vmIntrinsics::_VectorMaskOp:
return inline_vector_mask_operation();
case vmIntrinsics::_VectorShuffleToVector:
return inline_vector_shuffle_to_vector();
case vmIntrinsics::_VectorLoadOp:
return inline_vector_mem_operation(/*is_store=*/false);
case vmIntrinsics::_VectorLoadMaskedOp:
return inline_vector_mem_masked_operation(/*is_store*/false);
case vmIntrinsics::_VectorStoreOp:
return inline_vector_mem_operation(/*is_store=*/true);
case vmIntrinsics::_VectorStoreMaskedOp:
return inline_vector_mem_masked_operation(/*is_store=*/true);
case vmIntrinsics::_VectorGatherOp:
return inline_vector_gather_scatter(/*is_scatter*/ false);
case vmIntrinsics::_VectorScatterOp:
return inline_vector_gather_scatter(/*is_scatter*/ true);
case vmIntrinsics::_VectorReductionCoerced:
return inline_vector_reduction();
case vmIntrinsics::_VectorTest:
return inline_vector_test();
case vmIntrinsics::_VectorBlend:
return inline_vector_blend();
case vmIntrinsics::_VectorRearrange:
return inline_vector_rearrange();
case vmIntrinsics::_VectorCompare:
return inline_vector_compare();
case vmIntrinsics::_VectorBroadcastInt:
return inline_vector_broadcast_int();
case vmIntrinsics::_VectorConvert:
return inline_vector_convert();
case vmIntrinsics::_VectorInsert:
return inline_vector_insert();
case vmIntrinsics::_VectorExtract:
return inline_vector_extract();
case vmIntrinsics::_VectorCompressExpand:
return inline_vector_compress_expand();
case vmIntrinsics::_IndexVector:
return inline_index_vector();
case vmIntrinsics::_getObjectSize:
return inline_getObjectSize();
case vmIntrinsics::_blackhole:
return inline_blackhole();
default:
// If you get here, it may be that someone has added a new intrinsic
// to the list in vmIntrinsics.hpp without implementing it here.
#ifndef PRODUCT
if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
}
#endif
return false;
}
}
Node* LibraryCallKit::try_to_predicate(int predicate) {
if (!jvms()->has_method()) {
// Root JVMState has a null method.
assert(map()->memory()->Opcode() == Op_Parm, "");
// Insert the memory aliasing node
set_all_memory(reset_memory());
}
assert(merged_memory(), "");
switch (intrinsic_id()) {
case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
return inline_cipherBlockChaining_AESCrypt_predicate(false);
case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
return inline_cipherBlockChaining_AESCrypt_predicate(true);
case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
return inline_electronicCodeBook_AESCrypt_predicate(false);
case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
return inline_electronicCodeBook_AESCrypt_predicate(true);
case vmIntrinsics::_counterMode_AESCrypt:
return inline_counterMode_AESCrypt_predicate();
case vmIntrinsics::_digestBase_implCompressMB:
return inline_digestBase_implCompressMB_predicate(predicate);
case vmIntrinsics::_galoisCounterMode_AESCrypt:
return inline_galoisCounterMode_AESCrypt_predicate();
default:
// If you get here, it may be that someone has added a new intrinsic
// to the list in vmIntrinsics.hpp without implementing it here.
#ifndef PRODUCT
if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
}
#endif
Node* slow_ctl = control();
set_control(top()); // No fast path intrinsic
return slow_ctl;
}
}
//------------------------------set_result-------------------------------
// Helper function for finishing intrinsics.
void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
record_for_igvn(region);
set_control(_gvn.transform(region));
set_result( _gvn.transform(value));
assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
}
//------------------------------generate_guard---------------------------
// Helper function for generating guarded fast-slow graph structures.
// The given 'test', if true, guards a slow path. If the test fails
// then a fast path can be taken. (We generally hope it fails.)
// In all cases, GraphKit::control() is updated to the fast path.
// The returned value represents the control for the slow path.
// The return value is never 'top'; it is either a valid control
// or NULL if it is obvious that the slow path can never be taken.
// Also, if region and the slow control are not NULL, the slow edge
// is appended to the region.
Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
if (stopped()) {
// Already short circuited.
return NULL;
}
// Build an if node and its projections.
// If test is true we take the slow path, which we assume is uncommon.
if (_gvn.type(test) == TypeInt::ZERO) {
// The slow branch is never taken. No need to build this guard.
return NULL;
}
IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
Node* if_slow = _gvn.transform(new IfTrueNode(iff));
if (if_slow == top()) {
// The slow branch is never taken. No need to build this guard.
return NULL;
}
if (region != NULL)
region->add_req(if_slow);
Node* if_fast = _gvn.transform(new IfFalseNode(iff));
set_control(if_fast);
return if_slow;
}
inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
}
inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
return generate_guard(test, region, PROB_FAIR);
}
inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
Node* *pos_index) {
if (stopped())
return NULL; // already stopped
if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
return NULL; // index is already adequately typed
Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
if (is_neg != NULL && pos_index != NULL) {
// Emulate effect of Parse::adjust_map_after_if.
Node* ccast = new CastIINode(index, TypeInt::POS);
ccast->set_req(0, control());
(*pos_index) = _gvn.transform(ccast);
}
return is_neg;
}
// Make sure that 'position' is a valid limit index, in [0..length].
// There are two equivalent plans for checking this:
// A. (offset + copyLength) unsigned<= arrayLength
// B. offset <= (arrayLength - copyLength)
// We require that all of the values above, except for the sum and
// difference, are already known to be non-negative.
// Plan A is robust in the face of overflow, if offset and copyLength
// are both hugely positive.
//
// Plan B is less direct and intuitive, but it does not overflow at
// all, since the difference of two non-negatives is always
// representable. Whenever Java methods must perform the equivalent
// check they generally use Plan B instead of Plan A.
// For the moment we use Plan A.
inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
Node* subseq_length,
Node* array_length,
RegionNode* region) {
if (stopped())
return NULL; // already stopped
bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
if (zero_offset && subseq_length->eqv_uncast(array_length))
return NULL; // common case of whole-array copy
Node* last = subseq_length;
if (!zero_offset) // last += offset
last = _gvn.transform(new AddINode(last, offset));
Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
return is_over;
}
// Emit range checks for the given String.value byte array
void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
if (stopped()) {
return; // already stopped
}
RegionNode* bailout = new RegionNode(1);
record_for_igvn(bailout);
if (char_count) {
// Convert char count to byte count
count = _gvn.transform(new LShiftINode(count, intcon(1)));
}
// Offset and count must not be negative
generate_negative_guard(offset, bailout);
generate_negative_guard(count, bailout);
// Offset + count must not exceed length of array
generate_limit_guard(offset, count, load_array_length(array), bailout);
if (bailout->req() > 1) {
PreserveJVMState pjvms(this);
set_control(_gvn.transform(bailout));
uncommon_trap(Deoptimization::Reason_intrinsic,
Deoptimization::Action_maybe_recompile);
}
}
Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
bool is_immutable) {
ciKlass* thread_klass = env()->Thread_klass();
const Type* thread_type
= TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
Node* thread = _gvn.transform(new ThreadLocalNode());
Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
tls_output = thread;
Node* thread_obj_handle
= (is_immutable
? LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(),
TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
: make_load(NULL, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
thread_obj_handle = _gvn.transform(thread_obj_handle);
DecoratorSet decorators = IN_NATIVE;
if (is_immutable) {
decorators |= C2_IMMUTABLE_MEMORY;
}
return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
}
//--------------------------generate_current_thread--------------------
Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
/*is_immutable*/false);
}
//--------------------------generate_virtual_thread--------------------
Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
return current_thread_helper(tls_output, JavaThread::vthread_offset(),
!C->method()->changes_current_thread());
}
//------------------------------make_string_method_node------------------------
// Helper method for String intrinsic functions. This version is called with
// str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
// characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
// containing the lengths of str1 and str2.
Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
Node* result = NULL;
switch (opcode) {
case Op_StrIndexOf:
result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
str1_start, cnt1, str2_start, cnt2, ae);
break;
case Op_StrComp:
result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
str1_start, cnt1, str2_start, cnt2, ae);
break;
case Op_StrEquals:
// We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
// Use the constant length if there is one because optimized match rule may exist.
result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
break;
default:
ShouldNotReachHere();
return NULL;
}
// All these intrinsics have checks.
C->set_has_split_ifs(true); // Has chance for split-if optimization
clear_upper_avx();
return _gvn.transform(result);
}
//------------------------------inline_string_compareTo------------------------
bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
Node* arg1 = argument(0);
Node* arg2 = argument(1);
arg1 = must_be_not_null(arg1, true);
arg2 = must_be_not_null(arg2, true);
// Get start addr and length of first argument
Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
Node* arg1_cnt = load_array_length(arg1);
// Get start addr and length of second argument
Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
Node* arg2_cnt = load_array_length(arg2);
Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
set_result(result);
return true;
}
//------------------------------inline_string_equals------------------------
bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
Node* arg1 = argument(0);
Node* arg2 = argument(1);
// paths (plus control) merge
RegionNode* region = new RegionNode(3);
Node* phi = new PhiNode(region, TypeInt::BOOL);
if (!stopped()) {
arg1 = must_be_not_null(arg1, true);
arg2 = must_be_not_null(arg2, true);
// Get start addr and length of first argument
Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
Node* arg1_cnt = load_array_length(arg1);
// Get start addr and length of second argument
Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
Node* arg2_cnt = load_array_length(arg2);
// Check for arg1_cnt != arg2_cnt
Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
Node* if_ne = generate_slow_guard(bol, NULL);
if (if_ne != NULL) {
phi->init_req(2, intcon(0));
region->init_req(2, if_ne);
}
// Check for count == 0 is done by assembler code for StrEquals.
if (!stopped()) {
Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
phi->init_req(1, equals);
region->init_req(1, control());
}
}
// post merge
set_control(_gvn.transform(region));
record_for_igvn(region);
set_result(_gvn.transform(phi));
return true;
}
//------------------------------inline_array_equals----------------------------
bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
Node* arg1 = argument(0);
Node* arg2 = argument(1);
const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
clear_upper_avx();
return true;
}
//------------------------------inline_countPositives------------------------------
bool LibraryCallKit::inline_countPositives() {
if (too_many_traps(Deoptimization::Reason_intrinsic)) {
return false;
}
assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
// no receiver since it is static method
Node* ba = argument(0);
Node* offset = argument(1);
Node* len = argument(2);
ba = must_be_not_null(ba, true);
// Range checks
generate_string_range_check(ba, offset, len, false);
if (stopped()) {
return true;
}
Node* ba_start = array_element_address(ba, offset, T_BYTE);
Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
set_result(_gvn.transform(result));
return true;
}
bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
Node* index = argument(0);
Node* length = bt == T_INT ? argument(1) : argument(2);
if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
return false;
}
// check that length is positive
Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
{
BuildCutout unless(this, len_pos_bol, PROB_MAX);
uncommon_trap(Deoptimization::Reason_intrinsic,
Deoptimization::Action_make_not_entrant);
}
if (stopped()) {
// Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
return true;
}
// length is now known positive, add a cast node to make this explicit
jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
Node* casted_length = ConstraintCastNode::make(control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
casted_length = _gvn.transform(casted_length);
replace_in_map(length, casted_length);
length = casted_length;
// Use an unsigned comparison for the range check itself
Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
BoolTest::mask btest = BoolTest::lt;
Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
_gvn.set_type(rc, rc->Value(&_gvn));
if (!rc_bool->is_Con()) {
record_for_igvn(rc);
}
set_control(_gvn.transform(new IfTrueNode(rc)));
{
PreserveJVMState pjvms(this);
set_control(_gvn.transform(new IfFalseNode(rc)));
uncommon_trap(Deoptimization::Reason_range_check,
Deoptimization::Action_make_not_entrant);
}
if (stopped()) {
// Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
return true;
}
// index is now known to be >= 0 and < length, cast it
Node* result = ConstraintCastNode::make(control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
result = _gvn.transform(result);
set_result(result);
replace_in_map(index, result);
return true;
}
//------------------------------inline_string_indexOf------------------------
bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
return false;
}
Node* src = argument(0);
Node* tgt = argument(1);
// Make the merge point
RegionNode* result_rgn = new RegionNode(4);
Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
src = must_be_not_null(src, true);
tgt = must_be_not_null(tgt, true);
// Get start addr and length of source string
Node* src_start = array_element_address(src, intcon(0), T_BYTE);
Node* src_count = load_array_length(src);
// Get start addr and length of substring
Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
Node* tgt_count = load_array_length(tgt);
if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
// Divide src size by 2 if String is UTF16 encoded
src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
}
if (ae == StrIntrinsicNode::UU) {
// Divide substring size by 2 if String is UTF16 encoded
tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
}
Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
if (result != NULL) {
result_phi->init_req(3, result);
result_rgn->init_req(3, control());
}
set_control(_gvn.transform(result_rgn));
record_for_igvn(result_rgn);
set_result(_gvn.transform(result_phi));
return true;
}
//-----------------------------inline_string_indexOf-----------------------
bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
if (too_many_traps(Deoptimization::Reason_intrinsic)) {
return false;
}
if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
return false;
}
assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
Node* src = argument(0); // byte[]
Node* src_count = argument(1); // char count
Node* tgt = argument(2); // byte[]
Node* tgt_count = argument(3); // char count
Node* from_index = argument(4); // char index
src = must_be_not_null(src, true);
tgt = must_be_not_null(tgt, true);
// Multiply byte array index by 2 if String is UTF16 encoded
Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
src_count = _gvn.transform(new SubINode(src_count, from_index));
Node* src_start = array_element_address(src, src_offset, T_BYTE);
Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
// Range checks
generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
if (stopped()) {
return true;
}
RegionNode* region = new RegionNode(5);
Node* phi = new PhiNode(region, TypeInt::INT);
Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
if (result != NULL) {
// The result is index relative to from_index if substring was found, -1 otherwise.
// Generate code which will fold into cmove.
Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
Node* if_lt = generate_slow_guard(bol, NULL);
if (if_lt != NULL) {
// result == -1
phi->init_req(3, result);
region->init_req(3, if_lt);
}
if (!stopped()) {
result = _gvn.transform(new AddINode(result, from_index));
phi->init_req(4, result);
region->init_req(4, control());
}
}
set_control(_gvn.transform(region));
record_for_igvn(region);
set_result(_gvn.transform(phi));
clear_upper_avx();
return true;
}
// Create StrIndexOfNode with fast path checks
Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
// Check for substr count > string count
Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
Node* if_gt = generate_slow_guard(bol, NULL);
if (if_gt != NULL) {
phi->init_req(1, intcon(-1));
region->init_req(1, if_gt);
}
if (!stopped()) {
// Check for substr count == 0
cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
Node* if_zero = generate_slow_guard(bol, NULL);
if (if_zero != NULL) {
phi->init_req(2, intcon(0));
region->init_req(2, if_zero);
}
}
if (!stopped()) {
return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
}
return NULL;
}
//-----------------------------inline_string_indexOfChar-----------------------
bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
if (too_many_traps(Deoptimization::Reason_intrinsic)) {
return false;
}
if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
return false;
}
assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
Node* src = argument(0); // byte[]
Node* tgt = argument(1); // tgt is int ch
Node* from_index = argument(2);
Node* max = argument(3);
src = must_be_not_null(src, true);
Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
Node* src_start = array_element_address(src, src_offset, T_BYTE);
Node* src_count = _gvn.transform(new SubINode(max, from_index));
// Range checks
generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
if (stopped()) {
return true;
}
RegionNode* region = new RegionNode(3);
Node* phi = new PhiNode(region, TypeInt::INT);
Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, ae);
C->set_has_split_ifs(true); // Has chance for split-if optimization
_gvn.transform(result);
Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
Node* if_lt = generate_slow_guard(bol, NULL);
if (if_lt != NULL) {
// result == -1
phi->init_req(2, result);
region->init_req(2, if_lt);
}
if (!stopped()) {
result = _gvn.transform(new AddINode(result, from_index));
phi->init_req(1, result);
region->init_req(1, control());
}
set_control(_gvn.transform(region));
record_for_igvn(region);
set_result(_gvn.transform(phi));
return true;
}
//---------------------------inline_string_copy---------------------
// compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
// int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
// int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
// compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
// void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
// void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
bool LibraryCallKit::inline_string_copy(bool compress) {
if (too_many_traps(Deoptimization::Reason_intrinsic)) {
return false;
}
int nargs = 5; // 2 oops, 3 ints
assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
Node* src = argument(0);
Node* src_offset = argument(1);
Node* dst = argument(2);
Node* dst_offset = argument(3);
Node* length = argument(4);
// Check for allocation before we add nodes that would confuse
// tightly_coupled_allocation()
AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
// Figure out the size and type of the elements we will be copying.
const Type* src_type = src->Value(&_gvn);
const Type* dst_type = dst->Value(&_gvn);
BasicType src_elem = src_type->isa_aryptr()->elem()->array_element_basic_type();
BasicType dst_elem = dst_type->isa_aryptr()->elem()->array_element_basic_type();
assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
(!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
"Unsupported array types for inline_string_copy");
src = must_be_not_null(src, true);
dst = must_be_not_null(dst, true);
// Convert char[] offsets to byte[] offsets
bool convert_src = (compress && src_elem == T_BYTE);
bool convert_dst = (!compress && dst_elem == T_BYTE);
if (convert_src) {
src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
} else if (convert_dst) {
dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
}
// Range checks
generate_string_range_check(src, src_offset, length, convert_src);
generate_string_range_check(dst, dst_offset, length, convert_dst);
if (stopped()) {
return true;
}
Node* src_start = array_element_address(src, src_offset, src_elem);
Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
// 'src_start' points to src array + scaled offset
// 'dst_start' points to dst array + scaled offset
Node* count = NULL;
if (compress) {
count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
} else {
inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
}
if (alloc != NULL) {
if (alloc->maybe_set_complete(&_gvn)) {
// "You break it, you buy it."
InitializeNode* init = alloc->initialization();
assert(init->is_complete(), "we just did this");
init->set_complete_with_arraycopy();
assert(dst->is_CheckCastPP(), "sanity");
assert(dst->in(0)->in(0) == init, "dest pinned");
}
// Do not let stores that initialize this object be reordered with
// a subsequent store that would make this object accessible by
// other threads.
// Record what AllocateNode this StoreStore protects so that
// escape analysis can go from the MemBarStoreStoreNode to the
// AllocateNode and eliminate the MemBarStoreStoreNode if possible
// based on the escape status of the AllocateNode.
insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
}
if (compress) {
set_result(_gvn.transform(count));
}
clear_upper_avx();
return true;
}
#ifdef _LP64
#define XTOP ,top() /*additional argument*/
#else //_LP64
#define XTOP /*no additional argument*/
#endif //_LP64
//------------------------inline_string_toBytesU--------------------------
// public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
bool LibraryCallKit::inline_string_toBytesU() {
if (too_many_traps(Deoptimization::Reason_intrinsic)) {
return false;
}
// Get the arguments.
Node* value = argument(0);
Node* offset = argument(1);
Node* length = argument(2);
Node* newcopy = NULL;
// Set the original stack and the reexecute bit for the interpreter to reexecute
// the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
{ PreserveReexecuteState preexecs(this);
jvms()->set_should_reexecute(true);
// Check if a null path was taken unconditionally.
value = null_check(value);
RegionNode* bailout = new RegionNode(1);
record_for_igvn(bailout);
// Range checks
generate_negative_guard(offset, bailout);
generate_negative_guard(length, bailout);
generate_limit_guard(offset, length, load_array_length(value), bailout);
// Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
if (bailout->req() > 1) {
PreserveJVMState pjvms(this);
set_control(_gvn.transform(bailout));
uncommon_trap(Deoptimization::Reason_intrinsic,
Deoptimization::Action_maybe_recompile);
}
if (stopped()) {
return true;
}
Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
newcopy = new_array(klass_node, size, 0); // no arguments to push
AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
guarantee(alloc != NULL, "created above");
// Calculate starting addresses.
Node* src_start = array_element_address(value, offset, T_CHAR);
Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
// Check if src array address is aligned to HeapWordSize (dst is always aligned)
const TypeInt* toffset = gvn().type(offset)->is_int();
bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
// Figure out which arraycopy runtime method to call (disjoint, uninitialized).
const char* copyfunc_name = "arraycopy";
address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
OptoRuntime::fast_arraycopy_Type(),
copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
src_start, dst_start, ConvI2X(length) XTOP);
// Do not let reads from the cloned object float above the arraycopy.
if (alloc->maybe_set_complete(&_gvn)) {
// "You break it, you buy it."
InitializeNode* init = alloc->initialization();
assert(init->is_complete(), "we just did this");
init->set_complete_with_arraycopy();
assert(newcopy->is_CheckCastPP(), "sanity");
assert(newcopy->in(0)->in(0) == init, "dest pinned");
}
// Do not let stores that initialize this object be reordered with
// a subsequent store that would make this object accessible by
// other threads.
// Record what AllocateNode this StoreStore protects so that
// escape analysis can go from the MemBarStoreStoreNode to the
// AllocateNode and eliminate the MemBarStoreStoreNode if possible
// based on the escape status of the AllocateNode.
insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
} // original reexecute is set back here
C->set_has_split_ifs(true); // Has chance for split-if optimization
if (!stopped()) {
set_result(newcopy);
}
clear_upper_avx();
return true;
}
//------------------------inline_string_getCharsU--------------------------
// public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
bool LibraryCallKit::inline_string_getCharsU() {
if (too_many_traps(Deoptimization::Reason_intrinsic)) {
return false;
}
// Get the arguments.
Node* src = argument(0);
Node* src_begin = argument(1);
Node* src_end = argument(2); // exclusive offset (i < src_end)
Node* dst = argument(3);
Node* dst_begin = argument(4);
// Check for allocation before we add nodes that would confuse
// tightly_coupled_allocation()
AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
// Check if a null path was taken unconditionally.
src = null_check(src);
dst = null_check(dst);
if (stopped()) {
return true;
}
// Get length and convert char[] offset to byte[] offset
Node* length = _gvn.transform(new SubINode(src_end, src_begin));
src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
// Range checks
generate_string_range_check(src, src_begin, length, true);
generate_string_range_check(dst, dst_begin, length, false);
if (stopped()) {
return true;
}
if (!stopped()) {
// Calculate starting addresses.
Node* src_start = array_element_address(src, src_begin, T_BYTE);
Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
// Check if array addresses are aligned to HeapWordSize
const TypeInt* tsrc = gvn().type(src_begin)->is_int();
const TypeInt* tdst = gvn().type(dst_begin)->is_int();
bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
// Figure out which arraycopy runtime method to call (disjoint, uninitialized).
const char* copyfunc_name = "arraycopy";
address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
OptoRuntime::fast_arraycopy_Type(),
copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
src_start, dst_start, ConvI2X(length) XTOP);
// Do not let reads from the cloned object float above the arraycopy.
if (alloc != NULL) {
if (alloc->maybe_set_complete(&_gvn)) {
// "You break it, you buy it."
InitializeNode* init = alloc->initialization();
assert(init->is_complete(), "we just did this");
init->set_complete_with_arraycopy();
assert(dst->is_CheckCastPP(), "sanity");
assert(dst->in(0)->in(0) == init, "dest pinned");
}
// Do not let stores that initialize this object be reordered with
// a subsequent store that would make this object accessible by
// other threads.
// Record what AllocateNode this StoreStore protects so that
// escape analysis can go from the MemBarStoreStoreNode to the
// AllocateNode and eliminate the MemBarStoreStoreNode if possible
// based on the escape status of the AllocateNode.
insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
} else {
insert_mem_bar(Op_MemBarCPUOrder);
}
}
C->set_has_split_ifs(true); // Has chance for split-if optimization
return true;
}
//----------------------inline_string_char_access----------------------------
// Store/Load char to/from byte[] array.
// static void StringUTF16.putChar(byte[] val, int index, int c)
// static char StringUTF16.getChar(byte[] val, int index)
bool LibraryCallKit::inline_string_char_access(bool is_store) {
Node* value = argument(0);
Node* index = argument(1);
Node* ch = is_store ? argument(2) : NULL;
// This intrinsic accesses byte[] array as char[] array. Computing the offsets
// correctly requires matched array shapes.
assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
"sanity: byte[] and char[] bases agree");
assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
"sanity: byte[] and char[] scales agree");
// Bail when getChar over constants is requested: constant folding would
// reject folding mismatched char access over byte[]. A normal inlining for getChar
// Java method would constant fold nicely instead.
if (!is_store && value->is_Con() && index->is_Con()) {
return false;
}
// Save state and restore on bailout
uint old_sp = sp();
SafePointNode* old_map = clone_map();
value = must_be_not_null(value, true);
Node* adr = array_element_address(value, index, T_CHAR);
if (adr->is_top()) {
set_map(old_map);
set_sp(old_sp);
return false;
}
old_map->destruct(&_gvn);
if (is_store) {
access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
} else {
ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
set_result(ch);
}
return true;
}
//--------------------------round_double_node--------------------------------
// Round a double node if necessary.
Node* LibraryCallKit::round_double_node(Node* n) {
if (Matcher::strict_fp_requires_explicit_rounding) {
#ifdef IA32
if (UseSSE < 2) {
n = _gvn.transform(new RoundDoubleNode(NULL, n));
}
#else
Unimplemented();
#endif // IA32
}
return n;
}
//------------------------------inline_math-----------------------------------
// public static double Math.abs(double)
// public static double Math.sqrt(double)
// public static double Math.log(double)
// public static double Math.log10(double)
// public static double Math.round(double)
bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
Node* arg = round_double_node(argument(0));
Node* n = NULL;
switch (id) {
case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
case vmIntrinsics::_dsqrt:
case vmIntrinsics::_dsqrt_strict:
n = new SqrtDNode(C, control(), arg); break;
case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
default: fatal_unexpected_iid(id); break;
}
set_result(_gvn.transform(n));
return true;
}
//------------------------------inline_math-----------------------------------
// public static float Math.abs(float)
// public static int Math.abs(int)
// public static long Math.abs(long)
bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
Node* arg = argument(0);
Node* n = NULL;
switch (id) {
case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
default: fatal_unexpected_iid(id); break;
}
set_result(_gvn.transform(n));
return true;
}
//------------------------------runtime_math-----------------------------
bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
"must be (DD)D or (D)D type");
// Inputs
Node* a = round_double_node(argument(0));
Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
const TypePtr* no_memory_effects = NULL;
Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
no_memory_effects,
a, top(), b, b ? top() : NULL);
Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
#ifdef ASSERT
Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
assert(value_top == top(), "second value must be top");
#endif
set_result(value);
return true;
}
//------------------------------inline_math_pow-----------------------------
bool LibraryCallKit::inline_math_pow() {
Node* exp = round_double_node(argument(2));
const TypeD* d = _gvn.type(exp)->isa_double_constant();
if (d != NULL) {
if (d->getd() == 2.0) {
// Special case: pow(x, 2.0) => x * x
Node* base = round_double_node(argument(0));
set_result(_gvn.transform(new MulDNode(base, base)));
return true;
} else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
// Special case: pow(x, 0.5) => sqrt(x)
Node* base = round_double_node(argument(0));
Node* zero = _gvn.zerocon(T_DOUBLE);
RegionNode* region = new RegionNode(3);
Node* phi = new PhiNode(region, Type::DOUBLE);
Node* cmp = _gvn.transform(new CmpDNode(base, zero));
// According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
// So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
// -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
Node* if_pow = generate_slow_guard(test, NULL);
Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
phi->init_req(1, value_sqrt);
region->init_req(1, control());
if (if_pow != NULL) {
set_control(if_pow);
address target = StubRoutines::dpow() != NULL ? StubRoutines::dpow() :
CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
const TypePtr* no_memory_effects = NULL;
Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
no_memory_effects, base, top(), exp, top());
Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
#ifdef ASSERT
Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
assert(value_top == top(), "second value must be top");
#endif
phi->init_req(2, value_pow);
region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
}
C->set_has_split_ifs(true); // Has chance for split-if optimization
set_control(_gvn.transform(region));
record_for_igvn(region);
set_result(_gvn.transform(phi));
return true;
}
}
return StubRoutines::dpow() != NULL ?
runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
}
//------------------------------inline_math_native-----------------------------
bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
switch (id) {
case vmIntrinsics::_dsin:
return StubRoutines::dsin() != NULL ?
runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
case vmIntrinsics::_dcos:
return StubRoutines::dcos() != NULL ?
runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
case vmIntrinsics::_dtan:
return StubRoutines::dtan() != NULL ?
runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
case vmIntrinsics::_dexp:
return StubRoutines::dexp() != NULL ?
runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
case vmIntrinsics::_dlog:
return StubRoutines::dlog() != NULL ?
runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
case vmIntrinsics::_dlog10:
return StubRoutines::dlog10() != NULL ?
runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
case vmIntrinsics::_ceil:
case vmIntrinsics::_floor:
case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
case vmIntrinsics::_dsqrt:
case vmIntrinsics::_dsqrt_strict:
return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
case vmIntrinsics::_dpow: return inline_math_pow();
case vmIntrinsics::_dcopySign: return inline_double_math(id);
case vmIntrinsics::_fcopySign: return inline_math(id);
case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
// These intrinsics are not yet correctly implemented
case vmIntrinsics::_datan2:
return false;
default:
fatal_unexpected_iid(id);
return false;
}
}
//----------------------------inline_notify-----------------------------------*
bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
address func;
if (id == vmIntrinsics::_notify) {
func = OptoRuntime::monitor_notify_Java();
} else {
func = OptoRuntime::monitor_notifyAll_Java();
}
Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
make_slow_call_ex(call, env()->Throwable_klass(), false);
return true;
}
//----------------------------inline_min_max-----------------------------------
bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
set_result(generate_min_max(id, argument(0), argument(1)));
return true;
}
void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
Node* fast_path = _gvn.transform( new IfFalseNode(check));
Node* slow_path = _gvn.transform( new IfTrueNode(check) );
{
PreserveJVMState pjvms(this);
PreserveReexecuteState preexecs(this);
jvms()->set_should_reexecute(true);
set_control(slow_path);
set_i_o(i_o());
uncommon_trap(Deoptimization::Reason_intrinsic,
Deoptimization::Action_none);
}
set_control(fast_path);
set_result(math);
}
template <typename OverflowOp>
bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
typedef typename OverflowOp::MathOp MathOp;
MathOp* mathOp = new MathOp(arg1, arg2);
Node* operation = _gvn.transform( mathOp );
Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
inline_math_mathExact(operation, ofcheck);
return true;
}
bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
}
bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
}
bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
}
bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
}
bool LibraryCallKit::inline_math_negateExactI() {
return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
}
bool LibraryCallKit::inline_math_negateExactL() {
return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
}
bool LibraryCallKit::inline_math_multiplyExactI() {
return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
}
bool LibraryCallKit::inline_math_multiplyExactL() {
return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
}
bool LibraryCallKit::inline_math_multiplyHigh() {
set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
return true;
}
bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
return true;
}
Node*
LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
Node* result_val = NULL;
switch (id) {
case vmIntrinsics::_min:
case vmIntrinsics::_min_strict:
result_val = _gvn.transform(new MinINode(x0, y0));
break;
case vmIntrinsics::_max:
case vmIntrinsics::_max_strict:
result_val = _gvn.transform(new MaxINode(x0, y0));
break;
default:
fatal_unexpected_iid(id);
break;
}
return result_val;
}
inline int
LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
const TypePtr* base_type = TypePtr::NULL_PTR;
if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
if (base_type == NULL) {
// Unknown type.
return Type::AnyPtr;
} else if (base_type == TypePtr::NULL_PTR) {
// Since this is a NULL+long form, we have to switch to a rawptr.
base = _gvn.transform(new CastX2PNode(offset));
offset = MakeConX(0);
return Type::RawPtr;
} else if (base_type->base() == Type::RawPtr) {
return Type::RawPtr;
} else if (base_type->isa_oopptr()) {
// Base is never null => always a heap address.
if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
return Type::OopPtr;
}
// Offset is small => always a heap address.
const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
if (offset_type != NULL &&
base_type->offset() == 0 && // (should always be?)
offset_type->_lo >= 0 &&
!MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
return Type::OopPtr;
} else if (type == T_OBJECT) {
// off heap access to an oop doesn't make any sense. Has to be on
// heap.
return Type::OopPtr;
}
// Otherwise, it might either be oop+off or NULL+addr.
return Type::AnyPtr;
} else {
// No information:
return Type::AnyPtr;
}
}
Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
Node* uncasted_base = base;
int kind = classify_unsafe_addr(uncasted_base, offset, type);
if (kind == Type::RawPtr) {
return basic_plus_adr(top(), uncasted_base, offset);
} else if (kind == Type::AnyPtr) {
assert(base == uncasted_base, "unexpected base change");
if (can_cast) {
if (!_gvn.type(base)->speculative_maybe_null() &&
!too_many_traps(Deoptimization::Reason_speculate_null_check)) {
// According to profiling, this access is always on
// heap. Casting the base to not null and thus avoiding membars
// around the access should allow better optimizations
Node* null_ctl = top();
base = null_check_oop(base, &null_ctl, true, true, true);
assert(null_ctl->is_top(), "no null control here");
return basic_plus_adr(base, offset);
} else if (_gvn.type(base)->speculative_always_null() &&
!too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
// According to profiling, this access is always off
// heap.
base = null_assert(base);
Node* raw_base = _gvn.transform(new CastX2PNode(offset));
offset = MakeConX(0);
return basic_plus_adr(top(), raw_base, offset);
}
}
// We don't know if it's an on heap or off heap access. Fall back
// to raw memory access.
Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
return basic_plus_adr(top(), raw, offset);
} else {
assert(base == uncasted_base, "unexpected base change");
// We know it's an on heap access so base can't be null
if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
base = must_be_not_null(base, true);
}
return basic_plus_adr(base, offset);
}
}
//--------------------------inline_number_methods-----------------------------
// inline int Integer.numberOfLeadingZeros(int)
// inline int Long.numberOfLeadingZeros(long)
//
// inline int Integer.numberOfTrailingZeros(int)
// inline int Long.numberOfTrailingZeros(long)
//
// inline int Integer.bitCount(int)
// inline int Long.bitCount(long)
//
// inline char Character.reverseBytes(char)
// inline short Short.reverseBytes(short)
// inline int Integer.reverseBytes(int)
// inline long Long.reverseBytes(long)
bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
Node* arg = argument(0);
Node* n = NULL;
switch (id) {
case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
case vmIntrinsics::_reverse_i: n = new ReverseINode(0, arg); break;
case vmIntrinsics::_reverse_l: n = new ReverseLNode(0, arg); break;
default: fatal_unexpected_iid(id); break;
}
set_result(_gvn.transform(n));
return true;
}
//--------------------------inline_bitshuffle_methods-----------------------------
// inline int Integer.compress(int, int)
// inline int Integer.expand(int, int)
// inline long Long.compress(long, long)
// inline long Long.expand(long, long)
bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
Node* n = NULL;
switch (id) {
case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
default: fatal_unexpected_iid(id); break;
}
set_result(_gvn.transform(n));
return true;
}
//--------------------------inline_number_methods-----------------------------
// inline int Integer.compareUnsigned(int, int)
// inline int Long.compareUnsigned(long, long)
bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
Node* arg1 = argument(0);
Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
Node* n = NULL;
switch (id) {
case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
default: fatal_unexpected_iid(id); break;
}
set_result(_gvn.transform(n));
return true;
}
//--------------------------inline_unsigned_divmod_methods-----------------------------
// inline int Integer.divideUnsigned(int, int)
// inline int Integer.remainderUnsigned(int, int)
// inline long Long.divideUnsigned(long, long)
// inline long Long.remainderUnsigned(long, long)
bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
Node* n = NULL;
switch (id) {
case vmIntrinsics::_divideUnsigned_i: {
zero_check_int(argument(1));
// Compile-time detect of null-exception
if (stopped()) {
return true; // keep the graph constructed so far
}
n = new UDivINode(control(), argument(0), argument(1));
break;
}
case vmIntrinsics::_divideUnsigned_l: {
zero_check_long(argument(2));
// Compile-time detect of null-exception
if (stopped()) {
return true; // keep the graph constructed so far
}
n = new UDivLNode(control(), argument(0), argument(2));
break;
}
case vmIntrinsics::_remainderUnsigned_i: {
zero_check_int(argument(1));
// Compile-time detect of null-exception
if (stopped()) {
return true; // keep the graph constructed so far
}
n = new UModINode(control(), argument(0), argument(1));
break;
}
case vmIntrinsics::_remainderUnsigned_l: {
zero_check_long(argument(2));
// Compile-time detect of null-exception
if (stopped()) {
return true; // keep the graph constructed so far
}
n = new UModLNode(control(), argument(0), argument(2));
break;
}
default: fatal_unexpected_iid(id); break;
}
set_result(_gvn.transform(n));
return true;
}
//----------------------------inline_unsafe_access----------------------------
const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
// Attempt to infer a sharper value type from the offset and base type.
ciKlass* sharpened_klass = NULL;
// See if it is an instance field, with an object type.
if (alias_type->field() != NULL) {
if (alias_type->field()->type()->is_klass()) {
sharpened_klass = alias_type->field()->type()->as_klass();
}
}
const TypeOopPtr* result = NULL;
// See if it is a narrow oop array.
if (adr_type->isa_aryptr()) {
if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
if (elem_type != NULL && elem_type->is_loaded()) {
// Sharpen the value type.
result = elem_type;
}
}
}
// The sharpened class might be unloaded if there is no class loader
// contraint in place.
if (result == NULL && sharpened_klass != NULL && sharpened_klass->is_loaded()) {
// Sharpen the value type.
result = TypeOopPtr::make_from_klass(sharpened_klass);
}
if (result != NULL) {
#ifndef PRODUCT
if (C->print_intrinsics() || C->print_inlining()) {
tty->print(" from base type: "); adr_type->dump(); tty->cr();
tty->print(" sharpened value: "); result->dump(); tty->cr();
}
#endif
}
return result;
}
DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
switch (kind) {
case Relaxed:
return MO_UNORDERED;
case Opaque:
return MO_RELAXED;
case Acquire:
return MO_ACQUIRE;
case Release:
return MO_RELEASE;
case Volatile:
return MO_SEQ_CST;
default:
ShouldNotReachHere();
return 0;
}
}
bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
if (callee()->is_static()) return false; // caller must have the capability!
DecoratorSet decorators = C2_UNSAFE_ACCESS;
guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
if (is_reference_type(type)) {
decorators |= ON_UNKNOWN_OOP_REF;
}
if (unaligned) {
decorators |= C2_UNALIGNED;
}
#ifndef PRODUCT
{
ResourceMark rm;
// Check the signatures.
ciSignature* sig = callee()->signature();
#ifdef ASSERT
if (!is_store) {
// Object getReference(Object base, int/long offset), etc.
BasicType rtype = sig->return_type()->basic_type();
assert(rtype == type, "getter must return the expected value");
assert(sig->count() == 2, "oop getter has 2 arguments");
assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
} else {
// void putReference(Object base, int/long offset, Object x), etc.
assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
assert(sig->count() == 3, "oop putter has 3 arguments");
assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
assert(vtype == type, "putter must accept the expected value");
}
#endif // ASSERT
}
#endif //PRODUCT
C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
Node* receiver = argument(0); // type: oop
// Build address expression.
Node* heap_base_oop = top();
// The base is either a Java object or a value produced by Unsafe.staticFieldBase
Node* base = argument(1); // type: oop
// The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
Node* offset = argument(2); // type: long
// We currently rely on the cookies produced by Unsafe.xxxFieldOffset
// to be plain byte offsets, which are also the same as those accepted
// by oopDesc::field_addr.
assert(Unsafe_field_offset_to_byte_offset(11) == 11,
"fieldOffset must be byte-scaled");
// 32-bit machines ignore the high half!
offset = ConvL2X(offset);
// Save state and restore on bailout
uint old_sp = sp();
SafePointNode* old_map = clone_map();
Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
if (type != T_OBJECT) {
decorators |= IN_NATIVE; // off-heap primitive access
} else {
set_map(old_map);
set_sp(old_sp);
return false; // off-heap oop accesses are not supported
}
} else {
heap_base_oop = base; // on-heap or mixed access
}
// Can base be NULL? Otherwise, always on-heap access.
bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
if (!can_access_non_heap) {
decorators |= IN_HEAP;
}
Node* val = is_store ? argument(4) : NULL;
const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
if (adr_type == TypePtr::NULL_PTR) {
set_map(old_map);
set_sp(old_sp);
return false; // off-heap access with zero address
}
// Try to categorize the address.
Compile::AliasType* alias_type = C->alias_type(adr_type);
assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
if (alias_type->adr_type() == TypeInstPtr::KLASS ||
alias_type->adr_type() == TypeAryPtr::RANGE) {
set_map(old_map);
set_sp(old_sp);
return false; // not supported
}
bool mismatched = false;
BasicType bt = alias_type->basic_type();
if (bt != T_ILLEGAL) {
assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
if (bt == T_BYTE && adr_type->isa_aryptr()) {
// Alias type doesn't differentiate between byte[] and boolean[]).
// Use address type to get the element type.
bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
}
if (is_reference_type(bt, true)) {
// accessing an array field with getReference is not a mismatch
bt = T_OBJECT;
}
if ((bt == T_OBJECT) != (type == T_OBJECT)) {
// Don't intrinsify mismatched object accesses
set_map(old_map);
set_sp(old_sp);
return false;
}
mismatched = (bt != type);
} else if (alias_type->adr_type()->isa_oopptr()) {
mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
}
old_map->destruct(&_gvn);
assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
if (mismatched) {
decorators |= C2_MISMATCHED;
}
// First guess at the value type.
const Type *value_type = Type::get_const_basic_type(type);
// Figure out the memory ordering.
decorators |= mo_decorator_for_access_kind(kind);
if (!is_store && type == T_OBJECT) {
const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
if (tjp != NULL) {
value_type = tjp;
}
}
receiver = null_check(receiver);
if (stopped()) {
return true;
}
// Heap pointers get a null-check from the interpreter,
// as a courtesy. However, this is not guaranteed by Unsafe,
// and it is not possible to fully distinguish unintended nulls
// from intended ones in this API.
if (!is_store) {
Node* p = NULL;
// Try to constant fold a load from a constant field
ciField* field = alias_type->field();
if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
// final or stable field
p = make_constant_from_field(field, heap_base_oop);
}
if (p == NULL) { // Could not constant fold the load
p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
// Normalize the value returned by getBoolean in the following cases
if (type == T_BOOLEAN &&
(mismatched ||
heap_base_oop == top() || // - heap_base_oop is NULL or
(can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
// and the unsafe access is made to large offset
// (i.e., larger than the maximum offset necessary for any
// field access)
) {
IdealKit ideal = IdealKit(this);
#define __ ideal.
IdealVariable normalized_result(ideal);
__ declarations_done();
__ set(normalized_result, p);
__ if_then(p, BoolTest::ne, ideal.ConI(0));
__ set(normalized_result, ideal.ConI(1));
ideal.end_if();
final_sync(ideal);
p = __ value(normalized_result);
#undef __
}
}
if (type == T_ADDRESS) {
p = gvn().transform(new CastP2XNode(NULL, p));
p = ConvX2UL(p);
}
// The load node has the control of the preceding MemBarCPUOrder. All
// following nodes will have the control of the MemBarCPUOrder inserted at
// the end of this method. So, pushing the load onto the stack at a later
// point is fine.
set_result(p);
} else {
if (bt == T_ADDRESS) {
// Repackage the long as a pointer.
val = ConvL2X(val);
val = gvn().transform(new CastX2PNode(val));
}
access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
}
return true;
}
//----------------------------inline_unsafe_load_store----------------------------
// This method serves a couple of different customers (depending on LoadStoreKind):
//
// LS_cmp_swap:
//
// boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
// boolean compareAndSetInt( Object o, long offset, int expected, int x);
// boolean compareAndSetLong( Object o, long offset, long expected, long x);
//
// LS_cmp_swap_weak:
//
// boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
// boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
// boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
// boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
//
// boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
// boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
// boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
// boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
//
// boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
// boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
// boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
// boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
//
// LS_cmp_exchange:
//
// Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
// Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
// Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
//
// Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
// Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
// Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
//
// Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
// Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
// Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
//
// LS_get_add:
//
// int getAndAddInt( Object o, long offset, int delta)
// long getAndAddLong(Object o, long offset, long delta)
//
// LS_get_set:
//
// int getAndSet(Object o, long offset, int newValue)
// long getAndSet(Object o, long offset, long newValue)
// Object getAndSet(Object o, long offset, Object newValue)
//
bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
// This basic scheme here is the same as inline_unsafe_access, but
// differs in enough details that combining them would make the code
// overly confusing. (This is a true fact! I originally combined
// them, but even I was confused by it!) As much code/comments as
// possible are retained from inline_unsafe_access though to make
// the correspondences clearer. - dl
if (callee()->is_static()) return false; // caller must have the capability!
DecoratorSet decorators = C2_UNSAFE_ACCESS;
decorators |= mo_decorator_for_access_kind(access_kind);
#ifndef PRODUCT
BasicType rtype;
{
ResourceMark rm;
// Check the signatures.
ciSignature* sig = callee()->signature();
rtype = sig->return_type()->basic_type();
switch(kind) {
case LS_get_add:
case LS_get_set: {
// Check the signatures.
#ifdef ASSERT
assert(rtype == type, "get and set must return the expected type");
assert(sig->count() == 3, "get and set has 3 arguments");
assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
#endif // ASSERT
break;
}
case LS_cmp_swap:
case LS_cmp_swap_weak: {
// Check the signatures.
#ifdef ASSERT
assert(rtype == T_BOOLEAN, "CAS must return boolean");
assert(sig->count() == 4, "CAS has 4 arguments");
assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
#endif // ASSERT
break;
}
case LS_cmp_exchange: {
// Check the signatures.
#ifdef ASSERT
assert(rtype == type, "CAS must return the expected type");
assert(sig->count() == 4, "CAS has 4 arguments");
assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
#endif // ASSERT
break;
}
default:
ShouldNotReachHere();
}
}
#endif //PRODUCT
C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
// Get arguments:
Node* receiver = NULL;
Node* base = NULL;
Node* offset = NULL;
Node* oldval = NULL;
Node* newval = NULL;
switch(kind) {
case LS_cmp_swap:
case LS_cmp_swap_weak:
case LS_cmp_exchange: {
const bool two_slot_type = type2size[type] == 2;
receiver = argument(0); // type: oop
base = argument(1); // type: oop
offset = argument(2); // type: long
oldval = argument(4); // type: oop, int, or long
newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
break;
}
case LS_get_add:
case LS_get_set: {
receiver = argument(0); // type: oop
base = argument(1); // type: oop
offset = argument(2); // type: long
oldval = NULL;
newval = argument(4); // type: oop, int, or long
break;
}
default:
ShouldNotReachHere();
}
// Build field offset expression.
// We currently rely on the cookies produced by Unsafe.xxxFieldOffset
// to be plain byte offsets, which are also the same as those accepted
// by oopDesc::field_addr.
assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
// 32-bit machines ignore the high half of long offsets
offset = ConvL2X(offset);
// Save state and restore on bailout
uint old_sp = sp();
SafePointNode* old_map = clone_map();
Node* adr = make_unsafe_address(base, offset,type, false);
const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
Compile::AliasType* alias_type = C->alias_type(adr_type);
BasicType bt = alias_type->basic_type();
if (bt != T_ILLEGAL &&
(is_reference_type(bt) != (type == T_OBJECT))) {
// Don't intrinsify mismatched object accesses.
set_map(old_map);
set_sp(old_sp);
return false;
}
old_map->destruct(&_gvn);
// For CAS, unlike inline_unsafe_access, there seems no point in
// trying to refine types. Just use the coarse types here.
assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
const Type *value_type = Type::get_const_basic_type(type);
switch (kind) {
case LS_get_set:
case LS_cmp_exchange: {
if (type == T_OBJECT) {
const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
if (tjp != NULL) {
value_type = tjp;
}
}
break;
}
case LS_cmp_swap:
case LS_cmp_swap_weak:
case LS_get_add:
break;
default:
ShouldNotReachHere();
}
// Null check receiver.
receiver = null_check(receiver);
if (stopped()) {
return true;
}
int alias_idx = C->get_alias_index(adr_type);
if (is_reference_type(type)) {
decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
// Transformation of a value which could be NULL pointer (CastPP #NULL)
// could be delayed during Parse (for example, in adjust_map_after_if()).
// Execute transformation here to avoid barrier generation in such case.
if (_gvn.type(newval) == TypePtr::NULL_PTR)
newval = _gvn.makecon(TypePtr::NULL_PTR);
if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
// Refine the value to a null constant, when it is known to be null
oldval = _gvn.makecon(TypePtr::NULL_PTR);
}
}
Node* result = NULL;
switch (kind) {
case LS_cmp_exchange: {
result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
oldval, newval, value_type, type, decorators);
break;
}
case LS_cmp_swap_weak:
decorators |= C2_WEAK_CMPXCHG;
case LS_cmp_swap: {
result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
oldval, newval, value_type, type, decorators);
break;
}
case LS_get_set: {
result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
newval, value_type, type, decorators);
break;
}
case LS_get_add: {
result = access_atomic_add_at(base, adr, adr_type, alias_idx,
newval, value_type, type, decorators);
break;
}
default:
ShouldNotReachHere();
}
assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
set_result(result);
return true;
}
bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
// Regardless of form, don't allow previous ld/st to move down,
// then issue acquire, release, or volatile mem_bar.
insert_mem_bar(Op_MemBarCPUOrder);
switch(id) {
case vmIntrinsics::_loadFence:
insert_mem_bar(Op_LoadFence);
return true;
case vmIntrinsics::_storeFence:
insert_mem_bar(Op_StoreFence);
return true;
case vmIntrinsics::_storeStoreFence:
insert_mem_bar(Op_StoreStoreFence);
return true;
case vmIntrinsics::_fullFence:
insert_mem_bar(Op_MemBarVolatile);
return true;
default:
fatal_unexpected_iid(id);
return false;
}
}
bool LibraryCallKit::inline_onspinwait() {
insert_mem_bar(Op_OnSpinWait);
return true;
}
bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
if (!kls->is_Con()) {
return true;
}
const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
if (klsptr == NULL) {
return true;
}
ciInstanceKlass* ik = klsptr->instance_klass();
// don't need a guard for a klass that is already initialized
return !ik->is_initialized();
}
//----------------------------inline_unsafe_writeback0-------------------------
// public native void Unsafe.writeback0(long address)
bool LibraryCallKit::inline_unsafe_writeback0() {
if (!Matcher::has_match_rule(Op_CacheWB)) {
return false;
}
#ifndef PRODUCT
assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
ciSignature* sig = callee()->signature();
assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
#endif
null_check_receiver(); // null-check, then ignore
Node *addr = argument(1);
addr = new CastX2PNode(addr);
addr = _gvn.transform(addr);
Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
flush = _gvn.transform(flush);
set_memory(flush, TypeRawPtr::BOTTOM);
return true;
}
//----------------------------inline_unsafe_writeback0-------------------------
// public native void Unsafe.writeback0(long address)
bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
return false;
}
if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
return false;
}
#ifndef PRODUCT
assert(Matcher::has_match_rule(Op_CacheWB),
(is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
: "found match rule for CacheWBPostSync but not CacheWB"));
#endif
null_check_receiver(); // null-check, then ignore
Node *sync;
if (is_pre) {
sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
} else {
sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
}
sync = _gvn.transform(sync);
set_memory(sync, TypeRawPtr::BOTTOM);
return true;
}
//----------------------------inline_unsafe_allocate---------------------------
// public native Object Unsafe.allocateInstance(Class<?> cls);
bool LibraryCallKit::inline_unsafe_allocate() {
if (callee()->is_static()) return false; // caller must have the capability!
null_check_receiver(); // null-check, then ignore
Node* cls = null_check(argument(1));
if (stopped()) return true;
Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
kls = null_check(kls);
if (stopped()) return true; // argument was like int.class
Node* test = NULL;
if (LibraryCallKit::klass_needs_init_guard(kls)) {
// Note: The argument might still be an illegal value like
// Serializable.class or Object[].class. The runtime will handle it.
// But we must make an explicit check for initialization.
Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
// Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
// can generate code to load it as unsigned byte.
Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
Node* bits = intcon(InstanceKlass::fully_initialized);
test = _gvn.transform(new SubINode(inst, bits));
// The 'test' is non-zero if we need to take a slow path.
}
Node* obj = new_instance(kls, test);
set_result(obj);
return true;
}
//------------------------inline_native_time_funcs--------------
// inline code for System.currentTimeMillis() and System.nanoTime()
// these have the same type and signature
bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
const TypeFunc* tf = OptoRuntime::void_long_Type();
const TypePtr* no_memory_effects = NULL;
Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
#ifdef ASSERT
Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
assert(value_top == top(), "second value must be top");
#endif
set_result(value);
return true;
}
#ifdef JFR_HAVE_INTRINSICS
/**
* if oop->klass != null
* // normal class
* epoch = _epoch_state ? 2 : 1
* if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
* ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
* }
* id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
* else
* // primitive class
* if oop->array_klass != null
* id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
* else
* id = LAST_TYPE_ID + 1 // void class path
* if (!signaled)
* signaled = true
*/
bool LibraryCallKit::inline_native_classID() {
Node* cls = argument(0);
IdealKit ideal(this);
#define __ ideal.
IdealVariable result(ideal); __ declarations_done();
Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
basic_plus_adr(cls, java_lang_Class::klass_offset()),
TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
__ if_then(kls, BoolTest::ne, null()); {
Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
mask = _gvn.transform(new OrLNode(mask, epoch));
Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
float unlikely = PROB_UNLIKELY(0.999);
__ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
sync_kit(ideal);
make_runtime_call(RC_LEAF,
OptoRuntime::class_id_load_barrier_Type(),
CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
"class id load barrier",
TypePtr::BOTTOM,
kls);
ideal.sync_kit(this);
} __ end_if();
ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
} __ else_(); {
Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
__ if_then(array_kls, BoolTest::ne, null()); {
Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
} __ else_(); {
// void class case
ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
} __ end_if();
Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
__ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
} __ end_if();
} __ end_if();
final_sync(ideal);
set_result(ideal.value(result));
#undef __
return true;
}
/*
* The intrinsic is a model of this pseudo-code:
*
* JfrThreadLocal* const tl = Thread::jfr_thread_local()
* jobject h_event_writer = tl->java_event_writer();
* if (h_event_writer == NULL) {
* return NULL;
* }
* oop threadObj = Thread::threadObj();
* oop vthread = java_lang_Thread::vthread(threadObj);
* traceid tid;
* bool excluded;
* if (vthread != threadObj) { // i.e. current thread is virtual
* tid = java_lang_Thread::tid(vthread);
* u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
* excluded = vthread_epoch_raw & excluded_mask;
* if (!excluded) {
* traceid current_epoch = JfrTraceIdEpoch::current_generation();
* u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
* if (vthread_epoch != current_epoch) {
* write_checkpoint();
* }
* }
* } else {
* tid = java_lang_Thread::tid(threadObj);
* u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
* excluded = thread_epoch_raw & excluded_mask;
* }
* oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
* traceid tid_in_event_writer = getField(event_writer, "threadID");
* if (tid_in_event_writer != tid) {
* setField(event_writer, "threadID", tid);
* setField(event_writer, "excluded", excluded);
* }
* return event_writer
*/
bool LibraryCallKit::inline_native_getEventWriter() {
enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
// Save input memory and i_o state.
Node* input_memory_state = reset_memory();
set_all_memory(input_memory_state);
Node* input_io_state = i_o();
Node* excluded_mask = _gvn.intcon(32768);
Node* epoch_mask = _gvn.intcon(32767);
// TLS
Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
// Load the address of java event writer jobject handle from the jfr_thread_local structure.
Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
// Load the eventwriter jobject handle.
Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
// Null check the jobject handle.
Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
// False path, jobj is null.
Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
// True path, jobj is not null.
Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
set_control(jobj_is_not_null);
// Load the threadObj for the CarrierThread.
Node* threadObj = generate_current_thread(tls_ptr);
// Load the vthread.
Node* vthread = generate_virtual_thread(tls_ptr);
// If vthread != threadObj, this is a virtual thread.
Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
IfNode* iff_vthread_not_equal_threadObj =
create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
// False branch, fallback to threadObj.
Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
set_control(vthread_equal_threadObj);
// Load the tid field from the vthread object.
Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
// Load the raw epoch value from the threadObj.
Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
// Mask off the excluded information from the epoch.
Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
// True branch, this is a virtual thread.
Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
set_control(vthread_not_equal_threadObj);
// Load the tid field from the vthread object.
Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
// Load the raw epoch value from the vthread.
Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
// Mask off the excluded information from the epoch.
Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
// Branch on excluded to conditionalize updating the epoch for the virtual thread.
Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
// False branch, vthread is excluded, no need to write epoch info.
Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
// True branch, vthread is included, update epoch info.
Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
set_control(included);
// Get epoch value.
Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
// Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address())); | | |
