truffle: graal/GraalCompiler/src/com/sun/c1x/gen/LIRGenerator.java comparison

comparison graal/GraalCompiler/src/com/sun/c1x/gen/LIRGenerator.java @ 2509:16b9a8b5ad39

Renamings Runtime=>GraalRuntime and Compiler=>GraalCompiler

author	Thomas Wuerthinger <thomas@wuerthinger.net>
date	Wed, 27 Apr 2011 11:50:44 +0200
parents	graal/Compiler/src/com/sun/c1x/gen/LIRGenerator.java@9ec15d6914ca
children	4fdef1464592

comparison

equal deleted inserted replaced

-:fea94949e0a2
+:16b9a8b5ad39
+/*
+* Copyright (c) 2009, 2011, Oracle and/or its affiliates. All rights reserved.
+* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+*
+* This code is free software; you can redistribute it and/or modify it
+* under the terms of the GNU General Public License version 2 only, as
+* published by the Free Software Foundation.
+*
+* This code is distributed in the hope that it will be useful, but WITHOUT
+* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+* FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+* version 2 for more details (a copy is included in the LICENSE file that
+* accompanied this code).
+*
+* You should have received a copy of the GNU General Public License version
+* 2 along with this work; if not, write to the Free Software Foundation,
+* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+*
+* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+* or visit www.oracle.com if you need additional information or have any
+* questions.
+*/
+package com.sun.c1x.gen;
+import static com.sun.cri.bytecode.Bytecodes.*;
+import static com.sun.cri.bytecode.Bytecodes.MemoryBarriers.*;
+import static com.sun.cri.ci.CiCallingConvention.Type.*;
+import static com.sun.cri.ci.CiValue.*;
+import java.lang.reflect.*;
+import java.util.*;
+import com.sun.c1x.*;
+import com.sun.c1x.alloc.*;
+import com.sun.c1x.alloc.OperandPool.VariableFlag;
+import com.sun.c1x.asm.*;
+import com.sun.c1x.debug.*;
+import com.sun.c1x.globalstub.*;
+import com.sun.c1x.graph.*;
+import com.sun.c1x.ir.*;
+import com.sun.c1x.ir.Value.Flag;
+import com.sun.c1x.lir.FrameMap.StackBlock;
+import com.sun.c1x.lir.*;
+import com.sun.c1x.opt.*;
+import com.sun.c1x.util.*;
+import com.sun.c1x.value.*;
+import com.sun.c1x.value.FrameState.PhiProcedure;
+import com.sun.cri.bytecode.*;
+import com.sun.cri.bytecode.Bytecodes.MemoryBarriers;
+import com.sun.cri.ci.*;
+import com.sun.cri.ci.CiAddress.Scale;
+import com.sun.cri.ri.*;
+import com.sun.cri.xir.CiXirAssembler.XirConstant;
+import com.sun.cri.xir.CiXirAssembler.XirInstruction;
+import com.sun.cri.xir.CiXirAssembler.XirOperand;
+import com.sun.cri.xir.CiXirAssembler.XirParameter;
+import com.sun.cri.xir.CiXirAssembler.XirRegister;
+import com.sun.cri.xir.CiXirAssembler.XirTemp;
+import com.sun.cri.xir.*;
+/**
+* This class traverses the HIR instructions and generates LIR instructions from them.
+*
+* @author Thomas Wuerthinger
+* @author Ben L. Titzer
+* @author Marcelo Cintra
+* @author Doug Simon
+*/
+public abstract class LIRGenerator extends ValueVisitor {
+/**
+* Helper class for inserting memory barriers as necessary to implement the Java Memory Model
+* with respect to volatile field accesses.
+*
+* @see MemoryBarriers
+*/
+class VolatileMemoryAccess {
+/**
+* Inserts any necessary memory barriers before a volatile write as required by the JMM.
+*/
+void preVolatileWrite() {
+int barriers = compilation.target.arch.requiredBarriers(JMM_PRE_VOLATILE_WRITE);
+if (compilation.target.isMP && barriers != 0) {
+lir.membar(barriers);
+}
+}
+/**
+* Inserts any necessary memory barriers after a volatile write as required by the JMM.
+*/
+void postVolatileWrite() {
+int barriers = compilation.target.arch.requiredBarriers(JMM_POST_VOLATILE_WRITE);
+if (compilation.target.isMP && barriers != 0) {
+lir.membar(barriers);
+}
+}
+/**
+* Inserts any necessary memory barriers before a volatile read as required by the JMM.
+*/
+void preVolatileRead() {
+int barriers = compilation.target.arch.requiredBarriers(JMM_PRE_VOLATILE_READ);
+if (compilation.target.isMP && barriers != 0) {
+lir.membar(barriers);
+}
+}
+/**
+* Inserts any necessary memory barriers after a volatile read as required by the JMM.
+*/
+void postVolatileRead() {
+// Ensure field's data is loaded before any subsequent loads or stores.
+int barriers = compilation.target.arch.requiredBarriers(LOAD_LOAD | LOAD_STORE);
+if (compilation.target.isMP && barriers != 0) {
+lir.membar(barriers);
+}
+}
+}
+/**
+* Forces the result of a given instruction to be available in a given register,
+* inserting move instructions if necessary.
+*
+* @param instruction an instruction that produces a {@linkplain Value#operand() result}
+* @param register the {@linkplain CiRegister} in which the result of {@code instruction} must be available
+* @return {@code register} as an operand
+*/
+protected CiValue force(Value instruction, CiRegister register) {
+return force(instruction, register.asValue(instruction.kind));
+}
+/**
+* Forces the result of a given instruction to be available in a given operand,
+* inserting move instructions if necessary.
+*
+* @param instruction an instruction that produces a {@linkplain Value#operand() result}
+* @param operand the operand in which the result of {@code instruction} must be available
+* @return {@code operand}
+*/
+protected CiValue force(Value instruction, CiValue operand) {
+CiValue result = makeOperand(instruction);
+if (result != operand) {
+assert result.kind != CiKind.Illegal;
+if (!compilation.archKindsEqual(result.kind, operand.kind)) {
+// moves between different types need an intervening spill slot
+CiValue tmp = forceToSpill(result, operand.kind, false);
+lir.move(tmp, operand);
+} else {
+lir.move(result, operand);
+}
+}
+return operand;
+}
+protected CiValue load(Value val) {
+CiValue result = makeOperand(val);
+if (!result.isVariableOrRegister()) {
+CiVariable operand = newVariable(val.kind);
+lir.move(result, operand);
+return operand;
+}
+return result;
+}
+// the range of values in a lookupswitch or tableswitch statement
+private static final class SwitchRange {
+final int lowKey;
+int highKey;
+final BlockBegin sux;
+SwitchRange(int lowKey, BlockBegin sux) {
+this.lowKey = lowKey;
+this.highKey = lowKey;
+this.sux = sux;
+}
+}
+protected final C1XCompilation compilation;
+protected final IR ir;
+protected final XirSupport xirSupport;
+protected final RiXirGenerator xir;
+protected final boolean isTwoOperand;
+private BlockBegin currentBlock;
+public final OperandPool operands;
+private Value currentInstruction;
+private Value lastInstructionPrinted; // Debugging only
+private List<CiConstant> constants;
+private List<CiVariable> variablesForConstants;
+protected LIRList lir;
+final VolatileMemoryAccess vma;
+private ArrayList<DeoptimizationStub> deoptimizationStubs;
+public LIRGenerator(C1XCompilation compilation) {
+this.compilation = compilation;
+this.ir = compilation.hir();
+this.xir = compilation.compiler.xir;
+this.xirSupport = new XirSupport();
+this.isTwoOperand = compilation.target.arch.twoOperandMode();
+this.vma = new VolatileMemoryAccess();
+constants = new ArrayList<CiConstant>();
+variablesForConstants = new ArrayList<CiVariable>();
+this.operands = new OperandPool(compilation.target);
+// mark the liveness of all instructions if it hasn't already been done by the optimizer
+LivenessMarker livenessMarker = new LivenessMarker(ir);
+C1XMetrics.LiveHIRInstructions += livenessMarker.liveCount();
+}
+public ArrayList<DeoptimizationStub> deoptimizationStubs() {
+return deoptimizationStubs;
+}
+public static class DeoptimizationStub {
+public final Label label = new Label();
+public final LIRDebugInfo info;
+public DeoptimizationStub(FrameState state) {
+info = new LIRDebugInfo(state, null);
+}
+}
+public final void emitGuard(Guard x) {
+FrameState state = x.stateBefore();
+assert state != null : "deoptimize instruction always needs a state";
+if (deoptimizationStubs == null) {
+deoptimizationStubs = new ArrayList<DeoptimizationStub>();
+}
+// (tw) TODO: Try to reuse an existing stub if possible.
+// It is only allowed if there are no LIR instructions in between that can modify registers.
+DeoptimizationStub stub = new DeoptimizationStub(state);
+deoptimizationStubs.add(stub);
+lir.branch(x.condition.negate(), stub.label, stub.info);
+}
+public void doBlock(BlockBegin block) {
+blockDoProlog(block);
+this.currentBlock = block;
+for (Instruction instr = block; instr != null; instr = instr.next()) {
+if (instr.isLive()) {
+walkState(instr, instr.stateBefore());
+doRoot(instr);
+}
+}
+this.currentBlock = null;
+blockDoEpilog(block);
+}
+@Override
+public void visitArrayLength(ArrayLength x) {
+emitArrayLength(x);
+}
+public CiValue emitArrayLength(ArrayLength x) {
+XirArgument array = toXirArgument(x.array());
+XirSnippet snippet = xir.genArrayLength(site(x), array);
+emitXir(snippet, x, x.needsNullCheck() ? stateFor(x) : null, null, true);
+return x.operand();
+}
+@Override
+public void visitBase(Base x) {
+// emit phi-instruction move after safepoint since this simplifies
+// describing the state at the safepoint.
+moveToPhi(x.stateAfter());
+// all blocks with a successor must end with an unconditional jump
+// to the successor even if they are consecutive
+lir.jump(x.defaultSuccessor());
+}
+private void setOperandsForLocals(FrameState state) {
+CiCallingConvention args = compilation.frameMap().incomingArguments();
+int javaIndex = 0;
+for (int i = 0; i < args.locations.length; i++) {
+CiValue src = args.locations[i];
+assert src.isLegal() : "check";
+CiVariable dest = newVariable(src.kind.stackKind());
+lir.move(src, dest, src.kind);
+// Assign new location to Local instruction for this local
+Value instr = state.localAt(javaIndex);
+Local local = ((Local) instr);
+CiKind kind = src.kind.stackKind();
+assert kind == local.kind.stackKind() : "local type check failed";
+if (local.isLive()) {
+setResult(local, dest);
+}
+javaIndex += kind.jvmSlots;
+}
+}
+@Override
+public void visitResolveClass(ResolveClass i) {
+LIRDebugInfo info = stateFor(i);
+XirSnippet snippet = xir.genResolveClass(site(i), i.type, i.portion);
+emitXir(snippet, i, info, null, true);
+}
+@Override
+public void visitCheckCast(CheckCast x) {
+XirArgument obj = toXirArgument(x.object());
+XirSnippet snippet = xir.genCheckCast(site(x), obj, toXirArgument(x.targetClassInstruction), x.targetClass());
+emitXir(snippet, x, stateFor(x), null, true);
+}
+@Override
+public void visitInstanceOf(InstanceOf x) {
+XirArgument obj = toXirArgument(x.object());
+XirSnippet snippet = xir.genInstanceOf(site(x), obj, toXirArgument(x.targetClassInstruction), x.targetClass());
+emitXir(snippet, x, maybeStateFor(x), null, true);
+}
+@Override
+public void visitMonitorEnter(MonitorEnter x) {
+XirArgument obj = toXirArgument(x.object());
+XirArgument lockAddress = toXirArgument(x.lockAddress());
+XirSnippet snippet = xir.genMonitorEnter(site(x), obj, lockAddress);
+emitXir(snippet, x, maybeStateFor(x), stateFor(x, x.stateAfter()), null, true, null);
+}
+@Override
+public void visitMonitorExit(MonitorExit x) {
+XirArgument obj = toXirArgument(x.object());
+XirArgument lockAddress = toXirArgument(x.lockAddress());
+XirSnippet snippet = xir.genMonitorExit(site(x), obj, lockAddress);
+emitXir(snippet, x, maybeStateFor(x), null, true);
+}
+@Override
+public void visitStoreIndexed(StoreIndexed x) {
+XirArgument array = toXirArgument(x.array());
+XirArgument length = x.length() == null ? null : toXirArgument(x.length());
+XirArgument index = toXirArgument(x.index());
+XirArgument value = toXirArgument(x.value());
+XirSnippet snippet = xir.genArrayStore(site(x), array, index, length, value, x.elementKind(), null);
+emitXir(snippet, x, maybeStateFor(x), null, true);
+}
+@Override
+public void visitNewInstance(NewInstance x) {
+XirSnippet snippet = xir.genNewInstance(site(x), x.instanceClass());
+emitXir(snippet, x, stateFor(x), null, true);
+}
+@Override
+public void visitNewTypeArray(NewTypeArray x) {
+XirArgument length = toXirArgument(x.length());
+XirSnippet snippet = xir.genNewArray(site(x), length, x.elementKind(), null, null);
+emitXir(snippet, x, stateFor(x), null, true);
+}
+@Override
+public void visitNewObjectArray(NewObjectArray x) {
+XirArgument length = toXirArgument(x.length());
+XirSnippet snippet = xir.genNewArray(site(x), length, CiKind.Object, x.elementClass(), x.exactType());
+emitXir(snippet, x, stateFor(x), null, true);
+}
+@Override
+public void visitNewObjectArrayClone(NewObjectArrayClone x) {
+XirArgument length = toXirArgument(x.length());
+XirArgument referenceArray = toXirArgument(x.referenceArray());
+XirSnippet snippet = xir.genNewObjectArrayClone(site(x), length, referenceArray);
+emitXir(snippet, x, stateFor(x), null, true);
+}
+@Override
+public void visitNewMultiArray(NewMultiArray x) {
+XirArgument[] dims = new XirArgument[x.dimensions().length];
+for (int i = 0; i < dims.length; i++) {
+dims[i] = toXirArgument(x.dimensions()[i]);
+}
+XirSnippet snippet = xir.genNewMultiArray(site(x), dims, x.elementKind);
+emitXir(snippet, x, stateFor(x), null, true);
+}
+@Override
+public void visitConstant(Constant x) {
+if (canInlineAsConstant(x)) {
+//setResult(x, loadConstant(x));
+} else {
+CiValue res = x.operand();
+if (!(res.isLegal())) {
+res = x.asConstant();
+}
+if (res.isConstant()) {
+if (isUsedForValue(x)) {
+CiVariable reg = createResultVariable(x);
+lir.move(res, reg);
+} else {
+assert x.checkFlag(Value.Flag.LiveDeopt);
+x.setOperand(res);
+}
+} else {
+setResult(x, (CiVariable) res);
+}
+}
+}
+@Override
+public void visitExceptionObject(ExceptionObject x) {
+assert currentBlock.isExceptionEntry() : "ExceptionObject only allowed in exception handler block";
+assert currentBlock.next() == x : "ExceptionObject must be first instruction of block";
+// no moves are created for phi functions at the begin of exception
+// handlers, so assign operands manually here
+currentBlock.stateBefore().forEachLivePhi(currentBlock, new PhiProcedure() {
+public boolean doPhi(Phi phi) {
+operandForPhi(phi);
+return true;
+}
+});
+XirSnippet snippet = xir.genExceptionObject(site(x));
+emitXir(snippet, x, maybeStateFor(x), null, true);
+}
+@Override
+public void visitGoto(Goto x) {
+setNoResult(x);
+if (currentBlock.next() instanceof OsrEntry) {
+// need to free up storage used for OSR entry point
+CiValue osrBuffer = currentBlock.next().operand();
+callRuntime(CiRuntimeCall.OSRMigrationEnd, null, osrBuffer);
+emitXir(xir.genSafepoint(site(x)), x, stateFor(x, x.stateAfter()), null, false);
+}
+// emit phi-instruction moves after safepoint since this simplifies
+// describing the state at the safepoint.
+moveToPhi(x.stateAfter());
+lir.jump(x.defaultSuccessor());
+}
+@Override
+public void visitIfOp(IfOp i) {
+Value x = i.x();
+Value y = i.y();
+CiKind xtype = x.kind;
+CiKind ttype = i.trueValue().kind;
+assert xtype.isInt() || xtype.isObject() : "cannot handle others";
+assert ttype.isInt() || ttype.isObject() || ttype.isLong() || ttype.isWord() : "cannot handle others";
+assert ttype.equals(i.falseValue().kind) : "cannot handle others";
+CiValue left = load(x);
+CiValue right = null;
+if (!canInlineAsConstant(y)) {
+right = load(y);
+} else {
+right = makeOperand(y);
+}
+CiValue tVal = makeOperand(i.trueValue());
+CiValue fVal = makeOperand(i.falseValue());
+CiValue reg = createResultVariable(i);
+lir.cmp(i.condition(), left, right);
+lir.cmove(i.condition(), tVal, fVal, reg);
+}
+@Override
+public void visitIntrinsic(Intrinsic x) {
+Value[] vals = x.arguments();
+XirSnippet snippet;
+switch (x.intrinsic()) {
+case java_lang_Float$intBitsToFloat:
+case java_lang_Double$doubleToRawLongBits:
+case java_lang_Double$longBitsToDouble:
+case java_lang_Float$floatToRawIntBits: {
+visitFPIntrinsics(x);
+return;
+}
+case java_lang_System$currentTimeMillis: {
+assert x.numberOfArguments() == 0 : "wrong type";
+CiValue reg = callRuntimeWithResult(CiRuntimeCall.JavaTimeMillis, null, (CiValue[]) null);
+CiValue result = createResultVariable(x);
+lir.move(reg, result);
+return;
+}
+case java_lang_System$nanoTime: {
+assert x.numberOfArguments() == 0 : "wrong type";
+CiValue reg = callRuntimeWithResult(CiRuntimeCall.JavaTimeNanos, null, (CiValue[]) null);
+CiValue result = createResultVariable(x);
+lir.move(reg, result);
+return;
+}
+case java_lang_Object$init:
+visitRegisterFinalizer(x);
+return;
+case java_lang_Math$log:   // fall through
+case java_lang_Math$log10: // fall through
+case java_lang_Math$abs:   // fall through
+case java_lang_Math$sqrt:  // fall through
+case java_lang_Math$tan:   // fall through
+case java_lang_Math$sin:   // fall through
+case java_lang_Math$cos:
+genMathIntrinsic(x);
+return;
+case sun_misc_Unsafe$compareAndSwapObject:
+genCompareAndSwap(x, CiKind.Object);
+return;
+case sun_misc_Unsafe$compareAndSwapInt:
+genCompareAndSwap(x, CiKind.Int);
+return;
+case sun_misc_Unsafe$compareAndSwapLong:
+genCompareAndSwap(x, CiKind.Long);
+return;
+case java_lang_Thread$currentThread:
+snippet = xir.genCurrentThread(site(x));
+if (snippet != null) {
+emitXir(snippet, x, null, null, true);
+return;
+}
+break;
+case java_lang_Object$getClass:
+snippet = xir.genGetClass(site(x), toXirArgument(vals[0]));
+if (snippet != null) {
+emitXir(snippet, x, stateFor(x), null, true);
+return;
+}
+break;
+}
+XirArgument[] args = new XirArgument[vals.length];
+for (int i = 0; i < vals.length; i++) {
+args[i] = toXirArgument(vals[i]);
+}
+snippet = xir.genIntrinsic(site(x), args, x.target());
+if (snippet != null) {
+emitXir(snippet, x, x.stateBefore() == null ? null : stateFor(x), null, true);
+return;
+}
+x.setOperand(emitInvokeKnown(x.target(), x.stateBefore(), vals));
+}
+@Override
+public void visitInvoke(Invoke x) {
+RiMethod target = x.target();
+LIRDebugInfo info = stateFor(x, x.stateBefore());
+XirSnippet snippet = null;
+int opcode = x.opcode();
+XirArgument receiver;
+switch (opcode) {
+case INVOKESTATIC:
+snippet = xir.genInvokeStatic(site(x), target);
+break;
+case INVOKESPECIAL:
+receiver = toXirArgument(x.receiver());
+snippet = xir.genInvokeSpecial(site(x), receiver, target);
+break;
+case INVOKEVIRTUAL:
+receiver = toXirArgument(x.receiver());
+snippet = xir.genInvokeVirtual(site(x), receiver, target);
+break;
+case INVOKEINTERFACE:
+receiver = toXirArgument(x.receiver());
+snippet = xir.genInvokeInterface(site(x), receiver, target);
+break;
+}
+CiValue destinationAddress = null;
+// emitting the template earlier can ease pressure on register allocation, but the argument loading can destroy an
+// implicit calling convention between the XirSnippet and the call.
+if (!C1XOptions.InvokeSnippetAfterArguments) {
+destinationAddress = emitXir(snippet, x, info.copy(), x.target(), false);
+}
+CiValue resultOperand = resultOperandFor(x.kind);
+CiCallingConvention cc = compilation.frameMap().getCallingConvention(x.signature(), JavaCall);
+List<CiValue> pointerSlots = new ArrayList<CiValue>(2);
+List<CiValue> argList = visitInvokeArguments(cc, x.arguments(), pointerSlots);
+if (C1XOptions.InvokeSnippetAfterArguments) {
+destinationAddress = emitXir(snippet, x, info.copy(), null, x.target(), false, pointerSlots);
+}
+// emit direct or indirect call to the destination address
+if (destinationAddress instanceof CiConstant) {
+// Direct call
+assert ((CiConstant) destinationAddress).isDefaultValue() : "destination address should be zero";
+lir.callDirect(target, resultOperand, argList, info, snippet.marks, pointerSlots);
+} else {
+// Indirect call
+argList.add(destinationAddress);
+lir.callIndirect(target, resultOperand, argList, info, snippet.marks, pointerSlots);
+}
+if (resultOperand.isLegal()) {
+CiValue result = createResultVariable(x);
+lir.move(resultOperand, result);
+}
+}
+@Override
+public void visitNativeCall(NativeCall x) {
+LIRDebugInfo info = stateFor(x, x.stateBefore());
+CiValue resultOperand = resultOperandFor(x.kind);
+CiValue callAddress = load(x.address());
+CiKind[] signature = Util.signatureToKinds(x.signature, null);
+CiCallingConvention cc = compilation.frameMap().getCallingConvention(signature, NativeCall);
+List<CiValue> argList = visitInvokeArguments(cc, x.arguments, null);
+argList.add(callAddress);
+lir.callNative(x.nativeMethod.jniSymbol(), resultOperand, argList, info, null);
+if (resultOperand.isLegal()) {
+CiValue result = createResultVariable(x);
+lir.move(resultOperand, result);
+}
+}
+@Override
+public void visitTemplateCall(TemplateCall x) {
+CiValue resultOperand = resultOperandFor(x.kind);
+List<CiValue> argList;
+if (x.receiver() != null) {
+CiCallingConvention cc = compilation.frameMap().getCallingConvention(new CiKind[] {CiKind.Object}, JavaCall);
+argList = visitInvokeArguments(cc, new Value[] {x.receiver()}, null);
+} else {
+argList = new ArrayList<CiValue>();
+}
+if (x.address() != null) {
+CiValue callAddress = load(x.address());
+argList.add(callAddress);
+}
+lir.templateCall(resultOperand, argList);
+if (resultOperand.isLegal()) {
+CiValue result = createResultVariable(x);
+lir.move(resultOperand, result);
+}
+}
+@Override
+public void visitLoadRegister(LoadRegister x) {
+x.setOperand(x.register.asValue(x.kind));
+}
+@Override
+public void visitPause(Pause i) {
+lir.pause();
+}
+@Override
+public void visitBreakpointTrap(BreakpointTrap i) {
+lir.breakpoint();
+}
+protected CiAddress getAddressForPointerOp(PointerOp x, CiKind kind, CiValue pointer) {
+CiAddress addr;
+Value offset = x.offset();
+Value index = x.index();
+if (x.displacement() == null) {
+// address is [pointer + offset]
+if (offset.isConstant() && offset.kind.isInt()) {
+int displacement = x.offset().asConstant().asInt();
+addr = new CiAddress(kind, pointer, displacement);
+} else {
+addr = new CiAddress(kind, pointer, load(offset));
+}
+} else {
+// address is [pointer + disp + (index * scale)]
+assert (x.opcode & 0xff) == PGET || (x.opcode & 0xff) == PSET;
+if (!x.displacement().isConstant()) {
+CiVariable tmp = newVariable(CiKind.Word);
+arithmeticOpLong(Bytecodes.LADD, tmp, pointer, load(x.displacement()), null);
+int kindSize = compilation.target.sizeInBytes(kind);
+Scale scale = Scale.fromInt(kindSize);
+if (index.isConstant()) {
+addr = new CiAddress(kind, tmp, index.asConstant().asInt() * kindSize);
+} else {
+addr = new CiAddress(kind, tmp, load(index), scale, 0);
+}
+} else {
+int displacement = x.displacement().asConstant().asInt();
+int kindSize = compilation.target.sizeInBytes(kind);
+Scale scale = Scale.fromInt(kindSize);
+if (index.isConstant()) {
+displacement += index.asConstant().asInt() * kindSize;
+addr = new CiAddress(kind, pointer, displacement);
+} else {
+addr = new CiAddress(kind, pointer, load(index), scale, displacement);
+}
+}
+}
+return addr;
+}
+@Override
+public void visitAllocateStackHandle(StackHandle x) {
+CiValue value = load(x.value());
+CiValue src = forceToSpill(value, x.value().kind, true);
+CiValue dst = createResultVariable(x);
+CiConstant constant = x.value().isConstant() ? x.value().asConstant() : null;
+if (constant == null) {
+CiConstant zero = CiConstant.defaultValue(x.value().kind);
+lir.cmp(Condition.EQ, src, zero);
+}
+lir.lea(src, dst);
+if (constant != null) {
+if (constant.isDefaultValue()) {
+lir.move(value, dst);
+}
+} else {
+lir.cmove(Condition.EQ, CiConstant.ZERO, dst, dst);
+}
+}
+@Override
+public void visitLoadPointer(LoadPointer x) {
+LIRDebugInfo info = maybeStateFor(x);
+CiValue pointer = load(x.pointer());
+CiValue dst = createResultVariable(x);
+CiAddress src = getAddressForPointerOp(x, x.dataKind, pointer);
+lir.load(src, dst, info);
+}
+@Override
+public void visitStorePointer(StorePointer x) {
+LIRDebugInfo info = maybeStateFor(x);
+LIRItem value = new LIRItem(x.value(), this);
+CiValue pointer = load(x.pointer());
+value.loadItem(x.dataKind);
+CiAddress dst = getAddressForPointerOp(x, x.dataKind, pointer);
+lir.store(value.result(), dst, info);
+}
+@Override
+public void visitInfopoint(Infopoint x) {
+LIRDebugInfo info = stateFor(x);
+if (x.opcode == SAFEPOINT) {
+emitXir(xir.genSafepoint(site(x)), x, info, null, false);
+return;
+}
+assert x.opcode == HERE || x.opcode == INFO;
+CiValue result = x.kind.isVoid() ? CiValue.IllegalValue : createResultVariable(x);
+LIROpcode opcode = x.opcode == HERE ? LIROpcode.Here : LIROpcode.Info;
+lir.infopoint(opcode, result, info);
+}
+@Override
+public void visitStackAllocate(StackAllocate x) {
+CiValue result = createResultVariable(x);
+assert x.size().isConstant() : "ALLOCA bytecode 'size' operand is not a constant: " + x.size();
+StackBlock stackBlock = compilation.frameMap().reserveStackBlock(x.size().asConstant().asInt());
+lir.alloca(stackBlock, result);
+}
+@Override
+public void visitMonitorAddress(MonitorAddress x) {
+CiValue result = createResultVariable(x);
+lir.monitorAddress(x.monitor(), result);
+}
+@Override
+public void visitMemoryBarrier(MemoryBarrier x) {
+if (x.barriers != 0) {
+lir.membar(x.barriers);
+}
+}
+@Override
+public void visitUnsafeCast(UnsafeCast i) {
+assert !i.redundant : "redundant UnsafeCasts must be eliminated by the front end";
+CiValue src = load(i.value());
+CiValue dst = createResultVariable(i);
+lir.move(src, dst);
+}
+/**
+* For note on volatile fields, see {@link #visitStoreField(StoreField)}.
+*/
+@Override
+public void visitLoadField(LoadField x) {
+RiField field = x.field();
+boolean needsPatching = x.needsPatching();
+LIRDebugInfo info = null;
+if (needsPatching || x.needsNullCheck()) {
+info = stateFor(x, x.stateBefore());
+assert info != null;
+}
+XirArgument receiver = toXirArgument(x.object());
+XirSnippet snippet = x.isStatic() ? xir.genGetStatic(site(x), receiver, field) : xir.genGetField(site(x), receiver, field);
+emitXir(snippet, x, info, null, true);
+if (x.isVolatile()) {
+vma.postVolatileRead();
+}
+}
+@Override
+public void visitLoadIndexed(LoadIndexed x) {
+XirArgument array = toXirArgument(x.array());
+XirArgument index = toXirArgument(x.index());
+XirArgument length = toXirArgument(x.length());
+XirSnippet snippet = xir.genArrayLoad(site(x), array, index, length, x.elementKind(), null);
+emitXir(snippet, x, maybeStateFor(x), null, true);
+}
+protected GlobalStub stubFor(CiRuntimeCall runtimeCall) {
+GlobalStub stub = compilation.compiler.lookupGlobalStub(runtimeCall);
+compilation.frameMap().usesGlobalStub(stub);
+return stub;
+}
+protected GlobalStub stubFor(GlobalStub.Id globalStub) {
+GlobalStub stub = compilation.compiler.lookupGlobalStub(globalStub);
+compilation.frameMap().usesGlobalStub(stub);
+return stub;
+}
+protected GlobalStub stubFor(XirTemplate template) {
+GlobalStub stub = compilation.compiler.lookupGlobalStub(template);
+compilation.frameMap().usesGlobalStub(stub);
+return stub;
+}
+@Override
+public void visitLocal(Local x) {
+if (x.operand().isIllegal()) {
+createResultVariable(x);
+}
+}
+@Override
+public void visitLookupSwitch(LookupSwitch x) {
+CiValue tag = load(x.value());
+setNoResult(x);
+if (x.isSafepoint()) {
+emitXir(xir.genSafepoint(site(x)), x, stateFor(x, x.stateAfter()), null, false);
+}
+// move values into phi locations
+moveToPhi(x.stateAfter());
+if (x.numberOfCases() == 0 || x.numberOfCases() < C1XOptions.SequentialSwitchLimit) {
+int len = x.numberOfCases();
+for (int i = 0; i < len; i++) {
+lir.cmp(Condition.EQ, tag, x.keyAt(i));
+lir.branch(Condition.EQ, CiKind.Int, x.suxAt(i));
+}
+lir.jump(x.defaultSuccessor());
+} else {
+visitSwitchRanges(createLookupRanges(x), tag, x.defaultSuccessor());
+}
+}
+@Override
+public void visitNullCheck(NullCheck x) {
+// TODO: this is suboptimal because it may result in an unnecessary move
+CiValue value = load(x.object());
+if (x.canTrap()) {
+LIRDebugInfo info = stateFor(x);
+lir.nullCheck(value, info);
+}
+x.setOperand(value);
+}
+@Override
+public void visitOsrEntry(OsrEntry x) {
+// construct our frame and model the production of incoming pointer
+// to the OSR buffer.
+lir.osrEntry(osrBufferPointer());
+CiValue result = createResultVariable(x);
+lir.move(osrBufferPointer(), result);
+}
+@Override
+public void visitPhi(Phi i) {
+Util.shouldNotReachHere();
+}
+@Override
+public void visitReturn(Return x) {
+if (x.kind.isVoid()) {
+XirSnippet epilogue = xir.genEpilogue(site(x), compilation.method);
+if (epilogue != null) {
+emitXir(epilogue, x, stateFor(x, x.stateAfter()), compilation.method, false);
+lir.returnOp(IllegalValue);
+}
+} else {
+CiValue operand = resultOperandFor(x.kind);
+CiValue result = force(x.result(), operand);
+XirSnippet epilogue = xir.genEpilogue(site(x), compilation.method);
+if (epilogue != null) {
+emitXir(epilogue, x, stateFor(x, x.stateAfter()), compilation.method, false);
+lir.returnOp(result);
+}
+}
+setNoResult(x);
+}
+protected XirArgument toXirArgument(CiValue v) {
+if (v == null) {
+return null;
+}
+return XirArgument.forInternalObject(v);
+}
+protected XirArgument toXirArgument(Value i) {
+if (i == null) {
+return null;
+}
+return XirArgument.forInternalObject(new LIRItem(i, this));
+}
+private CiValue allocateOperand(XirSnippet snippet, XirOperand op) {
+if (op instanceof XirParameter)  {
+XirParameter param = (XirParameter) op;
+return allocateOperand(snippet.arguments[param.parameterIndex], op, param.canBeConstant);
+} else if (op instanceof XirRegister) {
+XirRegister reg = (XirRegister) op;
+return reg.register;
+} else if (op instanceof XirTemp) {
+return newVariable(op.kind);
+} else {
+Util.shouldNotReachHere();
+return null;
+}
+}
+private CiValue allocateOperand(XirArgument arg, XirOperand var, boolean canBeConstant) {
+if (arg.constant != null) {
+return arg.constant;
+} else {
+assert arg.object != null;
+if (arg.object instanceof CiValue) {
+return (CiValue) arg.object;
+}
+assert arg.object instanceof LIRItem;
+LIRItem item = (LIRItem) arg.object;
+if (canBeConstant) {
+return item.instruction.operand();
+} else {
+item.loadItem(var.kind);
+return item.result();
+}
+}
+}
+protected CiValue emitXir(XirSnippet snippet, Instruction x, LIRDebugInfo info, RiMethod method, boolean setInstructionResult) {
+return emitXir(snippet, x, info, null, method, setInstructionResult, null);
+}
+protected CiValue emitXir(XirSnippet snippet, Instruction instruction, LIRDebugInfo info, LIRDebugInfo infoAfter, RiMethod method, boolean setInstructionResult, List<CiValue> pointerSlots) {
+if (C1XOptions.PrintXirTemplates) {
+TTY.println("Emit XIR template " + snippet.template.name);
+}
+final CiValue[] operands = new CiValue[snippet.template.variableCount];
+compilation.frameMap().reserveOutgoing(snippet.template.outgoingStackSize);
+XirOperand resultOperand = snippet.template.resultOperand;
+if (snippet.template.allocateResultOperand) {
+CiValue outputOperand = IllegalValue;
+// This snippet has a result that must be separately allocated
+// Otherwise it is assumed that the result is part of the inputs
+if (resultOperand.kind != CiKind.Void && resultOperand.kind != CiKind.Illegal) {
+if (setInstructionResult) {
+outputOperand = newVariable(instruction.kind);
+} else {
+outputOperand = newVariable(resultOperand.kind);
+}
+assert operands[resultOperand.index] == null;
+}
+operands[resultOperand.index] = outputOperand;
+if (C1XOptions.PrintXirTemplates) {
+TTY.println("Output operand: " + outputOperand);
+}
+}
+for (XirTemp t : snippet.template.temps) {
+if (t instanceof XirRegister) {
+XirRegister reg = (XirRegister) t;
+if (!t.reserve) {
+operands[t.index] = reg.register;
+}
+}
+}
+for (XirTemplate calleeTemplate : snippet.template.calleeTemplates) {
+// TODO Save these for use in X86LIRAssembler
+stubFor(calleeTemplate);
+}
+for (XirConstant c : snippet.template.constants) {
+assert operands[c.index] == null;
+operands[c.index] = c.value;
+}
+XirOperand[] inputOperands = snippet.template.inputOperands;
+XirOperand[] inputTempOperands = snippet.template.inputTempOperands;
+XirOperand[] tempOperands = snippet.template.tempOperands;
+CiValue[] operandArray = new CiValue[inputOperands.length + inputTempOperands.length + tempOperands.length];
+int[] operandIndicesArray = new int[inputOperands.length + inputTempOperands.length + tempOperands.length];
+for (int i = 0; i < inputOperands.length; i++) {
+XirOperand x = inputOperands[i];
+CiValue op = allocateOperand(snippet, x);
+operands[x.index] = op;
+operandArray[i] = op;
+operandIndicesArray[i] = x.index;
+if (C1XOptions.PrintXirTemplates) {
+TTY.println("Input operand: " + x);
+}
+}
+for (int i = 0; i < inputTempOperands.length; i++) {
+XirOperand x = inputTempOperands[i];
+CiValue op = allocateOperand(snippet, x);
+CiValue newOp = newVariable(op.kind);
+lir.move(op, newOp);
+operands[x.index] = newOp;
+operandArray[i + inputOperands.length] = newOp;
+operandIndicesArray[i + inputOperands.length] = x.index;
+if (C1XOptions.PrintXirTemplates) {
+TTY.println("InputTemp operand: " + x);
+}
+}
+for (int i = 0; i < tempOperands.length; i++) {
+XirOperand x = tempOperands[i];
+CiValue op = allocateOperand(snippet, x);
+operands[x.index] = op;
+operandArray[i + inputOperands.length + inputTempOperands.length] = op;
+operandIndicesArray[i + inputOperands.length + inputTempOperands.length] = x.index;
+if (C1XOptions.PrintXirTemplates) {
+TTY.println("Temp operand: " + x);
+}
+}
+for (CiValue operand : operands) {
+assert operand != null;
+}
+CiValue allocatedResultOperand = operands[resultOperand.index];
+if (!allocatedResultOperand.isVariableOrRegister()) {
+allocatedResultOperand = IllegalValue;
+}
+if (setInstructionResult && allocatedResultOperand.isLegal()) {
+if (instruction.operand().isIllegal()) {
+setResult(instruction, (CiVariable) allocatedResultOperand);
+} else {
+assert instruction.operand() == allocatedResultOperand;
+}
+}
+XirInstruction[] slowPath = snippet.template.slowPath;
+if (!operands[resultOperand.index].isConstant() || snippet.template.fastPath.length != 0 || (slowPath != null && slowPath.length > 0)) {
+// XIR instruction is only needed when the operand is not a constant!
+lir.xir(snippet, operands, allocatedResultOperand, inputTempOperands.length, tempOperands.length,
+operandArray, operandIndicesArray,
+(operands[resultOperand.index] == IllegalValue) ? -1 : resultOperand.index,
+info, infoAfter, method, pointerSlots);
+}
+return operands[resultOperand.index];
+}
+@Override
+public void visitIncrementRegister(IncrementRegister x) {
+CiValue reg = x.register.asValue(CiKind.Word);
+if (x.delta().isConstant()) {
+int delta = x.delta().asConstant().asInt();
+if (delta < 0) {
+lir.sub(reg, CiConstant.forInt(-delta), reg);
+} else {
+lir.add(reg, CiConstant.forInt(delta), reg);
+}
+} else {
+lir.add(reg, makeOperand(x.delta()), reg);
+}
+}
+@Override
+public void visitStoreRegister(StoreRegister x) {
+CiValue reg = x.register.asValue(x.kind);
+lir.move(makeOperand(x.value()), reg);
+}
+@Override
+public void visitStoreField(StoreField x) {
+RiField field = x.field();
+boolean needsPatching = x.needsPatching();
+LIRDebugInfo info = null;
+if (needsPatching || x.needsNullCheck()) {
+info = stateFor(x, x.stateBefore());
+}
+if (x.isVolatile()) {
+vma.preVolatileWrite();
+}
+XirArgument receiver = toXirArgument(x.object());
+XirArgument value = toXirArgument(x.value());
+XirSnippet snippet = x.isStatic() ? xir.genPutStatic(site(x), receiver, field, value) : xir.genPutField(site(x), receiver, field, value);
+emitXir(snippet, x, info, null, true);
+if (x.isVolatile()) {
+vma.postVolatileWrite();
+}
+}
+@Override
+public void visitTableSwitch(TableSwitch x) {
+LIRItem value = new LIRItem(x.value(), this);
+// Making a copy of the switch value is necessary when generating a jump table
+value.setDestroysRegister();
+value.loadItem();
+CiValue tag = value.result();
+setNoResult(x);
+if (x.isSafepoint()) {
+emitXir(xir.genSafepoint(site(x)), x, stateFor(x, x.stateAfter()), null, false);
+}
+// move values into phi locations
+moveToPhi(x.stateAfter());
+// TODO: tune the defaults for the controls used to determine what kind of translation to use
+if (x.numberOfCases() == 0 || x.numberOfCases() <= C1XOptions.SequentialSwitchLimit) {
+int loKey = x.lowKey();
+int len = x.numberOfCases();
+for (int i = 0; i < len; i++) {
+lir.cmp(Condition.EQ, tag, i + loKey);
+lir.branch(Condition.EQ, CiKind.Int, x.suxAt(i));
+}
+lir.jump(x.defaultSuccessor());
+} else {
+SwitchRange[] switchRanges = createLookupRanges(x);
+int rangeDensity = x.numberOfCases() / switchRanges.length;
+if (rangeDensity >= C1XOptions.RangeTestsSwitchDensity) {
+visitSwitchRanges(switchRanges, tag, x.defaultSuccessor());
+} else {
+List<BlockBegin> nonDefaultSuccessors = x.successors().subList(0, x.numberOfCases());
+BlockBegin[] targets = nonDefaultSuccessors.toArray(new BlockBegin[nonDefaultSuccessors.size()]);
+lir.tableswitch(tag, x.lowKey(), x.defaultSuccessor(), targets);
+}
+}
+}
+@Override
+public void visitThrow(Throw x) {
+setNoResult(x);
+CiValue exceptionOpr = load(x.exception());
+LIRDebugInfo info = stateFor(x, x.stateAfter());
+// check if the instruction has an xhandler in any of the nested scopes
+boolean unwind = false;
+if (x.exceptionHandlers().size() == 0) {
+// this throw is not inside an xhandler
+unwind = true;
+} else {
+// get some idea of the throw type
+boolean typeIsExact = true;
+RiType throwType = x.exception().exactType();
+if (throwType == null) {
+typeIsExact = false;
+throwType = x.exception().declaredType();
+}
+if (throwType != null && throwType.isResolved() && throwType.isInstanceClass()) {
+unwind = !ExceptionHandler.couldCatch(x.exceptionHandlers(), throwType, typeIsExact);
+}
+}
+assert !currentBlock.checkBlockFlag(BlockBegin.BlockFlag.DefaultExceptionHandler) || unwind : "should be no more handlers to dispatch to";
+// move exception oop into fixed register
+CiCallingConvention callingConvention = compilation.frameMap().getCallingConvention(new CiKind[]{CiKind.Object}, RuntimeCall);
+CiValue argumentOperand = callingConvention.locations[0];
+lir.move(exceptionOpr, argumentOperand);
+if (unwind) {
+lir.unwindException(exceptionPcOpr(), exceptionOpr, info);
+} else {
+lir.throwException(exceptionPcOpr(), argumentOperand, info);
+}
+}
+@Override
+public void visitUnsafeGetObject(UnsafeGetObject x) {
+CiKind kind = x.unsafeOpKind;
+CiValue off = load(x.offset());
+CiValue src = load(x.object());
+CiValue reg = createResultVariable(x);
+if (x.isVolatile()) {
+vma.preVolatileRead();
+}
+genGetObjectUnsafe(reg, src, off, kind, x.isVolatile());
+if (x.isVolatile()) {
+vma.postVolatileRead();
+}
+}
+@Override
+public void visitUnsafeGetRaw(UnsafeGetRaw x) {
+LIRItem idx = new LIRItem(this);
+CiValue base = load(x.base());
+if (x.hasIndex()) {
+idx.setInstruction(x.index());
+idx.loadNonconstant();
+}
+CiValue reg = createResultVariable(x);
+int log2scale = 0;
+if (x.hasIndex()) {
+assert x.index().kind.isInt() : "should not find non-int index";
+log2scale = x.log2Scale();
+}
+assert !x.hasIndex() || idx.instruction == x.index() : "should match";
+CiKind dstKind = x.unsafeOpKind;
+CiValue indexOp = idx.result();
+CiAddress addr = null;
+if (indexOp.isConstant()) {
+assert log2scale == 0 : "must not have a scale";
+CiConstant constantIndexOp = (CiConstant) indexOp;
+addr = new CiAddress(dstKind, base, constantIndexOp.asInt());
+} else {
+if (compilation.target.arch.isX86()) {
+addr = new CiAddress(dstKind, base, indexOp, CiAddress.Scale.fromInt(2 ^ log2scale), 0);
+} else if (compilation.target.arch.isSPARC()) {
+if (indexOp.isIllegal() || log2scale == 0) {
+addr = new CiAddress(dstKind, base, indexOp);
+} else {
+CiValue tmp = newVariable(CiKind.Int);
+lir.shiftLeft(indexOp, log2scale, tmp);
+addr = new CiAddress(dstKind, base, tmp);
+}
+} else {
+Util.shouldNotReachHere();
+}
+}
+if (x.mayBeUnaligned() && (dstKind == CiKind.Long || dstKind == CiKind.Double)) {
+lir.unalignedMove(addr, reg);
+} else {
+lir.move(addr, reg);
+}
+}
+@Override
+public void visitUnsafePrefetchRead(UnsafePrefetchRead x) {
+visitUnsafePrefetch(x, false);
+}
+@Override
+public void visitUnsafePrefetchWrite(UnsafePrefetchWrite x) {
+visitUnsafePrefetch(x, true);
+}
+@Override
+public void visitUnsafePutObject(UnsafePutObject x) {
+CiKind kind = x.unsafeOpKind;
+LIRItem data = new LIRItem(x.value(), this);
+CiValue src = load(x.object());
+data.loadItem(kind);
+CiValue off = load(x.offset());
+setNoResult(x);
+if (x.isVolatile()) {
+vma.preVolatileWrite();
+}
+genPutObjectUnsafe(src, off, data.result(), kind, x.isVolatile());
+if (x.isVolatile()) {
+vma.postVolatileWrite();
+}
+}
+@Override
+public void visitUnsafePutRaw(UnsafePutRaw x) {
+int log2scale = 0;
+CiKind kind = x.unsafeOpKind;
+if (x.hasIndex()) {
+assert x.index().kind.isInt() : "should not find non-int index";
+log2scale = x.log2scale();
+}
+LIRItem value = new LIRItem(x.value(), this);
+LIRItem idx = new LIRItem(this);
+CiValue base = load(x.base());
+if (x.hasIndex()) {
+idx.setInstruction(x.index());
+idx.loadItem();
+}
+value.loadItem(kind);
+setNoResult(x);
+CiValue indexOp = idx.result();
+if (log2scale != 0) {
+// temporary fix (platform dependent code without shift on Intel would be better)
+indexOp = newVariable(CiKind.Int);
+lir.move(idx.result(), indexOp);
+lir.shiftLeft(indexOp, log2scale, indexOp);
+}
+CiValue addr = new CiAddress(x.unsafeOpKind, base, indexOp);
+lir.move(value.result(), addr);
+}
+private void blockDoEpilog(BlockBegin block) {
+if (C1XOptions.PrintIRWithLIR) {
+TTY.println();
+}
+// clear out variables for local constants
+constants.clear();
+variablesForConstants.clear();
+}
+private void blockDoProlog(BlockBegin block) {
+if (C1XOptions.PrintIRWithLIR) {
+TTY.print(block.toString());
+}
+// set up the list of LIR instructions
+assert block.lir() == null : "LIR list already computed for this block";
+lir = new LIRList(this);
+block.setLir(lir);
+lir.branchDestination(block.label());
+if (block == ir.startBlock) {
+XirSnippet prologue = xir.genPrologue(null, compilation.method);
+if (prologue != null) {
+emitXir(prologue, null, null, null, false);
+}
+setOperandsForLocals(block.end().stateAfter());
+}
+}
+/**
+* Copies a given value into an operand that is forced to be a stack location.
+*
+* @param value a value to be forced onto the stack
+* @param kind the kind of new operand
+* @param mustStayOnStack specifies if the new operand must never be allocated to a register
+* @return the operand that is guaranteed to be a stack location when it is
+*         initially defined a by move from {@code value}
+*/
+CiValue forceToSpill(CiValue value, CiKind kind, boolean mustStayOnStack) {
+assert value.isLegal() : "value should not be illegal";
+assert kind.jvmSlots == value.kind.jvmSlots : "size mismatch";
+if (!value.isVariableOrRegister()) {
+// force into a variable that must start in memory
+CiValue operand = operands.newVariable(value.kind, mustStayOnStack ? VariableFlag.MustStayInMemory : VariableFlag.MustStartInMemory);
+lir.move(value, operand);
+return operand;
+}
+// create a spill location
+CiValue operand = operands.newVariable(kind, mustStayOnStack ? VariableFlag.MustStayInMemory : VariableFlag.MustStartInMemory);
+// move from register to spill
+lir.move(value, operand);
+return operand;
+}
+private CiVariable loadConstant(Constant x) {
+return loadConstant(x.asConstant(), x.kind);
+}
+protected CiVariable loadConstant(CiConstant c, CiKind kind) {
+// XXX: linear search might be kind of slow for big basic blocks
+int index = constants.indexOf(c);
+if (index != -1) {
+C1XMetrics.LoadConstantIterations += index;
+return variablesForConstants.get(index);
+}
+C1XMetrics.LoadConstantIterations += constants.size();
+CiVariable result = newVariable(kind);
+lir.move(c, result);
+constants.add(c);
+variablesForConstants.add(result);
+return result;
+}
+/**
+* Allocates a variable operand to hold the result of a given instruction.
+* This can only be performed once for any given instruction.
+*
+* @param x an instruction that produces a result
+* @return the variable assigned to hold the result produced by {@code x}
+*/
+protected CiVariable createResultVariable(Value x) {
+CiVariable operand = newVariable(x.kind);
+setResult(x, operand);
+return operand;
+}
+private void visitFPIntrinsics(Intrinsic x) {
+assert x.numberOfArguments() == 1 : "wrong type";
+CiValue reg = createResultVariable(x);
+CiValue value = load(x.argumentAt(0));
+CiValue tmp = forceToSpill(value, x.kind, false);
+lir.move(tmp, reg);
+}
+private void visitRegisterFinalizer(Intrinsic x) {
+assert x.numberOfArguments() == 1 : "wrong type";
+CiValue receiver = load(x.argumentAt(0));
+LIRDebugInfo info = stateFor(x, x.stateBefore());
+callRuntime(CiRuntimeCall.RegisterFinalizer, info, receiver);
+setNoResult(x);
+}
+private void visitSwitchRanges(SwitchRange[] x, CiValue value, BlockBegin defaultSux) {
+for (int i = 0; i < x.length; i++) {
+SwitchRange oneRange = x[i];
+int lowKey = oneRange.lowKey;
+int highKey = oneRange.highKey;
+BlockBegin dest = oneRange.sux;
+if (lowKey == highKey) {
+lir.cmp(Condition.EQ, value, lowKey);
+lir.branch(Condition.EQ, CiKind.Int, dest);
+} else if (highKey - lowKey == 1) {
+lir.cmp(Condition.EQ, value, lowKey);
+lir.branch(Condition.EQ, CiKind.Int, dest);
+lir.cmp(Condition.EQ, value, highKey);
+lir.branch(Condition.EQ, CiKind.Int, dest);
+} else {
+Label l = new Label();
+lir.cmp(Condition.LT, value, lowKey);
+lir.branch(Condition.LT, l);
+lir.cmp(Condition.LE, value, highKey);
+lir.branch(Condition.LE, CiKind.Int, dest);
+lir.branchDestination(l);
+}
+}
+lir.jump(defaultSux);
+}
+private void visitUnsafePrefetch(UnsafePrefetch x, boolean isStore) {
+LIRItem src = new LIRItem(x.object(), this);
+LIRItem off = new LIRItem(x.offset(), this);
+src.loadItem();
+if (!(off.result().isConstant() && canInlineAsConstant(x.offset()))) {
+off.loadItem();
+}
+setNoResult(x);
+CiAddress addr = genAddress(src.result(), off.result(), 0, 0, CiKind.Byte);
+lir.prefetch(addr, isStore);
+}
+protected void arithmeticOpFpu(int code, CiValue result, CiValue left, CiValue right, CiValue tmp) {
+CiValue leftOp = left;
+if (isTwoOperand && leftOp != result) {
+assert right != result : "malformed";
+lir.move(leftOp, result);
+leftOp = result;
+}
+switch (code) {
+case DADD:
+case FADD:
+lir.add(leftOp, right, result);
+break;
+case FMUL:
+case DMUL:
+lir.mul(leftOp, right, result);
+break;
+case DSUB:
+case FSUB:
+lir.sub(leftOp, right, result);
+break;
+case FDIV:
+case DDIV:
+lir.div(leftOp, right, result, null);
+break;
+default:
+Util.shouldNotReachHere();
+}
+}
+protected void arithmeticOpInt(int code, CiValue result, CiValue left, CiValue right, CiValue tmp) {
+CiValue leftOp = left;
+if (isTwoOperand && leftOp != result) {
+assert right != result : "malformed";
+lir.move(leftOp, result);
+leftOp = result;
+}
+switch (code) {
+case IADD:
+lir.add(leftOp, right, result);
+break;
+case IMUL:
+boolean didStrengthReduce = false;
+if (right.isConstant()) {
+CiConstant rightConstant = (CiConstant) right;
+int c = rightConstant.asInt();
+if (CiUtil.isPowerOf2(c)) {
+// do not need tmp here
+lir.shiftLeft(leftOp, CiUtil.log2(c), result);
+didStrengthReduce = true;
+} else {
+didStrengthReduce = strengthReduceMultiply(leftOp, c, result, tmp);
+}
+}
+// we couldn't strength reduce so just emit the multiply
+if (!didStrengthReduce) {
+lir.mul(leftOp, right, result);
+}
+break;
+case ISUB:
+lir.sub(leftOp, right, result);
+break;
+default:
+// idiv and irem are handled elsewhere
+Util.shouldNotReachHere();
+}
+}
+protected void arithmeticOpLong(int code, CiValue result, CiValue left, CiValue right, LIRDebugInfo info) {
+CiValue leftOp = left;
+if (isTwoOperand && leftOp != result) {
+assert right != result : "malformed";
+lir.move(leftOp, result);
+leftOp = result;
+}
+switch (code) {
+case LADD:
+lir.add(leftOp, right, result);
+break;
+case LMUL:
+lir.mul(leftOp, right, result);
+break;
+case LSUB:
+lir.sub(leftOp, right, result);
+break;
+default:
+// ldiv and lrem are handled elsewhere
+Util.shouldNotReachHere();
+}
+}
+protected final CiValue callRuntime(CiRuntimeCall runtimeCall, LIRDebugInfo info, CiValue... args) {
+// get a result register
+CiKind result = runtimeCall.resultKind;
+CiKind[] arguments = runtimeCall.arguments;
+CiValue physReg = result.isVoid() ? IllegalValue : resultOperandFor(result);
+List<CiValue> argumentList;
+if (arguments.length > 0) {
+// move the arguments into the correct location
+CiCallingConvention cc = compilation.frameMap().getCallingConvention(arguments, RuntimeCall);
+assert cc.locations.length == args.length : "argument count mismatch";
+for (int i = 0; i < args.length; i++) {
+CiValue arg = args[i];
+CiValue loc = cc.locations[i];
+if (loc.isRegister()) {
+lir.move(arg, loc);
+} else {
+assert loc.isStackSlot();
+CiStackSlot slot = (CiStackSlot) loc;
+if (slot.kind == CiKind.Long || slot.kind == CiKind.Double) {
+lir.unalignedMove(arg, slot);
+} else {
+lir.move(arg, slot);
+}
+}
+}
+argumentList = Arrays.asList(cc.locations);
+} else {
+// no arguments
+assert args == null || args.length == 0;
+argumentList = Util.uncheckedCast(Collections.emptyList());
+}
+lir.callRuntime(runtimeCall, physReg, argumentList, info);
+return physReg;
+}
+protected final CiVariable callRuntimeWithResult(CiRuntimeCall runtimeCall, LIRDebugInfo info, CiValue... args) {
+CiVariable result = newVariable(runtimeCall.resultKind);
+CiValue location = callRuntime(runtimeCall, info, args);
+lir.move(location, result);
+return result;
+}
+SwitchRange[] createLookupRanges(LookupSwitch x) {
+// we expect the keys to be sorted by increasing value
+List<SwitchRange> res = new ArrayList<SwitchRange>(x.numberOfCases());
+int len = x.numberOfCases();
+if (len > 0) {
+BlockBegin defaultSux = x.defaultSuccessor();
+int key = x.keyAt(0);
+BlockBegin sux = x.suxAt(0);
+SwitchRange range = new SwitchRange(key, sux);
+for (int i = 1; i < len; i++) {
+int newKey = x.keyAt(i);
+BlockBegin newSux = x.suxAt(i);
+if (key + 1 == newKey && sux == newSux) {
+// still in same range
+range.highKey = newKey;
+} else {
+// skip tests which explicitly dispatch to the default
+if (range.sux != defaultSux) {
+res.add(range);
+}
+range = new SwitchRange(newKey, newSux);
+}
+key = newKey;
+sux = newSux;
+}
+if (res.size() == 0 || res.get(res.size() - 1) != range) {
+res.add(range);
+}
+}
+return res.toArray(new SwitchRange[res.size()]);
+}
+SwitchRange[] createLookupRanges(TableSwitch x) {
+// XXX: try to merge this with the code for LookupSwitch
+List<SwitchRange> res = new ArrayList<SwitchRange>(x.numberOfCases());
+int len = x.numberOfCases();
+if (len > 0) {
+BlockBegin sux = x.suxAt(0);
+int key = x.lowKey();
+BlockBegin defaultSux = x.defaultSuccessor();
+SwitchRange range = new SwitchRange(key, sux);
+for (int i = 0; i < len; i++, key++) {
+BlockBegin newSux = x.suxAt(i);
+if (sux == newSux) {
+// still in same range
+range.highKey = key;
+} else {
+// skip tests which explicitly dispatch to the default
+if (sux != defaultSux) {
+res.add(range);
+}
+range = new SwitchRange(key, newSux);
+}
+sux = newSux;
+}
+if (res.size() == 0 || res.get(res.size() - 1) != range) {
+res.add(range);
+}
+}
+return res.toArray(new SwitchRange[res.size()]);
+}
+void doRoot(Instruction instr) {
+currentInstruction = instr;
+assert instr.isLive() : "use only with roots";
+assert !instr.hasSubst() : "shouldn't have missed substitution";
+if (C1XOptions.TraceLIRVisit) {
+TTY.println("Visiting    " + instr);
+}
+instr.accept(this);
+if (C1XOptions.TraceLIRVisit) {
+TTY.println("Operand for " + instr + " = " + instr.operand());
+}
+assert (instr.operand().isLegal()) || !isUsedForValue(instr) || instr.isConstant() || instr instanceof UnsafeCast : "operand was not set for live instruction";
+}
+private boolean isUsedForValue(Instruction instr) {
+return instr.checkFlag(Value.Flag.LiveValue);
+}
+protected void logicOp(int code, CiValue resultOp, CiValue leftOp, CiValue rightOp) {
+if (isTwoOperand && leftOp != resultOp) {
+assert rightOp != resultOp : "malformed";
+lir.move(leftOp, resultOp);
+leftOp = resultOp;
+}
+switch (code) {
+case IAND:
+case LAND:
+lir.logicalAnd(leftOp, rightOp, resultOp);
+break;
+case IOR:
+case LOR:
+lir.logicalOr(leftOp, rightOp, resultOp);
+break;
+case IXOR:
+case LXOR:
+lir.logicalXor(leftOp, rightOp, resultOp);
+break;
+default:
+Util.shouldNotReachHere();
+}
+}
+void moveToPhi(PhiResolver resolver, Value curVal, Value suxVal) {
+// move current value to referenced phi function
+if (suxVal instanceof Phi) {
+Phi phi = (Phi) suxVal;
+// curVal can be null without phi being null in conjunction with inlining
+if (phi.isLive() && curVal != null && curVal != phi) {
+assert curVal.isLive() : "value not live: " + curVal + ", suxVal=" + suxVal;
+assert !phi.isIllegal() : "illegal phi cannot be marked as live";
+if (curVal instanceof Phi) {
+operandForPhi((Phi) curVal);
+}
+CiValue operand = curVal.operand();
+if (operand.isIllegal()) {
+assert curVal instanceof Constant || curVal instanceof Local : "these can be produced lazily";
+operand = operandForInstruction(curVal);
+}
+resolver.move(operand, operandForPhi(phi));
+}
+}
+}
+protected void moveToPhi(FrameState curState) {
+// Moves all stack values into their phi position
+BlockBegin bb = currentBlock;
+if (bb.numberOfSux() == 1) {
+BlockBegin sux = bb.suxAt(0);
+assert sux.numberOfPreds() > 0 : "invalid CFG";
+// a block with only one predecessor never has phi functions
+if (sux.numberOfPreds() > 1) {
+PhiResolver resolver = new PhiResolver(this);
+FrameState suxState = sux.stateBefore();
+for (int index = 0; index < suxState.stackSize(); index++) {
+moveToPhi(resolver, curState.stackAt(index), suxState.stackAt(index));
+}
+// walk up the inlined scopes until locals match
+while (curState.scope() != suxState.scope()) {
+curState = curState.callerState();
+assert curState != null : "scopes don't match up";
+}
+for (int index = 0; index < suxState.localsSize(); index++) {
+moveToPhi(resolver, curState.localAt(index), suxState.localAt(index));
+}
+assert curState.scope().callerState == suxState.scope().callerState : "caller states must be equal";
+resolver.dispose();
+}
+}
+}
+/**
+* Creates a new {@linkplain CiVariable variable}.
+*
+* @param kind the kind of the variable
+* @return a new variable
+*/
+public CiVariable newVariable(CiKind kind) {
+return operands.newVariable(kind);
+}
+CiValue operandForInstruction(Value x) {
+CiValue operand = x.operand();
+if (operand.isIllegal()) {
+if (x instanceof Constant) {
+x.setOperand(x.asConstant());
+} else {
+assert x instanceof Phi || x instanceof Local : "only for Phi and Local";
+// allocate a variable for this local or phi
+createResultVariable(x);
+}
+}
+return x.operand();
+}
+private CiValue operandForPhi(Phi phi) {
+if (phi.operand().isIllegal()) {
+// allocate a variable for this phi
+CiVariable operand = newVariable(phi.kind);
+setResult(phi, operand);
+}
+return phi.operand();
+}
+protected void postGCWriteBarrier(CiValue addr, CiValue newVal) {
+XirSnippet writeBarrier = xir.genWriteBarrier(toXirArgument(addr));
+if (writeBarrier != null) {
+emitXir(writeBarrier, null, null, null, false);
+}
+}
+protected void preGCWriteBarrier(CiValue addrOpr, boolean patch, LIRDebugInfo info) {
+}
+protected void setNoResult(Instruction x) {
+assert !isUsedForValue(x) : "can't have use";
+x.clearOperand();
+}
+protected CiValue setResult(Value x, CiVariable operand) {
+x.setOperand(operand);
+if (C1XOptions.DetailedAsserts) {
+operands.recordResult(operand, x);
+}
+return operand;
+}
+protected void shiftOp(int code, CiValue resultOp, CiValue value, CiValue count, CiValue tmp) {
+if (isTwoOperand && value != resultOp) {
+assert count != resultOp : "malformed";
+lir.move(value, resultOp);
+value = resultOp;
+}
+assert count.isConstant() || count.isVariableOrRegister();
+switch (code) {
+case ISHL:
+case LSHL:
+lir.shiftLeft(value, count, resultOp, tmp);
+break;
+case ISHR:
+case LSHR:
+lir.shiftRight(value, count, resultOp, tmp);
+break;
+case IUSHR:
+case LUSHR:
+lir.unsignedShiftRight(value, count, resultOp, tmp);
+break;
+default:
+Util.shouldNotReachHere();
+}
+}
+protected void walkState(Instruction x, FrameState state) {
+if (state == null) {
+return;
+}
+for (int index = 0; index < state.stackSize(); index++) {
+walkStateValue(state.stackAt(index));
+}
+FrameState s = state;
+int bci = x.bci();
+while (s != null) {
+IRScope scope = s.scope();
+if (bci == Instruction.SYNCHRONIZATION_ENTRY_BCI) {
+assert x instanceof ExceptionObject ||
+x instanceof Throw ||
+x instanceof MonitorEnter ||
+x instanceof MonitorExit;
+}
+for (int index = 0; index < s.localsSize(); index++) {
+final Value value = s.localAt(index);
+if (value != null) {
+if (!value.isIllegal()) {
+walkStateValue(value);
+}
+}
+}
+bci = scope.callerBCI();
+s = s.callerState();
+}
+}
+private void walkStateValue(Value value) {
+if (value != null) {
+assert !value.hasSubst() : "missed substitution";
+assert value.isLive() : "value must be marked live in frame state";
+if (value instanceof Phi && !value.isIllegal()) {
+// phi's are special
+operandForPhi((Phi) value);
+} else if (value.operand().isIllegal() && !(value instanceof UnsafeCast)) {
+// instruction doesn't have an operand yet
+CiValue operand = makeOperand(value);
+assert operand.isLegal() : "must be evaluated now";
+}
+}
+}
+protected LIRDebugInfo maybeStateFor(Instruction x) {
+FrameState stateBefore = x.stateBefore();
+if (stateBefore == null) {
+return null;
+}
+return stateFor(x, stateBefore);
+}
+protected LIRDebugInfo stateFor(Instruction x) {
+assert x.stateBefore() != null : "must have state before instruction for " + x;
+return stateFor(x, x.stateBefore());
+}
+protected LIRDebugInfo stateFor(Instruction x, FrameState state) {
+if (compilation.placeholderState != null) {
+state = compilation.placeholderState;
+}
+return new LIRDebugInfo(state, x.exceptionHandlers());
+}
+List<CiValue> visitInvokeArguments(CiCallingConvention cc, Value[] args, List<CiValue> pointerSlots) {
+// for each argument, load it into the correct location
+List<CiValue> argList = new ArrayList<CiValue>(args.length);
+int j = 0;
+for (Value arg : args) {
+if (arg != null) {
+CiValue operand = cc.locations[j++];
+if (operand.isRegister()) {
+force(arg, operand);
+} else {
+LIRItem param = new LIRItem(arg, this);
+assert operand.isStackSlot();
+CiStackSlot slot = (CiStackSlot) operand;
+assert !slot.inCallerFrame();
+param.loadForStore(slot.kind);
+if (slot.kind == CiKind.Long || slot.kind == CiKind.Double) {
+lir.unalignedMove(param.result(), slot);
+} else {
+lir.move(param.result(), slot);
+}
+if (arg.kind == CiKind.Object && pointerSlots != null) {
+// This slot must be marked explicitedly in the pointer map.
+pointerSlots.add(slot);
+}
+}
+argList.add(operand);
+}
+}
+return argList;
+}
+/**
+* Ensures that an operand has been {@linkplain Value#setOperand(CiValue) initialized}
+* for storing the result of an instruction.
+*
+* @param instruction an instruction that produces a result value
+*/
+protected CiValue makeOperand(Value instruction) {
+assert instruction.isLive();
+CiValue operand = instruction.operand();
+if (operand.isIllegal()) {
+if (instruction instanceof Phi) {
+// a phi may not have an operand yet if it is for an exception block
+operand = operandForPhi((Phi) instruction);
+} else if (instruction instanceof Constant) {
+operand = operandForInstruction(instruction);
+}
+}
+// the value must be a constant or have a valid operand
+assert operand.isLegal() : "this root has not been visited yet";
+return operand;
+}
+/**
+* Gets the ABI specific operand used to return a value of a given kind from a method.
+*
+* @param kind the kind of value being returned
+* @return the operand representing the ABI defined location used return a value of kind {@code kind}
+*/
+protected CiValue resultOperandFor(CiKind kind) {
+if (kind == CiKind.Void) {
+return IllegalValue;
+}
+CiRegister returnRegister = compilation.registerConfig.getReturnRegister(kind);
+return returnRegister.asValue(kind);
+}
+protected XirSupport site(Value x) {
+return xirSupport.site(x);
+}
+public void maybePrintCurrentInstruction() {
+if (currentInstruction != null && lastInstructionPrinted != currentInstruction) {
+lastInstructionPrinted = currentInstruction;
+InstructionPrinter ip = new InstructionPrinter(TTY.out());
+ip.printInstructionListing(currentInstruction);
+}
+}
+protected abstract boolean canInlineAsConstant(Value i);
+protected abstract boolean canStoreAsConstant(Value i, CiKind kind);
+protected abstract CiValue exceptionPcOpr();
+protected abstract CiValue osrBufferPointer();
+protected abstract boolean strengthReduceMultiply(CiValue left, int constant, CiValue result, CiValue tmp);
+protected abstract CiAddress genAddress(CiValue base, CiValue index, int shift, int disp, CiKind kind);
+protected abstract void genCmpMemInt(Condition condition, CiValue base, int disp, int c, LIRDebugInfo info);
+protected abstract void genCmpRegMem(Condition condition, CiValue reg, CiValue base, int disp, CiKind kind, LIRDebugInfo info);
+protected abstract void genGetObjectUnsafe(CiValue dest, CiValue src, CiValue offset, CiKind kind, boolean isVolatile);
+protected abstract void genPutObjectUnsafe(CiValue src, CiValue offset, CiValue data, CiKind kind, boolean isVolatile);
+protected abstract void genCompareAndSwap(Intrinsic x, CiKind kind);
+protected abstract void genMathIntrinsic(Intrinsic x);
+/**
+* Implements site-specific information for the XIR interface.
+*/
+static class XirSupport implements XirSite {
+Value current;
+XirSupport() {
+}
+public CiCodePos getCodePos() {
+// TODO: get the code position of the current instruction if possible
+return null;
+}
+public boolean isNonNull(XirArgument argument) {
+if (argument.constant == null && argument.object instanceof LIRItem) {
+// check the flag on the original value
+return ((LIRItem) argument.object).instruction.isNonNull();
+}
+return false;
+}
+public boolean requiresNullCheck() {
+return current == null || current.needsNullCheck();
+}
+public boolean requiresBoundsCheck() {
+return current == null || !current.checkFlag(Value.Flag.NoBoundsCheck);
+}
+public boolean requiresReadBarrier() {
+return current == null || !current.checkFlag(Value.Flag.NoReadBarrier);
+}
+public boolean requiresWriteBarrier() {
+return current == null || !current.checkFlag(Value.Flag.NoWriteBarrier);
+}
+public boolean requiresArrayStoreCheck() {
+return current == null || !current.checkFlag(Value.Flag.NoStoreCheck);
+}
+public RiType getApproximateType(XirArgument argument) {
+return current == null ? null : current.declaredType();
+}
+public RiType getExactType(XirArgument argument) {
+return current == null ? null : current.exactType();
+}
+XirSupport site(Value v) {
+current = v;
+return this;
+}
+@Override
+public String toString() {
+return "XirSupport<" + current + ">";
+}
+}
+public void arrayCopy(RiType type, ArrayCopy arrayCopy, XirSnippet snippet) {
+emitXir(snippet, arrayCopy, stateFor(arrayCopy), null, false);
+}
+@Override
+public void visitArrayCopy(ArrayCopy arrayCopy) {
+Value src = arrayCopy.src();
+Value dest = arrayCopy.dest();
+Value srcPos = arrayCopy.srcPos();
+Value destPos = arrayCopy.destPos();
+Value length = arrayCopy.length();
+RiType srcType = src.declaredType();
+RiType destType = dest.declaredType();
+if ((srcType != null && srcType.isArrayClass()) || (destType != null && destType.isArrayClass())) {
+RiType type = (srcType == null) ? destType : srcType;
+if ((srcType == null || destType == null || srcType.kind() != destType.kind()) && type.kind() != CiKind.Object) {
+TypeEqualityCheck typeCheck = new TypeEqualityCheck(src, dest, arrayCopy.stateBefore(), Condition.EQ);
+visitTypeEqualityCheck(typeCheck);
+}
+boolean inputsSame = (src == dest);
+boolean inputsDifferent = !inputsSame && (src.checkFlag(Flag.ResultIsUnique) || dest.checkFlag(Flag.ResultIsUnique));
+boolean needsStoreCheck = type.componentType().kind() == CiKind.Object && destType != srcType;
+if (!needsStoreCheck) {
+arrayCopy.setFlag(Flag.NoStoreCheck);
+}
+XirSnippet snippet = xir.genArrayCopy(site(arrayCopy), toXirArgument(src), toXirArgument(srcPos), toXirArgument(dest), toXirArgument(destPos), toXirArgument(length), type.componentType(), inputsSame, inputsDifferent);
+arrayCopy(type, arrayCopy, snippet);
+return;
+}
+arrayCopySlow(arrayCopy);
+}
+private void arrayCopySlow(ArrayCopy arrayCopy) {
+emitInvokeKnown(arrayCopy.arrayCopyMethod, arrayCopy.stateBefore(), arrayCopy.src(), arrayCopy.srcPos(), arrayCopy.dest(), arrayCopy.destPos(), arrayCopy.length());
+}
+private CiValue emitInvokeKnown(RiMethod method, FrameState stateBefore, Value... args) {
+boolean isStatic = Modifier.isStatic(method.accessFlags());
+Invoke invoke = new Invoke(isStatic ? Bytecodes.INVOKESTATIC : Bytecodes.INVOKESPECIAL, method.signature().returnKind(), args, isStatic, method, null, stateBefore);
+visitInvoke(invoke);
+return invoke.operand();
+}
+@Override
+public void visitTypeEqualityCheck(TypeEqualityCheck typeEqualityCheck) {
+Value x = typeEqualityCheck.left();
+Value y = typeEqualityCheck.right();
+CiValue leftValue = emitXir(xir.genGetClass(site(typeEqualityCheck), toXirArgument(x)), typeEqualityCheck, stateFor(typeEqualityCheck), null, false);
+CiValue rightValue = emitXir(xir.genGetClass(site(typeEqualityCheck), toXirArgument(y)), typeEqualityCheck, stateFor(typeEqualityCheck), null, false);
+lir.cmp(typeEqualityCheck.condition.negate(), leftValue, rightValue);
+emitGuard(typeEqualityCheck);
+}
+}

Mercurial > hg > truffle

comparison graal/GraalCompiler/src/com/sun/c1x/gen/LIRGenerator.java @ 2509:16b9a8b5ad39