package org.jruby.ir.instructions;

// A generic IR instruction is of the form: v = OP(arg_array, attribute_array)

import org.jruby.RubyInstanceConfig;
import org.jruby.ir.IRScope;
import org.jruby.ir.IRVisitor;
import org.jruby.ir.Interp;
import org.jruby.ir.Operation;
import org.jruby.ir.operands.LocalVariable;
import org.jruby.ir.operands.Operand;
import org.jruby.ir.operands.Variable;
import org.jruby.ir.persistence.IRWriterEncoder;
import org.jruby.ir.transformations.inlining.CloneInfo;
import org.jruby.parser.StaticScope;
import org.jruby.runtime.DynamicScope;
import org.jruby.runtime.ThreadContext;
import org.jruby.runtime.builtin.IRubyObject;
import static org.jruby.util.StringSupport.EMPTY_STRING_ARRAY;

import java.lang.reflect.Field;
import java.lang.reflect.Modifier;
import java.util.ArrayList;
import java.util.List;
import java.util.Map;

//
// Specialized forms:
//   v = OP(arg1, arg2, attribute_array); Ex: v = ADD(v1, v2)
//   v = OP(arg, attribute_array);        Ex: v = NOT(v1)
//
// _attributes store information about the operands of the instruction that have
// been collected as part of analysis.  For more information, see documentation
// in Attribute.java
//
// Ex: v = BOXED_FIXNUM(n)
//     v = HAS_TYPE(Fixnum)
public abstract class Instr {
    public static final Operand[] EMPTY_OPERANDS = new Operand[0];

    private transient final Operation operation;
    // Is this instruction live or dead?  During optimization passes, if this instruction
    // causes no side-effects and the result of the instruction is not needed by anyone else,
    // we can remove this instruction altogether without affecting program correctness.
    private boolean isDead;

    public Instr(Operation operation) {
        this.operation = operation;
    }

    public String[] toStringNonOperandArgs() {
        return EMPTY_STRING_ARRAY;
    }

    public void encode(IRWriterEncoder e) {
        if (RubyInstanceConfig.IR_WRITING_DEBUG) System.out.println("Instr(" + getOperation() + "): " + this);
        e.encode(getOperation());
    }

    
Instructions are meant to be in a machine-readable format so offline tooling can parse the debugging output. The format is: (result_op '=')? instr '(' (operand ',' )* operand? ';' (extra_arg ',')* extra_arg? ')' extra_arg can either be plain value or in a key: value format.
Returns:a String
/** * Instructions are meant to be in a machine-readable format so offline tooling can parse the * debugging output. The format is: * * (result_op '=')? instr '(' (operand ',' )* operand? ';' (extra_arg ',')* extra_arg? ')' * extra_arg can either be plain value or in a key: value format. * @return a String */
@Override public String toString() { StringBuilder buf = new StringBuilder(isDead() ? "[DEAD]" : ""); if (this instanceof ResultInstr) buf.append(((ResultInstr) this).getResult()).append(" = "); Operand[] operands = getOperands(); buf.append(operation).append('('); toArgList(buf, operands); String[] extraArgs = toStringNonOperandArgs(); if (extraArgs.length >= 1) { if (operands.length > 0) buf.append(' '); buf.append(';'); toArgList(buf, extraArgs); } buf.append(')'); return buf.toString(); } private static final Field[] EMPTY_FIELD_ARRAY = new Field[0];
A ClassValue wrapping an array of dumpable Field references for a given Instr type.
/** * A ClassValue wrapping an array of dumpable Field references for a given Instr type. */
private static final ClassValue<Field[]> dumpableFields = new ClassValue<Field[]>() { @Override protected Field[] computeValue(Class type) { try { Class cls = type; ArrayList<Field> list = new ArrayList<>(); while (cls != Instr.class) { for (Field f : cls.getDeclaredFields()) { if (Modifier.isTransient(f.getModifiers())) continue; if (Modifier.isStatic(f.getModifiers())) continue; if (f.getName().startsWith("$")) continue; try { f.setAccessible(true); } catch (SecurityException se) { continue; } list.add(f); } cls = cls.getSuperclass(); } return list.toArray(new Field[list.size()]); } catch (Exception e) { return EMPTY_FIELD_ARRAY; } } }; public Field[] dumpableFields() { return dumpableFields.get(getClass()); } private StringBuilder toArgList(StringBuilder buf, Object[] args) { if (args.length <= 0) return buf; for (int i = 0; i < args.length - 1; i++) { buf.append(args[i]).append(", "); } buf.append(args[args.length - 1]); return buf; } @Interp public Operation getOperation() { return operation; } // Does this instruction have side effects as a result of its operation // This information is used in optimization phases to impact dead code elimination // and other optimization passes public boolean hasSideEffects() { return operation.hasSideEffects(); } // Can this instruction raise exceptions -- this superclass method has to be conservative and cannot affect program correctness. public boolean canRaiseException() { return operation.canRaiseException(); } // Can this instruction raise exceptions -- this superclass method has to be conservative and cannot affect program correctness. public boolean transfersControl() { return operation.transfersControl(); }
Does this instruction do anything the scope is interested in?
Returns:true if it modified the scope.
/** * Does this instruction do anything the scope is interested in? * @return true if it modified the scope. */
public boolean computeScopeFlags(IRScope scope) { return false; }
Can this instruction be deleted? LVA will preserve instructions based on whether operands (variables) are living but even if there are no living variables then the instruction itself may not be able to be removed during DCE for other reasons (like if it unconditionally has a side-effect)
/** * Can this instruction be deleted? LVA will preserve instructions based on whether operands (variables) * are living but even if there are no living variables then the instruction itself may not be able to be removed * during DCE for other reasons (like if it unconditionally has a side-effect) */
public boolean isDeletable() { return !(hasSideEffects() || operation.isDebugOp() || canRaiseException() || transfersControl()); } public boolean canBeDeletedFromScope(IRScope s) { if (!isDeletable()) { return false; } if (this instanceof ResultInstr) { Variable r = ((ResultInstr) this).getResult(); // An escaped binding needs to preserve lvars since // consumers of that binding may access lvars. if (s.bindingHasEscaped()) return !(r instanceof LocalVariable); } return true; } public void markDead() { isDead = true; } @Interp public boolean isDead() { return isDead; } /* Array of all operands for this instruction */ public abstract Operand[] getOperands(); public abstract void setOperand(int i, Operand operand); /* List of all variables used by all operands of this instruction */ public List<Variable> getUsedVariables() { ArrayList<Variable> vars = new ArrayList<>(); for (Operand operand : getOperands()) { operand.addUsedVariables(vars); } return vars; } public void renameVars(Map<Operand, Operand> renameMap) { simplifyOperands(renameMap, true); if (this instanceof ResultInstr) { ResultInstr ri = (ResultInstr)this; Variable oldVar = ri.getResult(); Variable newVar = (Variable)renameMap.get(oldVar); if (newVar != null) ri.updateResult(newVar); } }
Clone the instruction for use in an inlining context (either when a scope is inlined into another scope, or when a block has to be cloned because its associated call belongs to an inlined scope). This might renaming variables and labels to eliminate naming conflicts. The implementation might vary on the cloning mode.
Params:
  • info – This object manages renaming of variables and labels, handles args and return values.
Returns:a new instruction that can be used in the target scope.
/** * Clone the instruction for use in an inlining context (either when a scope is inlined into * another scope, or when a block has to be cloned because its associated call belongs to * an inlined scope). This might renaming variables and labels to eliminate naming conflicts. * * The implementation might vary on the cloning mode. * * @param info This object manages renaming of variables and labels, handles * args and return values. * @return a new instruction that can be used in the target scope. */
public abstract Instr clone(CloneInfo info); public Operand[] cloneOperands(CloneInfo info) { Operand[] operands = getOperands(); Operand[] newOperands = new Operand[operands.length]; for (int i = 0; i < operands.length; i++) { newOperands[i] = operands[i].cloneForInlining(info); } return newOperands; }
This method takes as input a map of operands to their values, and outputs If the value map provides a value for any of the instruction's operands this method is expected to replace the original operands with the simplified values. It is not required that it do so -- code correctness is not compromised by failure to simplify
/** * This method takes as input a map of operands to their values, and outputs * * If the value map provides a value for any of the instruction's operands * this method is expected to replace the original operands with the simplified values. * It is not required that it do so -- code correctness is not compromised by failure * to simplify */
public void simplifyOperands(Map<Operand, Operand> valueMap, boolean force) { Operand[] operands = getOperands(); for (int i = 0; i < operands.length; i++) { setOperand(i, operands[i].getSimplifiedOperand(valueMap, force)); } }
This method takes as input a map of operands to their values, and outputs the result of this instruction. If the value map provides a value for any of the instruction's operands the expectation is that the operand will be replaced with the simplified value. It is not required that it do so -- code correctness is not compromised by failure to simplify.
Params:
  • scope – where this instr exists
  • valueMap – Mapping from operands to their simplified values
Returns:simplified result / output of this instruction
/** * This method takes as input a map of operands to their values, and outputs * the result of this instruction. * * If the value map provides a value for any of the instruction's operands * the expectation is that the operand will be replaced with the simplified value. * It is not required that it do so -- code correctness is not compromised by failure * to simplify. * * @param scope where this instr exists * @param valueMap Mapping from operands to their simplified values * @return simplified result / output of this instruction */
public Operand simplifyAndGetResult(IRScope scope, Map<Operand, Operand> valueMap) { simplifyOperands(valueMap, false); return null; // By default, no simplifications! } @Interp public Object interpret(ThreadContext context, StaticScope currScope, DynamicScope currDynScope, IRubyObject self, Object[] temp) { throw new RuntimeException(this.getClass().getSimpleName() + " should not be directly interpreted"); } @Interp public int interpretAndGetNewIPC(ThreadContext context, DynamicScope currDynScope, StaticScope currScope, IRubyObject self, Object[] temp, int ipc) { throw new RuntimeException(this.getClass().getSimpleName() + " should not be directly interpreted"); } public void visit(IRVisitor visitor) { throw new RuntimeException(this.getClass().getSimpleName() + " has no compile logic"); } }