前言

本章对应官方教程第7章。本章的目的是支持变量var和`=运算符,我们需要在堆栈分配内存给变量,详细说明请查看官方教程第7章。

教程如下:

教你使用swift写编译器玩具(0)

教你使用swift写编译器玩具(1)

教你使用swift写编译器玩具(2)

教你使用swift写编译器玩具(3)

教你使用swift写编译器玩具(4)

教你使用swift写编译器玩具(5)

教你使用swift写编译器玩具(6)

教你使用swift写编译器玩具(7)

教你使用swift写编译器玩具(8)

仓库在这

开始

调整现有变量

首先我们需要改变namedValues的value类型。

var namedValues: [String: IRInstruction] = [:]

此外,我们需要一个辅助函数来创建Alloca

func createEntryBlockAlloca(function: Function, name: String) -> IRInstruction {
    let instruction = builder.buildAlloca(type: FloatType.double, count: 0, name: name)
    return instruction
}

在本章中,变量改为了存于堆栈中,因此变量的代码生成也需要从堆栈中加载。我们修改VariableExprASTcodeGen()方法。

func codeGen() -> IRValue? {
    let value = namedValues[name]
    guard value != nil else {
        fatalError("unknow variable name.")
    }
    return builder.buildLoad(value!, name: name)
}

现在我们需要更新定义变量的代码来设置Alloca。我们从ForExprASTcodeGen()方法开始。

func codeGen() -> IRValue? {
    let startVal = start.codeGen()
    guard startVal != nil else {
        return nil
    }
    
    //for循环,插在当前的block之后
    let theFunction = builder.insertBlock?.parent
    guard theFunction != nil else {
        return nil
    }
    //在entry block中为变量创建alloca
    let alloca = createEntryBlockAlloca(function: theFunction!, name: name)
  	把变量存储在alloca中
    builder.buildStore(startVal!, to: alloca)
    
    let loopBB = theFunction!.appendBasicBlock(named: "loop")
    builder.buildBr(loopBB)
    builder.positionAtEnd(of: loopBB)
    
    let oldVal = namedValues[name]
    namedValues[name] = alloca
    
    guard body.codeGen() != nil else {
        return nil
    }
    
    let stepVal: IRValue?
    if step != nil {
        stepVal = step!.codeGen()
        guard stepVal != nil else {
            return nil
        }
    } else {
        stepVal = FloatType.double.constant(1)
    }
    
    //循环终止条件
    var endCond = end.codeGen()
    guard endCond != nil else {
        return nil
    }
    //build条件时候要使用int类型
    endCond = builder.buildICmp(endCond!, IntType.int1.zero(), .notEqual, name: "loopCond")
    
  	//加载当前变量
    let curVal = builder.buildLoad(alloca)
  	//让下一个变量为当前变量+步长,即增长了
    let nextVal = builder.buildAdd(curVal, stepVal!, name: "nextVal")
  	//再重新存储到alloca中
    builder.buildStore(nextVal, to: alloca)
    
    //循环后的代码basic block
    let afterBB = theFunction?.appendBasicBlock(named: "afterLoop")
    builder.buildCondBr(condition: endCond!, then: loopBB, else: afterBB!)
    builder.positionAtEnd(of: afterBB!)
    
    if oldVal != nil {
        namedValues[name] = oldVal!
    } else {
        namedValues[name] = nil
    }
    
    //for循环解析总是返回0
    return FloatType.double.constant(0)
}

这其中最大的变化就是我们不再使用phi操作而是使用load/store来根据需要访问变量。

我们在FunctionAST中也需要改变codeGen()方法。

func codeGen() -> Function? {
    functionProtos[proto.name] = proto
    let theFunction = getFunction(named: proto.name)
    guard theFunction != nil else {
        return nil
    }
    
    //如果是操作符,把他放在全局的操作符表中
    if proto.isOperator {
        BinOpPrecedence[proto.operatorName!] = proto.precedence
    }
    
    let entry = theFunction!.appendBasicBlock(named: "entry")
    builder.positionAtEnd(of: entry)
    
    namedValues.removeAll()
    var arg = theFunction!.firstParameter
    while arg != nil {
      	//为参数创建alloca
        let alloca = createEntryBlockAlloca(function: theFunction!, name: arg!.name)
      	//把变量存到alloca中
        builder.buildStore(arg!, to: alloca)
      	//把变量放到符号表里
        namedValues[arg!.name] = alloca
        arg = arg?.next()
    }
    
    if let retValue = body.codeGen() {
        builder.buildRet(retValue)
        do {
            try theModule.verify()
            return theFunction
        } catch {
            print("\(error)")
        }
    }
    //函数体出现问题,移除函数
    theFunction!.eraseFromParent()
    if proto.isOperator {
        BinOpPrecedence[proto.operatorName!] = nil
    }
    return nil
}

添加新的运算符

我们需要在全局操作符表中加入=

var BinOpPrecedence: [String: UInt] = ["=": 2, "<": 10, "+": 20, "-": 20, "*": 40]

接下来我想大家都能想到,那就是去修改BinaryExprASTcodeGen()方法。

我们只需要在codeGen()方法最开始判断一下=即可。

if op == "=" {
    let lhse = lhs as? VariableExprAST
    guard lhse != nil else {
        fatalError("Destination of '=' must be a variable.")
    }
    let val = lhse?.codeGen()
    guard val != nil else {
        return nil
    }
  	//获取符号表中的变量
    let variable = namedValues[lhse!.name]
    guard variable != nil else {
        fatalError("Unknow variable name.")
    }
  	//为变量赋值
    builder.buildStore(val!, to: variable!)
    return val
}

用户定义的局部变量

就像之前所做过的扩展一样,我们先要在TokenLexerParser

enum Token {
		...
    case `var`
    ...
}
else if identifierStr == "var" {
		currentToken = CurrentToken(token: .var, val: "var")
}

接着我们构造VarExprASTAST Node。

class VarExprAST: ExprAST {
    
    let varNames: [(String, ExprAST?)]
    
    let body: ExprAST
    
    init(_ varNames: [(String, ExprAST?)], _ body: ExprAST) {
        self.varNames = varNames
        self.body = body
    }
    
}

我们允许通过var/in一次定义多个变量以及其初始化的值,并且我们允许在body中访问var/in定义的变量。

之后我们需要定义Parser的方法。

/// 解析Var变量
///
/// - Returns: AST
private func parseVarExpr() -> ExprAST? {
    lexer.nextToken()
    var varNames: [(String, ExprAST?)] = []
    guard lexer.currentToken!.token == .identifier else {
        fatalError("Expected identifier after val.")
    }
    while true {
        let name = lexer.currentToken!.val
        lexer.nextToken()
        
        let expr: ExprAST? = nil
        if lexer.currentToken!.val == "=" {
            lexer.nextToken()
          	//解析"="右边
            let expr = parseExpression()
            guard expr != nil else {
                return nil
            }
        }
        
        varNames.append((name, expr))
        
      	//看看还有没有下一个
        if lexer.currentToken!.val != "," {
            break
        }
        lexer.nextToken()
        if lexer.currentToken!.token != .identifier {
            fatalError("Expected identifier list after var.")
        }
    }
    if lexer.currentToken!.token != .in {
        fatalError("Expected 'in' keyword after 'var'.")
    }
    lexer.nextToken()
  	//解析body
    let body = parseExpression()
    guard body != nil else {
        return nil
    }
    return VarExprAST(varNames, body!)
}

最后需要添加的是我们VarExprAST中的codeGen()方法。

func codeGen() -> IRValue? {
    var oldBindings: [IRInstruction?] = []
    let theFunction = builder.insertBlock?.parent
    guard theFunction != nil else {
        return nil
    }
    //注册所有变量,并让他们初始化
    for v in varNames {
        let initVal: IRValue?
        if v.1 != nil {
            initVal = v.1?.codeGen()
            guard initVal != nil else {
                return nil
            }
        } else {
            //没有的话就默认0
            initVal = FloatType.double.constant(0)
        }
        
        let alloca = createEntryBlockAlloca(function: theFunction!, name: v.0)
        //初始化变量,把initVal存到alloca中
        builder.buildStore(initVal!, to: alloca)
      	//记录的目的是防止丢失外部变量名相同的变量,比如说外部有变量a,body里也有变量a
        oldBindings.append(namedValues[v.0])
        namedValues[v.0] = alloca
    }
    
    let bodyVal = body.codeGen()
    guard bodyVal != nil else {
        return nil
    }
    //恢复之前的变量绑定
    for i in 0..<varNames.count {
        namedValues[varNames[i].0] = oldBindings[i]
    }
    return bodyVal
}

测试

我们输入

def binary : 1 (x y) y;

def fibi(x)
    var a = 1, b = 1, c in
    (for i = 3, i < x in
        c = a + b :
        a = b :
        b = c) :
    b;

fibi(10);

输出

Read function definition:

define double @"binary:"(double %x, double %y) {
entry:
  %x1 = alloca double, i64 0
  store double %x, double* %x1
  %y2 = alloca double, i64 0
  store double %y, double* %y2
  %y3 = load double, double* %y2
  ret double %y3
}
Read function definition:

define double @fibi(double %x) {
entry:
  %x1 = alloca double, i64 0
  store double %x, double* %x1
  %a = alloca double, i64 0
  store double 0.000000e+00, double* %a
  %b = alloca double, i64 0
  store double 0.000000e+00, double* %b
  %c = alloca double, i64 0
  store double 0.000000e+00, double* %c
  %i = alloca double, i64 0
  store double 3.000000e+00, double* %i
  br label %loop

loop:                                             ; preds = %loop, %entry
  %c2 = load double, double* %c
  store double %c2, double* %c
  %a3 = load double, double* %a
  store double %a3, double* %a
  %binaryOp = call double @"binary:"(double %c2, double %a3)
  %b4 = load double, double* %b
  store double %b4, double* %b
  %binaryOp5 = call double @"binary:"(double %binaryOp, double %b4)
  %i6 = load double, double* %i
  %x7 = load double, double* %x1
  %boolCmp = fcmp olt double %i6, %x7
  %loopCond = icmp ne i1 %boolCmp, false
  %0 = load double, double* %i
  %nextVal = fadd double %0, 1.000000e+00
  store double %nextVal, double* %i
  br i1 %loopCond, label %loop, label %afterLoop

afterLoop:                                        ; preds = %loop
  %b8 = load double, double* %b
  %binaryOp9 = call double @"binary:"(double 0.000000e+00, double %b8)
  ret double %binaryOp9
}
Read top-level expression:

define double @__anon_expr() {
entry:
  %call = call double @fibi(double 1.000000e+01)
  ret double %call
}
0.0