diff options
Diffstat (limited to 'docs/tutorial/LangImpl7.html')
| -rw-r--r-- | docs/tutorial/LangImpl7.html | 2164 |
1 files changed, 2164 insertions, 0 deletions
diff --git a/docs/tutorial/LangImpl7.html b/docs/tutorial/LangImpl7.html new file mode 100644 index 0000000000..0b46ba58ec --- /dev/null +++ b/docs/tutorial/LangImpl7.html @@ -0,0 +1,2164 @@ +<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" + "http://www.w3.org/TR/html4/strict.dtd"> + +<html> +<head> + <title>Kaleidoscope: Extending the Language: Mutable Variables / SSA + construction</title> + <meta http-equiv="Content-Type" content="text/html; charset=utf-8"> + <meta name="author" content="Chris Lattner"> + <link rel="stylesheet" href="../llvm.css" type="text/css"> +</head> + +<body> + +<div class="doc_title">Kaleidoscope: Extending the Language: Mutable Variables</div> + +<ul> +<li><a href="index.html">Up to Tutorial Index</a></li> +<li>Chapter 7 + <ol> + <li><a href="#intro">Chapter 7 Introduction</a></li> + <li><a href="#why">Why is this a hard problem?</a></li> + <li><a href="#memory">Memory in LLVM</a></li> + <li><a href="#kalvars">Mutable Variables in Kaleidoscope</a></li> + <li><a href="#adjustments">Adjusting Existing Variables for + Mutation</a></li> + <li><a href="#assignment">New Assignment Operator</a></li> + <li><a href="#localvars">User-defined Local Variables</a></li> + <li><a href="#code">Full Code Listing</a></li> + </ol> +</li> +<li><a href="LangImpl8.html">Chapter 8</a>: Conclusion and other useful LLVM + tidbits</li> +</ul> + +<div class="doc_author"> + <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p> +</div> + +<!-- *********************************************************************** --> +<div class="doc_section"><a name="intro">Chapter 7 Introduction</a></div> +<!-- *********************************************************************** --> + +<div class="doc_text"> + +<p>Welcome to Chapter 7 of the "<a href="index.html">Implementing a language +with LLVM</a>" tutorial. In chapters 1 through 6, we've built a very +respectable, albeit simple, <a +href="http://en.wikipedia.org/wiki/Functional_programming">functional +programming language</a>. In our journey, we learned some parsing techniques, +how to build and represent an AST, how to build LLVM IR, and how to optimize +the resultant code as well as JIT compile it.</p> + +<p>While Kaleidoscope is interesting as a functional language, the fact that it +is functional makes it "too easy" to generate LLVM IR for it. In particular, a +functional language makes it very easy to build LLVM IR directly in <a +href="http://en.wikipedia.org/wiki/Static_single_assignment_form">SSA form</a>. +Since LLVM requires that the input code be in SSA form, this is a very nice +property and it is often unclear to newcomers how to generate code for an +imperative language with mutable variables.</p> + +<p>The short (and happy) summary of this chapter is that there is no need for +your front-end to build SSA form: LLVM provides highly tuned and well tested +support for this, though the way it works is a bit unexpected for some.</p> + +</div> + +<!-- *********************************************************************** --> +<div class="doc_section"><a name="why">Why is this a hard problem?</a></div> +<!-- *********************************************************************** --> + +<div class="doc_text"> + +<p> +To understand why mutable variables cause complexities in SSA construction, +consider this extremely simple C example: +</p> + +<div class="doc_code"> +<pre> +int G, H; +int test(_Bool Condition) { + int X; + if (Condition) + X = G; + else + X = H; + return X; +} +</pre> +</div> + +<p>In this case, we have the variable "X", whose value depends on the path +executed in the program. Because there are two different possible values for X +before the return instruction, a PHI node is inserted to merge the two values. +The LLVM IR that we want for this example looks like this:</p> + +<div class="doc_code"> +<pre> +@G = weak global i32 0 ; type of @G is i32* +@H = weak global i32 0 ; type of @H is i32* + +define i32 @test(i1 %Condition) { +entry: + br i1 %Condition, label %cond_true, label %cond_false + +cond_true: + %X.0 = load i32* @G + br label %cond_next + +cond_false: + %X.1 = load i32* @H + br label %cond_next + +cond_next: + %X.2 = phi i32 [ %X.1, %cond_false ], [ %X.0, %cond_true ] + ret i32 %X.2 +} +</pre> +</div> + +<p>In this example, the loads from the G and H global variables are explicit in +the LLVM IR, and they live in the then/else branches of the if statement +(cond_true/cond_false). In order to merge the incoming values, the X.2 phi node +in the cond_next block selects the right value to use based on where control +flow is coming from: if control flow comes from the cond_false block, X.2 gets +the value of X.1. Alternatively, if control flow comes from cond_true, it gets +the value of X.0. The intent of this chapter is not to explain the details of +SSA form. For more information, see one of the many <a +href="http://en.wikipedia.org/wiki/Static_single_assignment_form">online +references</a>.</p> + +<p>The question for this article is "who places the phi nodes when lowering +assignments to mutable variables?". The issue here is that LLVM +<em>requires</em> that its IR be in SSA form: there is no "non-ssa" mode for it. +However, SSA construction requires non-trivial algorithms and data structures, +so it is inconvenient and wasteful for every front-end to have to reproduce this +logic.</p> + +</div> + +<!-- *********************************************************************** --> +<div class="doc_section"><a name="memory">Memory in LLVM</a></div> +<!-- *********************************************************************** --> + +<div class="doc_text"> + +<p>The 'trick' here is that while LLVM does require all register values to be +in SSA form, it does not require (or permit) memory objects to be in SSA form. +In the example above, note that the loads from G and H are direct accesses to +G and H: they are not renamed or versioned. This differs from some other +compiler systems, which do try to version memory objects. In LLVM, instead of +encoding dataflow analysis of memory into the LLVM IR, it is handled with <a +href="../WritingAnLLVMPass.html">Analysis Passes</a> which are computed on +demand.</p> + +<p> +With this in mind, the high-level idea is that we want to make a stack variable +(which lives in memory, because it is on the stack) for each mutable object in +a function. To take advantage of this trick, we need to talk about how LLVM +represents stack variables. +</p> + +<p>In LLVM, all memory accesses are explicit with load/store instructions, and +it is carefully designed not to have (or need) an "address-of" operator. Notice +how the type of the @G/@H global variables is actually "i32*" even though the +variable is defined as "i32". What this means is that @G defines <em>space</em> +for an i32 in the global data area, but its <em>name</em> actually refers to the +address for that space. Stack variables work the same way, except that instead of +being declared with global variable definitions, they are declared with the +<a href="../LangRef.html#i_alloca">LLVM alloca instruction</a>:</p> + +<div class="doc_code"> +<pre> +define i32 @example() { +entry: + %X = alloca i32 ; type of %X is i32*. + ... + %tmp = load i32* %X ; load the stack value %X from the stack. + %tmp2 = add i32 %tmp, 1 ; increment it + store i32 %tmp2, i32* %X ; store it back + ... +</pre> +</div> + +<p>This code shows an example of how you can declare and manipulate a stack +variable in the LLVM IR. Stack memory allocated with the alloca instruction is +fully general: you can pass the address of the stack slot to functions, you can +store it in other variables, etc. In our example above, we could rewrite the +example to use the alloca technique to avoid using a PHI node:</p> + +<div class="doc_code"> +<pre> +@G = weak global i32 0 ; type of @G is i32* +@H = weak global i32 0 ; type of @H is i32* + +define i32 @test(i1 %Condition) { +entry: + %X = alloca i32 ; type of %X is i32*. + br i1 %Condition, label %cond_true, label %cond_false + +cond_true: + %X.0 = load i32* @G + store i32 %X.0, i32* %X ; Update X + br label %cond_next + +cond_false: + %X.1 = load i32* @H + store i32 %X.1, i32* %X ; Update X + br label %cond_next + +cond_next: + %X.2 = load i32* %X ; Read X + ret i32 %X.2 +} +</pre> +</div> + +<p>With this, we have discovered a way to handle arbitrary mutable variables +without the need to create Phi nodes at all:</p> + +<ol> +<li>Each mutable variable becomes a stack allocation.</li> +<li>Each read of the variable becomes a load from the stack.</li> +<li>Each update of the variable becomes a store to the stack.</li> +<li>Taking the address of a variable just uses the stack address directly.</li> +</ol> + +<p>While this solution has solved our immediate problem, it introduced another +one: we have now apparently introduced a lot of stack traffic for very simple +and common operations, a major performance problem. Fortunately for us, the +LLVM optimizer has a highly-tuned optimization pass named "mem2reg" that handles +this case, promoting allocas like this into SSA registers, inserting Phi nodes +as appropriate. If you run this example through the pass, for example, you'll +get:</p> + +<div class="doc_code"> +<pre> +$ <b>llvm-as < example.ll | opt -mem2reg | llvm-dis</b> +@G = weak global i32 0 +@H = weak global i32 0 + +define i32 @test(i1 %Condition) { +entry: + br i1 %Condition, label %cond_true, label %cond_false + +cond_true: + %X.0 = load i32* @G + br label %cond_next + +cond_false: + %X.1 = load i32* @H + br label %cond_next + +cond_next: + %X.01 = phi i32 [ %X.1, %cond_false ], [ %X.0, %cond_true ] + ret i32 %X.01 +} +</pre> +</div> + +<p>The mem2reg pass implements the standard "iterated dominance frontier" +algorithm for constructing SSA form and has a number of optimizations that speed +up (very common) degenerate cases. The mem2reg optimization pass is the answer to dealing +with mutable variables, and we highly recommend that you depend on it. Note that +mem2reg only works on variables in certain circumstances:</p> + +<ol> +<li>mem2reg is alloca-driven: it looks for allocas and if it can handle them, it +promotes them. It does not apply to global variables or heap allocations.</li> + +<li>mem2reg only looks for alloca instructions in the entry block of the +function. Being in the entry block guarantees that the alloca is only executed +once, which makes analysis simpler.</li> + +<li>mem2reg only promotes allocas whose uses are direct loads and stores. If +the address of the stack object is passed to a function, or if any funny pointer +arithmetic is involved, the alloca will not be promoted.</li> + +<li>mem2reg only works on allocas of <a +href="../LangRef.html#t_classifications">first class</a> +values (such as pointers, scalars and vectors), and only if the array size +of the allocation is 1 (or missing in the .ll file). mem2reg is not capable of +promoting structs or arrays to registers. Note that the "scalarrepl" pass is +more powerful and can promote structs, "unions", and arrays in many cases.</li> + +</ol> + +<p> +All of these properties are easy to satisfy for most imperative languages, and +we'll illustrate it below with Kaleidoscope. The final question you may be +asking is: should I bother with this nonsense for my front-end? Wouldn't it be +better if I just did SSA construction directly, avoiding use of the mem2reg +optimization pass? In short, we strongly recommend that you use this technique +for building SSA form, unless there is an extremely good reason not to. Using +this technique is:</p> + +<ul> +<li>Proven and well tested: llvm-gcc and clang both use this technique for local +mutable variables. As such, the most common clients of LLVM are using this to +handle a bulk of their variables. You can be sure that bugs are found fast and +fixed early.</li> + +<li>Extremely Fast: mem2reg has a number of special cases that make it fast in +common cases as well as fully general. For example, it has fast-paths for +variables that are only used in a single block, variables that only have one +assignment point, good heuristics to avoid insertion of unneeded phi nodes, etc. +</li> + +<li>Needed for debug info generation: <a href="../SourceLevelDebugging.html"> +Debug information in LLVM</a> relies on having the address of the variable +exposed so that debug info can be attached to it. This technique dovetails +very naturally with this style of debug info.</li> +</ul> + +<p>If nothing else, this makes it much easier to get your front-end up and +running, and is very simple to implement. Lets extend Kaleidoscope with mutable +variables now! +</p> + +</div> + +<!-- *********************************************************************** --> +<div class="doc_section"><a name="kalvars">Mutable Variables in +Kaleidoscope</a></div> +<!-- *********************************************************************** --> + +<div class="doc_text"> + +<p>Now that we know the sort of problem we want to tackle, lets see what this +looks like in the context of our little Kaleidoscope language. We're going to +add two features:</p> + +<ol> +<li>The ability to mutate variables with the '=' operator.</li> +<li>The ability to define new variables.</li> +</ol> + +<p>While the first item is really what this is about, we only have variables +for incoming arguments as well as for induction variables, and redefining those only +goes so far :). Also, the ability to define new variables is a +useful thing regardless of whether you will be mutating them. Here's a +motivating example that shows how we could use these:</p> + +<div class="doc_code"> +<pre> +# Define ':' for sequencing: as a low-precedence operator that ignores operands +# and just returns the RHS. +def binary : 1 (x y) y; + +# Recursive fib, we could do this before. +def fib(x) + if (x < 3) then + 1 + else + fib(x-1)+fib(x-2); + +# Iterative fib. +def fibi(x) + <b>var a = 1, b = 1, c in</b> + (for i = 3, i < x in + <b>c = a + b</b> : + <b>a = b</b> : + <b>b = c</b>) : + b; + +# Call it. +fibi(10); +</pre> +</div> + +<p> +In order to mutate variables, we have to change our existing variables to use +the "alloca trick". Once we have that, we'll add our new operator, then extend +Kaleidoscope to support new variable definitions. +</p> + +</div> + +<!-- *********************************************************************** --> +<div class="doc_section"><a name="adjustments">Adjusting Existing Variables for +Mutation</a></div> +<!-- *********************************************************************** --> + +<div class="doc_text"> + +<p> +The symbol table in Kaleidoscope is managed at code generation time by the +'<tt>NamedValues</tt>' map. This map currently keeps track of the LLVM "Value*" +that holds the double value for the named variable. In order to support +mutation, we need to change this slightly, so that it <tt>NamedValues</tt> holds +the <em>memory location</em> of the variable in question. Note that this +change is a refactoring: it changes the structure of the code, but does not +(by itself) change the behavior of the compiler. All of these changes are +isolated in the Kaleidoscope code generator.</p> + +<p> +At this point in Kaleidoscope's development, it only supports variables for two +things: incoming arguments to functions and the induction variable of 'for' +loops. For consistency, we'll allow mutation of these variables in addition to +other user-defined variables. This means that these will both need memory +locations. +</p> + +<p>To start our transformation of Kaleidoscope, we'll change the NamedValues +map so that it maps to AllocaInst* instead of Value*. Once we do this, the C++ +compiler will tell us what parts of the code we need to update:</p> + +<div class="doc_code"> +<pre> +static std::map<std::string, AllocaInst*> NamedValues; +</pre> +</div> + +<p>Also, since we will need to create these alloca's, we'll use a helper +function that ensures that the allocas are created in the entry block of the +function:</p> + +<div class="doc_code"> +<pre> +/// CreateEntryBlockAlloca - Create an alloca instruction in the entry block of +/// the function. This is used for mutable variables etc. +static AllocaInst *CreateEntryBlockAlloca(Function *TheFunction, + const std::string &VarName) { + IRBuilder<> TmpB(&TheFunction->getEntryBlock(), + TheFunction->getEntryBlock().begin()); + return TmpB.CreateAlloca(Type::getDoubleTy(getGlobalContext()), 0, + VarName.c_str()); +} +</pre> +</div> + +<p>This funny looking code creates an IRBuilder object that is pointing at +the first instruction (.begin()) of the entry block. It then creates an alloca +with the expected name and returns it. Because all values in Kaleidoscope are +doubles, there is no need to pass in a type to use.</p> + +<p>With this in place, the first functionality change we want to make is to +variable references. In our new scheme, variables live on the stack, so code +generating a reference to them actually needs to produce a load from the stack +slot:</p> + +<div class="doc_code"> +<pre> +Value *VariableExprAST::Codegen() { + // Look this variable up in the function. + Value *V = NamedValues[Name]; + if (V == 0) return ErrorV("Unknown variable name"); + + <b>// Load the value. + return Builder.CreateLoad(V, Name.c_str());</b> +} +</pre> +</div> + +<p>As you can see, this is pretty straightforward. Now we need to update the +things that define the variables to set up the alloca. We'll start with +<tt>ForExprAST::Codegen</tt> (see the <a href="#code">full code listing</a> for +the unabridged code):</p> + +<div class="doc_code"> +<pre> + Function *TheFunction = Builder.GetInsertBlock()->getParent(); + + <b>// Create an alloca for the variable in the entry block. + AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);</b> + + // Emit the start code first, without 'variable' in scope. + Value *StartVal = Start->Codegen(); + if (StartVal == 0) return 0; + + <b>// Store the value into the alloca. + Builder.CreateStore(StartVal, Alloca);</b> + ... + + // Compute the end condition. + Value *EndCond = End->Codegen(); + if (EndCond == 0) return EndCond; + + <b>// Reload, increment, and restore the alloca. This handles the case where + // the body of the loop mutates the variable. + Value *CurVar = Builder.CreateLoad(Alloca); + Value *NextVar = Builder.CreateAdd(CurVar, StepVal, "nextvar"); + Builder.CreateStore(NextVar, Alloca);</b> + ... +</pre> +</div> + +<p>This code is virtually identical to the code <a +href="LangImpl5.html#forcodegen">before we allowed mutable variables</a>. The +big difference is that we no longer have to construct a PHI node, and we use +load/store to access the variable as needed.</p> + +<p>To support mutable argument variables, we need to also make allocas for them. +The code for this is also pretty simple:</p> + +<div class="doc_code"> +<pre> +/// CreateArgumentAllocas - Create an alloca for each argument and register the +/// argument in the symbol table so that references to it will succeed. +void PrototypeAST::CreateArgumentAllocas(Function *F) { + Function::arg_iterator AI = F->arg_begin(); + for (unsigned Idx = 0, e = Args.size(); Idx != e; ++Idx, ++AI) { + // Create an alloca for this variable. + AllocaInst *Alloca = CreateEntryBlockAlloca(F, Args[Idx]); + + // Store the initial value into the alloca. + Builder.CreateStore(AI, Alloca); + + // Add arguments to variable symbol table. + NamedValues[Args[Idx]] = Alloca; + } +} +</pre> +</div> + +<p>For each argument, we make an alloca, store the input value to the function +into the alloca, and register the alloca as the memory location for the +argument. This method gets invoked by <tt>FunctionAST::Codegen</tt> right after +it sets up the entry block for the function.</p> + +<p>The final missing piece is adding the mem2reg pass, which allows us to get +good codegen once again:</p> + +<div class="doc_code"> +<pre> + // Set up the optimizer pipeline. Start with registering info about how the + // target lays out data structures. + OurFPM.add(new TargetData(*TheExecutionEngine->getTargetData())); + <b>// Promote allocas to registers. + OurFPM.add(createPromoteMemoryToRegisterPass());</b> + // Do simple "peephole" optimizations and bit-twiddling optzns. + OurFPM.add(createInstructionCombiningPass()); + // Reassociate expressions. + OurFPM.add(createReassociatePass()); +</pre> +</div> + +<p>It is interesting to see what the code looks like before and after the +mem2reg optimization runs. For example, this is the before/after code for our +recursive fib function. Before the optimization:</p> + +<div class="doc_code"> +<pre> +define double @fib(double %x) { +entry: + <b>%x1 = alloca double + store double %x, double* %x1 + %x2 = load double* %x1</b> + %cmptmp = fcmp ult double %x2, 3.000000e+00 + %booltmp = uitofp i1 %cmptmp to double + %ifcond = fcmp one double %booltmp, 0.000000e+00 + br i1 %ifcond, label %then, label %else + +then: ; preds = %entry + br label %ifcont + +else: ; preds = %entry + <b>%x3 = load double* %x1</b> + %subtmp = fsub double %x3, 1.000000e+00 + %calltmp = call double @fib( double %subtmp ) + <b>%x4 = load double* %x1</b> + %subtmp5 = fsub double %x4, 2.000000e+00 + %calltmp6 = call double @fib( double %subtmp5 ) + %addtmp = fadd double %calltmp, %calltmp6 + br label %ifcont + +ifcont: ; preds = %else, %then + %iftmp = phi double [ 1.000000e+00, %then ], [ %addtmp, %else ] + ret double %iftmp +} +</pre> +</div> + +<p>Here there is only one variable (x, the input argument) but you can still +see the extremely simple-minded code generation strategy we are using. In the +entry block, an alloca is created, and the initial input value is stored into +it. Each reference to the variable does a reload from the stack. Also, note +that we didn't modify the if/then/else expression, so it still inserts a PHI +node. While we could make an alloca for it, it is actually easier to create a +PHI node for it, so we still just make the PHI.</p> + +<p>Here is the code after the mem2reg pass runs:</p> + +<div class="doc_code"> +<pre> +define double @fib(double %x) { +entry: + %cmptmp = fcmp ult double <b>%x</b>, 3.000000e+00 + %booltmp = uitofp i1 %cmptmp to double + %ifcond = fcmp one double %booltmp, 0.000000e+00 + br i1 %ifcond, label %then, label %else + +then: + br label %ifcont + +else: + %subtmp = fsub double <b>%x</b>, 1.000000e+00 + %calltmp = call double @fib( double %subtmp ) + %subtmp5 = fsub double <b>%x</b>, 2.000000e+00 + %calltmp6 = call double @fib( double %subtmp5 ) + %addtmp = fadd double %calltmp, %calltmp6 + br label %ifcont + +ifcont: ; preds = %else, %then + %iftmp = phi double [ 1.000000e+00, %then ], [ %addtmp, %else ] + ret double %iftmp +} +</pre> +</div> + +<p>This is a trivial case for mem2reg, since there are no redefinitions of the +variable. The point of showing this is to calm your tension about inserting +such blatent inefficiencies :).</p> + +<p>After the rest of the optimizers run, we get:</p> + +<div class="doc_code"> +<pre> +define double @fib(double %x) { +entry: + %cmptmp = fcmp ult double %x, 3.000000e+00 + %booltmp = uitofp i1 %cmptmp to double + %ifcond = fcmp ueq double %booltmp, 0.000000e+00 + br i1 %ifcond, label %else, label %ifcont + +else: + %subtmp = fsub double %x, 1.000000e+00 + %calltmp = call double @fib( double %subtmp ) + %subtmp5 = fsub double %x, 2.000000e+00 + %calltmp6 = call double @fib( double %subtmp5 ) + %addtmp = fadd double %calltmp, %calltmp6 + ret double %addtmp + +ifcont: + ret double 1.000000e+00 +} +</pre> +</div> + +<p>Here we see that the simplifycfg pass decided to clone the return instruction +into the end of the 'else' block. This allowed it to eliminate some branches +and the PHI node.</p> + +<p>Now that all symbol table references are updated to use stack variables, +we'll add the assignment operator.</p> + +</div> + +<!-- *********************************************************************** --> +<div class="doc_section"><a name="assignment">New Assignment Operator</a></div> +<!-- *********************************************************************** --> + +<div class="doc_text"> + +<p>With our current framework, adding a new assignment operator is really +simple. We will parse it just like any other binary operator, but handle it +internally (instead of allowing the user to define it). The first step is to +set a precedence:</p> + +<div class="doc_code"> +<pre> + int main() { + // Install standard binary operators. + // 1 is lowest precedence. + <b>BinopPrecedence['='] = 2;</b> + BinopPrecedence['<'] = 10; + BinopPrecedence['+'] = 20; + BinopPrecedence['-'] = 20; +</pre> +</div> + +<p>Now that the parser knows the precedence of the binary operator, it takes +care of all the parsing and AST generation. We just need to implement codegen +for the assignment operator. This looks like:</p> + +<div class="doc_code"> +<pre> +Value *BinaryExprAST::Codegen() { + // Special case '=' because we don't want to emit the LHS as an expression. + if (Op == '=') { + // Assignment requires the LHS to be an identifier. + VariableExprAST *LHSE = dynamic_cast<VariableExprAST*>(LHS); + if (!LHSE) + return ErrorV("destination of '=' must be a variable"); +</pre> +</div> + +<p>Unlike the rest of the binary operators, our assignment operator doesn't +follow the "emit LHS, emit RHS, do computation" model. As such, it is handled +as a special case before the other binary operators are handled. The other +strange thing is that it requires the LHS to be a variable. It is invalid to +have "(x+1) = expr" - only things like "x = expr" are allowed. +</p> + +<div class="doc_code"> +<pre> + // Codegen the RHS. + Value *Val = RHS->Codegen(); + if (Val == 0) return 0; + + // Look up the name. + Value *Variable = NamedValues[LHSE->getName()]; + if (Variable == 0) return ErrorV("Unknown variable name"); + + Builder.CreateStore(Val, Variable); + return Val; + } + ... +</pre> +</div> + +<p>Once we have the variable, codegen'ing the assignment is straightforward: +we emit the RHS of the assignment, create a store, and return the computed +value. Returning a value allows for chained assignments like "X = (Y = Z)".</p> + +<p>Now that we have an assignment operator, we can mutate loop variables and +arguments. For example, we can now run code like this:</p> + +<div class="doc_code"> +<pre> +# Function to print a double. +extern printd(x); + +# Define ':' for sequencing: as a low-precedence operator that ignores operands +# and just returns the RHS. +def binary : 1 (x y) y; + +def test(x) + printd(x) : + x = 4 : + printd(x); + +test(123); +</pre> +</div> + +<p>When run, this example prints "123" and then "4", showing that we did +actually mutate the value! Okay, we have now officially implemented our goal: +getting this to work requires SSA construction in the general case. However, +to be really useful, we want the ability to define our own local variables, lets +add this next! +</p> + +</div> + +<!-- *********************************************************************** --> +<div class="doc_section"><a name="localvars">User-defined Local +Variables</a></div> +<!-- *********************************************************************** --> + +<div class="doc_text"> + +<p>Adding var/in is just like any other other extensions we made to +Kaleidoscope: we extend the lexer, the parser, the AST and the code generator. +The first step for adding our new 'var/in' construct is to extend the lexer. +As before, this is pretty trivial, the code looks like this:</p> + +<div class="doc_code"> +<pre> +enum Token { + ... + <b>// var definition + tok_var = -13</b> +... +} +... +static int gettok() { +... + if (IdentifierStr == "in") return tok_in; + if (IdentifierStr == "binary") return tok_binary; + if (IdentifierStr == "unary") return tok_unary; + <b>if (IdentifierStr == "var") return tok_var;</b> + return tok_identifier; +... +</pre> +</div> + +<p>The next step is to define the AST node that we will construct. For var/in, +it looks like this:</p> + +<div class="doc_code"> +<pre> +/// VarExprAST - Expression class for var/in +class VarExprAST : public ExprAST { + std::vector<std::pair<std::string, ExprAST*> > VarNames; + ExprAST *Body; +public: + VarExprAST(const std::vector<std::pair<std::string, ExprAST*> > &varnames, + ExprAST *body) + : VarNames(varnames), Body(body) {} + + virtual Value *Codegen(); +}; +</pre> +</div> + +<p>var/in allows a list of names to be defined all at once, and each name can +optionally have an initializer value. As such, we capture this information in +the VarNames vector. Also, var/in has a body, this body is allowed to access +the variables defined by the var/in.</p> + +<p>With this in place, we can define the parser pieces. The first thing we do is add +it as a primary expression:</p> + +<div class="doc_code"> +<pre> +/// primary +/// ::= identifierexpr +/// ::= numberexpr +/// ::= parenexpr +/// ::= ifexpr +/// ::= forexpr +<b>/// ::= varexpr</b> +static ExprAST *ParsePrimary() { + switch (CurTok) { + default: return Error("unknown token when expecting an expression"); + case tok_identifier: return ParseIdentifierExpr(); + case tok_number: return ParseNumberExpr(); + case '(': return ParseParenExpr(); + case tok_if: return ParseIfExpr(); + case tok_for: return ParseForExpr(); + <b>case tok_var: return ParseVarExpr();</b> + } +} +</pre> +</div> + +<p>Next we define ParseVarExpr:</p> + +<div class="doc_code"> +<pre> +/// varexpr ::= 'var' identifier ('=' expression)? +// (',' identifier ('=' expression)?)* 'in' expression +static ExprAST *ParseVarExpr() { + getNextToken(); // eat the var. + + std::vector<std::pair<std::string, ExprAST*> > VarNames; + + // At least one variable name is required. + if (CurTok != tok_identifier) + return Error("expected identifier after var"); +</pre> +</div> + +<p>The first part of this code parses the list of identifier/expr pairs into the +local <tt>VarNames</tt> vector. + +<div class="doc_code"> +<pre> + while (1) { + std::string Name = IdentifierStr; + getNextToken(); // eat identifier. + + // Read the optional initializer. + ExprAST *Init = 0; + if (CurTok == '=') { + getNextToken(); // eat the '='. + + Init = ParseExpression(); + if (Init == 0) return 0; + } + + VarNames.push_back(std::make_pair(Name, Init)); + + // End of var list, exit loop. + if (CurTok != ',') break; + getNextToken(); // eat the ','. + + if (CurTok != tok_identifier) + return Error("expected identifier list after var"); + } +</pre> +</div> + +<p>Once all the variables are parsed, we then parse the body and create the +AST node:</p> + +<div class="doc_code"> +<pre> + // At this point, we have to have 'in'. + if (CurTok != tok_in) + return Error("expected 'in' keyword after 'var'"); + getNextToken(); // eat 'in'. + + ExprAST *Body = ParseExpression(); + if (Body == 0) return 0; + + return new VarExprAST(VarNames, Body); +} +</pre> +</div> + +<p>Now that we can parse and represent the code, we need to support emission of +LLVM IR for it. This code starts out with:</p> + +<div class="doc_code"> +<pre> +Value *VarExprAST::Codegen() { + std::vector<AllocaInst *> OldBindings; + + Function *TheFunction = Builder.GetInsertBlock()->getParent(); + + // Register all variables and emit their initializer. + for (unsigned i = 0, e = VarNames.size(); i != e; ++i) { + const std::string &VarName = VarNames[i].first; + ExprAST *Init = VarNames[i].second; +</pre> +</div> + +<p>Basically it loops over all the variables, installing them one at a time. +For each variable we put into the symbol table, we remember the previous value +that we replace in OldBindings.</p> + +<div class="doc_code"> +<pre> + // Emit the initializer before adding the variable to scope, this prevents + // the initializer from referencing the variable itself, and permits stuff + // like this: + // var a = 1 in + // var a = a in ... # refers to outer 'a'. + Value *InitVal; + if (Init) { + InitVal = Init->Codegen(); + if (InitVal == 0) return 0; + } else { // If not specified, use 0.0. + InitVal = ConstantFP::get(getGlobalContext(), APFloat(0.0)); + } + + AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName); + Builder.CreateStore(InitVal, Alloca); + + // Remember the old variable binding so that we can restore the binding when + // we unrecurse. + OldBindings.push_back(NamedValues[VarName]); + + // Remember this binding. + NamedValues[VarName] = Alloca; + } +</pre> +</div> + +<p>There are more comments here than code. The basic idea is that we emit the +initializer, create the alloca, then update the symbol table to point to it. +Once all the variables are installed in the symbol table, we evaluate the body +of the var/in expression:</p> + +<div class="doc_code"> +<pre> + // Codegen the body, now that all vars are in scope. + Value *BodyVal = Body->Codegen(); + if (BodyVal == 0) return 0; +</pre> +</div> + +<p>Finally, before returning, we restore the previous variable bindings:</p> + +<div class="doc_code"> +<pre> + // Pop all our variables from scope. + for (unsigned i = 0, e = VarNames.size(); i != e; ++i) + NamedValues[VarNames[i].first] = OldBindings[i]; + + // Return the body computation. + return BodyVal; +} +</pre> +</div> + +<p>The end result of all of this is that we get properly scoped variable +definitions, and we even (trivially) allow mutation of them :).</p> + +<p>With this, we completed what we set out to do. Our nice iterative fib +example from the intro compiles and runs just fine. The mem2reg pass optimizes +all of our stack variables into SSA registers, inserting PHI nodes where needed, +and our front-end remains simple: no "iterated dominance frontier" computation +anywhere in sight.</p> + +</div> + +<!-- *********************************************************************** --> +<div class="doc_section"><a name="code">Full Code Listing</a></div> +<!-- *********************************************************************** --> + +<div class="doc_text"> + +<p> +Here is the complete code listing for our running example, enhanced with mutable +variables and var/in support. To build this example, use: +</p> + +<div class="doc_code"> +<pre> + # Compile + g++ -g toy.cpp `llvm-config --cppflags --ldflags --libs core jit native` -O3 -o toy + |