class NumberExprAST : public ExprAST {
double Val;
public:
- NumberExprAST(double val) : Val(val) {}
+ explicit NumberExprAST(double val) : Val(val) {}
virtual Value *Codegen();
};
...
</pre>
</div>
-<p>"Value" is the class used to represent a "register" in LLVM. The Codegen()
-method says to emit IR for that AST node and all things it depends on. The
+<p>The Codegen() method says to emit IR for that AST node and all things it
+depends on, and they all return an LLVM Value object.
+"Value" is the class used to represent a "<a
+href="http://en.wikipedia.org/wiki/Static_single_assignment_form">Static Single
+Assignment (SSA)</a> register" or "SSA value" in LLVM. The most distinct aspect
+of SSA values is that their value is computed as the related instruction
+executes, and it does not get a new value until (and if) the instruction
+re-executes. In order words, there is no way to "change" an SSA value. For
+more information, please read up on <a
+href="http://en.wikipedia.org/wiki/Static_single_assignment_form">Static Single
+Assignment</a> - the concepts are really quite natural once you grok them.</p>
+
+<p>The
second thing we want is an "Error" method like we used for parser, which will
be used to report errors found during code generation (for example, use of an
undeclared parameter):</p>
</pre>
</div>
-<p>In the LLVM IR, numeric constants are represented with the ConstantFP class,
-which holds the numeric value in an APFloat internally (APFloat has the
-capability of holding floating point constants of arbitrary precision). This
-code basically just creates and returns a ConstantFP. Note that in the LLVM IR
+<p>In the LLVM IR, numeric constants are represented with the
+<tt>ConstantFP</tt> class, which holds the numeric value in an <tt>APFloat</tt>
+internally (<tt>APFloat</tt> has the capability of holding floating point
+constants of <em>A</em>rbitrary <em>P</em>recision). This code basically just
+creates and returns a <tt>ConstantFP</tt>. Note that in the LLVM IR
that constants are all uniqued together and shared. For this reason, the API
-uses "the foo::get(...)" idiom instead of a "create" method or "new foo".</p>
+uses "the foo::get(..)" idiom instead of "new foo(..)" or "foo::create(..).</p>
<div class="doc_code">
<pre>
</pre>
</div>
-<p>References to variables is also quite simple here. In our system, we assume
-that the variable has already been emited somewhere and its value is available.
-In practice, the only values in the NamedValues map will be arguments. This
+<p>References to variables is also quite simple here. In the simple version
+of Kaleidoscope, we assume that the variable has already been emited somewhere
+and its value is available. In practice, the only values that can be in the
+<tt>NamedValues</tt> map are function arguments. This
code simply checks to see that the specified name is in the map (if not, an
unknown variable is being referenced) and returns the value for it.</p>
</pre>
</div>
+<p>Binary operators start to get more interesting. The basic idea here is that
+we recursively emit code for the left-hand side of the expression, then the
+right-hand side, then we compute the result of the binary expression. In this
+code, we do a simple switch on the opcode to create the right LLVM instruction.
+</p>
+<p>In this example, the LLVM builder class is starting to show its value.
+Because it knows where to insert the newly created instruction, you just have to
+specificy what instruction to create (e.g. with <tt>CreateAdd</tt>), which
+operands to use (<tt>L</tt> and <tt>R</tt> here) and optionally provide a name
+for the generated instruction. One nice thing about LLVM is that the name is
+just a hint: if there are multiple additions in a single function, the first
+will be named "addtmp" and the second will be "autorenamed" by adding a suffix,
+giving it a name like "addtmp42". Local value names for instructions are purely
+optional, but it makes it much easier to read the IR dumps.</p>
+
+<p><a href="../LangRef.html#instref">LLVM instructions</a> are constrained to
+have very strict type properties: for example, the Left and Right operators of
+an <a href="../LangRef.html#i_add">add instruction</a> have to have the same
+type, and that the result of the add matches the operands. Because all values
+in Kaleidoscope are doubles, this makes for very simple code for add, sub and
+mul.</p>
+
+<p>On the other hand, LLVM specifies that the <a
+href="../LangRef.html#i_fcmp">fcmp instruction</a> always returns an 'i1' value
+(a one bit integer). However, Kaleidoscope wants the value to be a 0.0 or 1.0
+value. In order to get these semantics, we combine the fcmp instruction with
+a <a href="../LangRef.html#i_uitofp">uitofp instruction</a>. This instruction
+converts its input integer into a floating point value by treating the input
+as an unsigned value. In contrast, if we used the <a
+href="../LangRef.html#i_sitofp">sitofp instruction</a>, the Kaleidoscope '<'
+operator would return 0.0 and -1.0, depending on the input value.</p>
<div class="doc_code">
<pre>
</pre>
</div>
-<h1> more todo</h1>
+<p>Code generation for function calls is quite straight-forward with LLVM. The
+code above first looks the name of the function up in the LLVM Module's symbol
+table. Recall that the LLVM Module is the container that holds all of the
+functions we are JIT'ing. By giving each function the same name as what the
+user specifies, we can use the LLVM symbol table to resolve function names for
+us.</p>
+
+<p>Once we have the function to call, we recursively codegen each argument that
+is to be passed in, and create an LLVM <a href="../LangRef.html#i_call">call
+instruction</a>. Note that LLVM uses the native C calling conventions by
+default, allowing these calls to call into standard library functions like
+"sin" and "cos" with no additional effort.</p>
+
+<p>This wraps up our handling of the four basic expressions that we have so far
+in Kaleidoscope. Feel free to go in and add some more. For example, by
+browsing the <a href="../LangRef.html">LLVM language reference</a> you'll find
+several other interesting instructions that are really easy to plug into our
+basic framework.</p>
+
+</div>
+
+<!-- *********************************************************************** -->
+<div class="doc_section"><a name="funcs">Function Code Generation</a></div>
+<!-- *********************************************************************** -->
+
+<div class="doc_text">
+
+<p>Code generation for prototypes and functions has to handle a number of
+details, which make their code less beautiful and elegant than expression code
+generation, but they illustrate some important points. First, lets talk about
+code generation for prototypes: this is used both for function bodies as well
+as external function declarations. The code starts with:</p>
+
+<div class="doc_code">
+<pre>
+Function *PrototypeAST::Codegen() {
+ // Make the function type: double(double,double) etc.
+ std::vector<const Type*> Doubles(Args.size(), Type::DoubleTy);
+ FunctionType *FT = FunctionType::get(Type::DoubleTy, Doubles, false);
+
+ Function *F = new Function(FT, Function::ExternalLinkage, Name, TheModule);
+</pre>
+</div>
+
+<p>This code packs a lot of power into a few lines. The first step is to create
+the <tt>FunctionType</tt> that should be used for a given Prototype. Since all
+function arguments in Kaleidoscope are of type double, the first line creates
+a vector of "N" LLVM Double types. It then uses the <tt>FunctionType::get</tt>
+method to create a function type that takes "N" doubles as arguments, returns
+one double as a result, and that is not vararg (the false parameter indicates
+this). Note that Types in LLVM are uniqued just like Constants are, so you
+don't "new" a type, you "get" it.</p>
+
+<p>The final line above actually creates the function that the prototype will
+correspond to. This indicates which type, linkage, and name to use, and which
+module to insert into. "<a href="LangRef.html#linkage">external linkage</a>"
+means that the function may be defined outside the current module and/or that it
+is callable by functions outside the module. The Name passed in is the name the
+user specified: since "<tt>TheModule</tt>" is specified, this name is registered
+in "<tt>TheModule</tt>"s symbol table, which is used by the function call code
+above.</p>
+
+<div class="doc_code">
+<pre>
+ // If F conflicted, there was already something named 'Name'. If it has a
+ // body, don't allow redefinition or reextern.
+ if (F->getName() != Name) {
+ // Delete the one we just made and get the existing one.
+ F->eraseFromParent();
+ F = TheModule->getFunction(Name);
+</pre>
+</div>
+
+<p>The Module symbol table works just like the Function symbol table when it
+comes to name conflicts: if a new function is created with a name was previously
+added to the symbol table, it will get implicitly renamed when added to the
+Module. The code above exploits this fact to tell if there was a previous
+definition of this function.</p>
+
+<p>In Kaleidoscope, I choose to allow redefinitions of functions in two cases:
+first, we want to allow 'extern'ing a function more than once, so long as the
+prototypes for the externs match (since all arguments have the same type, we
+just have to check that the number of arguments match). Second, we want to
+allow 'extern'ing a function and then definining a body for it. This is useful
+when defining mutually recursive functions.</p>
+
+<p>In order to implement this, the code above first checks to see if there is
+a collision on the name of the function. If so, it deletes the function we just
+created (by calling <tt>eraseFromParent</tt>) and then calling
+<tt>getFunction</tt> to get the existing function with the specified name. Note
+that many APIs in LLVM have "erase" forms and "remove" forms. The "remove" form
+unlinks the object from its parent (e.g. a Function from a Module) and returns
+it. The "erase" form unlinks the object and then deletes it.</p>
+
+<div class="doc_code">
+<pre>
+ // If F already has a body, reject this.
+ if (!F->empty()) {
+ ErrorF("redefinition of function");
+ return 0;
+ }
+
+ // If F took a different number of args, reject.
+ if (F->arg_size() != Args.size()) {
+ ErrorF("redefinition of function with different # args");
+ return 0;
+ }
+ }
+</pre>
+</div>
+
+<p>In order to verify the logic above, we first check to see if the preexisting
+function is "empty". In this case, empty means that it has no basic blocks in
+it, which means it has no body. If it has no body, this means its a forward
+declaration. Since we don't allow anything after a full definition of the
+function, the code rejects this case. If the previous reference to a function
+was an 'extern', we simply verify that the number of arguments for that
+definition and this one match up. If not, we emit an error.</p>
+
+<div class="doc_code">
+<pre>
+ // Set names for all arguments.
+ unsigned Idx = 0;
+ for (Function::arg_iterator AI = F->arg_begin(); Idx != Args.size();
+ ++AI, ++Idx) {
+ AI->setName(Args[Idx]);
+
+ // Add arguments to variable symbol table.
+ NamedValues[Args[Idx]] = AI;
+ }
+ return F;
+}
+</pre>
+</div>
+
+<p>The last bit of code for prototypes loops over all of the arguments in the
+function, setting the name of the LLVM Argument objects to match and registering
+the arguments in the <tt>NamedValues</tt> map for future use by the
+<tt>VariableExprAST</tt> AST node. Once this is set up, it returns the Function
+object to the caller. Note that we don't check for conflicting
+argument names here (e.g. "extern foo(a b a)"). Doing so would be very
+straight-forward.</p>
+
+<div class="doc_code">
+<pre>
+Function *FunctionAST::Codegen() {
+ NamedValues.clear();
+
+ Function *TheFunction = Proto->Codegen();
+ if (TheFunction == 0)
+ return 0;
+</pre>
+</div>
+
+<p>Code generation for function definitions starts out simply enough: first we
+codegen the prototype and verify that it is ok. We also clear out the
+<tt>NamedValues</tt> map to make sure that there isn't anything in it from the
+last function we compiled.</p>
+
+<div class="doc_code">
+<pre>
+ // Create a new basic block to start insertion into.
+ BasicBlock *BB = new BasicBlock("entry", TheFunction);
+ Builder.SetInsertPoint(BB);
+
+ if (Value *RetVal = Body->Codegen()) {
+ // Finish off the function.
+ Builder.CreateRet(RetVal);
+ return TheFunction;
+ }
+</pre>
+</div>
+
+<p>Now we get to the point where the <tt>Builder</tt> is set up. The first
+line creates a new <a href="http://en.wikipedia.org/wiki/Basic_block">basic
+block</a> (named "entry"), which is inserted into <tt>TheFunction</tt>. The
+second line then tells the builder that new instructions should be inserted into
+the end of the new basic block. Basic blocks in LLVM are an important part
+of functions that define the <a
+href="http://en.wikipedia.org/wiki/Control_flow_graph">Control Flow Graph</a>.
+Since we don't have any control flow, our functions will only contain one
+block so far. We'll fix this in a future installment :).</p>
+
+<p>Once the insertion point is set up, we call the <tt>CodeGen()</tt> method for
+the root expression of the function. If no error happens, this emits code to
+compute the expression into the entry block and returns the value that was
+computed. Assuming no error, we then create an LLVM <a
+href="../LangRef.html#i_ret">ret instruction</a>. This completes the function,
+which is then returned.</p>
+
+<div class="doc_code">
+<pre>
+ // Error reading body, remove function.
+ TheFunction->eraseFromParent();
+ return 0;
+}
+</pre>
+</div>
+
+<p>The only piece left here is handling of the error case. For simplicity, we
+simply handle this by deleting the function we produced with the
+<tt>eraseFromParent</tt> method. This allows the user to redefine a function
+that they incorrectly typed in before: if we didn't delete it, it would live in
+the symbol table, with a body, preventing future redefinition.</p>
+
+<p>This code does have a bug though. Since the <tt>PrototypeAST::Codegen</tt>
+can return a previously defined forward declaration, this can actually delete
+a forward declaration. There are a number of ways to fix this bug, see what you
+can come up with! Here is a testcase:</p>
+
+<div class="doc_code">
+<pre>
+extern foo(a b); # ok, defines foo.
+def foo(a b) c; # error, 'c' is invalid.
+def bar() foo(1, 2); # error, unknown function "foo"
+</pre>
+</div>
</div>
<!-- *********************************************************************** -->
-<div class="doc_section"><a name="code">Conclusions and the Full Code</a></div>
+<div class="doc_section"><a name="driver">Driver Changes and
+Closing Thoughts</a></div>
<!-- *********************************************************************** -->
<div class="doc_text">
+<p>
+For now, code generation to LLVM doesn't really get us much, except that we can
+look at the pretty IR calls. The sample code inserts calls to Codegen into the
+"<tt>HandleDefinition</tt>", "<tt>HandleExtern</tt>" etc functions, and then
+dumps out the LLVM IR. This gives a nice way to look at the LLVM IR for simple
+functions. For example:
+</p>
+
+<div class="doc_code">
+<pre>
+ready> <b>4+5</b>;
+ready> Read top-level expression:
+define double @""() {
+entry:
+ %addtmp = add double 4.000000e+00, 5.000000e+00
+ ret double %addtmp
+}
+</pre>
+</div>
+
+<p>Note how the parser turns the top-level expression into anonymous functions
+for us. This will be handy when we add JIT support in the next chapter. Also
+note that the code is very literally transcribed, no optimizations are being
+performed. We will add optimizations explicitly in the next chapter.</p>
+
+<div class="doc_code">
+<pre>
+ready> <b>def foo(a b) a*a + 2*a*b + b*b;</b>
+ready> Read function definition:
+define double @foo(double %a, double %b) {
+entry:
+ %multmp = mul double %a, %a
+ %multmp1 = mul double 2.000000e+00, %a
+ %multmp2 = mul double %multmp1, %b
+ %addtmp = add double %multmp, %multmp2
+ %multmp3 = mul double %b, %b
+ %addtmp4 = add double %addtmp, %multmp3
+ ret double %addtmp4
+}
+</pre>
+</div>
+
+<p>This shows some simple arithmetic. Notice the striking similarity to the
+LLVM builder calls that we use to create the instructions.</p>
+
+<div class="doc_code">
+<pre>
+ready> <b>def bar(a) foo(a, 4.0) + bar(31337);</b>
+ready> Read function definition:
+define double @bar(double %a) {
+entry:
+ %calltmp = call double @foo( double %a, double 4.000000e+00 )
+ %calltmp1 = call double @bar( double 3.133700e+04 )
+ %addtmp = add double %calltmp, %calltmp1
+ ret double %addtmp
+}
+</pre>
+</div>
+
+<p>This shows some function calls. Note that the runtime of this function might
+be fairly high. In the future we'll add conditional control flow to make
+recursion actually be useful :).</p>
+
+<div class="doc_code">
+<pre>
+ready> <b>extern cos(x);</b>
+ready> Read extern:
+declare double @cos(double)
+
+ready> <b>cos(1.234);</b>
+ready> Read top-level expression:
+define double @""() {
+entry:
+ %calltmp = call double @cos( double 1.234000e+00 ) ; <double> [#uses=1]
+ ret double %calltmp
+}
+</pre>
+</div>
+
+<p>This shows an extern for the libm "cos" function, and a call to it.</p>
+
+
+<div class="doc_code">
+<pre>
+ready> <b>^D</b>
+; ModuleID = 'my cool jit'
+
+define double @""() {
+entry:
+ %addtmp = add double 4.000000e+00, 5.000000e+00
+ ret double %addtmp
+}
+
+define double @foo(double %a, double %b) {
+entry:
+ %multmp = mul double %a, %a
+ %multmp1 = mul double 2.000000e+00, %a
+ %multmp2 = mul double %multmp1, %b
+ %addtmp = add double %multmp, %multmp2
+ %multmp3 = mul double %b, %b
+ %addtmp4 = add double %addtmp, %multmp3
+ ret double %addtmp4
+}
+
+define double @bar(double %a) {
+entry:
+ %calltmp = call double @foo( double %a, double 4.000000e+00 )
+ %calltmp1 = call double @bar( double 3.133700e+04 )
+ %addtmp = add double %calltmp, %calltmp1
+ ret double %addtmp
+}
+
+declare double @cos(double)
+
+define double @""() {
+entry:
+ %calltmp = call double @cos( double 1.234000e+00 )
+ ret double %calltmp
+}
+</pre>
+</div>
+
+<p>When you quit the current demo, it dumps out the IR for the entire module
+generated. Here you can see the big picture with all the functions referencing
+each other.</p>
+
+<p>This wraps up this chapter of the Kaleidoscope tutorial. Up next we'll
+describe how to <a href="LangImpl4.html">add JIT codegen and optimizer
+support</a> to this so we can actually start running code!</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 the
+LLVM code generator. Because this uses the LLVM libraries, we need to link
+them in. To do this, we use the <a
+href="http://llvm.org/cmds/llvm-config.html">llvm-config</a> tool to inform
+our makefile/command line about which options to use:</p>
+
+<div class="doc_code">
+<pre>
+ # Compile
+ g++ -g toy.cpp `llvm-config --cppflags` `llvm-config --ldflags` \
+ `llvm-config --libs core` -I ~/llvm/include/ -o toy
+ # Run
+ ./toy
+</pre>
+</div>
+
+<p>Here is the code:</p>
+
<div class="doc_code">
<pre>
// To build this:
-// g++ -g toy.cpp `llvm-config --cppflags` `llvm-config --ldflags` \
-// `llvm-config --libs core` -I ~/llvm/include/
-// ./a.out
// See example below.
#include "llvm/DerivedTypes.h"
class NumberExprAST : public ExprAST {
double Val;
public:
- NumberExprAST(double val) : Val(val) {}
+ explicit NumberExprAST(double val) : Val(val) {}
virtual Value *Codegen();
};
class VariableExprAST : public ExprAST {
std::string Name;
public:
- VariableExprAST(const std::string &name) : Name(name) {}
+ explicit VariableExprAST(const std::string &name) : Name(name) {}
virtual Value *Codegen();
};
Function *PrototypeAST::Codegen() {
// Make the function type: double(double,double) etc.
- FunctionType *FT =
- FunctionType::get(Type::DoubleTy, std::vector<const Type*>(Args.size(),
- Type::DoubleTy),
- false);
+ std::vector<const Type*> Doubles(Args.size(), Type::DoubleTy);
+ FunctionType *FT = FunctionType::get(Type::DoubleTy, Doubles, false);
Function *F = new Function(FT, Function::ExternalLinkage, Name, TheModule);
return 0;
// Create a new basic block to start insertion into.
- Builder.SetInsertPoint(new BasicBlock("entry", TheFunction));
+ BasicBlock *BB = new BasicBlock("entry", TheFunction);
+ Builder.SetInsertPoint(BB);
if (Value *RetVal = Body->Codegen()) {
// Finish off the function.
TheModule->dump();
return 0;
}
-
-/* Examples:
-
-def fib(x)
- if (x < 3) then
- 1
- else
- fib(x-1)+fib(x-2);
-
-fib(10);
-
-*/
</pre>
</div>
</div>
projects / oota-llvm.git / blobdiff