| blob | 90925a904f83e8059a68a75e2c4d6893a748e48e |
4 <html>
5 <head>
6 <title>Kaleidoscope: Extending the Language: Mutable Variables / SSA
7 construction</title>
11 </head>
13 <body>
17 <ul>
20 <ol>
26 Mutation</a></li>
30 </ol>
31 </li>
33 tidbits</li>
34 </ul>
38 </div>
40 <!-- *********************************************************************** -->
42 <!-- *********************************************************************** -->
48 respectable, albeit simple, <a
50 programming language</a>. In our journey, we learned some parsing techniques,
51 how to build and represent an AST, how to build LLVM IR, and how to optimize
52 the resultant code as well as JIT compile it.</p>
54 <p>While Kaleidoscope is interesting as a functional language, the fact that it
55 is functional makes it "too easy" to generate LLVM IR for it. In particular, a
56 functional language makes it very easy to build LLVM IR directly in <a
58 Since LLVM requires that the input code be in SSA form, this is a very nice
59 property and it is often unclear to newcomers how to generate code for an
60 imperative language with mutable variables.</p>
62 <p>The short (and happy) summary of this chapter is that there is no need for
63 your front-end to build SSA form: LLVM provides highly tuned and well tested
64 support for this, though the way it works is a bit unexpected for some.</p>
66 </div>
68 <!-- *********************************************************************** -->
70 <!-- *********************************************************************** -->
74 <p>
75 To understand why mutable variables cause complexities in SSA construction,
76 consider this extremely simple C example:
77 </p>
80 <pre>
81 int G, H;
82 int test(_Bool Condition) {
83 int X;
84 if (Condition)
85 X = G;
86 else
87 X = H;
88 return X;
89 }
90 </pre>
91 </div>
94 executed in the program. Because there are two different possible values for X
95 before the return instruction, a PHI node is inserted to merge the two values.
96 The LLVM IR that we want for this example looks like this:</p>
99 <pre>
100 @G = weak global i32 0 ; type of @G is i32*
101 @H = weak global i32 0 ; type of @H is i32*
103 define i32 @test(i1 %Condition) {
104 entry:
105 br i1 %Condition, label %cond_true, label %cond_false
107 cond_true:
108 %X.0 = load i32* @G
109 br label %cond_next
111 cond_false:
112 %X.1 = load i32* @H
113 br label %cond_next
115 cond_next:
117 ret i32 %X.2
118 }
119 </pre>
120 </div>
122 <p>In this example, the loads from the G and H global variables are explicit in
123 the LLVM IR, and they live in the then/else branches of the if statement
124 (cond_true/cond_false). In order to merge the incoming values, the X.2 phi node
125 in the cond_next block selects the right value to use based on where control
126 flow is coming from: if control flow comes from the cond_false block, X.2 gets
127 the value of X.1. Alternatively, if control flow comes from cond_true, it gets
128 the value of X.0. The intent of this chapter is not to explain the details of
129 SSA form. For more information, see one of the many <a
134 assignments to mutable variables?". The issue here is that LLVM
136 However, SSA construction requires non-trivial algorithms and data structures,
137 so it is inconvenient and wasteful for every front-end to have to reproduce this
138 logic.</p>
140 </div>
142 <!-- *********************************************************************** -->
144 <!-- *********************************************************************** -->
148 <p>The 'trick' here is that while LLVM does require all register values to be
149 in SSA form, it does not require (or permit) memory objects to be in SSA form.
150 In the example above, note that the loads from G and H are direct accesses to
151 G and H: they are not renamed or versioned. This differs from some other
152 compiler systems, which do try to version memory objects. In LLVM, instead of
153 encoding dataflow analysis of memory into the LLVM IR, it is handled with <a
155 demand.</p>
157 <p>
158 With this in mind, the high-level idea is that we want to make a stack variable
159 (which lives in memory, because it is on the stack) for each mutable object in
160 a function. To take advantage of this trick, we need to talk about how LLVM
161 represents stack variables.
162 </p>
164 <p>In LLVM, all memory accesses are explicit with load/store instructions, and
165 it is carefully designed not to have (or need) an "address-of" operator. Notice
166 how the type of the @G/@H global variables is actually "i32*" even though the
169 address for that space. Stack variables work the same way, except that instead of
170 being declared with global variable definitions, they are declared with the
174 <pre>
175 define i32 @example() {
176 entry:
177 %X = alloca i32 ; type of %X is i32*.
178 ...
179 %tmp = load i32* %X ; load the stack value %X from the stack.
180 %tmp2 = add i32 %tmp, 1 ; increment it
181 store i32 %tmp2, i32* %X ; store it back
182 ...
183 </pre>
184 </div>
186 <p>This code shows an example of how you can declare and manipulate a stack
187 variable in the LLVM IR. Stack memory allocated with the alloca instruction is
188 fully general: you can pass the address of the stack slot to functions, you can
189 store it in other variables, etc. In our example above, we could rewrite the
190 example to use the alloca technique to avoid using a PHI node:</p>
193 <pre>
194 @G = weak global i32 0 ; type of @G is i32*
195 @H = weak global i32 0 ; type of @H is i32*
197 define i32 @test(i1 %Condition) {
198 entry:
199 %X = alloca i32 ; type of %X is i32*.
200 br i1 %Condition, label %cond_true, label %cond_false
202 cond_true:
203 %X.0 = load i32* @G
204 store i32 %X.0, i32* %X ; Update X
205 br label %cond_next
207 cond_false:
208 %X.1 = load i32* @H
209 store i32 %X.1, i32* %X ; Update X
210 br label %cond_next
212 cond_next:
213 %X.2 = load i32* %X ; Read X
214 ret i32 %X.2
215 }
216 </pre>
217 </div>
219 <p>With this, we have discovered a way to handle arbitrary mutable variables
220 without the need to create Phi nodes at all:</p>
222 <ol>
227 </ol>
229 <p>While this solution has solved our immediate problem, it introduced another
230 one: we have now apparently introduced a lot of stack traffic for very simple
231 and common operations, a major performance problem. Fortunately for us, the
232 LLVM optimizer has a highly-tuned optimization pass named "mem2reg" that handles
233 this case, promoting allocas like this into SSA registers, inserting Phi nodes
234 as appropriate. If you run this example through the pass, for example, you'll
235 get:</p>
238 <pre>
240 @G = weak global i32 0
241 @H = weak global i32 0
243 define i32 @test(i1 %Condition) {
244 entry:
245 br i1 %Condition, label %cond_true, label %cond_false
247 cond_true:
248 %X.0 = load i32* @G
249 br label %cond_next
251 cond_false:
252 %X.1 = load i32* @H
253 br label %cond_next
255 cond_next:
257 ret i32 %X.01
258 }
259 </pre>
260 </div>
263 algorithm for constructing SSA form and has a number of optimizations that speed
264 up (very common) degenerate cases. The mem2reg optimization pass is the answer to dealing
265 with mutable variables, and we highly recommend that you depend on it. Note that
266 mem2reg only works on variables in certain circumstances:</p>
268 <ol>
269 <li>mem2reg is alloca-driven: it looks for allocas and if it can handle them, it
270 promotes them. It does not apply to global variables or heap allocations.</li>
272 <li>mem2reg only looks for alloca instructions in the entry block of the
273 function. Being in the entry block guarantees that the alloca is only executed
274 once, which makes analysis simpler.</li>
276 <li>mem2reg only promotes allocas whose uses are direct loads and stores. If
277 the address of the stack object is passed to a function, or if any funny pointer
278 arithmetic is involved, the alloca will not be promoted.</li>
282 values (such as pointers, scalars and vectors), and only if the array size
283 of the allocation is 1 (or missing in the .ll file). mem2reg is not capable of
284 promoting structs or arrays to registers. Note that the "scalarrepl" pass is
287 </ol>
289 <p>
290 All of these properties are easy to satisfy for most imperative languages, and
291 we'll illustrate it below with Kaleidoscope. The final question you may be
292 asking is: should I bother with this nonsense for my front-end? Wouldn't it be
293 better if I just did SSA construction directly, avoiding use of the mem2reg
294 optimization pass? In short, we strongly recommend that you use this technique
295 for building SSA form, unless there is an extremely good reason not to. Using
296 this technique is:</p>
298 <ul>
299 <li>Proven and well tested: llvm-gcc and clang both use this technique for local
300 mutable variables. As such, the most common clients of LLVM are using this to
301 handle a bulk of their variables. You can be sure that bugs are found fast and
302 fixed early.</li>
304 <li>Extremely Fast: mem2reg has a number of special cases that make it fast in
305 common cases as well as fully general. For example, it has fast-paths for
306 variables that are only used in a single block, variables that only have one
307 assignment point, good heuristics to avoid insertion of unneeded phi nodes, etc.
308 </li>
311 Debug information in LLVM</a> relies on having the address of the variable
312 exposed so that debug info can be attached to it. This technique dovetails
313 very naturally with this style of debug info.</li>
314 </ul>
316 <p>If nothing else, this makes it much easier to get your front-end up and
317 running, and is very simple to implement. Lets extend Kaleidoscope with mutable
318 variables now!
319 </p>
321 </div>
323 <!-- *********************************************************************** -->
325 Kaleidoscope</a></div>
326 <!-- *********************************************************************** -->
330 <p>Now that we know the sort of problem we want to tackle, lets see what this
331 looks like in the context of our little Kaleidoscope language. We're going to
332 add two features:</p>
334 <ol>
337 </ol>
339 <p>While the first item is really what this is about, we only have variables
340 for incoming arguments as well as for induction variables, and redefining those only
341 goes so far :). Also, the ability to define new variables is a
342 useful thing regardless of whether you will be mutating them. Here's a
343 motivating example that shows how we could use these:</p>
346 <pre>
347 # Define ':' for sequencing: as a low-precedence operator that ignores operands
348 # and just returns the RHS.
349 def binary : 1 (x y) y;
351 # Recursive fib, we could do this before.
352 def fib(x)
354 1
355 else
358 # Iterative fib.
359 def fibi(x)
365 b;
367 # Call it.
368 fibi(10);
369 </pre>
370 </div>
372 <p>
373 In order to mutate variables, we have to change our existing variables to use
374 the "alloca trick". Once we have that, we'll add our new operator, then extend
375 Kaleidoscope to support new variable definitions.
376 </p>
378 </div>
380 <!-- *********************************************************************** -->
382 Mutation</a></div>
383 <!-- *********************************************************************** -->
387 <p>
388 The symbol table in Kaleidoscope is managed at code generation time by the
390 that holds the double value for the named variable. In order to support
393 change is a refactoring: it changes the structure of the code, but does not
394 (by itself) change the behavior of the compiler. All of these changes are
395 isolated in the Kaleidoscope code generator.</p>
397 <p>
398 At this point in Kaleidoscope's development, it only supports variables for two
399 things: incoming arguments to functions and the induction variable of 'for'
400 loops. For consistency, we'll allow mutation of these variables in addition to
401 other user-defined variables. This means that these will both need memory
402 locations.
403 </p>
405 <p>To start our transformation of Kaleidoscope, we'll change the NamedValues
406 map so that it maps to AllocaInst* instead of Value*. Once we do this, the C++
407 compiler will tell us what parts of the code we need to update:</p>
410 <pre>
412 </pre>
413 </div>
415 <p>Also, since we will need to create these alloca's, we'll use a helper
416 function that ensures that the allocas are created in the entry block of the
417 function:</p>
420 <pre>
421 /// CreateEntryBlockAlloca - Create an alloca instruction in the entry block of
422 /// the function. This is used for mutable variables etc.
423 static AllocaInst *CreateEntryBlockAlloca(Function *TheFunction,
424 const std::string &VarName) {
426 TheFunction->getEntryBlock().begin());
427 return TmpB.CreateAlloca(Type::getDoubleTy(getGlobalContext()), 0, VarName.c_str());
428 }
429 </pre>
430 </div>
432 <p>This funny looking code creates an IRBuilder object that is pointing at
433 the first instruction (.begin()) of the entry block. It then creates an alloca
434 with the expected name and returns it. Because all values in Kaleidoscope are
435 doubles, there is no need to pass in a type to use.</p>
437 <p>With this in place, the first functionality change we want to make is to
438 variable references. In our new scheme, variables live on the stack, so code
439 generating a reference to them actually needs to produce a load from the stack
440 slot:</p>
443 <pre>
444 Value *VariableExprAST::Codegen() {
445 // Look this variable up in the function.
446 Value *V = NamedValues[Name];
449 <b>// Load the value.
450 return Builder.CreateLoad(V, Name.c_str());</b>
451 }
452 </pre>
453 </div>
455 <p>As you can see, this is pretty straightforward. Now we need to update the
456 things that define the variables to set up the alloca. We'll start with
458 the unabridged code):</p>
461 <pre>
462 Function *TheFunction = Builder.GetInsertBlock()->getParent();
464 <b>// Create an alloca for the variable in the entry block.
465 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);</b>
467 // Emit the start code first, without 'variable' in scope.
468 Value *StartVal = Start->Codegen();
471 <b>// Store the value into the alloca.
472 Builder.CreateStore(StartVal, Alloca);</b>
473 ...
475 // Compute the end condition.
476 Value *EndCond = End->Codegen();
477 if (EndCond == 0) return EndCond;
479 <b>// Reload, increment, and restore the alloca. This handles the case where
480 // the body of the loop mutates the variable.
481 Value *CurVar = Builder.CreateLoad(Alloca);
482 Value *NextVar = Builder.CreateAdd(CurVar, StepVal, "nextvar");
483 Builder.CreateStore(NextVar, Alloca);</b>
484 ...
485 </pre>
486 </div>
490 big difference is that we no longer have to construct a PHI node, and we use
491 load/store to access the variable as needed.</p>
493 <p>To support mutable argument variables, we need to also make allocas for them.
494 The code for this is also pretty simple:</p>
497 <pre>
498 /// CreateArgumentAllocas - Create an alloca for each argument and register the
499 /// argument in the symbol table so that references to it will succeed.
500 void PrototypeAST::CreateArgumentAllocas(Function *F) {
501 Function::arg_iterator AI = F->arg_begin();
502 for (unsigned Idx = 0, e = Args.size(); Idx != e; ++Idx, ++AI) {
503 // Create an alloca for this variable.
504 AllocaInst *Alloca = CreateEntryBlockAlloca(F, Args[Idx]);
506 // Store the initial value into the alloca.
507 Builder.CreateStore(AI, Alloca);
509 // Add arguments to variable symbol table.
510 NamedValues[Args[Idx]] = Alloca;
511 }
512 }
513 </pre>
514 </div>
516 <p>For each argument, we make an alloca, store the input value to the function
517 into the alloca, and register the alloca as the memory location for the
519 it sets up the entry block for the function.</p>
521 <p>The final missing piece is adding the mem2reg pass, which allows us to get
522 good codegen once again:</p>
525 <pre>
526 // Set up the optimizer pipeline. Start with registering info about how the
527 // target lays out data structures.
528 OurFPM.add(new TargetData(*TheExecutionEngine->getTargetData()));
529 <b>// Promote allocas to registers.
530 OurFPM.add(createPromoteMemoryToRegisterPass());</b>
531 // Do simple "peephole" optimizations and bit-twiddling optzns.
532 OurFPM.add(createInstructionCombiningPass());
533 // Reassociate expressions.
534 OurFPM.add(createReassociatePass());
535 </pre>
536 </div>
538 <p>It is interesting to see what the code looks like before and after the
539 mem2reg optimization runs. For example, this is the before/after code for our
540 recursive fib function. Before the optimization:</p>
543 <pre>
544 define double @fib(double %x) {
545 entry:
546 <b>%x1 = alloca double
547 store double %x, double* %x1
548 %x2 = load double* %x1</b>
549 %cmptmp = fcmp ult double %x2, 3.000000e+00
550 %booltmp = uitofp i1 %cmptmp to double
551 %ifcond = fcmp one double %booltmp, 0.000000e+00
552 br i1 %ifcond, label %then, label %else
554 then: ; preds = %entry
555 br label %ifcont
557 else: ; preds = %entry
559 %subtmp = sub double %x3, 1.000000e+00
560 %calltmp = call double @fib( double %subtmp )
562 %subtmp5 = sub double %x4, 2.000000e+00
563 %calltmp6 = call double @fib( double %subtmp5 )
564 %addtmp = add double %calltmp, %calltmp6
565 br label %ifcont
567 ifcont: ; preds = %else, %then
568 %iftmp = phi double [ 1.000000e+00, %then ], [ %addtmp, %else ]
569 ret double %iftmp
570 }
571 </pre>
572 </div>
574 <p>Here there is only one variable (x, the input argument) but you can still
575 see the extremely simple-minded code generation strategy we are using. In the
576 entry block, an alloca is created, and the initial input value is stored into
577 it. Each reference to the variable does a reload from the stack. Also, note
578 that we didn't modify the if/then/else expression, so it still inserts a PHI
579 node. While we could make an alloca for it, it is actually easier to create a
580 PHI node for it, so we still just make the PHI.</p>
585 <pre>
586 define double @fib(double %x) {
587 entry:
589 %booltmp = uitofp i1 %cmptmp to double
590 %ifcond = fcmp one double %booltmp, 0.000000e+00
591 br i1 %ifcond, label %then, label %else
593 then:
594 br label %ifcont
596 else:
598 %calltmp = call double @fib( double %subtmp )
600 %calltmp6 = call double @fib( double %subtmp5 )
601 %addtmp = add double %calltmp, %calltmp6
602 br label %ifcont
604 ifcont: ; preds = %else, %then
605 %iftmp = phi double [ 1.000000e+00, %then ], [ %addtmp, %else ]
606 ret double %iftmp
607 }
608 </pre>
609 </div>
611 <p>This is a trivial case for mem2reg, since there are no redefinitions of the
612 variable. The point of showing this is to calm your tension about inserting
613 such blatent inefficiencies :).</p>
618 <pre>
619 define double @fib(double %x) {
620 entry:
621 %cmptmp = fcmp ult double %x, 3.000000e+00
622 %booltmp = uitofp i1 %cmptmp to double
623 %ifcond = fcmp ueq double %booltmp, 0.000000e+00
624 br i1 %ifcond, label %else, label %ifcont
626 else:
627 %subtmp = sub double %x, 1.000000e+00
628 %calltmp = call double @fib( double %subtmp )
629 %subtmp5 = sub double %x, 2.000000e+00
630 %calltmp6 = call double @fib( double %subtmp5 )
631 %addtmp = add double %calltmp, %calltmp6
632 ret double %addtmp
634 ifcont:
635 ret double 1.000000e+00
636 }
637 </pre>
638 </div>
640 <p>Here we see that the simplifycfg pass decided to clone the return instruction
641 into the end of the 'else' block. This allowed it to eliminate some branches
642 and the PHI node.</p>
644 <p>Now that all symbol table references are updated to use stack variables,
645 we'll add the assignment operator.</p>
647 </div>
649 <!-- *********************************************************************** -->
651 <!-- *********************************************************************** -->
655 <p>With our current framework, adding a new assignment operator is really
656 simple. We will parse it just like any other binary operator, but handle it
657 internally (instead of allowing the user to define it). The first step is to
658 set a precedence:</p>
661 <pre>
662 int main() {
663 // Install standard binary operators.
664 // 1 is lowest precedence.
667 BinopPrecedence['+'] = 20;
668 BinopPrecedence['-'] = 20;
669 </pre>
670 </div>
672 <p>Now that the parser knows the precedence of the binary operator, it takes
673 care of all the parsing and AST generation. We just need to implement codegen
674 for the assignment operator. This looks like:</p>
677 <pre>
678 Value *BinaryExprAST::Codegen() {
679 // Special case '=' because we don't want to emit the LHS as an expression.
680 if (Op == '=') {
681 // Assignment requires the LHS to be an identifier.
683 if (!LHSE)
684 return ErrorV("destination of '=' must be a variable");
685 </pre>
686 </div>
688 <p>Unlike the rest of the binary operators, our assignment operator doesn't
689 follow the "emit LHS, emit RHS, do computation" model. As such, it is handled
690 as a special case before the other binary operators are handled. The other
691 strange thing is that it requires the LHS to be a variable. It is invalid to
693 </p>
696 <pre>
697 // Codegen the RHS.
698 Value *Val = RHS->Codegen();
701 // Look up the name.
702 Value *Variable = NamedValues[LHSE->getName()];
705 Builder.CreateStore(Val, Variable);
706 return Val;
707 }
708 ...
709 </pre>
710 </div>
712 <p>Once we have the variable, codegen'ing the assignment is straightforward:
713 we emit the RHS of the assignment, create a store, and return the computed
716 <p>Now that we have an assignment operator, we can mutate loop variables and
717 arguments. For example, we can now run code like this:</p>
720 <pre>
721 # Function to print a double.
722 extern printd(x);
724 # Define ':' for sequencing: as a low-precedence operator that ignores operands
725 # and just returns the RHS.
726 def binary : 1 (x y) y;
728 def test(x)
729 printd(x) :
730 x = 4 :
731 printd(x);
733 test(123);
734 </pre>
735 </div>
738 actually mutate the value! Okay, we have now officially implemented our goal:
739 getting this to work requires SSA construction in the general case. However,
740 to be really useful, we want the ability to define our own local variables, lets
741 add this next!
742 </p>
744 </div>
746 <!-- *********************************************************************** -->
748 Variables</a></div>
749 <!-- *********************************************************************** -->
753 <p>Adding var/in is just like any other other extensions we made to
754 Kaleidoscope: we extend the lexer, the parser, the AST and the code generator.
755 The first step for adding our new 'var/in' construct is to extend the lexer.
756 As before, this is pretty trivial, the code looks like this:</p>
759 <pre>
760 enum Token {
761 ...
762 <b>// var definition
764 ...
765 }
766 ...
767 static int gettok() {
768 ...
769 if (IdentifierStr == "in") return tok_in;
770 if (IdentifierStr == "binary") return tok_binary;
771 if (IdentifierStr == "unary") return tok_unary;
773 return tok_identifier;
774 ...
775 </pre>
776 </div>
778 <p>The next step is to define the AST node that we will construct. For var/in,
779 it looks like this:</p>
782 <pre>
783 /// VarExprAST - Expression class for var/in
784 class VarExprAST : public ExprAST {
786 ExprAST *Body;
787 public:
789 ExprAST *body)
790 : VarNames(varnames), Body(body) {}
792 virtual Value *Codegen();
793 };
794 </pre>
795 </div>
797 <p>var/in allows a list of names to be defined all at once, and each name can
798 optionally have an initializer value. As such, we capture this information in
799 the VarNames vector. Also, var/in has a body, this body is allowed to access
800 the variables defined by the var/in.</p>
802 <p>With this in place, we can define the parser pieces. The first thing we do is add
803 it as a primary expression:</p>
806 <pre>
807 /// primary
808 /// ::= identifierexpr
809 /// ::= numberexpr
810 /// ::= parenexpr
811 /// ::= ifexpr
812 /// ::= forexpr
814 static ExprAST *ParsePrimary() {
815 switch (CurTok) {
816 default: return Error("unknown token when expecting an expression");
817 case tok_identifier: return ParseIdentifierExpr();
818 case tok_number: return ParseNumberExpr();
819 case '(': return ParseParenExpr();
820 case tok_if: return ParseIfExpr();
821 case tok_for: return ParseForExpr();
823 }
824 }
825 </pre>
826 </div>
831 <pre>
832 /// varexpr ::= 'var' identifier ('=' expression)?
833 // (',' identifier ('=' expression)?)* 'in' expression
834 static ExprAST *ParseVarExpr() {
835 getNextToken(); // eat the var.
839 // At least one variable name is required.
840 if (CurTok != tok_identifier)
841 return Error("expected identifier after var");
842 </pre>
843 </div>
845 <p>The first part of this code parses the list of identifier/expr pairs into the
849 <pre>
850 while (1) {
851 std::string Name = IdentifierStr;
852 getNextToken(); // eat identifier.
854 // Read the optional initializer.
855 ExprAST *Init = 0;
856 if (CurTok == '=') {
857 getNextToken(); // eat the '='.
859 Init = ParseExpression();
861 }
863 VarNames.push_back(std::make_pair(Name, Init));
865 // End of var list, exit loop.
866 if (CurTok != ',') break;
867 getNextToken(); // eat the ','.
869 if (CurTok != tok_identifier)
870 return Error("expected identifier list after var");
871 }
872 </pre>
873 </div>
875 <p>Once all the variables are parsed, we then parse the body and create the
876 AST node:</p>
879 <pre>
880 // At this point, we have to have 'in'.
881 if (CurTok != tok_in)
882 return Error("expected 'in' keyword after 'var'");
883 getNextToken(); // eat 'in'.
885 ExprAST *Body = ParseExpression();
888 return new VarExprAST(VarNames, Body);
889 }
890 </pre>
891 </div>
893 <p>Now that we can parse and represent the code, we need to support emission of
894 LLVM IR for it. This code starts out with:</p>
897 <pre>
898 Value *VarExprAST::Codegen() {
901 Function *TheFunction = Builder.GetInsertBlock()->getParent();
903 // Register all variables and emit their initializer.
904 for (unsigned i = 0, e = VarNames.size(); i != e; ++i) {
905 const std::string &VarName = VarNames[i].first;
906 ExprAST *Init = VarNames[i].second;
907 </pre>
908 </div>
910 <p>Basically it loops over all the variables, installing them one at a time.
911 For each variable we put into the symbol table, we remember the previous value
912 that we replace in OldBindings.</p>
915 <pre>
916 // Emit the initializer before adding the variable to scope, this prevents
917 // the initializer from referencing the variable itself, and permits stuff
918 // like this:
919 // var a = 1 in
920 // var a = a in ... # refers to outer 'a'.
921 Value *InitVal;
922 if (Init) {
923 InitVal = Init->Codegen();
925 } else { // If not specified, use 0.0.
926 InitVal = ConstantFP::get(getGlobalContext(), APFloat(0.0));
927 }
929 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
930 Builder.CreateStore(InitVal, Alloca);
932 // Remember the old variable binding so that we can restore the binding when
933 // we unrecurse.
934 OldBindings.push_back(NamedValues[VarName]);
936 // Remember this binding.
937 NamedValues[VarName] = Alloca;
938 }
939 </pre>
940 </div>
942 <p>There are more comments here than code. The basic idea is that we emit the
943 initializer, create the alloca, then update the symbol table to point to it.
944 Once all the variables are installed in the symbol table, we evaluate the body
945 of the var/in expression:</p>
948 <pre>
949 // Codegen the body, now that all vars are in scope.
950 Value *BodyVal = Body->Codegen();
952 </pre>
953 </div>
958 <pre>
959 // Pop all our variables from scope.
960 for (unsigned i = 0, e = VarNames.size(); i != e; ++i)
961 NamedValues[VarNames[i].first] = OldBindings[i];
963 // Return the body computation.
964 return BodyVal;
965 }
966 </pre>
967 </div>
969 <p>The end result of all of this is that we get properly scoped variable
970 definitions, and we even (trivially) allow mutation of them :).</p>
972 <p>With this, we completed what we set out to do. Our nice iterative fib
973 example from the intro compiles and runs just fine. The mem2reg pass optimizes
974 all of our stack variables into SSA registers, inserting PHI nodes where needed,
975 and our front-end remains simple: no "iterated dominance frontier" computation
976 anywhere in sight.</p>
978 </div>
980 <!-- *********************************************************************** -->
982 <!-- *********************************************************************** -->
986 <p>
987 Here is the complete code listing for our running example, enhanced with mutable
988 variables and var/in support. To build this example, use:
989 </p>
992 <pre>
993 # Compile
994 g++ -g toy.cpp `llvm-config --cppflags --ldflags --libs core jit native` -O3 -o toy
995 # Run
996 ./toy
997 </pre>
998 </div>
1003 <pre>
1004 #include "llvm/DerivedTypes.h"
1005 #include "llvm/ExecutionEngine/ExecutionEngine.h"
1006 #include "llvm/LLVMContext.h"
1007 #include "llvm/Module.h"
1008 #include "llvm/ModuleProvider.h"
1009 #include "llvm/PassManager.h"
1010 #include "llvm/Analysis/Verifier.h"
1011 #include "llvm/Target/TargetData.h"
1012 #include "llvm/Transforms/Scalar.h"
1013 #include "llvm/Support/IRBuilder.h"
1018 using namespace llvm;
1020 //===----------------------------------------------------------------------===//
1021 // Lexer
1022 //===----------------------------------------------------------------------===//
1025 // of these for known things.
1026 enum Token {
1027 tok_eof = -1,
1029 // commands
1032 // primary
1035 // control
1039 // operators
1042 // var definition
1043 tok_var = -13
1044 };
1046 static std::string IdentifierStr; // Filled in if tok_identifier
1047 static double NumVal; // Filled in if tok_number
1049 /// gettok - Return the next token from standard input.
1050 static int gettok() {
1051 static int LastChar = ' ';
1053 // Skip any whitespace.
1054 while (isspace(LastChar))
1055 LastChar = getchar();
1057 if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]*
1058 IdentifierStr = LastChar;
1059 while (isalnum((LastChar = getchar())))
1060 IdentifierStr += LastChar;
1062 if (IdentifierStr == "def") return tok_def;
1063 if (IdentifierStr == "extern") return tok_extern;
1064 if (IdentifierStr == "if") return tok_if;
1065 if (IdentifierStr == "then") return tok_then;
1066 if (IdentifierStr == "else") return tok_else;
1067 if (IdentifierStr == "for") return tok_for;
1068 if (IdentifierStr == "in") return tok_in;
1069 if (IdentifierStr == "binary") return tok_binary;
1070 if (IdentifierStr == "unary") return tok_unary;
1071 if (IdentifierStr == "var") return tok_var;
1072 return tok_identifier;
1073 }
1076 std::string NumStr;
1077 do {
1078 NumStr += LastChar;
1079 LastChar = getchar();
1080 } while (isdigit(LastChar) || LastChar == '.');
1082 NumVal = strtod(NumStr.c_str(), 0);
1083 return tok_number;
1084 }
1086 if (LastChar == '#') {
1087 // Comment until end of line.
1088 do LastChar = getchar();
1091 if (LastChar != EOF)
1092 return gettok();
1093 }
1095 // Check for end of file. Don't eat the EOF.
1096 if (LastChar == EOF)
1097 return tok_eof;
1099 // Otherwise, just return the character as its ascii value.
1100 int ThisChar = LastChar;
1101 LastChar = getchar();
1102 return ThisChar;
1103 }
1105 //===----------------------------------------------------------------------===//
1106 // Abstract Syntax Tree (aka Parse Tree)
1107 //===----------------------------------------------------------------------===//
1109 /// ExprAST - Base class for all expression nodes.
1110 class ExprAST {
1111 public:
1112 virtual ~ExprAST() {}
1113 virtual Value *Codegen() = 0;
1114 };
1116 /// NumberExprAST - Expression class for numeric literals like "1.0".
1117 class NumberExprAST : public ExprAST {
1118 double Val;
1119 public:
1120 NumberExprAST(double val) : Val(val) {}
1121 virtual Value *Codegen();
1122 };
1124 /// VariableExprAST - Expression class for referencing a variable, like "a".
1125 class VariableExprAST : public ExprAST {
1126 std::string Name;
1127 public:
1128 VariableExprAST(const std::string &name) : Name(name) {}
1129 const std::string &getName() const { return Name; }
1130 virtual Value *Codegen();
1131 };
1133 /// UnaryExprAST - Expression class for a unary operator.
1134 class UnaryExprAST : public ExprAST {
1135 char Opcode;
1136 ExprAST *Operand;
1137 public:
1138 UnaryExprAST(char opcode, ExprAST *operand)
1139 : Opcode(opcode), Operand(operand) {}
1140 virtual Value *Codegen();
1141 };
1143 /// BinaryExprAST - Expression class for a binary operator.
1144 class BinaryExprAST : public ExprAST {
1145 char Op;
1146 ExprAST *LHS, *RHS;
1147 public:
1148 BinaryExprAST(char op, ExprAST *lhs, ExprAST *rhs)
1149 : Op(op), LHS(lhs), RHS(rhs) {}
1150 virtual Value *Codegen();
1151 };
1153 /// CallExprAST - Expression class for function calls.
1154 class CallExprAST : public ExprAST {
1155 std::string Callee;
1157 public:
1159 : Callee(callee), Args(args) {}
1160 virtual Value *Codegen();
1161 };
1163 /// IfExprAST - Expression class for if/then/else.
1164 class IfExprAST : public ExprAST {
1165 ExprAST *Cond, *Then, *Else;
1166 public:
1167 IfExprAST(ExprAST *cond, ExprAST *then, ExprAST *_else)
1168 : Cond(cond), Then(then), Else(_else) {}
1169 virtual Value *Codegen();
1170 };
1172 /// ForExprAST - Expression class for for/in.
1173 class ForExprAST : public ExprAST {
1174 std::string VarName;
1175 ExprAST *Start, *End, *Step, *Body;
1176 public:
1177 ForExprAST(const std::string &varname, ExprAST *start, ExprAST *end,
1178 ExprAST *step, ExprAST *body)
1179 : VarName(varname), Start(start), End(end), Step(step), Body(body) {}
1180 virtual Value *Codegen();
1181 };
1183 /// VarExprAST - Expression class for var/in
1184 class VarExprAST : public ExprAST {
1186 ExprAST *Body;
1187 public:
1189 ExprAST *body)
1190 : VarNames(varnames), Body(body) {}
1192 virtual Value *Codegen();
1193 };
1195 /// PrototypeAST - This class represents the "prototype" for a function,
1196 /// which captures its argument names as well as if it is an operator.
1197 class PrototypeAST {
1198 std::string Name;
1200 bool isOperator;
1201 unsigned Precedence; // Precedence if a binary op.
1202 public:
1204 bool isoperator = false, unsigned prec = 0)
1205 : Name(name), Args(args), isOperator(isoperator), Precedence(prec) {}
1210 char getOperatorName() const {
1211 assert(isUnaryOp() || isBinaryOp());
1212 return Name[Name.size()-1];
1213 }
1215 unsigned getBinaryPrecedence() const { return Precedence; }
1217 Function *Codegen();
1219 void CreateArgumentAllocas(Function *F);
1220 };
1222 /// FunctionAST - This class represents a function definition itself.
1223 class FunctionAST {
1224 PrototypeAST *Proto;
1225 ExprAST *Body;
1226 public:
1227 FunctionAST(PrototypeAST *proto, ExprAST *body)
1228 : Proto(proto), Body(body) {}
1230 Function *Codegen();
1231 };
1233 //===----------------------------------------------------------------------===//
1234 // Parser
1235 //===----------------------------------------------------------------------===//
1237 /// CurTok/getNextToken - Provide a simple token buffer. CurTok is the current
1238 /// token the parser it looking at. getNextToken reads another token from the
1239 /// lexer and updates CurTok with its results.
1240 static int CurTok;
1241 static int getNextToken() {
1242 return CurTok = gettok();
1243 }
1245 /// BinopPrecedence - This holds the precedence for each binary operator that is
1246 /// defined.
1249 /// GetTokPrecedence - Get the precedence of the pending binary operator token.
1250 static int GetTokPrecedence() {
1251 if (!isascii(CurTok))
1252 return -1;
1254 // Make sure it's a declared binop.
1255 int TokPrec = BinopPrecedence[CurTok];
1257 return TokPrec;
1258 }
1260 /// Error* - These are little helper functions for error handling.
1262 PrototypeAST *ErrorP(const char *Str) { Error(Str); return 0; }
1263 FunctionAST *ErrorF(const char *Str) { Error(Str); return 0; }
1265 static ExprAST *ParseExpression();
1267 /// identifierexpr
1268 /// ::= identifier
1269 /// ::= identifier '(' expression* ')'
1270 static ExprAST *ParseIdentifierExpr() {
1271 std::string IdName = IdentifierStr;
1273 getNextToken(); // eat identifier.
1275 if (CurTok != '(') // Simple variable ref.
1276 return new VariableExprAST(IdName);
1278 // Call.
1279 getNextToken(); // eat (
1281 if (CurTok != ')') {
1282 while (1) {
1283 ExprAST *Arg = ParseExpression();
1284 if (!Arg) return 0;
1285 Args.push_back(Arg);
1287 if (CurTok == ')') break;
1289 if (CurTok != ',')
1290 return Error("Expected ')' or ',' in argument list");
1291 getNextToken();
1292 }
1293 }
1295 // Eat the ')'.
1296 getNextToken();
1298 return new CallExprAST(IdName, Args);
1299 }
1301 /// numberexpr ::= number
1302 static ExprAST *ParseNumberExpr() {
1303 ExprAST *Result = new NumberExprAST(NumVal);
1304 getNextToken(); // consume the number
1305 return Result;
1306 }
1308 /// parenexpr ::= '(' expression ')'
1309 static ExprAST *ParseParenExpr() {
1310 getNextToken(); // eat (.
1311 ExprAST *V = ParseExpression();
1312 if (!V) return 0;
1314 if (CurTok != ')')
1315 return Error("expected ')'");
1316 getNextToken(); // eat ).
1317 return V;
1318 }
1320 /// ifexpr ::= 'if' expression 'then' expression 'else' expression
1321 static ExprAST *ParseIfExpr() {
1322 getNextToken(); // eat the if.
1324 // condition.
1325 ExprAST *Cond = ParseExpression();
1326 if (!Cond) return 0;
1328 if (CurTok != tok_then)
1329 return Error("expected then");
1330 getNextToken(); // eat the then
1332 ExprAST *Then = ParseExpression();
1335 if (CurTok != tok_else)
1336 return Error("expected else");
1338 getNextToken();
1340 ExprAST *Else = ParseExpression();
1341 if (!Else) return 0;
1343 return new IfExprAST(Cond, Then, Else);
1344 }
1346 /// forexpr ::= 'for' identifier '=' expr ',' expr (',' expr)? 'in' expression
1347 static ExprAST *ParseForExpr() {
1348 getNextToken(); // eat the for.
1350 if (CurTok != tok_identifier)
1351 return Error("expected identifier after for");
1353 std::string IdName = IdentifierStr;
1354 getNextToken(); // eat identifier.
1356 if (CurTok != '=')
1357 return Error("expected '=' after for");
1358 getNextToken(); // eat '='.
1361 ExprAST *Start = ParseExpression();
1363 if (CurTok != ',')
1364 return Error("expected ',' after for start value");
1365 getNextToken();
1367 ExprAST *End = ParseExpression();
1370 // The step value is optional.
1371 ExprAST *Step = 0;
1372 if (CurTok == ',') {
1373 getNextToken();
1374 Step = ParseExpression();
1376 }
1378 if (CurTok != tok_in)
1379 return Error("expected 'in' after for");
1380 getNextToken(); // eat 'in'.
1382 ExprAST *Body = ParseExpression();
1385 return new ForExprAST(IdName, Start, End, Step, Body);
1386 }
1388 /// varexpr ::= 'var' identifier ('=' expression)?
1389 // (',' identifier ('=' expression)?)* 'in' expression
1390 static ExprAST *ParseVarExpr() {
1391 getNextToken(); // eat the var.
1395 // At least one variable name is required.
1396 if (CurTok != tok_identifier)
1397 return Error("expected identifier after var");
1399 while (1) {
1400 std::string Name = IdentifierStr;
1401 getNextToken(); // eat identifier.
1403 // Read the optional initializer.
1404 ExprAST *Init = 0;
1405 if (CurTok == '=') {
1406 getNextToken(); // eat the '='.
1408 Init = ParseExpression();
1410 }
1412 VarNames.push_back(std::make_pair(Name, Init));
1414 // End of var list, exit loop.
1415 if (CurTok != ',') break;
1416 getNextToken(); // eat the ','.
1418 if (CurTok != tok_identifier)
1419 return Error("expected identifier list after var");
1420 }
1422 // At this point, we have to have 'in'.
1423 if (CurTok != tok_in)
1424 return Error("expected 'in' keyword after 'var'");
1425 getNextToken(); // eat 'in'.
1427 ExprAST *Body = ParseExpression();
1430 return new VarExprAST(VarNames, Body);
1431 }
1434 /// primary
1435 /// ::= identifierexpr
1436 /// ::= numberexpr
1437 /// ::= parenexpr
1438 /// ::= ifexpr
1439 /// ::= forexpr
1440 /// ::= varexpr
1441 static ExprAST *ParsePrimary() {
1442 switch (CurTok) {
1443 default: return Error("unknown token when expecting an expression");
1444 case tok_identifier: return ParseIdentifierExpr();
1445 case tok_number: return ParseNumberExpr();
1446 case '(': return ParseParenExpr();
1447 case tok_if: return ParseIfExpr();
1448 case tok_for: return ParseForExpr();
1449 case tok_var: return ParseVarExpr();
1450 }
1451 }
1453 /// unary
1454 /// ::= primary
1455 /// ::= '!' unary
1456 static ExprAST *ParseUnary() {
1457 // If the current token is not an operator, it must be a primary expr.
1458 if (!isascii(CurTok) || CurTok == '(' || CurTok == ',')
1459 return ParsePrimary();
1461 // If this is a unary operator, read it.
1462 int Opc = CurTok;
1463 getNextToken();
1464 if (ExprAST *Operand = ParseUnary())
1465 return new UnaryExprAST(Opc, Operand);
1466 return 0;
1467 }
1469 /// binoprhs
1470 /// ::= ('+' unary)*
1471 static ExprAST *ParseBinOpRHS(int ExprPrec, ExprAST *LHS) {
1472 // If this is a binop, find its precedence.
1473 while (1) {
1474 int TokPrec = GetTokPrecedence();
1476 // If this is a binop that binds at least as tightly as the current binop,
1477 // consume it, otherwise we are done.
1478 if (TokPrec < ExprPrec)
1479 return LHS;
1481 // Okay, we know this is a binop.
1482 int BinOp = CurTok;
1483 getNextToken(); // eat binop
1485 // Parse the unary expression after the binary operator.
1486 ExprAST *RHS = ParseUnary();
1487 if (!RHS) return 0;
1489 // If BinOp binds less tightly with RHS than the operator after RHS, let
1490 // the pending operator take RHS as its LHS.
1491 int NextPrec = GetTokPrecedence();
1492 if (TokPrec < NextPrec) {
1493 RHS = ParseBinOpRHS(TokPrec+1, RHS);
1495 }
1497 // Merge LHS/RHS.
1498 LHS = new BinaryExprAST(BinOp, LHS, RHS);
1499 }
1500 }
1502 /// expression
1503 /// ::= unary binoprhs
1504 ///
1505 static ExprAST *ParseExpression() {
1506 ExprAST *LHS = ParseUnary();
1507 if (!LHS) return 0;
1509 return ParseBinOpRHS(0, LHS);
1510 }
1512 /// prototype
1513 /// ::= id '(' id* ')'
1514 /// ::= binary LETTER number? (id, id)
1515 /// ::= unary LETTER (id)
1516 static PrototypeAST *ParsePrototype() {
1517 std::string FnName;
1520 unsigned BinaryPrecedence = 30;
1522 switch (CurTok) {
1523 default:
1524 return ErrorP("Expected function name in prototype");
1525 case tok_identifier:
1526 FnName = IdentifierStr;
1527 Kind = 0;
1528 getNextToken();
1529 break;
1530 case tok_unary:
1531 getNextToken();
1532 if (!isascii(CurTok))
1533 return ErrorP("Expected unary operator");
1534 FnName = "unary";
1535 FnName += (char)CurTok;
1536 Kind = 1;
1537 getNextToken();
1538 break;
1539 case tok_binary:
1540 getNextToken();
1541 if (!isascii(CurTok))
1542 return ErrorP("Expected binary operator");
1543 FnName = "binary";
1544 FnName += (char)CurTok;
1545 Kind = 2;
1546 getNextToken();
1548 // Read the precedence if present.
1549 if (CurTok == tok_number) {
1551 return ErrorP("Invalid precedecnce: must be 1..100");
1552 BinaryPrecedence = (unsigned)NumVal;
1553 getNextToken();
1554 }
1555 break;
1556 }
1558 if (CurTok != '(')
1559 return ErrorP("Expected '(' in prototype");
1562 while (getNextToken() == tok_identifier)
1563 ArgNames.push_back(IdentifierStr);
1564 if (CurTok != ')')
1565 return ErrorP("Expected ')' in prototype");
1567 // success.
1568 getNextToken(); // eat ')'.
1570 // Verify right number of names for operator.
1571 if (Kind && ArgNames.size() != Kind)
1572 return ErrorP("Invalid number of operands for operator");
1574 return new PrototypeAST(FnName, ArgNames, Kind != 0, BinaryPrecedence);
1575 }
1577 /// definition ::= 'def' prototype expression
1578 static FunctionAST *ParseDefinition() {
1579 getNextToken(); // eat def.
1580 PrototypeAST *Proto = ParsePrototype();
1583 if (ExprAST *E = ParseExpression())
1584 return new FunctionAST(Proto, E);
1585 return 0;
1586 }
1588 /// toplevelexpr ::= expression
1589 static FunctionAST *ParseTopLevelExpr() {
1590 if (ExprAST *E = ParseExpression()) {
1591 // Make an anonymous proto.
1593 return new FunctionAST(Proto, E);
1594 }
1595 return 0;
1596 }
1598 /// external ::= 'extern' prototype
1599 static PrototypeAST *ParseExtern() {
1600 getNextToken(); // eat extern.
1601 return ParsePrototype();
1602 }
1604 //===----------------------------------------------------------------------===//
1605 // Code Generation
1606 //===----------------------------------------------------------------------===//
1608 static Module *TheModule;
1609 static IRBuilder<> Builder(getGlobalContext());
1611 static FunctionPassManager *TheFPM;
1613 Value *ErrorV(const char *Str) { Error(Str); return 0; }
1615 /// CreateEntryBlockAlloca - Create an alloca instruction in the entry block of
1616 /// the function. This is used for mutable variables etc.
1617 static AllocaInst *CreateEntryBlockAlloca(Function *TheFunction,
1618 const std::string &VarName) {
1620 TheFunction->getEntryBlock().begin());
1621 return TmpB.CreateAlloca(Type::getDoubleTy(getGlobalContext()), 0, VarName.c_str());
1622 }
1625 Value *NumberExprAST::Codegen() {
1626 return ConstantFP::get(getGlobalContext(), APFloat(Val));
1627 }
1629 Value *VariableExprAST::Codegen() {
1630 // Look this variable up in the function.
1631 Value *V = NamedValues[Name];
1634 // Load the value.
1635 return Builder.CreateLoad(V, Name.c_str());
1636 }
1638 Value *UnaryExprAST::Codegen() {
1639 Value *OperandV = Operand->Codegen();
1643 if (F == 0)
1644 return ErrorV("Unknown unary operator");
1646 return Builder.CreateCall(F, OperandV, "unop");
1647 }
1650 Value *BinaryExprAST::Codegen() {
1651 // Special case '=' because we don't want to emit the LHS as an expression.
1652 if (Op == '=') {
1653 // Assignment requires the LHS to be an identifier.
1655 if (!LHSE)
1656 return ErrorV("destination of '=' must be a variable");
1657 // Codegen the RHS.
1658 Value *Val = RHS->Codegen();
1661 // Look up the name.
1662 Value *Variable = NamedValues[LHSE->getName()];
1665 Builder.CreateStore(Val, Variable);
1666 return Val;
1667 }
1670 Value *L = LHS->Codegen();
1671 Value *R = RHS->Codegen();
1674 switch (Op) {
1675 case '+': return Builder.CreateAdd(L, R, "addtmp");
1676 case '-': return Builder.CreateSub(L, R, "subtmp");
1677 case '*': return Builder.CreateMul(L, R, "multmp");
1678 case '<':
1679 L = Builder.CreateFCmpULT(L, R, "cmptmp");
1681 return Builder.CreateUIToFP(L, Type::getDoubleTy(getGlobalContext()), "booltmp");
1682 default: break;
1683 }
1685 // If it wasn't a builtin binary operator, it must be a user defined one. Emit
1686 // a call to it.
1690 Value *Ops[] = { L, R };
1692 }
1694 Value *CallExprAST::Codegen() {
1695 // Look up the name in the global module table.
1696 Function *CalleeF = TheModule->getFunction(Callee);
1697 if (CalleeF == 0)
1698 return ErrorV("Unknown function referenced");
1700 // If argument mismatch error.
1701 if (CalleeF->arg_size() != Args.size())
1702 return ErrorV("Incorrect # arguments passed");
1705 for (unsigned i = 0, e = Args.size(); i != e; ++i) {
1706 ArgsV.push_back(Args[i]->Codegen());
1708 }
1710 return Builder.CreateCall(CalleeF, ArgsV.begin(), ArgsV.end(), "calltmp");
1711 }
1713 Value *IfExprAST::Codegen() {
1714 Value *CondV = Cond->Codegen();
1717 // Convert condition to a bool by comparing equal to 0.0.
1718 CondV = Builder.CreateFCmpONE(CondV,
1719 ConstantFP::get(getGlobalContext(), APFloat(0.0)),
1720 "ifcond");
1722 Function *TheFunction = Builder.GetInsertBlock()->getParent();
1724 // Create blocks for the then and else cases. Insert the 'then' block at the
1725 // end of the function.
1726 BasicBlock *ThenBB = BasicBlock::Create(getGlobalContext(), "then", TheFunction);
1727 BasicBlock *ElseBB = BasicBlock::Create(getGlobalContext(), "else");
1728 BasicBlock *MergeBB = BasicBlock::Create(getGlobalContext(), "ifcont");
1730 Builder.CreateCondBr(CondV, ThenBB, ElseBB);
1732 // Emit then value.
1733 Builder.SetInsertPoint(ThenBB);
1735 Value *ThenV = Then->Codegen();
1738 Builder.CreateBr(MergeBB);
1739 // Codegen of 'Then' can change the current block, update ThenBB for the PHI.
1740 ThenBB = Builder.GetInsertBlock();
1742 // Emit else block.
1743 TheFunction->getBasicBlockList().push_back(ElseBB);
1744 Builder.SetInsertPoint(ElseBB);
1746 Value *ElseV = Else->Codegen();
1749 Builder.CreateBr(MergeBB);
1750 // Codegen of 'Else' can change the current block, update ElseBB for the PHI.
1751 ElseBB = Builder.GetInsertBlock();
1753 // Emit merge block.
1754 TheFunction->getBasicBlockList().push_back(MergeBB);
1755 Builder.SetInsertPoint(MergeBB);
1756 PHINode *PN = Builder.CreatePHI(Type::getDoubleTy(getGlobalContext()), "iftmp");
1758 PN->addIncoming(ThenV, ThenBB);
1759 PN->addIncoming(ElseV, ElseBB);
1760 return PN;
1761 }
1763 Value *ForExprAST::Codegen() {
1764 // Output this as:
1765 // var = alloca double
1766 // ...
1767 // start = startexpr
1768 // store start -> var
1769 // goto loop
1770 // loop:
1771 // ...
1772 // bodyexpr
1773 // ...
1774 // loopend:
1775 // step = stepexpr
1776 // endcond = endexpr
1777 //
1778 // curvar = load var
1779 // nextvar = curvar + step
1780 // store nextvar -> var
1781 // br endcond, loop, endloop
1782 // outloop:
1784 Function *TheFunction = Builder.GetInsertBlock()->getParent();
1786 // Create an alloca for the variable in the entry block.
1787 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
1789 // Emit the start code first, without 'variable' in scope.
1790 Value *StartVal = Start->Codegen();
1793 // Store the value into the alloca.
1794 Builder.CreateStore(StartVal, Alloca);
1796 // Make the new basic block for the loop header, inserting after current
1797 // block.
1798 BasicBlock *PreheaderBB = Builder.GetInsertBlock();
1799 BasicBlock *LoopBB = BasicBlock::Create(getGlobalContext(), "loop", TheFunction);
1801 // Insert an explicit fall through from the current block to the LoopBB.
1802 Builder.CreateBr(LoopBB);
1804 // Start insertion in LoopBB.
1805 Builder.SetInsertPoint(LoopBB);
1807 // Within the loop, the variable is defined equal to the PHI node. If it
1808 // shadows an existing variable, we have to restore it, so save it now.
1809 AllocaInst *OldVal = NamedValues[VarName];
1810 NamedValues[VarName] = Alloca;
1812 // Emit the body of the loop. This, like any other expr, can change the
1813 // current BB. Note that we ignore the value computed by the body, but don't
1814 // allow an error.
1816 return 0;
1818 // Emit the step value.
1819 Value *StepVal;
1820 if (Step) {
1821 StepVal = Step->Codegen();
1823 } else {
1824 // If not specified, use 1.0.
1825 StepVal = ConstantFP::get(getGlobalContext(), APFloat(1.0));
1826 }
1828 // Compute the end condition.
1829 Value *EndCond = End->Codegen();
1830 if (EndCond == 0) return EndCond;
1832 // Reload, increment, and restore the alloca. This handles the case where
1833 // the body of the loop mutates the variable.
1834 Value *CurVar = Builder.CreateLoad(Alloca, VarName.c_str());
1835 Value *NextVar = Builder.CreateAdd(CurVar, StepVal, "nextvar");
1836 Builder.CreateStore(NextVar, Alloca);
1838 // Convert condition to a bool by comparing equal to 0.0.
1839 EndCond = Builder.CreateFCmpONE(EndCond,
1840 ConstantFP::get(getGlobalContext(), APFloat(0.0)),
1841 "loopcond");
1843 // Create the "after loop" block and insert it.
1844 BasicBlock *LoopEndBB = Builder.GetInsertBlock();
1845 BasicBlock *AfterBB = BasicBlock::Create(getGlobalContext(), "afterloop", TheFunction);
1847 // Insert the conditional branch into the end of LoopEndBB.
1848 Builder.CreateCondBr(EndCond, LoopBB, AfterBB);
1850 // Any new code will be inserted in AfterBB.
1851 Builder.SetInsertPoint(AfterBB);
1853 // Restore the unshadowed variable.
1854 if (OldVal)
1855 NamedValues[VarName] = OldVal;
1856 else
1857 NamedValues.erase(VarName);
1860 // for expr always returns 0.0.
1861 return Constant::getNullValue(Type::getDoubleTy(getGlobalContext()));
1862 }
1864 Value *VarExprAST::Codegen() {
1867 Function *TheFunction = Builder.GetInsertBlock()->getParent();
1869 // Register all variables and emit their initializer.
1870 for (unsigned i = 0, e = VarNames.size(); i != e; ++i) {
1871 const std::string &VarName = VarNames[i].first;
1872 ExprAST *Init = VarNames[i].second;
1874 // Emit the initializer before adding the variable to scope, this prevents
1875 // the initializer from referencing the variable itself, and permits stuff
1876 // like this:
1877 // var a = 1 in
1878 // var a = a in ... # refers to outer 'a'.
1879 Value *InitVal;
1880 if (Init) {
1881 InitVal = Init->Codegen();
1883 } else { // If not specified, use 0.0.
1884 InitVal = ConstantFP::get(getGlobalContext(), APFloat(0.0));
1885 }
1887 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
1888 Builder.CreateStore(InitVal, Alloca);
1890 // Remember the old variable binding so that we can restore the binding when
1891 // we unrecurse.
1892 OldBindings.push_back(NamedValues[VarName]);
1894 // Remember this binding.
1895 NamedValues[VarName] = Alloca;
1896 }
1898 // Codegen the body, now that all vars are in scope.
1899 Value *BodyVal = Body->Codegen();
1902 // Pop all our variables from scope.
1903 for (unsigned i = 0, e = VarNames.size(); i != e; ++i)
1904 NamedValues[VarNames[i].first] = OldBindings[i];
1906 // Return the body computation.
1907 return BodyVal;
1908 }
1911 Function *PrototypeAST::Codegen() {
1912 // Make the function type: double(double,double) etc.
1914 FunctionType *FT = FunctionType::get(Type::getDoubleTy(getGlobalContext()), Doubles, false);
1916 Function *F = Function::Create(FT, Function::ExternalLinkage, Name, TheModule);
1918 // If F conflicted, there was already something named 'Name'. If it has a
1919 // body, don't allow redefinition or reextern.
1920 if (F->getName() != Name) {
1921 // Delete the one we just made and get the existing one.
1922 F->eraseFromParent();
1923 F = TheModule->getFunction(Name);
1925 // If F already has a body, reject this.
1926 if (!F->empty()) {
1927 ErrorF("redefinition of function");
1928 return 0;
1929 }
1931 // If F took a different number of args, reject.
1932 if (F->arg_size() != Args.size()) {
1933 ErrorF("redefinition of function with different # args");
1934 return 0;
1935 }
1936 }
1938 // Set names for all arguments.
1939 unsigned Idx = 0;
1940 for (Function::arg_iterator AI = F->arg_begin(); Idx != Args.size();
1941 ++AI, ++Idx)
1942 AI->setName(Args[Idx]);
1944 return F;
1945 }
1947 /// CreateArgumentAllocas - Create an alloca for each argument and register the
1948 /// argument in the symbol table so that references to it will succeed.
1949 void PrototypeAST::CreateArgumentAllocas(Function *F) {
1950 Function::arg_iterator AI = F->arg_begin();
1951 for (unsigned Idx = 0, e = Args.size(); Idx != e; ++Idx, ++AI) {
1952 // Create an alloca for this variable.
1953 AllocaInst *Alloca = CreateEntryBlockAlloca(F, Args[Idx]);
1955 // Store the initial value into the alloca.
1956 Builder.CreateStore(AI, Alloca);
1958 // Add arguments to variable symbol table.
1959 NamedValues[Args[Idx]] = Alloca;
1960 }
1961 }
1964 Function *FunctionAST::Codegen() {
1965 NamedValues.clear();
1967 Function *TheFunction = Proto->Codegen();
1968 if (TheFunction == 0)
1969 return 0;
1971 // If this is an operator, install it.
1972 if (Proto->isBinaryOp())
1975 // Create a new basic block to start insertion into.
1976 BasicBlock *BB = BasicBlock::Create(getGlobalContext(), "entry", TheFunction);
1977 Builder.SetInsertPoint(BB);
1979 // Add all arguments to the symbol table and create their allocas.
1980 Proto->CreateArgumentAllocas(TheFunction);
1982 if (Value *RetVal = Body->Codegen()) {
1983 // Finish off the function.
1984 Builder.CreateRet(RetVal);
1986 // Validate the generated code, checking for consistency.
1987 verifyFunction(*TheFunction);
1989 // Optimize the function.
1990 TheFPM->run(*TheFunction);
1992 return TheFunction;
1993 }
1995 // Error reading body, remove function.
1996 TheFunction->eraseFromParent();
1998 if (Proto->isBinaryOp())
1999 BinopPrecedence.erase(Proto->getOperatorName());
2000 return 0;
2001 }
2003 //===----------------------------------------------------------------------===//
2004 // Top-Level parsing and JIT Driver
2005 //===----------------------------------------------------------------------===//
2007 static ExecutionEngine *TheExecutionEngine;
2009 static void HandleDefinition() {
2010 if (FunctionAST *F = ParseDefinition()) {
2011 if (Function *LF = F->Codegen()) {
2012 fprintf(stderr, "Read function definition:");
2013 LF->dump();
2014 }
2015 } else {
2016 // Skip token for error recovery.
2017 getNextToken();
2018 }
2019 }
2021 static void HandleExtern() {
2022 if (PrototypeAST *P = ParseExtern()) {
2023 if (Function *F = P->Codegen()) {
2024 fprintf(stderr, "Read extern: ");
2025 F->dump();
2026 }
2027 } else {
2028 // Skip token for error recovery.
2029 getNextToken();
2030 }
2031 }
2033 static void HandleTopLevelExpression() {
2034 // Evaluate a top level expression into an anonymous function.
2035 if (FunctionAST *F = ParseTopLevelExpr()) {
2036 if (Function *LF = F->Codegen()) {
2037 // JIT the function, returning a function pointer.
2038 void *FPtr = TheExecutionEngine->getPointerToFunction(LF);
2040 // Cast it to the right type (takes no arguments, returns a double) so we
2041 // can call it as a native function.
2042 double (*FP)() = (double (*)())FPtr;
2043 fprintf(stderr, "Evaluated to %f\n", FP());
2044 }
2045 } else {
2046 // Skip token for error recovery.
2047 getNextToken();
2048 }
2049 }
2051 /// top ::= definition | external | expression | ';'
2052 static void MainLoop() {
2053 while (1) {
2054 fprintf(stderr, "ready> ");
2055 switch (CurTok) {
2056 case tok_eof: return;
2057 case ';': getNextToken(); break; // ignore top level semicolons.
2058 case tok_def: HandleDefinition(); break;
2059 case tok_extern: HandleExtern(); break;
2060 default: HandleTopLevelExpression(); break;
2061 }
2062 }
2063 }
2067 //===----------------------------------------------------------------------===//
2069 //===----------------------------------------------------------------------===//
2071 /// putchard - putchar that takes a double and returns 0.
2072 extern "C"
2073 double putchard(double X) {
2074 putchar((char)X);
2075 return 0;
2076 }
2079 extern "C"
2080 double printd(double X) {
2081 printf("%f\n", X);
2082 return 0;
2083 }
2085 //===----------------------------------------------------------------------===//
2086 // Main driver code.
2087 //===----------------------------------------------------------------------===//
2089 int main() {
2090 // Install standard binary operators.
2091 // 1 is lowest precedence.
2092 BinopPrecedence['='] = 2;
2094 BinopPrecedence['+'] = 20;
2095 BinopPrecedence['-'] = 20;
2096 BinopPrecedence['*'] = 40; // highest.
2098 // Prime the first token.
2099 fprintf(stderr, "ready> ");
2100 getNextToken();
2102 // Make the module, which holds all the code.
2103 TheModule = new Module("my cool jit", getGlobalContext());
2105 ExistingModuleProvider *OurModuleProvider =
2106 new ExistingModuleProvider(TheModule);
2108 // Create the JIT. This takes ownership of the module and module provider.
2109 TheExecutionEngine = EngineBuilder(OurModuleProvider).create();
2111 FunctionPassManager OurFPM(OurModuleProvider);
2113 // Set up the optimizer pipeline. Start with registering info about how the
2114 // target lays out data structures.
2115 OurFPM.add(new TargetData(*TheExecutionEngine->getTargetData()));
2116 // Do simple "peephole" optimizations and bit-twiddling optzns.
2117 OurFPM.add(createInstructionCombiningPass());
2118 // Reassociate expressions.
2119 OurFPM.add(createReassociatePass());
2120 // Eliminate Common SubExpressions.
2121 OurFPM.add(createGVNPass());
2122 // Simplify the control flow graph (deleting unreachable blocks, etc).
2123 OurFPM.add(createCFGSimplificationPass());
2125 // Set the global so the code gen can use this.
2126 TheFPM = &OurFPM;
2128 // Run the main "interpreter loop" now.
2129 MainLoop();
2131 TheFPM = 0;
2133 // Print out all of the generated code.
2134 TheModule->dump();
2136 return 0;
2137 }
2138 </pre>
2139 </div>
2142 </div>
2144 <!-- *********************************************************************** -->
2145 <hr>
2146 <address>
2155 </address>
2156 </body>
2157 </html>