void ASTCodeGenVisitor::Visit(DecrData* s) {
IRBuilder<> builder = builders_.top();
Value* ptr_val = builder.CreateLoad(ptr_);
Value* result = builder.CreateAdd(ptr_val, neg_one);
builder.CreateStore(result, ptr_);
VisitNextASTNode(s);
}
Value* Id::codeGen(CodeGenContext &context) {
if (context.locals().find(name) == context.locals().end()) {
return NULL;
}
IRBuilder<> *builder = context.currentBuilder();
return builder->CreateLoad(context.locals()[name], name);
}
int main(int argc, char **argv)
{
InitializeNativeTarget();
LLVMContext &Context = getGlobalContext();
Module *m = new Module("test", Context);
Type *intTy = Type::getInt64Ty(Context);
StructType* structTy = StructType::create(Context, "struct.list");
std::vector<Type*> fields;
fields.push_back(intTy);
fields.push_back(PointerType::get(structTy, 0));
if (structTy->isOpaque()) {
structTy->setBody(fields, false);
}
/*
* int f1(struct x *p) { return p->next->l1; }
*/
std::vector<Type*> args_type;
args_type.push_back(PointerType::get(structTy, 0));
FunctionType *fnTy = FunctionType::get(intTy, args_type, false);
Function *func = Function::Create(fnTy,
GlobalValue::ExternalLinkage, "f1", m);
Value *v = func->arg_begin();
BasicBlock *bb = BasicBlock::Create(Context, "EntryBlock", func);
IRBuilder<> *builder = new IRBuilder<>(bb);
v = builder->CreateStructGEP(v, 1);
v = builder->CreateLoad(v, "load0");
v = builder->CreateStructGEP(v, 0);
v = builder->CreateLoad(v, "load1");
builder->CreateRet(v);
(*m).dump();
{
ExecutionEngine *ee = EngineBuilder(m).
setEngineKind(EngineKind::JIT).create();
void *f = ee->getPointerToFunction(func);
typedef int (*func_t) (struct x *);
struct x o = {10, NULL}, v = {};
v.next = &o;
std::cout << ((func_t)f)(&v) << std::endl;
}
return 0;
}
void CNodeCodeGenVisitor::Visit(COutput* s) {
IRBuilder<> builder = builders_.top();
Value* offset_ptr = builder.CreateGEP(ptr_, GetPtrOffset(s->GetOffset()));
Value* ptr_value = builder.CreateLoad(offset_ptr);
builder.CreateCall(put_char_, ptr_value);
VisitNextCNode(s);
}
vector<Value*> AssertionSiteInstrumenter::CollectArgs(
Instruction *Before, const Automaton& A,
Module& Mod, IRBuilder<>& Builder) {
// Find named values to be passed to instrumentation.
std::map<string,Value*> ValuesInScope;
for (auto G = Mod.global_begin(); G != Mod.global_end(); G++)
ValuesInScope[G->getName()] = G;
auto *Fn = Before->getParent()->getParent();
for (auto& Arg : Fn->getArgumentList())
ValuesInScope[Arg.getName()] = &Arg;
auto& EntryBlock(*Fn->begin());
for (auto& I : EntryBlock) {
auto *Inst = dyn_cast<AllocaInst>(&I);
if (!Inst)
break;
ValuesInScope[Inst->getName()] = Builder.CreateLoad(Inst);
}
int ArgSize = 0;
for (auto& Arg : A.getAssertion().argument())
if (!Arg.free())
ArgSize = std::max(ArgSize + 1, Arg.index());
vector<Value*> Args(ArgSize, NULL);
for (auto& Arg : A.getAssertion().argument()) {
if (Arg.free())
continue;
string Name(BaseName(Arg));
if (ValuesInScope.find(Name) == ValuesInScope.end()) {
string s;
raw_string_ostream Out(s);
for (auto v : ValuesInScope) {
Out << " \"" << v.first << "\": ";
v.second->getType()->print(Out);
Out << "\n";
}
panic("assertion references non-existent variable '" + BaseName(Arg)
+ "'; was it defined under '#ifdef TESLA'?\n\n"
"Variables in scope are:\n" + Out.str());
}
Args[Arg.index()] =
GetArgumentValue(ValuesInScope[Name], Arg, Builder, true);
}
return Args;
}
Value*
code_emitter::lookup_value (value_t *value)
{
g_assert(value->index >= 0);
if (value_map.find(value) != value_map.end())
return value_map[value];
g_assert(compiler_is_permanent_const_value(value));
return builder->CreateLoad(emit_const_value_addr(value));
}
Value* WyvernFunction::getNamedValue (string name)
{
Value* value;
value = namedValues[name];
if (dynamic_cast<AllocaInst*> (value) == NULL)
return value;
return Builder.CreateLoad (value);
}
CodeGenBlock::CodeGenBlock(int args, int locals, CodeGenLexicalScope
*enclosingScope, CodeGenModule *Mod)
: CodeGenLexicalScope(Mod), parentScope(enclosingScope)
{
Value *enclosingContext = enclosingScope->getContext();
// Define the layout of a block
BlockTy = StructType::get(
Mod->Context,
IdTy, // 0 - isa.
IMPTy, // 1 - Function pointer.
Type::getInt32Ty(Mod->Context),// 2 - Number of args.
enclosingContext->getType(), // 3 - Context.
NULL);
std::vector<const Type*> argTy;
argTy.push_back(PointerType::getUnqual(BlockTy));
// FIXME: Broken on Etoile runtime - _cmd needs to be a GEP on _call
argTy.push_back(SelTy);
for (int i=0 ; i<args ; ++i)
{
argTy.push_back(IdTy);
}
FunctionType *BlockFunctionTy = FunctionType::get(IdTy, argTy, false);
IRBuilder<> *MethodBuilder = enclosingScope->getBuilder();
// Create the block object
// The NewBlock function gets a block from a pool. It should really be
// inlined.
Block = MethodBuilder->CreateAlloca(BlockTy);
Module *TheModule = CGM->getModule();
// Create the block function
CurrentFunction = Function::Create(BlockFunctionTy,
GlobalValue::InternalLinkage, "BlockFunction", TheModule);
InitialiseFunction(Args, Locals, locals);
// Set the isa pointer
Value *isa = MethodBuilder->CreateLoad(
TheModule->getGlobalVariable(".smalltalk_block_stack_class", true));
storeInStruct(MethodBuilder, Block, isa, 0);
// Store the block function in the object
storeInStruct(MethodBuilder, Block,
MethodBuilder->CreateBitCast(CurrentFunction, IMPTy), 1);
// Store the number of arguments
storeInStruct(MethodBuilder, Block,
ConstantInt::get(Type::getInt32Ty(Mod->Context), args), 2);
// Set the context
storeInStruct(MethodBuilder, Block, enclosingScope->getContext(), 3);
}
void CNodeCodeGenVisitor::Visit(CMul* s) {
IRBuilder<> builder = builders_.top();
int op_offset = s->GetOpOffset();
int target_offset = s->GetTargetOffset();
int amt = s->GetAmt();
Value* op_offset_ptr = builder.CreateGEP(ptr_, GetPtrOffset(op_offset));
Value* target_offset_ptr =
builder.CreateGEP(ptr_, GetPtrOffset(target_offset));
Value* mul_val = GetDataOffset(amt);
Value* op_val = builder.CreateLoad(op_offset_ptr);
Value* target_val = builder.CreateLoad(target_offset_ptr);
Value* mul_result = builder.CreateMul(op_val, mul_val);
Value* add_result = builder.CreateAdd(target_val, mul_result);
builder.CreateStore(add_result, target_offset_ptr);
VisitNextCNode(s);
}
void CNodeCodeGenVisitor::Visit(CAdd* s) {
IRBuilder<> builder = builders_.top();
Value* offset_ptr = builder.CreateGEP(ptr_, GetPtrOffset(s->GetOffset()));
Value* offset_val = builder.CreateLoad(offset_ptr);
Value* add_val = GetDataOffset(s->GetAmt());
Value* result = builder.CreateAdd(offset_val, add_val);
builder.CreateStore(result, offset_ptr);
VisitNextCNode(s);
}
Value* ColExpression::getValue() {
IRBuilder<>* builder = codegen::getBuilder();
DataType dt = codegen::getAttType(index);
Value* tupleptr = codegen::getTupleptr();
Value *indices[1];
indices[0] = ConstantInt::get(Type::getInt32Ty(getGlobalContext()), (uint64_t) index);
ArrayRef<Value*> indicesRef(indices);
Value *dataptr = builder->CreateInBoundsGEP(tupleptr, indicesRef);
switch(dt){
case DOUBLE:
return builder->CreateLoad(
builder->CreateBitCast(dataptr, Type::getDoublePtrTy(getGlobalContext()))
);
case LONG:
case STRING:
case DATE:
return builder->CreateLoad(dataptr);
}
}
/// compile_put - Emit code for '.'
void BrainFTraceRecorder::compile_put(BrainFTraceNode *node,
IRBuilder<>& builder) {
Value *Loaded = builder.CreateLoad(DataPtr);
Value *Print =
builder.CreateSExt(Loaded, IntegerType::get(Loaded->getContext(), 32));
builder.CreateCall(putchar_func, Print);
if (node->left != (BrainFTraceNode*)~0ULL)
compile_opcode(node->left, builder);
else {
HeaderPHI->addIncoming(DataPtr, builder.GetInsertBlock());
builder.CreateBr(Header);
}
}
void WinEHStatePass::unlinkExceptionRegistration(IRBuilder<> &Builder) {
// Clone Link into the current BB for better address mode folding.
if (auto *GEP = dyn_cast<GetElementPtrInst>(Link)) {
GEP = cast<GetElementPtrInst>(GEP->clone());
Builder.Insert(GEP);
Link = GEP;
}
Type *LinkTy = getEHLinkRegistrationType();
// [fs:00] = Link->Next
Value *Next =
Builder.CreateLoad(Builder.CreateStructGEP(LinkTy, Link, 0));
Constant *FSZero =
Constant::getNullValue(LinkTy->getPointerTo()->getPointerTo(257));
Builder.CreateStore(Next, FSZero);
}
/// compile_minus - Emit code for '-'
void BrainFTraceRecorder::compile_minus(BrainFTraceNode *node,
IRBuilder<>& builder) {
Value *CellValue = builder.CreateLoad(DataPtr);
Constant *One =
ConstantInt::get(IntegerType::getInt8Ty(Header->getContext()), 1);
Value *UpdatedValue = builder.CreateSub(CellValue, One);
builder.CreateStore(UpdatedValue, DataPtr);
if (node->left != (BrainFTraceNode*)~0ULL)
compile_opcode(node->left, builder);
else {
HeaderPHI->addIncoming(DataPtr, builder.GetInsertBlock());
builder.CreateBr(Header);
}
}
void WinEHStatePass::linkExceptionRegistration(IRBuilder<> &Builder,
Function *Handler) {
// Emit the .safeseh directive for this function.
Handler->addFnAttr("safeseh");
Type *LinkTy = getEHLinkRegistrationType();
// Handler = Handler
Value *HandlerI8 = Builder.CreateBitCast(Handler, Builder.getInt8PtrTy());
Builder.CreateStore(HandlerI8, Builder.CreateStructGEP(LinkTy, Link, 1));
// Next = [fs:00]
Constant *FSZero =
Constant::getNullValue(LinkTy->getPointerTo()->getPointerTo(257));
Value *Next = Builder.CreateLoad(FSZero);
Builder.CreateStore(Next, Builder.CreateStructGEP(LinkTy, Link, 0));
// [fs:00] = Link
Builder.CreateStore(Link, FSZero);
}
void AddressSanitizer::instrumentAddress(Instruction *OrigIns,
IRBuilder<> &IRB, Value *Addr,
uint32_t TypeSize, bool IsWrite) {
Value *AddrLong = IRB.CreatePointerCast(Addr, IntptrTy);
Type *ShadowTy = IntegerType::get(
*C, std::max(8U, TypeSize >> MappingScale));
Type *ShadowPtrTy = PointerType::get(ShadowTy, 0);
Value *ShadowPtr = memToShadow(AddrLong, IRB);
Value *CmpVal = Constant::getNullValue(ShadowTy);
Value *ShadowValue = IRB.CreateLoad(
IRB.CreateIntToPtr(ShadowPtr, ShadowPtrTy));
Value *Cmp = IRB.CreateICmpNE(ShadowValue, CmpVal);
Instruction *CheckTerm = splitBlockAndInsertIfThen(
cast<Instruction>(Cmp)->getNextNode(), Cmp);
IRBuilder<> IRB2(CheckTerm);
size_t Granularity = 1 << MappingScale;
if (TypeSize < 8 * Granularity) {
// Addr & (Granularity - 1)
Value *Lower3Bits = IRB2.CreateAnd(
AddrLong, ConstantInt::get(IntptrTy, Granularity - 1));
// (Addr & (Granularity - 1)) + size - 1
Value *LastAccessedByte = IRB2.CreateAdd(
Lower3Bits, ConstantInt::get(IntptrTy, TypeSize / 8 - 1));
// (uint8_t) ((Addr & (Granularity-1)) + size - 1)
LastAccessedByte = IRB2.CreateIntCast(
LastAccessedByte, IRB.getInt8Ty(), false);
// ((uint8_t) ((Addr & (Granularity-1)) + size - 1)) >= ShadowValue
Value *Cmp2 = IRB2.CreateICmpSGE(LastAccessedByte, ShadowValue);
CheckTerm = splitBlockAndInsertIfThen(CheckTerm, Cmp2);
}
IRBuilder<> IRB1(CheckTerm);
Instruction *Crash = generateCrashCode(IRB1, AddrLong, IsWrite, TypeSize);
Crash->setDebugLoc(OrigIns->getDebugLoc());
ReplaceInstWithInst(CheckTerm, new UnreachableInst(*C));
}
Value*
code_emitter::convert_complex_return_value (Value *result)
{
/* The result is complex, whose representation
differs between archs, and we need to transform
it into another arch-dependent
representation. */
if (sizeof(gpointer) == 4)
{
Value *local = complex_copy_var;
Value *local_ptr = builder->CreateBitCast(local, PointerType::getUnqual(Type::Int64Ty));
builder->CreateStore(result, local_ptr);
result = builder->CreateLoad(local);
}
else if (sizeof(gpointer) == 8)
result = builder->CreateExtractValue(result, 0);
else
g_assert_not_reached();
return result;
}
// Rewrite final suspend point handling. We do not use suspend index to
// represent the final suspend point. Instead we zero-out ResumeFnAddr in the
// coroutine frame, since it is undefined behavior to resume a coroutine
// suspended at the final suspend point. Thus, in the resume function, we can
// simply remove the last case (when coro::Shape is built, the final suspend
// point (if present) is always the last element of CoroSuspends array).
// In the destroy function, we add a code sequence to check if ResumeFnAddress
// is Null, and if so, jump to the appropriate label to handle cleanup from the
// final suspend point.
static void handleFinalSuspend(IRBuilder<> &Builder, Value *FramePtr,
coro::Shape &Shape, SwitchInst *Switch,
bool IsDestroy) {
assert(Shape.HasFinalSuspend);
auto FinalCaseIt = std::prev(Switch->case_end());
BasicBlock *ResumeBB = FinalCaseIt->getCaseSuccessor();
Switch->removeCase(FinalCaseIt);
if (IsDestroy) {
BasicBlock *OldSwitchBB = Switch->getParent();
auto *NewSwitchBB = OldSwitchBB->splitBasicBlock(Switch, "Switch");
Builder.SetInsertPoint(OldSwitchBB->getTerminator());
auto *GepIndex = Builder.CreateConstInBoundsGEP2_32(Shape.FrameTy, FramePtr,
0, 0, "ResumeFn.addr");
auto *Load = Builder.CreateLoad(GepIndex);
auto *NullPtr =
ConstantPointerNull::get(cast<PointerType>(Load->getType()));
auto *Cond = Builder.CreateICmpEQ(Load, NullPtr);
Builder.CreateCondBr(Cond, ResumeBB, NewSwitchBB);
OldSwitchBB->getTerminator()->eraseFromParent();
}
}
static void IncrementTimeCounter(Value* Inc, Value* calle, unsigned Index, GlobalVariable* Counters, IRBuilder<>& Builder, Value* Point)
{
LLVMContext &Context = Inc->getContext();
//In order to insert instruction after Point, we use the nextIns function.
Value* nextIns = getNextIns(Point);
Builder.SetInsertPoint(dyn_cast<Instruction>(nextIns));
// Create the getelementptr constant expression
std::vector<Constant*> Indices(2);
Indices[0] = Constant::getNullValue(Type::getInt32Ty(Context));
Indices[1] = ConstantInt::get(Type::getInt32Ty(Context), Index);
Constant *ElementPtr =
ConstantExpr::getGetElementPtr(Counters, Indices);
// Load, increment and store the value back.
// Use this formula: a = a + end_time - start_time
ArrayRef<Value*> args;
CallInst* Inc_end = Builder.CreateCall(calle, args, "");
Value* OldVal = Builder.CreateLoad(ElementPtr, "OldTimeCounter");
Value* TmpVal = Builder.CreateFSub(OldVal, Inc, "TmpTimeCounter");
Value* NewVal = Builder.CreateFAdd(TmpVal, Inc_end, "NewTimeCounter");
Builder.CreateStore(NewVal, ElementPtr);
}
void AddressSanitizer::instrumentAddress(AsanFunctionContext &AFC,
Instruction *OrigIns,
IRBuilder<> &IRB, Value *Addr,
uint32_t TypeSize, bool IsWrite) {
Value *AddrLong = IRB.CreatePointerCast(Addr, IntptrTy);
Type *ShadowTy = IntegerType::get(
*C, std::max(8U, TypeSize >> MappingScale));
Type *ShadowPtrTy = PointerType::get(ShadowTy, 0);
Value *ShadowPtr = memToShadow(AddrLong, IRB);
Value *CmpVal = Constant::getNullValue(ShadowTy);
Value *ShadowValue = IRB.CreateLoad(
IRB.CreateIntToPtr(ShadowPtr, ShadowPtrTy));
Value *Cmp = IRB.CreateICmpNE(ShadowValue, CmpVal);
size_t AccessSizeIndex = TypeSizeToSizeIndex(TypeSize);
size_t Granularity = 1 << MappingScale;
TerminatorInst *CrashTerm = 0;
if (ClAlwaysSlowPath || (TypeSize < 8 * Granularity)) {
TerminatorInst *CheckTerm = splitBlockAndInsertIfThen(Cmp, false);
assert(dyn_cast<BranchInst>(CheckTerm)->isUnconditional());
BasicBlock *NextBB = CheckTerm->getSuccessor(0);
IRB.SetInsertPoint(CheckTerm);
Value *Cmp2 = createSlowPathCmp(IRB, AddrLong, ShadowValue, TypeSize);
BasicBlock *CrashBlock = BasicBlock::Create(*C, "", &AFC.F, NextBB);
CrashTerm = new UnreachableInst(*C, CrashBlock);
BranchInst *NewTerm = BranchInst::Create(CrashBlock, NextBB, Cmp2);
ReplaceInstWithInst(CheckTerm, NewTerm);
} else {
CrashTerm = splitBlockAndInsertIfThen(Cmp, true);
}
Instruction *Crash =
generateCrashCode(CrashTerm, AddrLong, IsWrite, AccessSizeIndex);
Crash->setDebugLoc(OrigIns->getDebugLoc());
}
codegen_value ast::distribution::createConstructor(Module *module, IRBuilder<> &builder,
const string &ctor_name,
Type *parameter_type,
const vector<type_spec> ¶m_type_list,
Value *eval, Value *sample, Value *pdf, Value *emit,
Function *dtor) {
//create function accepting parameters as arguments
vector<Type*> arg_types;
for (auto it = param_type_list.begin(); it != param_type_list.end(); ++it) arg_types.push_back((*it)->llvm_type());
FunctionType *ft = FunctionType::get(state->types["dfunc"]->llvm_type(), arg_types, false);
Function *f = Function::Create(ft, Function::ExternalLinkage, ctor_name, module);
BasicBlock *bb = BasicBlock::Create(getGlobalContext(), "func_entry", f);
builder.SetInsertPoint(bb);
//setup arguments for the alloc call
Value *gd_scene = module->getNamedGlobal(".__gd_scene");
assert(gd_scene != NULL);
Value *scene_ptr = builder.CreateLoad(gd_scene);
//compute the shader flags
codegen_value flag_val = codegen_all_flags(module, builder);
return errors::codegen_call(flag_val, [&] (Value *&flag_bitmask) -> codegen_value {
//get memory for a new distribution object
Value *dfunc_ptr = state->types["dfunc"]->allocate(module, builder);
//initialize the object and dynamically allocate parameter memory (calling a builtin function)
Type* int_ptr_ty = Type::getInt32Ty(getGlobalContext())->getPointerTo();
vector<Type*> alloc_arg_types({state->types["scene_ptr"]->llvm_type(),
Type::getInt32Ty(getGlobalContext()), state->types["shader_flag"]->llvm_type(),
int_ptr_ty, int_ptr_ty, int_ptr_ty, int_ptr_ty,
dtor->getType(), dfunc_ptr->getType()});
FunctionType *alloc_type = FunctionType::get(Type::getInt32PtrTy(getGlobalContext()), alloc_arg_types, false);
Function *alloc_func = GetExternalFunction(module, "gd_builtin_alloc_dfunc", alloc_type);
int param_data_size = DataLayout(module).getTypeAllocSize(parameter_type);
Constant *param_size_arg = ConstantInt::get(getGlobalContext(), APInt(8*sizeof(int), param_data_size));
vector<Value*> alloc_args({scene_ptr, param_size_arg, flag_bitmask,
builder.CreatePointerCast(eval, int_ptr_ty),
builder.CreatePointerCast(sample, int_ptr_ty),
builder.CreatePointerCast(pdf, int_ptr_ty),
builder.CreatePointerCast(emit, int_ptr_ty),
dtor, dfunc_ptr});
Value *param_ptr = builder.CreatePointerCast(builder.CreateCall(alloc_func, alloc_args),
parameter_type->getPointerTo(), "dfunc_param_ptr");
//set each parameter
auto arg_it = f->arg_begin();
unsigned int field_idx = 0;
for (auto it = param_type_list.begin(); it != param_type_list.end(); ++it, ++arg_it, ++field_idx) {
Value *param_copy = (*it)->copy(arg_it, module, builder);
(*it)->store(param_copy, builder.CreateStructGEP(param_ptr, field_idx), module, builder);
}
//return the object
Value *rt_val = builder.CreateLoad(dfunc_ptr, "dist_ref");
builder.CreateRet(rt_val);
return f;
});
}
void ASTCodeGenVisitor::Visit(Output* s) {
IRBuilder<> builder = builders_.top();
Value* output = builder.CreateLoad(ptr_);
builder.CreateCall(put_char_, output);
VisitNextASTNode(s);
}
/// compile_if - Emit code for '['
void BrainFTraceRecorder::compile_if(BrainFTraceNode *node,
IRBuilder<>& builder) {
BasicBlock *ZeroChild = 0;
BasicBlock *NonZeroChild = 0;
BasicBlock *Parent = builder.GetInsertBlock();
LLVMContext &Context = Header->getContext();
// If both directions of the branch go back to the trace-head, just
// jump there directly.
if (node->left == (BrainFTraceNode*)~0ULL &&
node->right == (BrainFTraceNode*)~0ULL) {
HeaderPHI->addIncoming(DataPtr, builder.GetInsertBlock());
builder.CreateBr(Header);
return;
}
// Otherwise, there are two cases to handle for each direction:
// ~0ULL - A branch back to the trace head
// 0 - A branch out of the trace
// * - A branch to a node we haven't compiled yet.
// Go ahead and generate code for both targets.
if (node->left == (BrainFTraceNode*)~0ULL) {
NonZeroChild = Header;
HeaderPHI->addIncoming(DataPtr, Parent);
} else if (node->left == 0) {
NonZeroChild = BasicBlock::Create(Context,
"exit_left_"+utostr(node->pc),
Header->getParent());
builder.SetInsertPoint(NonZeroChild);
// Set the extension leaf, which is a pointer to the leaf of the trace
// tree from which we are side exiting.
ConstantInt *ExtLeaf = ConstantInt::get(int_type, (intptr_t)node);
builder.CreateStore(ExtLeaf, ext_leaf);
ConstantInt *NewPc = ConstantInt::get(int_type, node->pc+1);
Value *BytecodeIndex =
builder.CreateConstInBoundsGEP1_32(bytecode_array, node->pc+1);
Value *Target = builder.CreateLoad(BytecodeIndex);
CallInst *Call =cast<CallInst>(builder.CreateCall2(Target, NewPc, DataPtr));
Call->setTailCall();
builder.CreateRetVoid();
} else {
NonZeroChild = BasicBlock::Create(Context,
utostr(node->left->pc),
Header->getParent());
builder.SetInsertPoint(NonZeroChild);
compile_opcode(node->left, builder);
}
if (node->right == (BrainFTraceNode*)~0ULL) {
ZeroChild = Header;
HeaderPHI->addIncoming(DataPtr, Parent);
} else if (node->right == 0) {
ZeroChild = BasicBlock::Create(Context,
"exit_right_"+utostr(node->pc),
Header->getParent());
builder.SetInsertPoint(ZeroChild);
// Set the extension leaf, which is a pointer to the leaf of the trace
// tree from which we are side exiting.
ConstantInt *ExtLeaf = ConstantInt::get(int_type, (intptr_t)node);
builder.CreateStore(ExtLeaf, ext_leaf);
ConstantInt *NewPc = ConstantInt::get(int_type, JumpMap[node->pc]+1);
Value *BytecodeIndex =
builder.CreateConstInBoundsGEP1_32(bytecode_array, JumpMap[node->pc]+1);
Value *Target = builder.CreateLoad(BytecodeIndex);
CallInst *Call =cast<CallInst>(builder.CreateCall2(Target, NewPc, DataPtr));
Call->setTailCall();
builder.CreateRetVoid();
} else {
ZeroChild = BasicBlock::Create(Context,
utostr(node->right->pc),
Header->getParent());
builder.SetInsertPoint(ZeroChild);
compile_opcode(node->right, builder);
}
// Generate the test and branch to select between the targets.
builder.SetInsertPoint(Parent);
Value *Loaded = builder.CreateLoad(DataPtr);
Value *Cmp = builder.CreateICmpEQ(Loaded,
ConstantInt::get(Loaded->getType(), 0));
builder.CreateCondBr(Cmp, ZeroChild, NonZeroChild);
}
JITFunction*
CompilePipeline(Pipeline *pipeline, Thread *thread) {
size_t i = 0;
size_t size = 0;
std::unique_ptr<Module> owner = make_unique<Module>("PipelineFunction", thread->context);
Module *module = owner.get();
std::string fname = std::string("PipelineFunction") + std::to_string(thread->functions++);
size_t input_count = pipeline->inputData->objects.size();
size_t output_count = pipeline->outputData->objects.size();
size_t function_arg_count = 6;
size_t arg_count = input_count + output_count + (function_arg_count - 2);
size_t start_addr = input_count + output_count;
size_t end_addr = start_addr + 1;
size_t result_sizes_addr = end_addr + 1;
size_t thread_nr_addr = result_sizes_addr + 1;
IRBuilder<> *builder = &thread->builder;
auto passmanager = CreatePassManager(module, thread->jit.get());
module->setDataLayout(thread->jit->getTargetMachine().createDataLayout());
Type *int8_tpe = Type::getInt8Ty(thread->context);
Type *int8ptr_tpe = PointerType::get(int8_tpe, 0);
Type *int8ptrptr_tpe = PointerType::get(int8ptr_tpe, 0);
Type *int64_tpe = Type::getInt64Ty(thread->context);
Type *int64ptr_tpe = PointerType::get(int64_tpe, 0);
JITInformation info;
// arguments of the function
// the arguments are (void **result, void** inputs, size_t start, size_t end);
// note that we don't actually use void**, we use int8**, because LLVM does not support void pointers
std::vector<Type*> arguments(function_arg_count);
i = 0;
arguments[i++] = int8ptrptr_tpe; // void** results
arguments[i++] = int8ptrptr_tpe; // void** inputs
arguments[i++] = int64_tpe; // size_t start
arguments[i++] = int64_tpe; // size_t end
arguments[i++] = int64ptr_tpe; // size_t* result_sizes
arguments[i++] = int64_tpe; // size_t thread_nr
assert(i == function_arg_count);
/*for(auto inputs = pipeline->inputData->objects.begin(); inputs != pipeline->inputData->objects.end(); inputs++, i++) {
arguments[i] = PointerType::get(getLLVMType(thread->context, inputs->type), 0);
}
for(auto outputs = pipeline->outputData->objects.begin(); outputs != pipeline->outputData->objects.end(); outputs++, i++) {
arguments[i] = PointerType::get(getLLVMType(thread->context, outputs->type), 0);
}*/
// create the LLVM function
FunctionType *prototype = FunctionType::get(int64_tpe, arguments, false);
Function *function = Function::Create(prototype, GlobalValue::ExternalLinkage, fname, module);
function->setCallingConv(CallingConv::C);
// create the basic blocks
BasicBlock *loop_entry = BasicBlock::Create(thread->context, "entry", function, 0);
BasicBlock *loop_cond = BasicBlock::Create(thread->context, "for.cond", function, 0);
BasicBlock *loop_body = BasicBlock::Create(thread->context, "for.body", function, 0);
BasicBlock *loop_inc = BasicBlock::Create(thread->context, "for.inc", function, 0);
BasicBlock *loop_end = BasicBlock::Create(thread->context, "for.end", function, 0);
info.builder = &thread->builder;
info.context = &thread->context;
info.function = function;
info.loop_entry = loop_entry;
info.loop_cond = loop_cond;
info.loop_body = loop_body;
info.loop_inc = loop_inc;
info.loop_end = loop_end;
info.current = loop_body;
#ifndef _NOTDEBUG
// argument names (for debug purposes only)
std::vector<std::string> argument_names(arg_count);
i = 0;
for(auto inputs = pipeline->inputData->objects.begin(); inputs != pipeline->inputData->objects.end(); inputs++, i++) {
argument_names[i] = std::string("inputs") + std::to_string(i);
}
for(auto outputs = pipeline->outputData->objects.begin(); outputs != pipeline->outputData->objects.end(); outputs++, i++) {
argument_names[i] = std::string("outputs") + std::to_string(i - input_count);
}
argument_names[i++] = "start";
argument_names[i++] = "end";
argument_names[i++] = "result_sizes";
argument_names[i++] = "thread_nr";
#endif
std::vector<AllocaInst*> argument_addresses(arg_count);
builder->SetInsertPoint(loop_entry);
{
// allocate space for the arguments
auto args = function->arg_begin();
i = 0;
for(auto outputs = pipeline->outputData->objects.begin(); outputs != pipeline->outputData->objects.end(); outputs++, i++) {
Type *column_type = PointerType::get(getLLVMType(thread->context, outputs->source->type), 0);
Value *voidptrptr = builder->CreateGEP(int8ptr_tpe, &*args, ConstantInt::get(int64_tpe, i, true));
Value *voidptr = builder->CreateLoad(voidptrptr, "voidptr");
Value *columnptr = builder->CreatePointerCast(voidptr, column_type);
argument_addresses[i] = builder->CreateAlloca(column_type, nullptr, argument_names[i]);
builder->CreateStore(columnptr, argument_addresses[i]);
outputs->alloca_address = (void*) argument_addresses[i];
if (size == 0 || size == 1) {
assert(outputs->source->size >= 0);
size = outputs->source->size;
}
assert(size == outputs->source->size || outputs->source->size == 1);
}
args++;
for(auto inputs = pipeline->inputData->objects.begin(); inputs != pipeline->inputData->objects.end(); inputs++, i++) {
Type *column_type = PointerType::get(getLLVMType(thread->context, inputs->source->type), 0);
Value *voidptrptr = builder->CreateGEP(int8ptr_tpe, &*args, ConstantInt::get(int64_tpe, i - output_count, true));
Value *voidptr = builder->CreateLoad(voidptrptr, "voidptr");
Value *columnptr = builder->CreatePointerCast(voidptr, column_type);
argument_addresses[i] = builder->CreateAlloca(column_type, nullptr, argument_names[i]);
builder->CreateStore(columnptr, argument_addresses[i]);
inputs->alloca_address = (void*) argument_addresses[i];
if (size == 0 || size == 1) {
assert(inputs->source->size >= 0);
size = inputs->source->size;
}
assert(size == inputs->source->size || inputs->source->size == 1);
}
args++;
argument_addresses[i] = builder->CreateAlloca(arguments[2], nullptr, argument_names[i]);
builder->CreateStore(&*args, argument_addresses[i]);
args++; i++;
argument_addresses[i] = builder->CreateAlloca(arguments[3], nullptr, argument_names[i]);
builder->CreateStore(&*args, argument_addresses[i]);
args++; i++;
argument_addresses[i] = builder->CreateAlloca(arguments[4], nullptr, argument_names[i]);
builder->CreateStore(&*args, argument_addresses[i]);
args++; i++;
argument_addresses[i] = builder->CreateAlloca(arguments[5], nullptr, argument_names[i]);
builder->CreateStore(&*args, argument_addresses[i]);
args++; i++;
assert(args == function->arg_end());
assert(i == arg_count);
info.index_addr = argument_addresses[start_addr];
info.thread_addr = argument_addresses[thread_nr_addr];
PerformInitialization(info, pipeline->operation);
builder->CreateBr(loop_cond);
}
// for loop condition: index < end
builder->SetInsertPoint(loop_cond);
{
LoadInst *index = builder->CreateLoad(argument_addresses[start_addr], "index");
LoadInst *end = builder->CreateLoad(argument_addresses[end_addr], "end");
Value *condition = builder->CreateICmpSLT(index, end, "index < end");
builder->CreateCondBr(condition, loop_body, loop_end);
}
// loop body: perform the computation
builder->SetInsertPoint(loop_body);
{
LoadInst *index = builder->CreateLoad(argument_addresses[start_addr], "index");
info.index = index;
info.index_addr = argument_addresses[start_addr];
// perform the computation over the given index
// we don't use the return value because the final assignment has already taken place
Value *v = PerformOperation(info, thread->builder, thread->context, pipeline->operation, pipeline->inputData, pipeline->outputData);
if (v == NULL) {
// failed to perform operation
printf("Failed to compile pipeline %s\n", pipeline->name);
return NULL;
}
builder->CreateBr(loop_inc);
}
// loop increment: index++
builder->SetInsertPoint(loop_inc);
{
LoadInst *index = builder->CreateLoad(argument_addresses[start_addr], "index");
Value *incremented_index = builder->CreateAdd(index, ConstantInt::get(int64_tpe, 1, true), "index++");
builder->CreateStore(incremented_index, argument_addresses[start_addr]);
builder->CreateBr(loop_cond);
}
// loop end: return; (nothing happens here because we have no return value)
builder->SetInsertPoint(loop_end);
{
// return the output size of each of the columns
int i = 0;
Value *result_sizes = builder->CreateLoad(argument_addresses[result_sizes_addr], "result_sizes[]");
for(auto it = pipeline->outputData->objects.begin(); it != pipeline->outputData->objects.end(); it++) {
Value* output_count;
if (it->index_addr) {
output_count = builder->CreateLoad((Value*) it->index_addr, "count");
} else {
output_count = ConstantInt::get(int64_tpe, 1, true);
}
Value *output_addr = builder->CreateGEP(int64_tpe, result_sizes, ConstantInt::get(int64_tpe, i, true));
builder->CreateStore(output_count, output_addr);
i++;
}
builder->CreateRet(ConstantInt::get(int64_tpe, 0, true));
}
#ifndef _NOTDEBUG
verifyFunction(*function);
verifyModule(*module);
#endif
//printf("LLVM for pipeline %s\n", pipeline->name);
module->dump();
passmanager->run(*function);
// dump generated LLVM code
//module->dump();
auto handle = thread->jit->addModule(std::move(owner));
jit_function compiled_function = (jit_function) thread->jit->findSymbol(fname).getAddress();
if (!compiled_function) {
printf("Error creating function.\n");
return NULL;
}
JITFunction *jf = CreateJITFunction(thread, pipeline);
jf->size = size;
jf->function = compiled_function;
jf->jit = thread->jit.get();
jf->handle = handle;
assert(jf->function);
return jf;
}