ldc/gen/llvmhelpers.cpp
Frits van Bommel 1958e17734 Rewrite StructLiteralExp::toElem to store individual fields instead of
generating a constant to fill the entire struct with a single `store`.
This is much more efficient at compile time (fixing #320) and vastly reduces
the size of the emitted code. Since LLVM no longer needs to keep the data for
all fields in "registers" until the store happens, it should also be more
efficient at run time in cases where the fields aren't assigned with constants.

There's also some code clean-up by removing duplicated logic.
2009-06-06 20:16:13 +02:00

1439 lines
46 KiB
C++

#include "gen/llvmhelpers.h"
#include "gen/llvm.h"
#include "llvm/Target/TargetMachineRegistry.h"
#include "mars.h"
#include "init.h"
#include "id.h"
#include "expression.h"
#include "template.h"
#include "module.h"
#include "gen/tollvm.h"
#include "gen/irstate.h"
#include "gen/runtime.h"
#include "gen/logger.h"
#include "gen/arrays.h"
#include "gen/dvalue.h"
#include "gen/complex.h"
#include "gen/classes.h"
#include "gen/functions.h"
#include "gen/typeinf.h"
#include "gen/todebug.h"
#include "gen/cl_options.h"
#include "gen/nested.h"
#include "ir/irmodule.h"
#include <stack>
/****************************************************************************************/
/*////////////////////////////////////////////////////////////////////////////////////////
// DYNAMIC MEMORY HELPERS
////////////////////////////////////////////////////////////////////////////////////////*/
LLValue* DtoNew(Type* newtype)
{
// get runtime function
llvm::Function* fn = LLVM_D_GetRuntimeFunction(gIR->module, "_d_allocmemoryT");
// get type info
LLConstant* ti = DtoTypeInfoOf(newtype);
assert(isaPointer(ti));
// call runtime allocator
LLValue* mem = gIR->CreateCallOrInvoke(fn, ti, ".gc_mem").getInstruction();
// cast
return DtoBitCast(mem, getPtrToType(DtoType(newtype)), ".gc_mem");
}
void DtoDeleteMemory(LLValue* ptr)
{
// get runtime function
llvm::Function* fn = LLVM_D_GetRuntimeFunction(gIR->module, "_d_delmemory");
// build args
LLSmallVector<LLValue*,1> arg;
arg.push_back(DtoBitCast(ptr, getVoidPtrType(), ".tmp"));
// call
gIR->CreateCallOrInvoke(fn, arg.begin(), arg.end());
}
void DtoDeleteClass(LLValue* inst)
{
// get runtime function
llvm::Function* fn = LLVM_D_GetRuntimeFunction(gIR->module, "_d_delclass");
// build args
LLSmallVector<LLValue*,1> arg;
arg.push_back(DtoBitCast(inst, fn->getFunctionType()->getParamType(0), ".tmp"));
// call
gIR->CreateCallOrInvoke(fn, arg.begin(), arg.end());
}
void DtoDeleteInterface(LLValue* inst)
{
// get runtime function
llvm::Function* fn = LLVM_D_GetRuntimeFunction(gIR->module, "_d_delinterface");
// build args
LLSmallVector<LLValue*,1> arg;
arg.push_back(DtoBitCast(inst, fn->getFunctionType()->getParamType(0), ".tmp"));
// call
gIR->CreateCallOrInvoke(fn, arg.begin(), arg.end());
}
void DtoDeleteArray(DValue* arr)
{
// get runtime function
llvm::Function* fn = LLVM_D_GetRuntimeFunction(gIR->module, "_d_delarray");
// build args
LLSmallVector<LLValue*,2> arg;
arg.push_back(DtoArrayLen(arr));
arg.push_back(DtoBitCast(DtoArrayPtr(arr), getVoidPtrType(), ".tmp"));
// call
gIR->CreateCallOrInvoke(fn, arg.begin(), arg.end());
}
/****************************************************************************************/
/*////////////////////////////////////////////////////////////////////////////////////////
// ALLOCA HELPERS
////////////////////////////////////////////////////////////////////////////////////////*/
llvm::AllocaInst* DtoAlloca(Type* type, const char* name)
{
const llvm::Type* lltype = DtoType(type);
llvm::AllocaInst* ai = new llvm::AllocaInst(lltype, name, gIR->topallocapoint());
ai->setAlignment(type->alignsize());
return ai;
}
llvm::AllocaInst* DtoArrayAlloca(Type* type, unsigned arraysize, const char* name)
{
const llvm::Type* lltype = DtoType(type);
llvm::AllocaInst* ai = new llvm::AllocaInst(
lltype, DtoConstUint(arraysize), name, gIR->topallocapoint());
ai->setAlignment(type->alignsize());
return ai;
}
llvm::AllocaInst* DtoRawAlloca(const llvm::Type* lltype, size_t alignment, const char* name)
{
llvm::AllocaInst* ai = new llvm::AllocaInst(lltype, name, gIR->topallocapoint());
if (alignment)
ai->setAlignment(alignment);
return ai;
}
/****************************************************************************************/
/*////////////////////////////////////////////////////////////////////////////////////////
// ASSERT HELPER
////////////////////////////////////////////////////////////////////////////////////////*/
void DtoAssert(Module* M, Loc loc, DValue* msg)
{
std::vector<LLValue*> args;
// func
const char* fname = msg ? "_d_assert_msg" : "_d_assert";
llvm::Function* fn = LLVM_D_GetRuntimeFunction(gIR->module, fname);
// msg param
if (msg)
{
args.push_back(msg->getRVal());
}
// file param
// we might be generating for an imported template function
const char* cur_file = M->srcfile->name->toChars();
if (loc.filename && strcmp(loc.filename, cur_file) != 0)
{
args.push_back(DtoConstString(loc.filename));
}
else
{
IrModule* irmod = getIrModule(M);
args.push_back(DtoLoad(irmod->fileName));
}
// line param
LLConstant* c = DtoConstUint(loc.linnum);
args.push_back(c);
// call
gIR->CreateCallOrInvoke(fn, args.begin(), args.end());
// end debug info
if (global.params.symdebug)
DtoDwarfFuncEnd(gIR->func()->decl);
// after assert is always unreachable
gIR->ir->CreateUnreachable();
}
/****************************************************************************************/
/*////////////////////////////////////////////////////////////////////////////////////////
// LABEL HELPER
////////////////////////////////////////////////////////////////////////////////////////*/
LabelStatement* DtoLabelStatement(Identifier* ident)
{
FuncDeclaration* fd = gIR->func()->decl;
FuncDeclaration::LabelMap::iterator iter = fd->labmap.find(ident->toChars());
if (iter == fd->labmap.end())
{
if (fd->returnLabel && fd->returnLabel->ident->equals(ident))
{
assert(fd->returnLabel->statement);
return fd->returnLabel->statement;
}
return NULL;
}
return iter->second;
}
/****************************************************************************************/
/*////////////////////////////////////////////////////////////////////////////////////////
// GOTO HELPER
////////////////////////////////////////////////////////////////////////////////////////*/
void DtoGoto(Loc loc, Identifier* target, TryFinallyStatement* sourceFinally)
{
assert(!gIR->scopereturned());
LabelStatement* lblstmt = DtoLabelStatement(target);
if(!lblstmt) {
error(loc, "the label %s does not exist", target->toChars());
fatal();
}
// if the target label is inside inline asm, error
if(lblstmt->asmLabel) {
error(loc, "cannot goto to label %s inside an inline asm block", target->toChars());
fatal();
}
// find target basic block
std::string labelname = gIR->func()->getScopedLabelName(target->toChars());
llvm::BasicBlock*& targetBB = gIR->func()->labelToBB[labelname];
if (targetBB == NULL)
targetBB = llvm::BasicBlock::Create("label_" + labelname, gIR->topfunc());
// emit code for finallys between goto and label
DtoEnclosingHandlers(loc, lblstmt);
// goto into finally blocks is forbidden by the spec
// but should work fine
if(lblstmt->enclosingFinally != sourceFinally) {
error(loc, "spec disallows goto into or out of finally block");
fatal();
}
llvm::BranchInst::Create(targetBB, gIR->scopebb());
}
/****************************************************************************************/
/*////////////////////////////////////////////////////////////////////////////////////////
// TRY-FINALLY, VOLATILE AND SYNCHRONIZED HELPER
////////////////////////////////////////////////////////////////////////////////////////*/
void EnclosingSynchro::emitCode(IRState * p)
{
if (s->exp)
DtoLeaveMonitor(s->exp->toElem(p)->getRVal());
else
DtoLeaveCritical(s->llsync);
}
////////////////////////////////////////////////////////////////////////////////////////
void EnclosingVolatile::emitCode(IRState * p)
{
// store-load barrier
DtoMemoryBarrier(false, false, true, false);
}
////////////////////////////////////////////////////////////////////////////////////////
void EnclosingTryFinally::emitCode(IRState * p)
{
if (tf->finalbody)
{
llvm::BasicBlock* oldpad = p->func()->landingPad;
p->func()->landingPad = landingPad;
tf->finalbody->toIR(p);
p->func()->landingPad = oldpad;
}
}
////////////////////////////////////////////////////////////////////////////////////////
void DtoEnclosingHandlers(Loc loc, Statement* target)
{
// labels are a special case: they are not required to enclose the current scope
// for them we use the enclosing scope handler as a reference point
LabelStatement* lblstmt = dynamic_cast<LabelStatement*>(target);
if (lblstmt)
target = lblstmt->enclosingScopeExit;
// figure out up until what handler we need to emit
IrFunction::TargetScopeVec::reverse_iterator targetit = gIR->func()->targetScopes.rbegin();
IrFunction::TargetScopeVec::reverse_iterator it_end = gIR->func()->targetScopes.rend();
while(targetit != it_end) {
if (targetit->s == target) {
break;
}
++targetit;
}
if (target && targetit == it_end) {
if (lblstmt)
error(loc, "cannot goto into try, volatile or synchronized statement at %s", target->loc.toChars());
else
error(loc, "internal error, cannot find jump path to statement at %s", target->loc.toChars());
return;
}
//
// emit code for enclosing handlers
//
// since the labelstatements possibly inside are private
// and might already exist push a label scope
gIR->func()->pushUniqueLabelScope("enclosing");
IrFunction::TargetScopeVec::reverse_iterator it = gIR->func()->targetScopes.rbegin();
while (it != targetit) {
if (it->enclosinghandler)
it->enclosinghandler->emitCode(gIR);
++it;
}
gIR->func()->popLabelScope();
}
/****************************************************************************************/
/*////////////////////////////////////////////////////////////////////////////////////////
// SYNCHRONIZED SECTION HELPERS
////////////////////////////////////////////////////////////////////////////////////////*/
void DtoEnterCritical(LLValue* g)
{
LLFunction* fn = LLVM_D_GetRuntimeFunction(gIR->module, "_d_criticalenter");
gIR->CreateCallOrInvoke(fn, g);
}
void DtoLeaveCritical(LLValue* g)
{
LLFunction* fn = LLVM_D_GetRuntimeFunction(gIR->module, "_d_criticalexit");
gIR->CreateCallOrInvoke(fn, g);
}
void DtoEnterMonitor(LLValue* v)
{
LLFunction* fn = LLVM_D_GetRuntimeFunction(gIR->module, "_d_monitorenter");
v = DtoBitCast(v, fn->getFunctionType()->getParamType(0));
gIR->CreateCallOrInvoke(fn, v);
}
void DtoLeaveMonitor(LLValue* v)
{
LLFunction* fn = LLVM_D_GetRuntimeFunction(gIR->module, "_d_monitorexit");
v = DtoBitCast(v, fn->getFunctionType()->getParamType(0));
gIR->CreateCallOrInvoke(fn, v);
}
/****************************************************************************************/
/*////////////////////////////////////////////////////////////////////////////////////////
// ASSIGNMENT HELPER (store this in that)
////////////////////////////////////////////////////////////////////////////////////////*/
// is this a good approach at all ?
void DtoAssign(Loc& loc, DValue* lhs, DValue* rhs)
{
Logger::println("DtoAssign(...);\n");
LOG_SCOPE;
Type* t = lhs->getType()->toBasetype();
Type* t2 = rhs->getType()->toBasetype();
if (t->ty == Tstruct) {
if (!t->equals(t2)) {
// FIXME: use 'rhs' for something !?!
DtoAggrZeroInit(lhs->getLVal());
}
else {
DtoAggrCopy(lhs->getLVal(), rhs->getRVal());
}
}
else if (t->ty == Tarray) {
// lhs is slice
if (DSliceValue* s = lhs->isSlice()) {
if (DSliceValue* s2 = rhs->isSlice()) {
DtoArrayCopySlices(s, s2);
}
else if (t->nextOf()->toBasetype()->equals(t2)) {
DtoArrayInit(loc, s, rhs);
}
else {
DtoArrayCopyToSlice(s, rhs);
}
}
// rhs is slice
else if (DSliceValue* s = rhs->isSlice()) {
assert(s->getType()->toBasetype() == lhs->getType()->toBasetype());
DtoSetArray(lhs->getLVal(),DtoArrayLen(s),DtoArrayPtr(s));
}
// null
else if (rhs->isNull()) {
DtoSetArrayToNull(lhs->getLVal());
}
// reference assignment
else if (t2->ty == Tarray) {
DtoStore(rhs->getRVal(), lhs->getLVal());
}
// some implicitly converting ref assignment
else {
DtoSetArray(lhs->getLVal(), DtoArrayLen(rhs), DtoArrayPtr(rhs));
}
}
else if (t->ty == Tsarray) {
// T[n] = T[n]
if (DtoType(lhs->getType()) == DtoType(rhs->getType())) {
DtoStaticArrayCopy(lhs->getLVal(), rhs->getRVal());
}
// T[n] = T
else if (t->nextOf()->toBasetype()->equals(t2)) {
DtoArrayInit(loc, lhs, rhs);
}
// T[n] = T[] - generally only generated by frontend in rare cases
else if (t2->ty == Tarray && t->nextOf()->toBasetype()->equals(t2->nextOf()->toBasetype())) {
DtoMemCpy(lhs->getLVal(), DtoArrayPtr(rhs), DtoArrayLen(rhs));
} else {
assert(0 && "Unimplemented static array assign!");
}
}
else if (t->ty == Tdelegate) {
LLValue* l = lhs->getLVal();
LLValue* r = rhs->getRVal();
if (Logger::enabled())
Logger::cout() << "assign\nlhs: " << *l << "rhs: " << *r << '\n';
DtoStore(r, l);
}
else if (t->ty == Tclass) {
assert(t2->ty == Tclass);
LLValue* l = lhs->getLVal();
LLValue* r = rhs->getRVal();
if (Logger::enabled())
{
Logger::cout() << "l : " << *l << '\n';
Logger::cout() << "r : " << *r << '\n';
}
r = DtoBitCast(r, l->getType()->getContainedType(0));
DtoStore(r, l);
}
else if (t->iscomplex()) {
LLValue* dst = lhs->getLVal();
LLValue* src = DtoCast(loc, rhs, lhs->getType())->getRVal();
DtoStore(src, dst);
}
else {
LLValue* l = lhs->getLVal();
LLValue* r = rhs->getRVal();
if (Logger::enabled())
Logger::cout() << "assign\nlhs: " << *l << "rhs: " << *r << '\n';
const LLType* lit = l->getType()->getContainedType(0);
if (r->getType() != lit) {
r = DtoCast(loc, rhs, lhs->getType())->getRVal();
if (Logger::enabled())
Logger::cout() << "really assign\nlhs: " << *l << "rhs: " << *r << '\n';
assert(r->getType() == l->getType()->getContainedType(0));
}
gIR->ir->CreateStore(r, l);
}
}
/****************************************************************************************/
/*////////////////////////////////////////////////////////////////////////////////////////
// NULL VALUE HELPER
////////////////////////////////////////////////////////////////////////////////////////*/
DValue* DtoNullValue(Type* type)
{
Type* basetype = type->toBasetype();
TY basety = basetype->ty;
const LLType* lltype = DtoType(basetype);
// complex, needs to be first since complex are also floating
if (basetype->iscomplex())
{
const LLType* basefp = DtoComplexBaseType(basetype);
LLValue* res = DtoAggrPair(DtoType(type), LLConstant::getNullValue(basefp), LLConstant::getNullValue(basefp));
return new DImValue(type, res);
}
// integer, floating, pointer and class have no special representation
else if (basetype->isintegral() || basetype->isfloating() || basety == Tpointer || basety == Tclass)
{
return new DConstValue(type, LLConstant::getNullValue(lltype));
}
// dynamic array
else if (basety == Tarray)
{
LLValue* len = DtoConstSize_t(0);
LLValue* ptr = getNullPtr(getPtrToType(DtoType(basetype->nextOf())));
return new DSliceValue(type, len, ptr);
}
// delegate
else if (basety == Tdelegate)
{
return new DNullValue(type, LLConstant::getNullValue(lltype));
}
// unknown
llvm::cout << "unsupported: null value for " << type->toChars() << '\n';
assert(0);
return 0;
}
/****************************************************************************************/
/*////////////////////////////////////////////////////////////////////////////////////////
// CASTING HELPERS
////////////////////////////////////////////////////////////////////////////////////////*/
DValue* DtoCastInt(Loc& loc, DValue* val, Type* _to)
{
const LLType* tolltype = DtoType(_to);
Type* to = _to->toBasetype();
Type* from = val->getType()->toBasetype();
assert(from->isintegral());
size_t fromsz = from->size();
size_t tosz = to->size();
LLValue* rval = val->getRVal();
if (rval->getType() == tolltype) {
return new DImValue(_to, rval);
}
if (to->ty == Tbool) {
LLValue* zero = LLConstantInt::get(rval->getType(), 0, false);
rval = gIR->ir->CreateICmpNE(rval, zero, "tmp");
}
else if (to->isintegral()) {
if (fromsz < tosz || from->ty == Tbool) {
if (Logger::enabled())
Logger::cout() << "cast to: " << *tolltype << '\n';
if (from->isunsigned() || from->ty == Tbool) {
rval = new llvm::ZExtInst(rval, tolltype, "tmp", gIR->scopebb());
} else {
rval = new llvm::SExtInst(rval, tolltype, "tmp", gIR->scopebb());
}
}
else if (fromsz > tosz) {
rval = new llvm::TruncInst(rval, tolltype, "tmp", gIR->scopebb());
}
else {
rval = DtoBitCast(rval, tolltype);
}
}
else if (to->iscomplex()) {
return DtoComplex(loc, to, val);
}
else if (to->isfloating()) {
if (from->isunsigned()) {
rval = new llvm::UIToFPInst(rval, tolltype, "tmp", gIR->scopebb());
}
else {
rval = new llvm::SIToFPInst(rval, tolltype, "tmp", gIR->scopebb());
}
}
else if (to->ty == Tpointer) {
if (Logger::enabled())
Logger::cout() << "cast pointer: " << *tolltype << '\n';
rval = gIR->ir->CreateIntToPtr(rval, tolltype, "tmp");
}
else {
error(loc, "invalid cast from '%s' to '%s'", val->getType()->toChars(), _to->toChars());
fatal();
}
return new DImValue(_to, rval);
}
DValue* DtoCastPtr(Loc& loc, DValue* val, Type* to)
{
const LLType* tolltype = DtoType(to);
Type* totype = to->toBasetype();
Type* fromtype = val->getType()->toBasetype();
assert(fromtype->ty == Tpointer || fromtype->ty == Tfunction);
LLValue* rval;
if (totype->ty == Tpointer || totype->ty == Tclass) {
LLValue* src = val->getRVal();
if (Logger::enabled())
Logger::cout() << "src: " << *src << "to type: " << *tolltype << '\n';
rval = DtoBitCast(src, tolltype);
}
else if (totype->ty == Tbool) {
LLValue* src = val->getRVal();
LLValue* zero = LLConstant::getNullValue(src->getType());
rval = gIR->ir->CreateICmpNE(src, zero, "tmp");
}
else if (totype->isintegral()) {
rval = new llvm::PtrToIntInst(val->getRVal(), tolltype, "tmp", gIR->scopebb());
}
else {
error(loc, "invalid cast from '%s' to '%s'", val->getType()->toChars(), to->toChars());
fatal();
}
return new DImValue(to, rval);
}
DValue* DtoCastFloat(Loc& loc, DValue* val, Type* to)
{
if (val->getType() == to)
return val;
const LLType* tolltype = DtoType(to);
Type* totype = to->toBasetype();
Type* fromtype = val->getType()->toBasetype();
assert(fromtype->isfloating());
size_t fromsz = fromtype->size();
size_t tosz = totype->size();
LLValue* rval;
if (totype->ty == Tbool) {
rval = val->getRVal();
LLValue* zero = LLConstant::getNullValue(rval->getType());
rval = gIR->ir->CreateFCmpUNE(rval, zero, "tmp");
}
else if (totype->iscomplex()) {
return DtoComplex(loc, to, val);
}
else if (totype->isfloating()) {
if (fromsz == tosz) {
rval = val->getRVal();
assert(rval->getType() == tolltype);
}
else if (fromsz < tosz) {
rval = new llvm::FPExtInst(val->getRVal(), tolltype, "tmp", gIR->scopebb());
}
else if (fromsz > tosz) {
rval = new llvm::FPTruncInst(val->getRVal(), tolltype, "tmp", gIR->scopebb());
}
else {
error(loc, "invalid cast from '%s' to '%s'", val->getType()->toChars(), to->toChars());
fatal();
}
}
else if (totype->isintegral()) {
if (totype->isunsigned()) {
rval = new llvm::FPToUIInst(val->getRVal(), tolltype, "tmp", gIR->scopebb());
}
else {
rval = new llvm::FPToSIInst(val->getRVal(), tolltype, "tmp", gIR->scopebb());
}
}
else {
error(loc, "invalid cast from '%s' to '%s'", val->getType()->toChars(), to->toChars());
fatal();
}
return new DImValue(to, rval);
}
DValue* DtoCastDelegate(Loc& loc, DValue* val, Type* to)
{
if (to->toBasetype()->ty == Tdelegate)
{
return DtoPaintType(loc, val, to);
}
else if (to->toBasetype()->ty == Tbool)
{
return new DImValue(to, DtoDelegateEquals(TOKnotequal, val->getRVal(), NULL));
}
else
{
error(loc, "invalid cast from '%s' to '%s'", val->getType()->toChars(), to->toChars());
fatal();
}
}
DValue* DtoCast(Loc& loc, DValue* val, Type* to)
{
Type* fromtype = val->getType()->toBasetype();
Type* totype = to->toBasetype();
if (fromtype->equals(totype))
return val;
Logger::println("Casting from '%s' to '%s'", fromtype->toChars(), to->toChars());
LOG_SCOPE;
if (fromtype->isintegral()) {
return DtoCastInt(loc, val, to);
}
else if (fromtype->iscomplex()) {
return DtoCastComplex(loc, val, to);
}
else if (fromtype->isfloating()) {
return DtoCastFloat(loc, val, to);
}
else if (fromtype->ty == Tclass) {
return DtoCastClass(val, to);
}
else if (fromtype->ty == Tarray || fromtype->ty == Tsarray) {
return DtoCastArray(loc, val, to);
}
else if (fromtype->ty == Tpointer || fromtype->ty == Tfunction) {
return DtoCastPtr(loc, val, to);
}
else if (fromtype->ty == Tdelegate) {
return DtoCastDelegate(loc, val, to);
}
else {
error(loc, "invalid cast from '%s' to '%s'", val->getType()->toChars(), to->toChars());
fatal();
}
}
//////////////////////////////////////////////////////////////////////////////////////////
DValue* DtoPaintType(Loc& loc, DValue* val, Type* to)
{
Type* from = val->getType()->toBasetype();
Logger::println("repainting from '%s' to '%s'", from->toChars(), to->toChars());
if (from->ty == Tarray)
{
Type* at = to->toBasetype();
assert(at->ty == Tarray);
Type* elem = at->nextOf()->pointerTo();
if (DSliceValue* slice = val->isSlice())
{
return new DSliceValue(to, slice->len, DtoBitCast(slice->ptr, DtoType(elem)));
}
else if (val->isLVal())
{
LLValue* ptr = val->getLVal();
ptr = DtoBitCast(ptr, DtoType(at->pointerTo()));
return new DVarValue(to, ptr);
}
else
{
LLValue *len, *ptr;
len = DtoArrayLen(val);
ptr = DtoArrayPtr(val);
ptr = DtoBitCast(ptr, DtoType(elem));
return new DImValue(to, DtoAggrPair(len, ptr, "tmp"));
}
}
else if (from->ty == Tdelegate)
{
Type* dgty = to->toBasetype();
assert(dgty->ty == Tdelegate);
if (val->isLVal())
{
LLValue* ptr = val->getLVal();
assert(isaPointer(ptr));
ptr = DtoBitCast(ptr, getPtrToType(DtoType(dgty)));
if (Logger::enabled())
Logger::cout() << "dg ptr: " << *ptr << '\n';
return new DVarValue(to, ptr);
}
else
{
LLValue* dg = val->getRVal();
LLValue* context = gIR->ir->CreateExtractValue(dg, 0, ".context");
LLValue* funcptr = gIR->ir->CreateExtractValue(dg, 1, ".funcptr");
funcptr = DtoBitCast(funcptr, DtoType(dgty)->getContainedType(1));
LLValue* aggr = DtoAggrPair(context, funcptr, "tmp");
if (Logger::enabled())
Logger::cout() << "dg: " << *aggr << '\n';
return new DImValue(to, aggr);
}
}
else if (from->ty == Tpointer || from->ty == Tclass || from->ty == Taarray)
{
Type* b = to->toBasetype();
assert(b->ty == Tpointer || b->ty == Tclass || b->ty == Taarray);
LLValue* ptr = DtoBitCast(val->getRVal(), DtoType(b));
return new DImValue(to, ptr);
}
else
{
assert(!val->isLVal());
assert(DtoType(to) == DtoType(to));
return new DImValue(to, val->getRVal());
}
}
/****************************************************************************************/
/*////////////////////////////////////////////////////////////////////////////////////////
// TEMPLATE HELPERS
////////////////////////////////////////////////////////////////////////////////////////*/
TemplateInstance* DtoIsTemplateInstance(Dsymbol* s)
{
if (!s) return NULL;
if (s->isTemplateInstance() && !s->isTemplateMixin())
return s->isTemplateInstance();
else if (s->parent)
return DtoIsTemplateInstance(s->parent);
return NULL;
}
/****************************************************************************************/
/*////////////////////////////////////////////////////////////////////////////////////////
// PROCESSING QUEUE HELPERS
////////////////////////////////////////////////////////////////////////////////////////*/
void DtoResolveDsymbol(Dsymbol* dsym)
{
if (StructDeclaration* sd = dsym->isStructDeclaration()) {
DtoResolveStruct(sd);
}
else if (ClassDeclaration* cd = dsym->isClassDeclaration()) {
DtoResolveClass(cd);
}
else if (FuncDeclaration* fd = dsym->isFuncDeclaration()) {
DtoResolveFunction(fd);
}
else if (TypeInfoDeclaration* fd = dsym->isTypeInfoDeclaration()) {
DtoResolveTypeInfo(fd);
}
else {
error(dsym->loc, "unsupported dsymbol: %s", dsym->toChars());
assert(0 && "unsupported dsymbol for DtoResolveDsymbol");
}
}
//////////////////////////////////////////////////////////////////////////////////////////
void DtoConstInitGlobal(VarDeclaration* vd)
{
vd->codegen(Type::sir);
if (vd->ir.initialized) return;
vd->ir.initialized = gIR->dmodule;
Logger::println("DtoConstInitGlobal(%s) @ %s", vd->toChars(), vd->loc.toChars());
LOG_SCOPE;
Dsymbol* par = vd->toParent();
// build the initializer
LLConstant* initVal = DtoConstInitializer(vd->loc, vd->type, vd->init);
// set the initializer if appropriate
IrGlobal* glob = vd->ir.irGlobal;
llvm::GlobalVariable* gvar = llvm::cast<llvm::GlobalVariable>(glob->value);
// refine the global's opaque type to the type of the initializer
llvm::cast<LLOpaqueType>(glob->type.get())->refineAbstractTypeTo(initVal->getType());
assert(!glob->constInit);
glob->constInit = initVal;
// assign the initializer
llvm::GlobalVariable* globalvar = llvm::cast<llvm::GlobalVariable>(glob->value);
if (!(vd->storage_class & STCextern) && mustDefineSymbol(vd))
{
if (Logger::enabled())
{
Logger::println("setting initializer");
Logger::cout() << "global: " << *gvar << '\n';
#if 0
Logger::cout() << "init: " << *initVal << '\n';
#endif
}
gvar->setInitializer(initVal);
// do debug info
if (global.params.symdebug)
{
LLGlobalVariable* gv = DtoDwarfGlobalVariable(gvar, vd).getGV();
// keep a reference so GDCE doesn't delete it !
gIR->usedArray.push_back(llvm::ConstantExpr::getBitCast(gv, getVoidPtrType()));
}
}
}
/****************************************************************************************/
/*////////////////////////////////////////////////////////////////////////////////////////
// DECLARATION EXP HELPER
////////////////////////////////////////////////////////////////////////////////////////*/
DValue* DtoDeclarationExp(Dsymbol* declaration)
{
Logger::print("DtoDeclarationExp: %s\n", declaration->toChars());
LOG_SCOPE;
// variable declaration
if (VarDeclaration* vd = declaration->isVarDeclaration())
{
Logger::println("VarDeclaration");
// if aliassym is set, this VarDecl is redone as an alias to another symbol
// this seems to be done to rewrite Tuple!(...) v;
// as a TupleDecl that contains a bunch of individual VarDecls
if (vd->aliassym)
return DtoDeclarationExp(vd->aliassym);
// static
if (vd->isDataseg())
{
vd->codegen(Type::sir);
}
else
{
if (global.params.llvmAnnotate)
DtoAnnotation(declaration->toChars());
Logger::println("vdtype = %s", vd->type->toChars());
// referenced by nested delegate?
#if DMDV2
if (vd->nestedrefs.dim) {
#else
if (vd->nestedref) {
#endif
Logger::println("has nestedref set");
assert(vd->ir.irLocal);
DtoNestedInit(vd);
}
// normal stack variable, allocate storage on the stack if it has not already been done
else if(!vd->ir.irLocal) {
const LLType* lltype = DtoType(vd->type);
llvm::Value* allocainst;
if(gTargetData->getTypeSizeInBits(lltype) == 0)
allocainst = llvm::ConstantPointerNull::get(getPtrToType(lltype));
else
allocainst = DtoAlloca(vd->type, vd->toChars());
//allocainst->setAlignment(vd->type->alignsize()); // TODO
vd->ir.irLocal = new IrLocal(vd);
vd->ir.irLocal->value = allocainst;
if (global.params.symdebug)
{
DtoDwarfLocalVariable(allocainst, vd);
}
}
else
{
assert(vd->ir.irLocal->value);
}
if (Logger::enabled())
Logger::cout() << "llvm value for decl: " << *vd->ir.irLocal->value << '\n';
DValue* ie = DtoInitializer(vd->ir.irLocal->value, vd->init);
}
return new DVarValue(vd->type, vd, vd->ir.getIrValue());
}
// struct declaration
else if (StructDeclaration* s = declaration->isStructDeclaration())
{
Logger::println("StructDeclaration");
s->codegen(Type::sir);
}
// function declaration
else if (FuncDeclaration* f = declaration->isFuncDeclaration())
{
Logger::println("FuncDeclaration");
f->codegen(Type::sir);
}
// alias declaration
else if (AliasDeclaration* a = declaration->isAliasDeclaration())
{
Logger::println("AliasDeclaration - no work");
// do nothing
}
// enum
else if (EnumDeclaration* e = declaration->isEnumDeclaration())
{
Logger::println("EnumDeclaration - no work");
// do nothing
}
// class
else if (ClassDeclaration* e = declaration->isClassDeclaration())
{
Logger::println("ClassDeclaration");
e->codegen(Type::sir);
}
// typedef
else if (TypedefDeclaration* tdef = declaration->isTypedefDeclaration())
{
Logger::println("TypedefDeclaration");
DtoTypeInfoOf(tdef->type, false);
}
// attribute declaration
else if (AttribDeclaration* a = declaration->isAttribDeclaration())
{
Logger::println("AttribDeclaration");
for (int i=0; i < a->decl->dim; ++i)
{
DtoDeclarationExp((Dsymbol*)a->decl->data[i]);
}
}
// mixin declaration
else if (TemplateMixin* m = declaration->isTemplateMixin())
{
Logger::println("TemplateMixin");
for (int i=0; i < m->members->dim; ++i)
{
Dsymbol* mdsym = (Dsymbol*)m->members->data[i];
DtoDeclarationExp(mdsym);
}
}
// tuple declaration
else if (TupleDeclaration* tupled = declaration->isTupleDeclaration())
{
Logger::println("TupleDeclaration");
if(!tupled->isexp) {
error(declaration->loc, "don't know how to handle non-expression tuple decls yet");
assert(0);
}
assert(tupled->objects);
for (int i=0; i < tupled->objects->dim; ++i)
{
DsymbolExp* exp = (DsymbolExp*)tupled->objects->data[i];
DtoDeclarationExp(exp->s);
}
}
// unsupported declaration
else
{
error(declaration->loc, "Unimplemented Declaration type for DeclarationExp. kind: %s", declaration->kind());
assert(0);
}
return NULL;
}
// does pretty much the same as DtoDeclarationExp, except it doesn't initialize, and only handles var declarations
LLValue* DtoRawVarDeclaration(VarDeclaration* var, LLValue* addr)
{
// we don't handle globals with this one
assert(!var->isDataseg());
// we don't handle aliases either
assert(!var->aliassym);
// alloca if necessary
LLValue* allocaval = NULL;
if (!addr && (!var->ir.irLocal || !var->ir.irLocal->value))
{
addr = DtoAlloca(var->type, var->toChars());
// add debug info
if (global.params.symdebug)
DtoDwarfLocalVariable(addr, var);
}
// referenced by nested function?
#if DMDV2
if (var->nestedrefs.dim)
#else
if (var->nestedref)
#endif
{
assert(var->ir.irLocal);
if(!var->ir.irLocal->value)
{
assert(addr);
var->ir.irLocal->value = addr;
}
else
assert(!addr || addr == var->ir.irLocal->value);
DtoNestedInit(var);
}
// normal local variable
else
{
// if this already has storage, it must've been handled already
if (var->ir.irLocal && var->ir.irLocal->value) {
if (addr && addr != var->ir.irLocal->value) {
// This can happen, for example, in scope(exit) blocks which
// are translated to IR multiple times.
// That *should* only happen after the first one is completely done
// though, so just set the address.
IF_LOG {
Logger::println("Replacing LLVM address of %s", var->toChars());
LOG_SCOPE;
Logger::cout() << "Old val: " << *var->ir.irLocal->value << '\n';
Logger::cout() << "New val: " << *addr << '\n';
}
var->ir.irLocal->value = addr;
}
return addr;
}
assert(!var->ir.isSet());
assert(addr);
var->ir.irLocal = new IrLocal(var);
var->ir.irLocal->value = addr;
}
// return the alloca
return var->ir.irLocal->value;
}
/****************************************************************************************/
/*////////////////////////////////////////////////////////////////////////////////////////
// INITIALIZER HELPERS
////////////////////////////////////////////////////////////////////////////////////////*/
LLConstant* DtoConstInitializer(Loc loc, Type* type, Initializer* init)
{
LLConstant* _init = 0; // may return zero
if (!init)
{
Logger::println("const default initializer for %s", type->toChars());
_init = DtoConstExpInit(loc, type, type->defaultInit());
}
else if (ExpInitializer* ex = init->isExpInitializer())
{
Logger::println("const expression initializer");
_init = DtoConstExpInit(loc, type, ex->exp);;
}
else if (StructInitializer* si = init->isStructInitializer())
{
Logger::println("const struct initializer");
si->ad->codegen(Type::sir);
return si->ad->ir.irStruct->createStructInitializer(si);
}
else if (ArrayInitializer* ai = init->isArrayInitializer())
{
Logger::println("const array initializer");
_init = DtoConstArrayInitializer(ai);
}
else if (init->isVoidInitializer())
{
Logger::println("const void initializer");
const LLType* ty = DtoType(type);
_init = llvm::Constant::getNullValue(ty);
}
else {
Logger::println("unsupported const initializer: %s", init->toChars());
}
return _init;
}
//////////////////////////////////////////////////////////////////////////////////////////
DValue* DtoInitializer(LLValue* target, Initializer* init)
{
if (!init)
return 0;
else if (ExpInitializer* ex = init->isExpInitializer())
{
Logger::println("expression initializer");
assert(ex->exp);
return ex->exp->toElem(gIR);
}
else if (init->isVoidInitializer())
{
// do nothing
}
else {
Logger::println("unsupported initializer: %s", init->toChars());
assert(0);
}
return 0;
}
//////////////////////////////////////////////////////////////////////////////////////////
static LLConstant* expand_to_sarray(Type *base, Expression* exp)
{
Logger::println("building type %s from expression (%s) of type %s", base->toChars(), exp->toChars(), exp->type->toChars());
const LLType* dstTy = DtoType(base);
if (Logger::enabled())
Logger::cout() << "final llvm type requested: " << *dstTy << '\n';
LLConstant* val = exp->toConstElem(gIR);
Type* expbase = exp->type->toBasetype();
Logger::println("expbase: %s", expbase->toChars());
Type* t = base->toBasetype();
LLSmallVector<size_t, 4> dims;
while(1)
{
Logger::println("t: %s", t->toChars());
if (t->equals(expbase))
break;
assert(t->ty == Tsarray);
TypeSArray* tsa = (TypeSArray*)t;
dims.push_back(tsa->dim->toInteger());
assert(t->nextOf());
t = t->nextOf()->toBasetype();
}
size_t i = dims.size();
assert(i);
std::vector<LLConstant*> inits;
while (i--)
{
const LLArrayType* arrty = LLArrayType::get(val->getType(), dims[i]);
inits.clear();
inits.insert(inits.end(), dims[i], val);
val = LLConstantArray::get(arrty, inits);
}
return val;
}
LLConstant* DtoConstExpInit(Loc loc, Type* type, Expression* exp)
{
Type* expbase = exp->type->toBasetype();
Type* base = type->toBasetype();
// if not the same basetypes, we won't get the same llvm types either
if (!expbase->equals(base))
{
if (base->ty == Tsarray)
{
if (base->nextOf()->toBasetype()->ty == Tvoid) {
error(loc, "static arrays of voids have no default initializer");
fatal();
}
Logger::println("type is a static array, building constant array initializer to single value");
return expand_to_sarray(base, exp);
}
else
{
error("cannot yet convert default initializer %s to type %s to %s", exp->toChars(), exp->type->toChars(), type->toChars());
fatal();
}
assert(0);
}
return exp->toConstElem(gIR);
}
//////////////////////////////////////////////////////////////////////////////////////////
void DtoAnnotation(const char* str)
{
std::string s("CODE: ");
s.append(str);
char* p = &s[0];
while (*p)
{
if (*p == '"')
*p = '\'';
++p;
}
// create a noop with the code as the result name!
// FIXME: this is const folded and eliminated immediately ... :/
gIR->ir->CreateAnd(DtoConstSize_t(0),DtoConstSize_t(0),s.c_str());
}
//////////////////////////////////////////////////////////////////////////////////////////
LLConstant* DtoTypeInfoOf(Type* type, bool base)
{
#if DMDV2
// FIXME: this is probably wrong, but it makes druntime's genobj.d compile!
type = type->mutableOf()->merge(); // needed.. getTypeInfo does the same
#else
type = type->merge(); // needed.. getTypeInfo does the same
#endif
type->getTypeInfo(NULL);
TypeInfoDeclaration* tidecl = type->vtinfo;
assert(tidecl);
tidecl->codegen(Type::sir);
assert(tidecl->ir.irGlobal != NULL);
assert(tidecl->ir.irGlobal->value != NULL);
LLConstant* c = isaConstant(tidecl->ir.irGlobal->value);
assert(c != NULL);
if (base)
return llvm::ConstantExpr::getBitCast(c, DtoType(Type::typeinfo->type));
return c;
}
//////////////////////////////////////////////////////////////////////////////////////////
void DtoOverloadedIntrinsicName(TemplateInstance* ti, TemplateDeclaration* td, std::string& name)
{
Logger::println("DtoOverloadedIntrinsicName");
LOG_SCOPE;
Logger::println("template instance: %s", ti->toChars());
Logger::println("template declaration: %s", td->toChars());
Logger::println("intrinsic name: %s", td->intrinsicName.c_str());
// for now use the size in bits of the first template param in the instance
assert(ti->tdtypes.dim == 1);
Type* T = (Type*)ti->tdtypes.data[0];
char prefix = T->isreal() ? 'f' : T->isintegral() ? 'i' : 0;
if (!prefix) {
ti->error("has invalid template parameter for intrinsic: %s", T->toChars());
fatal(); // or LLVM asserts
}
char tmp[21]; // probably excessive, but covers a uint64_t
sprintf(tmp, "%lu", (unsigned long) gTargetData->getTypeSizeInBits(DtoType(T)));
// replace # in name with bitsize
name = td->intrinsicName;
std::string needle("#");
size_t pos;
while(std::string::npos != (pos = name.find(needle))) {
if (pos > 0 && name[pos-1] == prefix) {
// Properly prefixed, insert bitwidth.
name.replace(pos, 1, tmp);
} else {
if (pos && (name[pos-1] == 'i' || name[pos-1] == 'f')) {
// Wrong type character.
ti->error("has invalid parameter type for intrinsic %s: %s is not a%s type",
name.c_str(), T->toChars(),
(name[pos-1] == 'i' ? "n integral" : " floating-point"));
} else {
// Just plain wrong. (Error in declaration, not instantiation)
td->error("has an invalid intrinsic name: %s", name.c_str());
}
fatal(); // or LLVM asserts
}
}
Logger::println("final intrinsic name: %s", name.c_str());
}
//////////////////////////////////////////////////////////////////////////////////////////
bool mustDefineSymbol(Dsymbol* s)
{
if (FuncDeclaration* fd = s->isFuncDeclaration())
{
// we can't (and probably shouldn't?) define functions
// that weren't semantic3'ed
if (fd->semanticRun < 4)
return false;
if (fd->isArrayOp)
return true;
}
TemplateInstance* tinst = DtoIsTemplateInstance(s);
if (tinst)
{
if (!opts::singleObj)
return true;
if (!tinst->emittedInModule)
{
gIR->seenTemplateInstances.insert(tinst);
tinst->emittedInModule = gIR->dmodule;
}
return tinst->emittedInModule == gIR->dmodule;
}
return s->getModule() == gIR->dmodule;
}
//////////////////////////////////////////////////////////////////////////////////////////
bool needsTemplateLinkage(Dsymbol* s)
{
return DtoIsTemplateInstance(s) && mustDefineSymbol(s);
}
//////////////////////////////////////////////////////////////////////////////////////////
bool hasUnalignedFields(Type* t)
{
t = t->toBasetype();
if (t->ty == Tsarray) {
assert(t->nextOf()->size() % t->nextOf()->alignsize() == 0);
return hasUnalignedFields(t->nextOf());
} else if (t->ty != Tstruct)
return false;
TypeStruct* ts = (TypeStruct*)t;
if (ts->unaligned)
return (ts->unaligned == 2);
StructDeclaration* sym = ts->sym;
// go through all the fields and try to find something unaligned
ts->unaligned = 2;
for (int i = 0; i < sym->fields.dim; i++)
{
VarDeclaration* f = (VarDeclaration*)sym->fields.data[i];
unsigned a = f->type->alignsize() - 1;
if (((f->offset + a) & ~a) != f->offset)
return true;
else if (f->type->toBasetype()->ty == Tstruct && hasUnalignedFields(f->type))
return true;
}
ts->unaligned = 1;
return false;
}
//////////////////////////////////////////////////////////////////////////////////////////
IrModule * getIrModule(Module * M)
{
if (M == NULL)
M = gIR->func()->decl->getModule();
assert(M && "null module");
if (!M->ir.irModule)
M->ir.irModule = new IrModule(M, M->srcfile->toChars());
return M->ir.irModule;
}
//////////////////////////////////////////////////////////////////////////////////////////
size_t realignOffset(size_t offset, Type* type)
{
size_t alignsize = type->alignsize();
size_t alignedoffset = (offset + alignsize - 1) & ~(alignsize - 1);
// if the aligned offset already matches the input offset
// don't waste time checking things are ok!
if (alignedoffset == offset)
return alignedoffset;
// we cannot get the llvm alignment if the type is still opaque, this can happen in some
// forward reference situations, so when this happens we fall back to manual padding.
const llvm::Type* T = DtoType(type);
if (llvm::isa<llvm::OpaqueType>(T))
{
return offset;
}
// then we check against the llvm alignment
size_t alignsize2 = gTargetData->getABITypeAlignment(T);
// if it differs we need to insert manual padding as well
if (alignsize != alignsize2)
{
assert(alignsize > alignsize2 && "this is not good, the D and LLVM "
"type alignments differ, but LLVM's is bigger! This will break "
"aggregate type mapping");
// don't try and align the offset, and let the mappers pad 100% manually
return offset;
}
// ok, we're good, llvm will align properly!
return alignedoffset;
}
//////////////////////////////////////////////////////////////////////////////////////////