diff --git a/driver/toobj.cpp b/driver/toobj.cpp
index 087c3da487..8015e031af 100644
--- a/driver/toobj.cpp
+++ b/driver/toobj.cpp
@@ -185,9 +185,9 @@ public:
         // from being repeated in its parameters. Might need to be
         // extended, but GEPs/PHIs are the most common ones.
         os << ", type = " << *val.getType();
-      } else if (isa<AllocaInst>(&val)) {
+      } else if (auto ai = dyn_cast<AllocaInst>(&val)) {
         os << ", size/byte = "
-           << DL.getTypeAllocSize(val.getType()->getContainedType(0));
+           << DL.getTypeAllocSize(ai->getAllocatedType());
       }
       os << ']';
     }
diff --git a/gen/abi-generic.h b/gen/abi-generic.h
index 60b7c6d2f9..f94d3298ed 100644
--- a/gen/abi-generic.h
+++ b/gen/abi-generic.h
@@ -105,7 +105,8 @@ struct RemoveStructPadding : ABIRewrite {
     LLValue *lval = DtoAlloca(dty, ".RemoveStructPadding_dump");
     // Make sure the padding is zero, so struct comparisons work.
     // TODO: Only do this if there's padding, and/or only initialize padding.
-    DtoMemSetZero(lval, DtoConstSize_t(getTypeAllocSize(DtoType(dty))));
+    DtoMemSetZero(DtoType(dty), lval,
+                  DtoConstSize_t(getTypeAllocSize(DtoType(dty))));
     DtoPaddedStruct(dty->toBasetype(), v, lval);
     return lval;
   }
diff --git a/gen/abi-x86-64.cpp b/gen/abi-x86-64.cpp
index a00dc93b00..764ace778f 100644
--- a/gen/abi-x86-64.cpp
+++ b/gen/abi-x86-64.cpp
@@ -364,7 +364,7 @@ void X86_64TargetABI::vaCopy(DLValue *dest, DValue *src) {
            DtoBitCast(DtoLVal(dest), getPtrToType(valistmem->getType())));
   // Then fill the new struct with a bitcopy of the source struct.
   // `src` is a __va_list_tag* pointer to the source struct.
-  DtoMemCpy(valistmem, DtoRVal(src));
+  DtoMemCpy(getValistType(), valistmem, DtoRVal(src));
 }
 
 LLValue *X86_64TargetABI::prepareVaArg(DLValue *ap) {
diff --git a/gen/abi.cpp b/gen/abi.cpp
index c81f1d861e..506c51f861 100644
--- a/gen/abi.cpp
+++ b/gen/abi.cpp
@@ -208,7 +208,7 @@ LLValue *TargetABI::prepareVaStart(DLValue *ap) {
 void TargetABI::vaCopy(DLValue *dest, DValue *src) {
   LLValue *llDest = DtoLVal(dest);
   if (src->isLVal()) {
-    DtoMemCpy(llDest, DtoLVal(src));
+    DtoMemCpy(DtoType(dest->type), llDest, DtoLVal(src));
   } else {
     DtoStore(DtoRVal(src), llDest);
   }
diff --git a/gen/arrays.cpp b/gen/arrays.cpp
index cb79aebb38..3840326164 100644
--- a/gen/arrays.cpp
+++ b/gen/arrays.cpp
@@ -28,14 +28,12 @@
 #include "ir/irfunction.h"
 #include "ir/irmodule.h"
 
-static void DtoSetArray(DValue *array, LLValue *dim, LLValue *ptr);
+static void DtoSetArray(DValue *array, DValue *rhs);
 
 ////////////////////////////////////////////////////////////////////////////////
 
 namespace {
-LLValue *DtoSlice(LLValue *ptr, LLValue *length, LLType *elemType = nullptr) {
-  if (!elemType)
-    elemType = ptr->getType()->getContainedType(0);
+LLValue *DtoSlice(LLValue *ptr, LLValue *length, LLType *elemType) {
   elemType = i1ToI8(voidToI8(elemType));
   return DtoAggrPair(length, DtoBitCast(ptr, elemType->getPointerTo()));
 }
@@ -44,7 +42,7 @@ LLValue *DtoSlice(Expression *e) {
   DValue *dval = toElem(e);
   if (dval->type->toBasetype()->ty == TY::Tsarray) {
     // Convert static array to slice
-    return DtoSlice(DtoLVal(dval), DtoArrayLen(dval));
+    return DtoSlice(DtoLVal(dval), DtoArrayLen(dval), DtoType(dval->type->nextOf()));
   }
   return DtoRVal(dval);
 }
@@ -218,7 +216,7 @@ void DtoArrayAssign(const Loc &loc, DValue *lhs, DValue *rhs, EXP op,
     if (rhs->isNull()) {
       DtoSetArrayToNull(lhs);
     } else {
-      DtoSetArray(lhs, DtoArrayLen(rhs), DtoArrayPtr(rhs));
+      DtoSetArray(lhs, rhs);
     }
     return;
   }
@@ -254,7 +252,7 @@ void DtoArrayAssign(const Loc &loc, DValue *lhs, DValue *rhs, EXP op,
       const size_t elementSize = getTypeAllocSize(DtoMemType(elemType));
       if (rhs->isNull()) {
         LLValue *lhsSize = computeSize(lhsLength, elementSize);
-        DtoMemSetZero(lhsPtr, lhsSize);
+        DtoMemSetZero(getI8Type(), lhsPtr, lhsSize);
       } else {
         bool knownInBounds =
             isConstructing || (t->ty == TY::Tsarray && t2->ty == TY::Tsarray);
@@ -274,16 +272,16 @@ void DtoArrayAssign(const Loc &loc, DValue *lhs, DValue *rhs, EXP op,
     } else if (isConstructing) {
       LLFunction *fn = getRuntimeFunction(loc, gIR->module, "_d_arrayctor");
       gIR->CreateCallOrInvoke(fn, DtoTypeInfoOf(loc, elemType),
-                              DtoSlice(rhsPtr, rhsLength),
-                              DtoSlice(lhsPtr, lhsLength));
+                              DtoSlice(rhsPtr, rhsLength, getI8Type()),
+                              DtoSlice(lhsPtr, lhsLength, getI8Type()));
     } else { // assigning
       LLValue *tmpSwap = DtoAlloca(elemType, "arrayAssign.tmpSwap");
       LLFunction *fn = getRuntimeFunction(
           loc, gIR->module,
           !canSkipPostblit ? "_d_arrayassign_l" : "_d_arrayassign_r");
       gIR->CreateCallOrInvoke(
-          fn, DtoTypeInfoOf(loc, elemType), DtoSlice(rhsPtr, rhsLength),
-          DtoSlice(lhsPtr, lhsLength), DtoBitCast(tmpSwap, getVoidPtrType()));
+          fn, DtoTypeInfoOf(loc, elemType), DtoSlice(rhsPtr, rhsLength, getI8Type()),
+          DtoSlice(lhsPtr, lhsLength, getI8Type()), DtoBitCast(tmpSwap, getVoidPtrType()));
     }
   } else {
     // scalar rhs:
@@ -295,7 +293,7 @@ void DtoArrayAssign(const Loc &loc, DValue *lhs, DValue *rhs, EXP op,
     if (!needsDestruction && !needsPostblit) {
       // fast version
       const size_t lhsElementSize =
-          getTypeAllocSize(realLhsPtr->getType()->getContainedType(0));
+          getTypeAllocSize(DtoMemType(lhs->type->nextOf()));
       LLType *rhsType = DtoMemType(t2);
       const size_t rhsSize = getTypeAllocSize(rhsType);
       LLValue *actualPtr = DtoBitCast(realLhsPtr, rhsType->getPointerTo());
@@ -323,13 +321,13 @@ void DtoArrayAssign(const Loc &loc, DValue *lhs, DValue *rhs, EXP op,
 
 ////////////////////////////////////////////////////////////////////////////////
 
-static void DtoSetArray(DValue *array, LLValue *dim, LLValue *ptr) {
+static void DtoSetArray(DValue *array, DValue *rhs) {
   IF_LOG Logger::println("SetArray");
   LLValue *arr = DtoLVal(array);
-  LLType *s = isaStruct(arr->getType()->getContainedType(0));
+  LLType *s = DtoType(array->type);
   assert(s);
-  DtoStore(dim, DtoGEP(s, arr, 0u, 0));
-  DtoStore(ptr, DtoGEP(s, arr, 0, 1));
+  DtoStore(DtoArrayLen(rhs), DtoGEP(s, arr, 0u, 0));
+  DtoStore(DtoArrayPtr(rhs), DtoGEP(s, arr, 0, 1));
 }
 
 ////////////////////////////////////////////////////////////////////////////////
@@ -829,7 +827,7 @@ DSliceValue *DtoCatAssignArray(const Loc &loc, DValue *arr, Expression *exp) {
   LLValue *newArray = gIR->CreateCallOrInvoke(
       fn, DtoTypeInfoOf(loc, arrayType),
       DtoBitCast(DtoLVal(arr), fn->getFunctionType()->getParamType(1)),
-      DtoAggrPaint(DtoSlice(exp), fn->getFunctionType()->getParamType(2)),
+      DtoSlicePaint(DtoSlice(exp), fn->getFunctionType()->getParamType(2)),
       ".appendedArray");
 
   return getSlice(arrayType, newArray);
@@ -921,10 +919,12 @@ DSliceValue *DtoCatArrays(const Loc &loc, Type *arrayType, Expression *exp1,
 
     auto loadArray = [fn](Expression* e, int paramTypeIdx) {
       DValue * dval = toElem(e);
-      LLValue *val = DtoLoad(DtoSlicePtrType(dval), DtoSlicePtr(e));
-      return DtoAggrPaint(val, fn->getFunctionType()->getParamType(paramTypeIdx));
+      LLValue *val = DtoLoad(DtoSlicePtrType(dval), DtoSlicePtr(dval));
+      return DtoSlicePaint(val, fn->getFunctionType()->getParamType(paramTypeIdx));
     };
+    // byte[] x
     args.push_back(loadArray(exp1,1));
+    // byte[] y
     args.push_back(loadArray(exp2,2));
   }
 
@@ -988,9 +988,9 @@ LLValue *DtoArrayEqCmp_impl(const Loc &loc, const char *func, DValue *l,
   LLSmallVector<LLValue *, 3> args;
 
   // get values, reinterpret cast to void[]
-  args.push_back(DtoAggrPaint(DtoRVal(l),
+  args.push_back(DtoSlicePaint(DtoRVal(l),
                               DtoArrayType(LLType::getInt8Ty(gIR->context()))));
-  args.push_back(DtoAggrPaint(DtoRVal(r),
+  args.push_back(DtoSlicePaint(DtoRVal(r),
                               DtoArrayType(LLType::getInt8Ty(gIR->context()))));
 
   // pass array typeinfo ?
@@ -1066,12 +1066,12 @@ bool validCompareWithMemcmp(DValue *l, DValue *r) {
 
 // Create a call instruction to memcmp.
 llvm::CallInst *callMemcmp(const Loc &loc, IRState &irs, LLValue *l_ptr,
-                           LLValue *r_ptr, LLValue *numElements) {
+                           LLValue *r_ptr, LLValue *numElements, LLType *elemty) {
   assert(l_ptr && r_ptr && numElements);
   LLFunction *fn = getRuntimeFunction(loc, gIR->module, "memcmp");
   assert(fn);
   auto sizeInBytes = numElements;
-  size_t elementSize = getTypeAllocSize(l_ptr->getType()->getContainedType(0));
+  size_t elementSize = getTypeAllocSize(elemty);
   if (elementSize != 1) {
     sizeInBytes = irs.ir->CreateMul(sizeInBytes, DtoConstSize_t(elementSize));
   }
@@ -1101,7 +1101,7 @@ LLValue *DtoArrayEqCmp_memcmp(const Loc &loc, DValue *l, DValue *r,
       (r->type->toBasetype()->ty == TY::Tsarray);
   if (staticArrayComparison) {
     // TODO: simply codegen when comparing static arrays with different length (int[3] == int[2])
-    return callMemcmp(loc, irs, l_ptr, r_ptr, l_length);
+    return callMemcmp(loc, irs, l_ptr, r_ptr, l_length, DtoMemType(l->type->nextOf()));
   }
 
   // First compare the array lengths
@@ -1118,7 +1118,7 @@ LLValue *DtoArrayEqCmp_memcmp(const Loc &loc, DValue *l, DValue *r,
   // The array comparison is UB for non-zero length, and memcmp will correctly
   // return 0 (equality) when the length is zero.
   irs.ir->SetInsertPoint(memcmpBB);
-  auto memcmpAnswer = callMemcmp(loc, irs, l_ptr, r_ptr, l_length);
+  auto memcmpAnswer = callMemcmp(loc, irs, l_ptr, r_ptr, l_length, DtoMemType(l->type->nextOf()));
   irs.ir->CreateBr(memcmpEndBB);
 
   // Merge the result of length check and memcmp call into a phi node.
diff --git a/gen/asmstmt.cpp b/gen/asmstmt.cpp
index 100466c1c8..9f0019651e 100644
--- a/gen/asmstmt.cpp
+++ b/gen/asmstmt.cpp
@@ -267,10 +267,12 @@ void AsmStatement_toIR(InlineAsmStatement *stmt, IRState *irs) {
       arg_map[i] = --input_idx;
       asmStmt->in.ops.push_back(arg_val);
       input_constraints.push_back(cns);
+      asmStmt->in.dTypes.push_back(arg->expr->type);
     } else {
       arg_map[i] = n_outputs++;
       asmStmt->out.ops.push_back(arg_val);
       output_constraints.push_back(cns);
+      asmStmt->out.dTypes.push_back(arg->expr->type);
     }
   }
 
@@ -367,6 +369,7 @@ void AsmStatement_toIR(InlineAsmStatement *stmt, IRState *irs) {
       // numbering.
       input_constraints.push_back(ss.str());
       asmStmt->in.ops.push_back(asmStmt->out.ops[n]);
+      asmStmt->in.dTypes.push_back(asmStmt->out.dTypes[n]);
     }
     asmStmt->out.c += oc;
     asmStmt->out.c += ",";
@@ -549,6 +552,7 @@ void CompoundAsmStatement_toIR(CompoundAsmStatement *stmt, IRState *p) {
       // FIXME: Store the value -> label mapping somewhere, so it can be
       // referenced later
       outSetterStmt->in.ops.push_back(DtoConstUint(n_goto));
+      outSetterStmt->in.dTypes.push_back(Type::tuns32);
       outSetterStmt->in.c += "i,";
       code << asmGotoEnd;
 
@@ -564,6 +568,7 @@ void CompoundAsmStatement_toIR(CompoundAsmStatement *stmt, IRState *p) {
       // setup variable for output from asm
       outSetterStmt->out.c = "=*m,";
       outSetterStmt->out.ops.push_back(jump_target);
+      outSetterStmt->out.dTypes.push_back(Type::tuns32);
 
       asmblock->s.push_back(outSetterStmt);
     } else {
@@ -586,8 +591,10 @@ void CompoundAsmStatement_toIR(CompoundAsmStatement *stmt, IRState *p) {
 
   // build asm block
   struct ArgBlock {
+    ArgBlock() = default;
     std::vector<LLValue *> args;
     std::vector<LLType *> types;
+    std::vector<Type *> dTypes;
     std::string c;
   } in, out;
   std::string clobbers;
@@ -603,6 +610,7 @@ void CompoundAsmStatement_toIR(CompoundAsmStatement *stmt, IRState *p) {
     for (size_t j = 0; j < onn; ++j) {
       out.args.push_back(a->out.ops[j]);
       out.types.push_back(a->out.ops[j]->getType());
+      out.dTypes.push_back(a->out.dTypes[j]);
     }
     if (!a->out.c.empty()) {
       out.c += a->out.c;
@@ -619,6 +627,7 @@ void CompoundAsmStatement_toIR(CompoundAsmStatement *stmt, IRState *p) {
     for (size_t j = 0; j < inn; ++j) {
       in.args.push_back(a->in.ops[j]);
       in.types.push_back(a->in.ops[j]->getType());
+      in.dTypes.push_back(a->in.dTypes[j]);
     }
     if (!a->in.c.empty()) {
       in.c += a->in.c;
@@ -668,6 +677,33 @@ void CompoundAsmStatement_toIR(CompoundAsmStatement *stmt, IRState *p) {
   std::vector<LLValue *> args;
   args.insert(args.end(), out.args.begin(), out.args.end());
   args.insert(args.end(), in.args.begin(), in.args.end());
+    
+  auto constraintInfo = llvm::InlineAsm::ParseConstraints(out.c);
+  assert(constraintInfo.size() >= out.dTypes.size() + in.dTypes.size());
+  std::vector<LLType *> indirectTypes;
+  indirectTypes.reserve(out.dTypes.size() + in.dTypes.size());
+  size_t i = asmblock->retn;
+    
+  for (Type *t : out.dTypes) {
+    assert(constraintInfo[i].Type == llvm::InlineAsm::ConstraintPrefix::isOutput);
+    if (constraintInfo[i].isIndirect) {
+      if (TypePointer *pt = t->isTypePointer())
+        indirectTypes.push_back(DtoMemType(pt->nextOf()));
+      else
+        indirectTypes.push_back(DtoMemType(t));
+    }
+    i++;
+  }
+  for (Type *t : in.dTypes) {
+    assert(constraintInfo[i].Type == llvm::InlineAsm::ConstraintPrefix::isInput);
+    if (constraintInfo[i].isIndirect) {
+      if (TypePointer *pt = t->isTypePointer())
+        indirectTypes.push_back(DtoType(pt->nextOf()));
+      else
+        indirectTypes.push_back(DtoType(t));
+    }
+    i++;
+  }
 
   IF_LOG {
     Logger::cout() << "Arguments:" << '\n';
@@ -685,9 +721,21 @@ void CompoundAsmStatement_toIR(CompoundAsmStatement *stmt, IRState *p) {
     Logger::undent();
   }
 
+  for (; i < constraintInfo.size(); i++) {
+    if (!constraintInfo[i].isIndirect)
+      continue;
+    llvm::errs() << "unhandled indirect constraint in" << out.c << "\nindex i = " << i << '\n';
+    llvm::errs() << "function type = " << *fty << '\n';
+    for (size_t j = 0; j < indirectTypes.size(); j++) {
+      llvm::errs() << " " << *(indirectTypes[j]) << '\n';
+    }
+          
+    llvm_unreachable("unhandled indirect constraint");
+  }
+
   llvm::InlineAsm *ia = llvm::InlineAsm::get(fty, code, out.c, true);
 
-  auto call = p->createInlineAsmCall(stmt->loc, ia, args, {});
+  auto call = p->createInlineAsmCall(stmt->loc, ia, args, indirectTypes);
   if (!retty->isVoidTy()) {
     call->setName("asm");
   }
diff --git a/gen/classes.cpp b/gen/classes.cpp
index ba6e9aec53..5fecad4ae9 100644
--- a/gen/classes.cpp
+++ b/gen/classes.cpp
@@ -80,7 +80,7 @@ DValue *DtoNewClass(const Loc &loc, TypeClass *tc, NewExp *newexp) {
   LLValue *mem;
   bool doInit = true;
   if (newexp->onstack) {
-    mem = DtoRawAlloca(DtoType(tc)->getContainedType(0), tc->sym->alignsize,
+    mem = DtoRawAlloca(getIrAggr(tc->sym)->getLLStructType(), tc->sym->alignsize,
                        ".newclass_alloca");
   } else {
     const bool useEHAlloc = global.params.ehnogc && newexp->thrownew;
@@ -145,15 +145,14 @@ void DtoInitClass(TypeClass *tc, LLValue *dst) {
 
   // Set vtable field. Doing this seperately might be optimized better.
   LLValue *tmp = DtoGEP(st, dst, 0u, 0, "vtbl");
-  LLValue *val =
-      DtoBitCast(irClass->getVtblSymbol(), tmp->getType()->getContainedType(0));
+  LLValue *val = irClass->getVtblSymbol();
   DtoStore(val, tmp);
 
   // For D classes, set the monitor field to null.
   const bool isCPPclass = tc->sym->isCPPclass() ? true : false;
   if (!isCPPclass) {
     tmp = DtoGEP(st, dst, 0, 1, "monitor");
-    val = LLConstant::getNullValue(tmp->getType()->getContainedType(0));
+    val = LLConstant::getNullValue(getVoidPtrType());
     DtoStore(val, tmp);
   }
 
diff --git a/gen/dcompute/target.h b/gen/dcompute/target.h
index b1ca77210c..176f8d5dd4 100644
--- a/gen/dcompute/target.h
+++ b/gen/dcompute/target.h
@@ -27,7 +27,7 @@ class DComputeTarget {
 public:
   llvm::LLVMContext &ctx;
   int tversion; // OpenCL or CUDA CC version:major*100 + minor*10
-  enum ID { Host = 0, OpenCL = 1, CUDA = 2 };
+  enum class ID { Host = 0, OpenCL = 1, CUDA = 2 };
   ID target;    // ID for codegen time conditional compilation.
   const char *short_name;
   const char *binSuffix;
diff --git a/gen/dcompute/targetCUDA.cpp b/gen/dcompute/targetCUDA.cpp
index b1a5ebe09c..f242f322a5 100644
--- a/gen/dcompute/targetCUDA.cpp
+++ b/gen/dcompute/targetCUDA.cpp
@@ -26,7 +26,7 @@ class TargetCUDA : public DComputeTarget {
 public:
   TargetCUDA(llvm::LLVMContext &c, int sm)
       : DComputeTarget(
-            c, sm, CUDA, "cuda", "ptx", createNVPTXABI(),
+            c, sm, ID::CUDA, "cuda", "ptx", createNVPTXABI(),
 
             // Map from nominal DCompute address space to NVPTX address space.
             // see $LLVM_ROOT/docs/docs/NVPTXUsage.rst section Address Spaces
diff --git a/gen/dcompute/targetOCL.cpp b/gen/dcompute/targetOCL.cpp
index 5634df24df..7c70293d2d 100644
--- a/gen/dcompute/targetOCL.cpp
+++ b/gen/dcompute/targetOCL.cpp
@@ -42,7 +42,7 @@ class TargetOCL : public DComputeTarget {
   bool usedImage;
 public:
   TargetOCL(llvm::LLVMContext &c, int oclversion)
-      : DComputeTarget(c, oclversion, OpenCL, "ocl", "spv", createSPIRVABI(),
+      : DComputeTarget(c, oclversion, ID::OpenCL, "ocl", "spv", createSPIRVABI(),
                        // Map from nomimal DCompute address space to OpenCL
                        // address. For OpenCL this is a no-op.
                        {{0, 1, 2, 3, 4}}) {
diff --git a/gen/dibuilder.h b/gen/dibuilder.h
index 9888c21001..64a50d32e7 100644
--- a/gen/dibuilder.h
+++ b/gen/dibuilder.h
@@ -213,16 +213,6 @@ public:
     addr.push_back(offset);
   }
 
-  template <typename T> void OpOffset(T &addr, llvm::Value *val, int index) {
-    if (!global.params.symdebug) {
-      return;
-    }
-
-    llvm::StructType *type = isaStruct(val->getType()->getContainedType(0));
-    assert(type);
-    OpOffset(addr, type, index);
-  }
-
   template <typename T> void OpDeref(T &addr) {
     if (!global.params.symdebug) {
       return;
diff --git a/gen/dvalue.cpp b/gen/dvalue.cpp
index 3b7f5aaa3f..02b4166dd0 100644
--- a/gen/dvalue.cpp
+++ b/gen/dvalue.cpp
@@ -131,7 +131,7 @@ DRValue *DLValue::getRVal() {
     return nullptr;
   }
 
-  LLValue *rval = DtoLoad(val->getType()->getPointerElementType(), val);
+  LLValue *rval = DtoLoad(DtoMemType(type), val);
 
   const auto ty = type->toBasetype()->ty;
   if (ty == TY::Tbool) {
@@ -220,3 +220,35 @@ void DBitFieldLValue::store(LLValue *value) {
   const auto newVal = gIR->ir->CreateOr(maskedOldVal, bfVal);
   gIR->ir->CreateAlignedStore(newVal, ptr, LLMaybeAlign(1));
 }
+
+DDcomputeLValue::DDcomputeLValue(Type *t, llvm::Type * llt, LLValue *v) : DLValue(t, v) {
+    lltype = llt;
+}
+DRValue *DDcomputeLValue::getRVal() {
+  if (DtoIsInMemoryOnly(type)) {
+    llvm_unreachable("getRVal() for memory-only type");
+    return nullptr;
+  }
+  
+  LLValue *rval = DtoLoad(lltype, val);
+  
+  const auto ty = type->toBasetype()->ty;
+  if (ty == TY::Tbool) {
+    assert(rval->getType() == llvm::Type::getInt8Ty(gIR->context()));
+  
+    if (isOptimizationEnabled()) {
+      // attach range metadata for i8 being loaded: [0, 2)
+      llvm::MDBuilder mdBuilder(gIR->context());
+      llvm::cast<llvm::LoadInst>(rval)->setMetadata(
+          llvm::LLVMContext::MD_range,
+          mdBuilder.createRange(llvm::APInt(8, 0), llvm::APInt(8, 2)));
+    }
+  
+    // truncate to i1
+    rval = gIR->ir->CreateTrunc(rval, llvm::Type::getInt1Ty(gIR->context()));
+  } else if (ty == TY::Tarray) {
+    return new DSliceValue(type, rval);
+  }
+  
+  return new DImValue(type, rval);
+}
diff --git a/gen/dvalue.h b/gen/dvalue.h
index fefd85d36e..b5131afcab 100644
--- a/gen/dvalue.h
+++ b/gen/dvalue.h
@@ -37,6 +37,7 @@ class DNullValue;
 class DLValue;
 class DSpecialRefValue;
 class DBitFieldLValue;
+class DDcomputeLValue;
 class DSliceValue;
 class DFuncValue;
 
@@ -61,6 +62,7 @@ public:
   virtual DLValue *isLVal() { return nullptr; }
   virtual DSpecialRefValue *isSpecialRef() { return nullptr; }
   virtual DBitFieldLValue *isBitFieldLVal() { return nullptr; }
+  virtual DDcomputeLValue *isDDcomputeLVal() { return nullptr; }
 
   virtual DRValue *isRVal() { return nullptr; }
   virtual DImValue *isIm() { return nullptr; }
@@ -199,5 +201,15 @@ private:
                                     // byte the highest bit is in
 };
 
+/// Represents lvalues of address spaced pointers and pointers
+/// to address spaces other then 0
+class DDcomputeLValue : public DLValue {
+public:
+  llvm::Type *lltype;
+  DDcomputeLValue *isDDcomputeLVal() override { return this; }
+    DDcomputeLValue(Type *t, llvm::Type * llt, llvm::Value *v);
+  DRValue *getRVal() override;
+};
+
 inline llvm::Value *DtoRVal(DValue *v) { return v->getRVal()->val; }
 llvm::Value *DtoLVal(DValue *v);
diff --git a/gen/functions.cpp b/gen/functions.cpp
index f4afb6a12e..a596c6b0e4 100644
--- a/gen/functions.cpp
+++ b/gen/functions.cpp
@@ -488,7 +488,7 @@ void applyTargetMachineAttributes(llvm::Function &func,
   const auto dcompute = gIR->dcomputetarget;
 
   // TODO: (correctly) apply these for NVPTX (but not for SPIRV).
-  if (dcompute && dcompute->target == DComputeTarget::OpenCL)
+  if (dcompute && dcompute->target == DComputeTarget::ID::OpenCL)
     return;
   const auto cpu = dcompute ? "" : target.getTargetCPU();
   const auto features = dcompute ? "" : target.getTargetFeatureString();
diff --git a/gen/irstate.cpp b/gen/irstate.cpp
index 581e38ef41..1160b98885 100644
--- a/gen/irstate.cpp
+++ b/gen/irstate.cpp
@@ -136,7 +136,7 @@ LLConstant *
 IRState::setGlobalVarInitializer(LLGlobalVariable *&globalVar,
                                  LLConstant *initializer,
                                  Dsymbol *symbolForLinkageAndVisibility) {
-  if (initializer->getType() == globalVar->getType()->getContainedType(0)) {
+  if (initializer->getType() == globalVar->getValueType()) {
     defineGlobal(globalVar, initializer, symbolForLinkageAndVisibility);
     return globalVar;
   }
@@ -273,18 +273,14 @@ IRState::createInlineAsmCall(const Loc &loc, llvm::InlineAsm *ia,
   // a non-indirect output constraint (=> return value of call) shifts the
   // constraint/argument index mapping
   ptrdiff_t i = call->getType()->isVoidTy() ? 0 : -1;
-  size_t indirectIdx = 0, indirectLen = indirectTypes.size();
+  size_t indirectIdx = 0;
     
   for (const auto &constraintInfo : ia->ParseConstraints()) {
     if (constraintInfo.isIndirect) {
-      llvm::Type *indirectType = indirectLen != 0 ?
-                                    indirectTypes[indirectIdx] :
-                                    args[i]->getType()->getPointerElementType();
-        
       call->addParamAttr(i, llvm::Attribute::get(
                                 context(),
                                 llvm::Attribute::ElementType,
-                                indirectType));
+                                indirectTypes[indirectIdx]));
       ++indirectIdx;
     }
     ++i;
diff --git a/gen/irstate.h b/gen/irstate.h
index 07851e5e49..fbd69cd4a3 100644
--- a/gen/irstate.h
+++ b/gen/irstate.h
@@ -80,6 +80,7 @@ struct IRAsmStmt {
   struct Operands {
     std::string c; // contraint
     std::vector<LLValue *> ops;
+    std::vector<Type *> dTypes;
   };
   Operands out, in;
 
diff --git a/gen/llvmhelpers.cpp b/gen/llvmhelpers.cpp
index 769e077948..61338b9528 100644
--- a/gen/llvmhelpers.cpp
+++ b/gen/llvmhelpers.cpp
@@ -414,7 +414,7 @@ void DtoAssign(const Loc &loc, DValue *lhs, DValue *rhs, EXP op,
       // time as to not emit an invalid (overlapping) memcpy on trivial
       // struct self-assignments like 'A a; a = a;'.
       if (src != dst)
-        DtoMemCpy(dst, src);
+        DtoMemCpy(DtoType(lhs->type), dst, src);
     }
   } else if (t->ty == TY::Tarray || t->ty == TY::Tsarray) {
     DtoArrayAssign(loc, lhs, rhs, op, canSkipPostblit);
@@ -434,7 +434,7 @@ void DtoAssign(const Loc &loc, DValue *lhs, DValue *rhs, EXP op,
       Logger::cout() << "l : " << *l << '\n';
       Logger::cout() << "r : " << *r << '\n';
     }
-    r = DtoBitCast(r, l->getType()->getContainedType(0));
+    r = DtoBitCast(r, DtoType(lhs->type));
     DtoStore(r, l);
   } else if (t->iscomplex()) {
     LLValue *dst = DtoLVal(lhs);
@@ -447,7 +447,7 @@ void DtoAssign(const Loc &loc, DValue *lhs, DValue *rhs, EXP op,
       Logger::cout() << "lhs: " << *l << '\n';
       Logger::cout() << "rhs: " << *r << '\n';
     }
-    LLType *lit = l->getType()->getContainedType(0);
+    LLType *lit = DtoType(lhs->type);
     if (r->getType() != lit) {
       r = DtoRVal(DtoCast(loc, rhs, lhs->type));
       IF_LOG {
@@ -1882,10 +1882,12 @@ DLValue *DtoIndexAggregate(LLValue *src, AggregateDeclaration *ad,
   unsigned off = irTypeAggr->getMemberLocation(vd, isFieldIdx);
 
   LLValue *ptr = src;
+  LLType * ty = nullptr;
   if (!isFieldIdx) {
     // Cast to void* to apply byte-wise offset from object start.
     ptr = DtoBitCast(ptr, getVoidPtrType());
     ptr = DtoGEP1(llvm::Type::getInt8Ty(gIR->context()), ptr, off);
+    ty = DtoType(vd->type);
   } else {
     if (ad->structsize == 0) { // can happen for extern(C) structs
       assert(off == 0);
@@ -1904,6 +1906,7 @@ DLValue *DtoIndexAggregate(LLValue *src, AggregateDeclaration *ad,
       }
       ptr = DtoBitCast(ptr, pst);
       ptr = DtoGEP(st, ptr, 0, off);
+      ty = isaStruct(st)->getElementType(off);
     }
   }
 
@@ -1911,6 +1914,10 @@ DLValue *DtoIndexAggregate(LLValue *src, AggregateDeclaration *ad,
   ptr = DtoBitCast(ptr, DtoPtrToType(vd->type));
 
   IF_LOG Logger::cout() << "Pointer: " << *ptr << '\n';
+  if (auto p = isaPointer(ty)) {
+    if (p->getAddressSpace())
+      return new DDcomputeLValue(vd->type, p, ptr);
+  }
   return new DLValue(vd->type, ptr);
 }
 
diff --git a/gen/naked.cpp b/gen/naked.cpp
index cce45c102c..05c68d6541 100644
--- a/gen/naked.cpp
+++ b/gen/naked.cpp
@@ -503,17 +503,20 @@ llvm::CallInst *DtoInlineAsmExpr(const Loc &loc, llvm::StringRef code,
   llvm::FunctionType *FT =
       llvm::FunctionType::get(returnType, operandTypes, false);
 
-  // make sure the constraints are valid
-  if (!llvm::InlineAsm::
 #if LDC_LLVM_VER < 1500
-      Verify
+  // make sure the constraints are valid
+  if (!llvm::InlineAsm::Verify(FT, constraints)) {
+      error(loc, "inline asm constraints are invalid");
+      fatal();
+  }
 #else
-      verify
-#endif
-      (FT, constraints)) {
+  if (auto err = llvm::InlineAsm::verify(FT, constraints)) {
     error(loc, "inline asm constraints are invalid");
+    llvm::errs() << err;
     fatal();
   }
+#endif
+
 
   // build asm call
   bool sideeffect = true;
diff --git a/gen/nested.cpp b/gen/nested.cpp
index cc8abc4819..e3d9c06c82 100644
--- a/gen/nested.cpp
+++ b/gen/nested.cpp
@@ -149,7 +149,7 @@ DValue *DtoNestedVariable(const Loc &loc, Type *astype, VarDeclaration *vd,
 
   const auto offsetToNthField = [&val, &dwarfAddrOps, &currFrame](unsigned fieldIndex,
                                                       const char *name = "") {
-    gIR->DBuilder.OpOffset(dwarfAddrOps, val, fieldIndex);
+    gIR->DBuilder.OpOffset(dwarfAddrOps, currFrame, fieldIndex);
     val = DtoGEP(currFrame, val, 0, fieldIndex, name);
   };
 
@@ -538,7 +538,7 @@ void DtoCreateNestedContext(FuncGenState &funcGen) {
           // The parameter value is an alloca'd stack slot.
           // Copy to the nesting frame and leave the alloca for
           // the optimizers to clean up.
-          DtoMemCpy(gep, parm->value);
+          DtoMemCpy(frameType->getContainedType(irLocal->nestedIndex), gep, parm->value);
           gep->takeName(parm->value);
           parm->value = gep; // update variable lvalue
         }
diff --git a/gen/passes/GarbageCollect2Stack.cpp b/gen/passes/GarbageCollect2Stack.cpp
index adb7ab9d04..320e6cb78f 100644
--- a/gen/passes/GarbageCollect2Stack.cpp
+++ b/gen/passes/GarbageCollect2Stack.cpp
@@ -322,7 +322,7 @@ public:
     }
 
     // Should be i8.
-    Ty = CB->getType()->getContainedType(0);
+    Ty = llvm::Type::getInt8Ty(CB->getContext());
     return true;
   }
 
diff --git a/gen/semantic-dcompute.cpp b/gen/semantic-dcompute.cpp
index c40ca586e6..157656a02f 100644
--- a/gen/semantic-dcompute.cpp
+++ b/gen/semantic-dcompute.cpp
@@ -203,7 +203,7 @@ struct DComputeSemanticAnalyser : public StoppableVisitor {
     if (auto ce = stmt->condition->isCallExp()) {
       if (ce->f && ce->f->ident == Id::dcReflect) {
         auto arg1 = (DComputeTarget::ID)(*ce->arguments)[0]->toInteger();
-        if (arg1 == DComputeTarget::Host)
+        if (arg1 == DComputeTarget::ID::Host)
           stop = true;
       }
     }
diff --git a/gen/statements.cpp b/gen/statements.cpp
index a6ad29ab1a..29b9fcb942 100644
--- a/gen/statements.cpp
+++ b/gen/statements.cpp
@@ -364,7 +364,7 @@ public:
     auto arg2 = (*ce->arguments)[1]->toInteger();
     auto dct = irs->dcomputetarget;
     if (!dct) {
-      return arg1 == DComputeTarget::Host;
+      return arg1 == DComputeTarget::ID::Host;
     } else {
       return arg1 == dct->target &&
              (!arg2 || arg2 == static_cast<dinteger_t>(dct->tversion));
diff --git a/gen/tocall.cpp b/gen/tocall.cpp
index 6b353d6427..a1eff2c33b 100644
--- a/gen/tocall.cpp
+++ b/gen/tocall.cpp
@@ -188,7 +188,7 @@ static void addExplicitArguments(std::vector<LLValue *> &args, AttrSet &attrs,
         Logger::cout() << "expects: " << *paramType << '\n';
       }
       if (isaStruct(llVal)) {
-        llVal = DtoAggrPaint(llVal, paramType);
+        llVal = DtoSlicePaint(llVal, paramType);
       } else {
         llVal = DtoBitCast(llVal, paramType);
       }
@@ -295,7 +295,7 @@ bool DtoLowerMagicIntrinsic(IRState *p, FuncDeclaration *fndecl, CallExp *e,
     assert(ap);
     // variadic extern(D) function with implicit _argptr?
     if (LLValue *argptrMem = p->func()->_argptr) {
-      DtoMemCpy(DtoLVal(ap), argptrMem); // ap = _argptr
+      DtoMemCpy(DtoType(ap->type), DtoLVal(ap), argptrMem); // ap = _argptr
     } else {
       LLValue *llAp = gABI->prepareVaStart(ap);
       p->ir->CreateCall(GET_INTRINSIC_DECL(vastart), llAp, "");
@@ -395,7 +395,7 @@ bool DtoLowerMagicIntrinsic(IRState *p, FuncDeclaration *fndecl, CallExp *e,
 
     DValue *dval = toElem(exp1);
     LLValue *ptr = DtoRVal(exp2);
-    LLType *pointeeType = ptr->getType()->getContainedType(0);
+    LLType *pointeeType = DtoType(exp2->type->isTypePointer()->nextOf());
 
     LLValue *val = nullptr;
     if (pointeeType->isIntegerTy()) {
@@ -482,7 +482,7 @@ bool DtoLowerMagicIntrinsic(IRState *p, FuncDeclaration *fndecl, CallExp *e,
     const bool isWeak = (*e->arguments)[5]->toInteger() != 0;
 
     LLValue *ptr = DtoRVal(exp1);
-    LLType *pointeeType = ptr->getType()->getContainedType(0);
+    LLType *pointeeType = DtoType(exp1->type->isTypePointer()->nextOf());
     DValue *dcmp = toElem(exp2);
     DValue *dval = toElem(exp3);
 
@@ -708,11 +708,9 @@ private:
     }
 
     size_t index = args.size();
-    LLType *llArgType = *(llArgTypesBegin + index);
-
     LLValue *pointer = sretPointer;
     if (!pointer) {
-      pointer = DtoRawAlloca(llArgType->getContainedType(0),
+      pointer = DtoRawAlloca(DtoType(tf->nextOf()),
                              DtoAlignment(resulttype), ".sret_tmp");
     }
 
@@ -958,7 +956,7 @@ DValue *DtoCallFunction(const Loc &loc, Type *resulttype, DValue *fnval,
       if (tf->isref()) {
         retllval = DtoBitCast(retllval, DtoType(rbase->pointerTo()));
       } else {
-        retllval = DtoAggrPaint(retllval, DtoType(rbase));
+        retllval = DtoSlicePaint(retllval, DtoType(rbase));
       }
       break;
 
diff --git a/gen/toconstelem.cpp b/gen/toconstelem.cpp
index e2676d7ba4..e62e2f8bac 100644
--- a/gen/toconstelem.cpp
+++ b/gen/toconstelem.cpp
@@ -341,7 +341,7 @@ public:
       result = base;
     } else {
       const unsigned elemSize =
-          gDataLayout->getTypeStoreSize(base->getType()->getContainedType(0));
+          gDataLayout->getTypeStoreSize(DtoType(e->var->type));
 
       IF_LOG Logger::println("adding offset: %llu (elem size: %u)",
                              static_cast<unsigned long long>(e->offset),
diff --git a/gen/toir.cpp b/gen/toir.cpp
index c3710c9331..61ddfbda56 100644
--- a/gen/toir.cpp
+++ b/gen/toir.cpp
@@ -65,8 +65,9 @@ bool walkPostorder(Expression *e, StoppableVisitor *v);
 
 static LLValue *write_zeroes(LLValue *mem, unsigned start, unsigned end) {
   mem = DtoBitCast(mem, getVoidPtrType());
-  LLValue *gep = DtoGEP1(LLType::getInt8Ty(gIR->context()), mem, start, ".padding");
-  DtoMemSetZero(gep, DtoConstSize_t(end - start));
+  LLType *i8 = LLType::getInt8Ty(gIR->context());
+  LLValue *gep = DtoGEP1(i8, mem, start, ".padding");
+  DtoMemSetZero(i8, gep, DtoConstSize_t(end - start));
   return mem;
 }
 
@@ -546,7 +547,7 @@ public:
       Logger::println("performing aggregate zero initialization");
       assert(e->e2->toInteger() == 0);
       LLValue *lval = DtoLVal(lhs);
-      DtoMemSetZero(lval);
+      DtoMemSetZero(DtoType(lhs->type), lval);
       TypeStruct *ts = static_cast<TypeStruct *>(e->e1->type);
       if (ts->sym->isNested() && ts->sym->vthis)
         DtoResolveNestedContext(e->loc, ts->sym, lval);
@@ -843,7 +844,7 @@ public:
     if (e->offset == 0) {
       offsetValue = baseValue;
     } else {
-      LLType *elemType = baseValue->getType()->getContainedType(0);
+      LLType *elemType = DtoType(base->type);
       if (elemType->isSized()) {
         uint64_t elemSize = gDataLayout->getTypeAllocSize(elemType);
         if (e->offset % elemSize == 0) {
@@ -948,6 +949,14 @@ public:
     result = new DLValue(e->type, DtoBitCast(V, DtoPtrToType(e->type)));
   }
 
+  static llvm::PointerType * getWithSamePointeeType(llvm::PointerType *p, unsigned as) {
+#if LDC_LLVM_VER >= 1300
+    return llvm::PointerType::getWithSamePointeeType(p, as);
+#else
+    return p->getPointerElementType()->getPointerTo(as);
+#endif
+  }
+
   //////////////////////////////////////////////////////////////////////////////
 
   void visit(DotVarExp *e) override {
@@ -985,14 +994,24 @@ public:
         llvm_unreachable("Unknown DotVarExp type for VarDeclaration.");
       }
 
-      auto ptr = DtoLVal(DtoIndexAggregate(aggrPtr, ad, vd));
+      auto ptr = DtoIndexAggregate(aggrPtr, ad, vd);
 
-      // special case for bit fields (no real lvalues)
+      // special case for bit fields (no real lvalues), and address spaced pointers
       if (auto bf = vd->isBitFieldDeclaration()) {
-        result = new DBitFieldLValue(e->type, ptr, bf);
+        result = new DBitFieldLValue(e->type, DtoLVal(ptr), bf);
+      } else if (auto d = ptr->isDDcomputeLVal()) {
+        LLType *ptrty = nullptr;
+        if (llvm::PointerType *p = isaPointer(d->lltype)) {
+          unsigned as = p->getAddressSpace();
+          ptrty = getWithSamePointeeType(isaPointer(DtoType(e->type)), as);
+        }
+        else
+           ptrty = DtoType(e->type);
+        result = new DDcomputeLValue(e->type, i1ToI8(ptrty),
+                                     DtoBitCast(DtoLVal(d), DtoPtrToType(e->type)));
       } else {
-        ptr = DtoBitCast(ptr, DtoPtrToType(e->type));
-        result = new DLValue(e->type, ptr);
+        LLValue *p = DtoBitCast(DtoLVal(ptr), DtoPtrToType(e->type));
+        result = new DLValue(e->type, p);
       }
     } else if (FuncDeclaration *fdecl = e->var->isFuncDeclaration()) {
       // This is a bit more convoluted than it would need to be, because it
@@ -1622,7 +1641,7 @@ public:
         DtoDeleteClass(e->loc, dval); // sets dval to null
       } else if (dval->isLVal()) {
         LLValue *lval = DtoLVal(dval);
-        DtoStore(LLConstant::getNullValue(lval->getType()->getContainedType(0)),
+        DtoStore(LLConstant::getNullValue(DtoType(dval->type)),
                  lval);
       }
     }
@@ -2301,12 +2320,12 @@ public:
         dstMem = DtoAlloca(e->type, ".structliteral");
 
       if (sd->zeroInit()) {
-        DtoMemSetZero(dstMem);
+        DtoMemSetZero(DtoType(e->type), dstMem);
       } else {
         LLValue *initsym = getIrAggr(sd)->getInitSymbol();
         initsym = DtoBitCast(initsym, DtoType(e->type->pointerTo()));
         assert(dstMem->getType() == initsym->getType());
-        DtoMemCpy(dstMem, initsym);
+        DtoMemCpy(DtoType(e->type), dstMem, initsym);
       }
 
       return new DLValue(e->type, dstMem);
@@ -2456,7 +2475,7 @@ public:
           LLGlobalValue::InternalLinkage, initval, ".aaKeysStorage");
       LLConstant *slice = DtoGEP(initval->getType(), globalstore, 0u, 0u);
       slice = DtoConstSlice(DtoConstSize_t(e->keys->length), slice);
-      LLValue *keysArray = DtoAggrPaint(slice, funcTy->getParamType(1));
+      LLValue *keysArray = DtoSlicePaint(slice, funcTy->getParamType(1));
 
       initval = arrayConst(valuesInits, vtype);
       globalstore = new LLGlobalVariable(gIR->module, initval->getType(), false,
@@ -2464,7 +2483,7 @@ public:
                                          initval, ".aaValuesStorage");
       slice = DtoGEP(initval->getType(), globalstore, 0u, 0u);
       slice = DtoConstSlice(DtoConstSize_t(e->keys->length), slice);
-      LLValue *valuesArray = DtoAggrPaint(slice, funcTy->getParamType(2));
+      LLValue *valuesArray = DtoSlicePaint(slice, funcTy->getParamType(2));
 
       LLValue *aa = gIR->CreateCallOrInvoke(func, aaTypeInfo, keysArray,
                                             valuesArray, "aa");
@@ -2574,7 +2593,7 @@ public:
       DValue *ep = toElem(el);
       LLValue *gep = DtoGEP(st, val, 0, i);
       if (DtoIsInMemoryOnly(el->type)) {
-        DtoMemCpy(gep, DtoLVal(ep));
+        DtoMemCpy(st->getContainedType(i), gep, DtoLVal(ep));
       } else if (el->type->ty != TY::Tvoid) {
         DtoStoreZextI8(DtoRVal(ep), gep);
       } else {
@@ -2651,7 +2670,7 @@ public:
       Type *srcElementType = tsrc->nextOf();
 
       if (DtoMemType(elementType) == DtoMemType(srcElementType)) {
-        DtoMemCpy(dstMem, arrayPtr);
+        DtoMemCpy(dstType, dstMem, arrayPtr);
       } else {
         for (unsigned i = 0; i < N; ++i) {
           LLValue *gep = DtoGEP1(DtoMemType(e1->type->nextOf()), arrayPtr, i);
@@ -2821,7 +2840,7 @@ bool toInPlaceConstruction(DLValue *lhs, Expression *rhs) {
         DtoResolveStruct(sd);
         if (sd->zeroInit()) {
           Logger::println("success, zeroing out");
-          DtoMemSetZero(DtoLVal(lhs));
+          DtoMemSetZero(DtoType(lhs->type) ,DtoLVal(lhs));
           return true;
         }
       }
diff --git a/gen/tollvm.cpp b/gen/tollvm.cpp
index 6c95968db9..4225262b61 100644
--- a/gen/tollvm.cpp
+++ b/gen/tollvm.cpp
@@ -178,7 +178,7 @@ LLType *DtoType(Type *t) {
       // This is an enum forward reference that is only legal when referenced
       // through an indirection (e.g. "enum E; void foo(E* p);"). For lack of a
       // better choice, make the outer indirection a void pointer.
-      return getVoidPtrType()->getContainedType(0);
+      return getI8Type();
     }
     return DtoType(bt);
   }
@@ -397,13 +397,13 @@ void DtoMemSet(LLValue *dst, LLValue *val, LLValue *nbytes, unsigned align) {
 
 ////////////////////////////////////////////////////////////////////////////////
 
-void DtoMemSetZero(LLValue *dst, LLValue *nbytes, unsigned align) {
+void DtoMemSetZero(LLType *type, LLValue *dst, LLValue *nbytes, unsigned align) {
   DtoMemSet(dst, DtoConstUbyte(0), nbytes, align);
 }
 
-void DtoMemSetZero(LLValue *dst, unsigned align) {
-  uint64_t n = getTypeStoreSize(dst->getType()->getContainedType(0));
-  DtoMemSetZero(dst, DtoConstSize_t(n), align);
+void DtoMemSetZero(LLType *type, LLValue *dst, unsigned align) {
+  uint64_t n = getTypeStoreSize(type);
+  DtoMemSetZero(type, dst, DtoConstSize_t(n), align);
 }
 
 ////////////////////////////////////////////////////////////////////////////////
@@ -418,10 +418,9 @@ void DtoMemCpy(LLValue *dst, LLValue *src, LLValue *nbytes, unsigned align) {
   gIR->ir->CreateMemCpy(dst, A, src, A, nbytes, false /*isVolatile*/);
 }
 
-void DtoMemCpy(LLValue *dst, LLValue *src, bool withPadding, unsigned align) {
-  LLType *pointee = dst->getType()->getContainedType(0);
+void DtoMemCpy(LLType *type, LLValue *dst, LLValue *src, bool withPadding, unsigned align) {
   uint64_t n =
-      withPadding ? getTypeAllocSize(pointee) : getTypeStoreSize(pointee);
+      withPadding ? getTypeAllocSize(type) : getTypeStoreSize(type);
   DtoMemCpy(dst, src, DtoConstSize_t(n), align);
 }
 
@@ -543,7 +542,6 @@ void DtoVolatileStore(LLValue *src, LLValue *dst) {
 void DtoStoreZextI8(LLValue *src, LLValue *dst) {
   if (src->getType()->isIntegerTy(1)) {
     llvm::Type *i8 = llvm::Type::getInt8Ty(gIR->context());
-    assert(dst->getType()->getContainedType(0) == i8);
     src = gIR->ir->CreateZExt(src, i8);
   }
   gIR->ir->CreateStore(src, dst);
@@ -767,7 +765,7 @@ LLValue *DtoAggrPair(LLValue *V1, LLValue *V2, const char *name) {
   return DtoAggrPair(t, V1, V2, name);
 }
 
-LLValue *DtoAggrPaint(LLValue *aggr, LLType *as) {
+LLValue *DtoSlicePaint(LLValue *aggr, LLType *as) {
   if (aggr->getType() == as) {
     return aggr;
   }
diff --git a/gen/tollvm.h b/gen/tollvm.h
index b4e7eac531..57dbc7acc9 100644
--- a/gen/tollvm.h
+++ b/gen/tollvm.h
@@ -161,7 +161,7 @@ unsigned int getABITypeAlign(LLType *t);
 LLValue *DtoAggrPair(LLType *type, LLValue *V1, LLValue *V2,
                      const char *name = "");
 LLValue *DtoAggrPair(LLValue *V1, LLValue *V2, const char *name = "");
-LLValue *DtoAggrPaint(LLValue *aggr, LLType *as);
+LLValue *DtoSlicePaint(LLValue *aggr, LLType *as);
 
 /**
  * Generates a call to llvm.memset.i32 (or i64 depending on architecture).
@@ -178,7 +178,7 @@ void DtoMemSet(LLValue *dst, LLValue *val, LLValue *nbytes, unsigned align = 1);
  * @param nbytes Number of bytes to overwrite.
  * @param align The minimum alignment of the destination memory.
  */
-void DtoMemSetZero(LLValue *dst, LLValue *nbytes, unsigned align = 1);
+void DtoMemSetZero(LLType *type, LLValue *dst, LLValue *nbytes, unsigned align = 1);
 
 /**
  * The same as DtoMemSetZero but figures out the size itself based on the
@@ -186,7 +186,7 @@ void DtoMemSetZero(LLValue *dst, LLValue *nbytes, unsigned align = 1);
  * @param dst Destination memory.
  * @param align The minimum alignment of the destination memory.
  */
-void DtoMemSetZero(LLValue *dst, unsigned align = 1);
+void DtoMemSetZero(LLType *type, LLValue *dst, unsigned align = 1);
 
 /**
  * Generates a call to llvm.memcpy.i32 (or i64 depending on architecture).
@@ -205,7 +205,7 @@ void DtoMemCpy(LLValue *dst, LLValue *src, LLValue *nbytes, unsigned align = 1);
  * @param withPadding Use the dst pointee's padded size, not its store size.
  * @param align The minimum alignment of the source and destination memory.
  */
-void DtoMemCpy(LLValue *dst, LLValue *src, bool withPadding = false,
+void DtoMemCpy(LLType *type, LLValue *dst, LLValue *src, bool withPadding = false,
                unsigned align = 1);
 
 /**
diff --git a/ir/irclass.cpp b/ir/irclass.cpp
index cfdd7f2413..0447baec85 100644
--- a/ir/irclass.cpp
+++ b/ir/irclass.cpp
@@ -426,7 +426,7 @@ LLConstant *IrClass::getClassInfoInit() {
   }
 
   // build the initializer
-  LLType *initType = getClassInfoSymbol()->getType()->getContainedType(0);
+  LLType *initType = getClassInfoSymbol()->getValueType();
   constTypeInfo = b.get_constant(isaStruct(initType));
 
   return constTypeInfo;
@@ -567,7 +567,7 @@ LLConstant *IrClass::getInterfaceVtblInit(BaseClass *b,
                                   needsCOMDAT());
       const auto callee = irFunc->getLLVMCallee();
       thunk = LLFunction::Create(
-          isaFunction(callee->getType()->getContainedType(0)), lwc.first,
+          callee->getFunctionType(), lwc.first,
           thunkIRMangle, &gIR->module);
       setLinkage(lwc, thunk);
       thunk->copyAttributesFrom(callee);
diff --git a/ir/irvar.cpp b/ir/irvar.cpp
index 7c329020d9..97339d2f9e 100644
--- a/ir/irvar.cpp
+++ b/ir/irvar.cpp
@@ -51,6 +51,9 @@ LLValue *IrGlobal::getValue(bool define) {
   return value;
 }
 
+llvm::Type *IrGlobal::getType() {
+  return llvm::dyn_cast<llvm::GlobalVariable>(value)->getValueType();
+}
 void IrGlobal::declare() {
   Logger::println("Declaring global: %s", V->toChars());
   LOG_SCOPE
diff --git a/ir/irvar.h b/ir/irvar.h
index 4ef6ee5b02..ff6732cac6 100644
--- a/ir/irvar.h
+++ b/ir/irvar.h
@@ -38,7 +38,7 @@ struct IrGlobal : IrVar {
   bool nakedUse = false;
 
   llvm::Value *getValue(bool define = false);
-  llvm::Type *getType() { return value->getType()->getContainedType(0); }
+  llvm::Type *getType();
 
 private:
   void declare();
diff --git a/runtime/druntime b/runtime/druntime
index 39fccadfd7..673ff369d1 160000
--- a/runtime/druntime
+++ b/runtime/druntime
@@ -1 +1 @@
-Subproject commit 39fccadfd706c56688ccece37d3bed008eef8315
+Subproject commit 673ff369d1d7f0fc1a8392413cce99380158989a
diff --git a/tests/codegen/gh4134.d b/tests/codegen/gh4134.d
new file mode 100644
index 0000000000..dc91a5b0f1
--- /dev/null
+++ b/tests/codegen/gh4134.d
@@ -0,0 +1,14 @@
+// https://github.com/ldc-developers/ldc/issues/4134
+// RUN: %ldc -run %s
+
+int i;
+
+string getString() {
+    ++i;
+    return "Abc";
+}
+
+void main() {
+    const r = getString() ~ getString();
+    assert(i == 2);
+}