[Draft] Address space lowering for SYCL

clang/include/clang/AST/Type.h

@@ -488,7 +488,12 @@ class Qualifiers {
   /// Returns true if the address space in these qualifiers is equal to or
   /// a superset of the address space in the argument qualifiers.
   bool isAddressSpaceSupersetOf(Qualifiers other) const {
-    return isAddressSpaceSupersetOf(getAddressSpace(), other.getAddressSpace());
+    return isAddressSpaceSupersetOf(getAddressSpace(),
+                                    other.getAddressSpace()) ||
+           (!hasAddressSpace() &&
+            (other.getAddressSpace() == LangAS::opencl_private ||
+             other.getAddressSpace() == LangAS::opencl_local ||
+             other.getAddressSpace() == LangAS::opencl_global));
   }
 
   /// Determines if these qualifiers compatibly include another set.

clang/lib/AST/ItaniumMangle.cpp

@@ -2381,7 +2381,7 @@ void CXXNameMangler::mangleQualifiers(Qualifiers Quals, const DependentAddressSp
     if (Context.getASTContext().addressSpaceMapManglingFor(AS)) {
       //  <target-addrspace> ::= "AS" <address-space-number>
       unsigned TargetAS = Context.getASTContext().getTargetAddressSpace(AS);
-      if (TargetAS != 0)
+      if (TargetAS != 0 || (Context.getASTContext().getLangOpts().SYCLIsDevice))
         ASString = "AS" + llvm::utostr(TargetAS);
     } else {
       switch (AS) {

clang/lib/Basic/Targets.cpp

@@ -567,17 +567,29 @@ TargetInfo *AllocateTarget(const llvm::Triple &Triple,
     }
 
   case llvm::Triple::spir: {
-    if (Triple.getOS() != llvm::Triple::UnknownOS ||
-        Triple.getEnvironment() != llvm::Triple::UnknownEnvironment)
-      return nullptr;
+    if (Triple.getEnvironment() == llvm::Triple::SYCLDevice) {
+      switch (os) {
+      case llvm::Triple::Linux:
+        return new LinuxTargetInfo<SPIR32SYCLDeviceTargetInfo>(Triple, Opts);
+      default:
+        return new SPIR32SYCLDeviceTargetInfo(Triple, Opts);
+      }
+    }
     return new SPIR32TargetInfo(Triple, Opts);
   }
+
   case llvm::Triple::spir64: {
-    if (Triple.getOS() != llvm::Triple::UnknownOS ||
-        Triple.getEnvironment() != llvm::Triple::UnknownEnvironment)
-      return nullptr;
+    if (Triple.getEnvironment() == llvm::Triple::SYCLDevice) {
+      switch (os) {
+      case llvm::Triple::Linux:
+        return new LinuxTargetInfo<SPIR64SYCLDeviceTargetInfo>(Triple, Opts);
+      default:
+        return new SPIR64SYCLDeviceTargetInfo(Triple, Opts);
+      }
+    }
     return new SPIR64TargetInfo(Triple, Opts);
   }
+
   case llvm::Triple::wasm32:
     if (Triple.getSubArch() != llvm::Triple::NoSubArch ||
         Triple.getVendor() != llvm::Triple::UnknownVendor ||

clang/lib/Basic/Targets/SPIR.h

@@ -36,18 +36,31 @@ static const unsigned SPIRAddrSpaceMap[] = {
     0  // ptr64
 };
 
+static const unsigned SYCLAddrSpaceMap[] = {
+    4, // Default
+    1, // opencl_global
+    3, // opencl_local
+    2, // opencl_constant
+    0, // opencl_private
+    4, // opencl_generic
+    0, // cuda_device
+    0, // cuda_constant
+    0, // cuda_shared
+    0, // ptr32_sptr
+    0, // ptr32_uptr
+    0  // ptr64
+};
+
 class LLVM_LIBRARY_VISIBILITY SPIRTargetInfo : public TargetInfo {
 public:
   SPIRTargetInfo(const llvm::Triple &Triple, const TargetOptions &)
       : TargetInfo(Triple) {
-    assert(getTriple().getOS() == llvm::Triple::UnknownOS &&
-           "SPIR target must use unknown OS");
-    assert(getTriple().getEnvironment() == llvm::Triple::UnknownEnvironment &&
-           "SPIR target must use unknown environment type");
     TLSSupported = false;
     VLASupported = false;
     LongWidth = LongAlign = 64;
-    AddrSpaceMap = &SPIRAddrSpaceMap;
+    AddrSpaceMap = (Triple.getEnvironment() == llvm::Triple::SYCLDevice)
+                       ? &SYCLAddrSpaceMap
+                       : &SPIRAddrSpaceMap;
     UseAddrSpaceMapMangling = true;
     HasLegalHalfType = true;
     HasFloat16 = true;
@@ -132,6 +145,43 @@ class LLVM_LIBRARY_VISIBILITY SPIR64TargetInfo : public SPIRTargetInfo {
   void getTargetDefines(const LangOptions &Opts,
                         MacroBuilder &Builder) const override;
 };
+
+class LLVM_LIBRARY_VISIBILITY SPIR32SYCLDeviceTargetInfo
+    : public SPIR32TargetInfo {
+public:
+  SPIR32SYCLDeviceTargetInfo(const llvm::Triple &Triple,
+                             const TargetOptions &Opts)
+      : SPIR32TargetInfo(Triple, Opts) {
+    // This is workaround for exception_ptr class.
+    // Exceptions is not allowed in sycl device code but we should be able
+    // to parse host code. So we allow compilation of exception_ptr but
+    // if exceptions are used in device code we should emit a diagnostic.
+    MaxAtomicInlineWidth = 32;
+    // This is workaround for mutex class.
+    // I'm not sure about this hack but I guess that mutex_class is same
+    // problem.
+    TLSSupported = true;
+  }
+};
+
+class LLVM_LIBRARY_VISIBILITY SPIR64SYCLDeviceTargetInfo
+    : public SPIR64TargetInfo {
+public:
+  SPIR64SYCLDeviceTargetInfo(const llvm::Triple &Triple,
+                             const TargetOptions &Opts)
+      : SPIR64TargetInfo(Triple, Opts) {
+    // This is workaround for exception_ptr class.
+    // Exceptions is not allowed in sycl device code but we should be able
+    // to parse host code. So we allow compilation of exception_ptr but
+    // if exceptions are used in device code we should emit a diagnostic.
+    MaxAtomicInlineWidth = 64;
+    // This is workaround for mutex class.
+    // I'm not sure about this hack but I guess that mutex_class is same
+    // problem.
+    TLSSupported = true;
+  }
+};
+
 } // namespace targets
 } // namespace clang
 #endif // LLVM_CLANG_LIB_BASIC_TARGETS_SPIR_H

clang/lib/CodeGen/CGCall.cpp

@@ -4558,6 +4558,17 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
             V->getType()->isIntegerTy())
           V = Builder.CreateZExt(V, ArgInfo.getCoerceToType());
 
+        if (FirstIRArg < IRFuncTy->getNumParams()) {
+          const auto *LHSPtrTy =
+              dyn_cast_or_null<llvm::PointerType>(V->getType());
+          const auto *RHSPtrTy = dyn_cast_or_null<llvm::PointerType>(
+              IRFuncTy->getParamType(FirstIRArg));
+          if (LHSPtrTy && RHSPtrTy &&
+              LHSPtrTy->getAddressSpace() != RHSPtrTy->getAddressSpace())
+            V = Builder.CreateAddrSpaceCast(V,
+                                            IRFuncTy->getParamType(FirstIRArg));
+        }
+
         // If the argument doesn't match, perform a bitcast to coerce it.  This
         // can happen due to trivial type mismatches.
         if (FirstIRArg < IRFuncTy->getNumParams() &&
@@ -4784,6 +4795,20 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
   if (!CallArgs.getCleanupsToDeactivate().empty())
     deactivateArgCleanupsBeforeCall(*this, CallArgs);
 
+  // Addrspace cast to generic if necessary
+  for (unsigned i = 0; i < IRFuncTy->getNumParams(); ++i) {
+    if (auto *PtrTy = dyn_cast<llvm::PointerType>(IRCallArgs[i]->getType())) {
+      auto *ExpectedPtrType =
+          cast<llvm::PointerType>(IRFuncTy->getParamType(i));
+      unsigned ValueAS = PtrTy->getAddressSpace();
+      unsigned ExpectedAS = ExpectedPtrType->getAddressSpace();
+      if (ValueAS != ExpectedAS) {
+        IRCallArgs[i] = Builder.CreatePointerBitCastOrAddrSpaceCast(
+            IRCallArgs[i], ExpectedPtrType);
+      }
+    }
+  }
+
   // Assert that the arguments we computed match up.  The IR verifier
   // will catch this, but this is a common enough source of problems
   // during IRGen changes that it's way better for debugging to catch

clang/lib/CodeGen/CGClass.cpp

@@ -336,7 +336,7 @@ Address CodeGenFunction::GetAddressOfBaseClass(
       EmitTypeCheck(TCK_Upcast, Loc, Value.getPointer(),
                     DerivedTy, DerivedAlign, SkippedChecks);
     }
-    return Builder.CreateBitCast(Value, BasePtrTy);
+    return Builder.CreatePointerBitCastOrAddrSpaceCast(Value, BasePtrTy);
   }
 
   llvm::BasicBlock *origBB = nullptr;

clang/lib/CodeGen/CGDecl.cpp

@@ -350,7 +350,7 @@ CodeGenFunction::AddInitializerToStaticVarDecl(const VarDecl &D,
                                   OldGV->getLinkage(), Init, "",
                                   /*InsertBefore*/ OldGV,
                                   OldGV->getThreadLocalMode(),
-                           CGM.getContext().getTargetAddressSpace(D.getType()));
+                                  OldGV->getType()->getPointerAddressSpace());
     GV->setVisibility(OldGV->getVisibility());
     GV->setDSOLocal(OldGV->isDSOLocal());
     GV->setComdat(OldGV->getComdat());

clang/lib/CodeGen/CGDeclCXX.cpp

@@ -27,10 +27,12 @@ using namespace CodeGen;
 
 static void EmitDeclInit(CodeGenFunction &CGF, const VarDecl &D,
                          ConstantAddress DeclPtr) {
-  assert(
-      (D.hasGlobalStorage() ||
-       (D.hasLocalStorage() && CGF.getContext().getLangOpts().OpenCLCPlusPlus)) &&
-      "VarDecl must have global or local (in the case of OpenCL) storage!");
+  assert((D.hasGlobalStorage() ||
+          (D.hasLocalStorage() &&
+           (CGF.getContext().getLangOpts().OpenCLCPlusPlus ||
+            CGF.getContext().getLangOpts().SYCLIsDevice))) &&
+         "VarDecl must have global or local (in the case of OpenCL and SYCL) "
+         "storage!");
   assert(!D.getType()->isReferenceType() &&
          "Should not call EmitDeclInit on a reference!");
 
@@ -161,13 +163,15 @@ void CodeGenFunction::EmitInvariantStart(llvm::Constant *Addr, CharUnits Size) {
   // Grab the llvm.invariant.start intrinsic.
   llvm::Intrinsic::ID InvStartID = llvm::Intrinsic::invariant_start;
   // Overloaded address space type.
-  llvm::Type *ObjectPtr[1] = {Int8PtrTy};
+  llvm::Type *ResTy = llvm::PointerType::getInt8PtrTy(
+      CGM.getLLVMContext(), Addr->getType()->getPointerAddressSpace());
+  llvm::Type *ObjectPtr[1] = {ResTy};
   llvm::Function *InvariantStart = CGM.getIntrinsic(InvStartID, ObjectPtr);
 
   // Emit a call with the size in bytes of the object.
   uint64_t Width = Size.getQuantity();
-  llvm::Value *Args[2] = { llvm::ConstantInt::getSigned(Int64Ty, Width),
-                           llvm::ConstantExpr::getBitCast(Addr, Int8PtrTy)};
+  llvm::Value *Args[2] = {llvm::ConstantInt::getSigned(Int64Ty, Width),
+                          llvm::ConstantExpr::getBitCast(Addr, ResTy)};
   Builder.CreateCall(InvariantStart, Args);
 }
 

clang/lib/CodeGen/CGExpr.cpp

@@ -1118,10 +1118,8 @@ Address CodeGenFunction::EmitPointerWithAlignment(const Expr *E,
                                       CodeGenFunction::CFITCK_UnrelatedCast,
                                       CE->getBeginLoc());
         }
-        return CE->getCastKind() != CK_AddressSpaceConversion
-                   ? Builder.CreateBitCast(Addr, ConvertType(E->getType()))
-                   : Builder.CreateAddrSpaceCast(Addr,
-                                                 ConvertType(E->getType()));
+        return Builder.CreatePointerBitCastOrAddrSpaceCast(
+            Addr, ConvertType(E->getType()));
       }
       break;
 
@@ -1813,6 +1811,16 @@ void CodeGenFunction::EmitStoreOfScalar(llvm::Value *Value, Address Addr,
     return;
   }
 
+  if (auto *PtrTy = dyn_cast<llvm::PointerType>(Value->getType())) {
+    auto *ExpectedPtrType =
+        cast<llvm::PointerType>(Addr.getType()->getElementType());
+    unsigned ValueAS = PtrTy->getAddressSpace();
+    unsigned ExpectedAS = ExpectedPtrType->getAddressSpace();
+    if (ValueAS != ExpectedAS) {
+      Value =
+          Builder.CreatePointerBitCastOrAddrSpaceCast(Value, ExpectedPtrType);
+    }
+  }
   llvm::StoreInst *Store = Builder.CreateStore(Value, Addr, Volatile);
   if (isNontemporal) {
     llvm::MDNode *Node =