diff --git a/clang/lib/Frontend/InitPreprocessor.cpp b/clang/lib/Frontend/InitPreprocessor.cpp
index cc09932d58b7..1d6c8ffb57b8 100755
--- a/clang/lib/Frontend/InitPreprocessor.cpp
+++ b/clang/lib/Frontend/InitPreprocessor.cpp
@@ -1117,8 +1117,22 @@ static void InitializePredefinedMacros(const TargetInfo &TI,
     Builder.defineMacro("SYCL_EXTERNAL", "__attribute__((sycl_device))");
     // Defines a macro that switches on SPIR intrinsics in SYCL runtime, used
     // by Xilinx FPGA devices for the moment
-    if (LangOpts.SYCLXOCCDevice)
+    if (LangOpts.SYCLXOCCDevice) {
       Builder.defineMacro("__SYCL_SPIR_DEVICE__");
+      switch (TI.getTriple().getSubArch()) {
+      case llvm::Triple::FPGASubArch_sw_emu:
+        Builder.defineMacro("__SYCL_XILINX_SW_EMU_MODE__");
+        break;
+      case llvm::Triple::FPGASubArch_hw_emu:
+        Builder.defineMacro("__SYCL_XILINX_HW_EMU_MODE__");
+        break;
+      case llvm::Triple::FPGASubArch_hw:
+        Builder.defineMacro("__SYCL_XILINX_HW_MODE__");
+        break;
+      default:
+        break;
+      }
+    }
 
     if (TI.getTriple().isNVPTX()) {
         Builder.defineMacro("__SYCL_NVPTX__", "1");
diff --git a/llvm/lib/SYCL/LowerSYCLMetaData.cpp b/llvm/lib/SYCL/LowerSYCLMetaData.cpp
index 6b8f728a20a3..71d29b407671 100644
--- a/llvm/lib/SYCL/LowerSYCLMetaData.cpp
+++ b/llvm/lib/SYCL/LowerSYCLMetaData.cpp
@@ -94,11 +94,10 @@ struct LSMDState {
         ResultMD.push_back(MDNode::get(
             Ctx, {MDString::get(Ctx, "llvm.loop.pipeline.enable"),
                    ConstantAsMetadata::get(
-                       ConstantInt::get(Type::getInt32Ty(Ctx), 1)),
+                       ConstantInt::get(Type::getInt32Ty(Ctx), -1)),
                    ConstantAsMetadata::get(
                        ConstantInt::getFalse(Type::getInt1Ty(Ctx))),
-                   ConstantAsMetadata::get(
-                       ConstantInt::get(Type::getInt8Ty(Ctx), -1))}));
+                }));
         MDNode *MDN = MDNode::getDistinct(Ctx, ResultMD);
         BB->getTerminator()->setMetadata(LLVMContext::MD_loop, MDN);
         BB->getTerminator()
@@ -162,6 +161,9 @@ struct LowerSYCLMetaData : public ModulePass {
   bool runOnModule(Module &M) override {
     return LSMDState(M).run();
   }
+  virtual StringRef getPassName() const override {
+    return "LowerSYCLMetaData";
+  }
 };
 }
 
diff --git a/llvm/lib/SYCL/PrepareSYCLOpt.cpp b/llvm/lib/SYCL/PrepareSYCLOpt.cpp
index 072629261592..518a113ec17b 100644
--- a/llvm/lib/SYCL/PrepareSYCLOpt.cpp
+++ b/llvm/lib/SYCL/PrepareSYCLOpt.cpp
@@ -15,6 +15,7 @@
 #include <regex>
 #include <string>
 
+#include "llvm/Analysis/ValueTracking.h"
 #include "llvm/IR/CallingConv.h"
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/Intrinsics.h"
@@ -48,6 +49,8 @@ struct PrepareSYCLOpt : public ModulePass {
         assert(F.use_empty());
         continue;
       }
+      if (F.isIntrinsic())
+        continue;
       F.setCallingConv(CallingConv::SPIR_FUNC);
       for (Value* V : F.users()) {
         if (auto* Call = dyn_cast<CallBase>(V))
@@ -72,14 +75,48 @@ struct PrepareSYCLOpt : public ModulePass {
       I->eraseFromParent();
   }
 
+  /// This will change array partition such that after the O3 pipeline it
+  /// matched very closely what v++ generates.
+  /// This will change the type of the alloca referenced by the array partition
+  /// into an array. and change the argument received by xlx_array_partition
+  /// into a pointer on an array.
+  void lowerArrayPartition(Module &M) {
+    Function* Func = Intrinsic::getDeclaration(&M, Intrinsic::sideeffect);
+    for (Use& U : Func->uses()) {
+      auto* Usr = dyn_cast<CallBase>(U.getUser());
+      if (!Usr)
+        continue;
+      if (!Usr->getOperandBundle("xlx_array_partition"))
+        continue;
+      Use& Ptr = U.getUser()->getOperandUse(0);
+      Value* Obj = getUnderlyingObject(Ptr);
+      if (!isa<AllocaInst>(Obj))
+        return;
+      auto* Alloca = cast<AllocaInst>(Obj);
+      auto *Replacement =
+          new AllocaInst(Ptr->getType()->getPointerElementType(), 0,
+                         ConstantInt::get(Type::getInt32Ty(M.getContext()), 1),
+                         Align(128), "");
+      Replacement->insertAfter(Alloca);
+      Instruction* Cast = BitCastInst::Create(
+          Instruction::BitCast, Replacement, Alloca->getType());
+      Cast->insertAfter(Replacement);
+      Alloca->replaceAllUsesWith(Cast);
+      Value* Zero = ConstantInt::get(Type::getInt32Ty(M.getContext()), 0);
+      Instruction* GEP = GetElementPtrInst::Create(nullptr, Replacement, {Zero});
+      GEP->insertAfter(Cast);
+      Ptr.set(GEP);
+    }
+  }
+
   bool runOnModule(Module &M) override {
     turnNonKernelsIntoPrivate(M);
     setCallingConventions(M);
+    lowerArrayPartition(M);
     removeAnnotationsIntrisic(M);
     return true;
   }
 };
-
 }
 
 namespace llvm {
diff --git a/sycl/doc/GettingStartedXilinxFPGA.md b/sycl/doc/GettingStartedXilinxFPGA.md
index 1b1a99a98e89..14395364c768 100644
--- a/sycl/doc/GettingStartedXilinxFPGA.md
+++ b/sycl/doc/GettingStartedXilinxFPGA.md
@@ -472,6 +472,16 @@ sudo rmmod xocl
 sudo modprobe xocl
 ```
 
+## Xilinx Macros
+
+``__SYCL_XILINX_SW_EMU_MODE__`` will be defined when compiling device code in sw_emu mode
+
+``__SYCL_XILINX_HW_EMU_MODE__`` will be defined when compiling device code in hw_emu mode
+
+``__SYCL_XILINX_HW_MODE__`` will be defined when compiling device code in hw mode
+
+when compiling host code none of them will be defined.
+
 ## Xilinx FPGA SYCL compiler architecture
 
 [Architecture of the Xilinx SYCL
diff --git a/sycl/include/CL/sycl/xilinx/fpga/opt_decorate_func.hpp b/sycl/include/CL/sycl/xilinx/fpga/opt_decorate_func.hpp
index bed968bdc311..c51c88e66292 100644
--- a/sycl/include/CL/sycl/xilinx/fpga/opt_decorate_func.hpp
+++ b/sycl/include/CL/sycl/xilinx/fpga/opt_decorate_func.hpp
@@ -57,7 +57,7 @@ void dataflow(T&& functor) {
 */
 template<typename T>
 __SYCL_DEVICE_ANNOTATE("xilinx_pipeline")
-ALWAYS_INLINE void pipeline(T&& functor) {
+__SYCL_ALWAYS_INLINE void pipeline(T&& functor) {
   /// the std::forward can make a difference when the operator() is l or r value
   /// specified.
   std::forward<T>(functor)();
diff --git a/sycl/include/CL/sycl/xilinx/fpga/partition_array.hpp b/sycl/include/CL/sycl/xilinx/fpga/partition_array.hpp
index d2bab320d628..983d0745c225 100644
--- a/sycl/include/CL/sycl/xilinx/fpga/partition_array.hpp
+++ b/sycl/include/CL/sycl/xilinx/fpga/partition_array.hpp
@@ -52,8 +52,11 @@ namespace partition {
   /// This fuction is currently empty but the LowerSYCLMetaData Pass will fill
   /// it with the required IR.
   template<typename Ptr>
+#if defined(__SYCL_XILINX_HW_EMU_MODE__) || defined(__SYCL_XILINX_HW_MODE__)
   __SYCL_DEVICE_ANNOTATE("xilinx_partition_array")
-  ALWAYS_INLINE inline void xilinx_partition_array(Ptr, int, int, int) {}
+#endif
+  __SYCL_ALWAYS_INLINE
+  inline void xilinx_partition_array(Ptr, int, int, int) {}
 
   /** Represent a cyclic partition.