diff --git a/clang/include/clang/Driver/Action.h b/clang/include/clang/Driver/Action.h
index 3965ad1df7814..8e4e1ab1df274 100644
--- a/clang/include/clang/Driver/Action.h
+++ b/clang/include/clang/Driver/Action.h
@@ -687,8 +687,13 @@ class SYCLPostLinkJobAction : public JobAction {
 
   bool getRTSetsSpecConstants() const { return RTSetsSpecConsts; }
 
+  void setDeadCodeRemoval(bool Val) { DeadCodeRemoval = Val; }
+
+  bool getDeadCodeRemoval() const { return DeadCodeRemoval; }
+
 private:
   bool RTSetsSpecConsts = true;
+  bool DeadCodeRemoval = false;
 };
 
 class PartialLinkJobAction : public JobAction {
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index 7d4e1487adeed..1e425214caead 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -3550,6 +3550,8 @@ def fsycl_esimd : Flag<["-"], "fsycl-explicit-simd">, Group<sycl_Group>, Flags<[
   HelpText<"Enable SYCL explicit SIMD extension">;
 def fno_sycl_esimd : Flag<["-"], "fno-sycl-explicit-simd">, Group<sycl_Group>,
   HelpText<"Disable SYCL explicit SIMD extension">, Flags<[NoArgumentUnused, CoreOption]>;
+def fno_sycl_devicelib : Flag<["-"], "fno-sycl-devicelib">, Group<sycl_Group>,
+  HelpText<"Disable SYCL device library link with kernel code">, Flags<[NoArgumentUnused, CoreOption]>;
 defm sycl_early_optimizations : OptOutFFlag<"sycl-early-optimizations", "Enable", "Disable", " standard optimization pipeline for SYCL device compiler", [CoreOption]>;
 def fsycl_dead_args_optimization : Flag<["-"], "fsycl-dead-args-optimization">,
   Group<sycl_Group>, Flags<[NoArgumentUnused, CoreOption]>, HelpText<"Enables "
diff --git a/clang/lib/Driver/Driver.cpp b/clang/lib/Driver/Driver.cpp
index 836219453b682..75625ef660590 100644
--- a/clang/lib/Driver/Driver.cpp
+++ b/clang/lib/Driver/Driver.cpp
@@ -5,7 +5,6 @@
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
-
 #include "clang/Driver/Driver.h"
 #include "InputInfo.h"
 #include "ToolChains/AIX.h"
@@ -2714,6 +2713,14 @@ static SmallVector<const char *, 16> getLinkerArgs(Compilation &C,
   return LibArgs;
 }
 
+static bool IsSYCLDeviceLibObj(std::string ObjFilePath) {
+  StringRef ObjFileName = llvm::sys::path::filename(ObjFilePath);
+  bool Ret = (ObjFileName.startswith("libsycl-") && ObjFileName.endswith(".o"))
+                 ? true
+                 : false;
+  return Ret;
+}
+
 // Goes through all of the arguments, including inputs expected for the
 // linker directly, to determine if we need to perform additional work for
 // static offload libraries.
@@ -3746,7 +3753,6 @@ class OffloadingActionBuilder final {
     }
 
     ActionBuilderReturnCode addDeviceDepences(Action *HostAction) override {
-
       // If this is an input action replicate it for each SYCL toolchain.
       if (auto *IA = dyn_cast<InputAction>(HostAction)) {
         SYCLDeviceActions.clear();
@@ -3789,7 +3795,10 @@ class OffloadingActionBuilder final {
           if (IA->getType() == types::TY_Object) {
             if (!isObjectFile(FileName))
               return ABRT_Inactive;
-            if (Args.hasArg(options::OPT_fintelfpga))
+            // For SYCL device libraries, don't need to add them to
+            // FPGAObjectInputs as there is no fpga dep files inside.
+            if (Args.hasArg(options::OPT_fintelfpga) &&
+                !IsSYCLDeviceLibObj(FileName))
               FPGAObjectInputs.push_back(IA);
           }
           // When creating FPGA device fat objects, all host objects are
@@ -3853,6 +3862,53 @@ class OffloadingActionBuilder final {
       SYCLDeviceActions.clear();
     }
 
+    void addSYCLDeviceLibs(const ToolChain *TC, ActionList &DeviceLinkObjects,
+                           bool isSpirvAOT, bool isMSVCEnv) {
+      enum SYCLDeviceLibType {
+        sycl_devicelib_wrapper,
+        sycl_devicelib_fallback
+      };
+      StringRef LibLoc, LibSysUtils;
+      if (isMSVCEnv) {
+        LibLoc = Args.MakeArgString(TC->getDriver().Dir + "/../bin");
+        LibSysUtils = "libsycl-msvc";
+      } else {
+        LibLoc = Args.MakeArgString(TC->getDriver().Dir + "/../lib");
+        LibSysUtils = "libsycl-glibc";
+      }
+      SmallVector<StringRef, 4> sycl_device_wrapper_libs = {
+          LibSysUtils, "libsycl-complex", "libsycl-complex-fp64",
+          "libsycl-cmath", "libsycl-cmath-fp64"};
+      // For AOT compilation, we need to link sycl_device_fallback_libs as
+      // default too.
+      SmallVector<StringRef, 4> sycl_device_fallback_libs = {
+          "libsycl-fallback-cassert", "libsycl-fallback-complex",
+          "libsycl-fallback-complex-fp64", "libsycl-fallback-cmath",
+          "libsycl-fallback-cmath-fp64"};
+      auto addInputs = [&](SYCLDeviceLibType t) {
+        auto sycl_libs = (t == sycl_devicelib_wrapper)
+                             ? sycl_device_wrapper_libs
+                             : sycl_device_fallback_libs;
+        for (const StringRef &Lib : sycl_libs) {
+          SmallString<128> LibName(LibLoc);
+          llvm::sys::path::append(LibName, Lib);
+          llvm::sys::path::replace_extension(LibName, ".o");
+          Arg *InputArg = MakeInputArg(Args, C.getDriver().getOpts(),
+                                       Args.MakeArgString(LibName));
+          auto *SYCLDeviceLibsInputAction =
+              C.MakeAction<InputAction>(*InputArg, types::TY_Object);
+          auto *SYCLDeviceLibsUnbundleAction =
+              C.MakeAction<OffloadUnbundlingJobAction>(
+                  SYCLDeviceLibsInputAction);
+          addDeviceDepences(SYCLDeviceLibsUnbundleAction);
+          DeviceLinkObjects.push_back(SYCLDeviceLibsUnbundleAction);
+        }
+      };
+      addInputs(sycl_devicelib_wrapper);
+      if (isSpirvAOT)
+        addInputs(sycl_devicelib_fallback);
+    }
+
     void appendLinkDependences(OffloadAction::DeviceDependences &DA) override {
       assert(ToolChains.size() == DeviceLinkerInputs.size() &&
              "Toolchains and linker inputs sizes do not match.");
@@ -3862,7 +3918,6 @@ class OffloadingActionBuilder final {
 
       unsigned I = 0;
       for (auto &LI : DeviceLinkerInputs) {
-
         auto TripleIt = llvm::find_if(SYCLTripleList, [&](auto &SYCLTriple) {
           return SYCLTriple == (*TC)->getTriple();
         });
@@ -3932,6 +3987,11 @@ class OffloadingActionBuilder final {
         }
         ActionList DeviceLibObjects;
         ActionList LinkObjects;
+        auto TT = SYCLTripleList[I];
+        auto isNVPTX = (*TC)->getTriple().isNVPTX();
+        bool isSpirvAOT = TT.getSubArch() == llvm::Triple::SPIRSubArch_fpga ||
+                          TT.getSubArch() == llvm::Triple::SPIRSubArch_gen ||
+                          TT.getSubArch() == llvm::Triple::SPIRSubArch_x86_64;
         for (const auto &Input : LI) {
           // FPGA aoco does not go through the link, everything else does.
           if (Input->getType() == types::TY_FPGA_AOCO)
@@ -3939,6 +3999,13 @@ class OffloadingActionBuilder final {
           else
             LinkObjects.push_back(Input);
         }
+
+        // For SYCL compilation, add SYCL device libraries as default.
+        if (!isNVPTX && !Args.hasArg(options::OPT_fno_sycl_devicelib)) {
+          addSYCLDeviceLibs(
+              *TC, LinkObjects, isSpirvAOT,
+              C.getDefaultToolChain().getTriple().isWindowsMSVCEnvironment());
+        }
         // The linkage actions subgraph leading to the offload wrapper.
         // [cond] Means incoming/outgoing dependence is created only when cond
         //        is true. A function of:
@@ -3993,7 +4060,6 @@ class OffloadingActionBuilder final {
         Action *DeviceLinkAction =
             C.MakeAction<LinkJobAction>(LinkObjects, types::TY_LLVM_BC);
         // setup some flags upfront
-        auto isNVPTX = (*TC)->getTriple().isNVPTX();
 
         if (isNVPTX && DeviceCodeSplit) {
           // TODO Temporary limitation, need to support code splitting for PTX
@@ -4005,10 +4071,6 @@ class OffloadingActionBuilder final {
           D.Diag(diag::err_drv_unsupported_opt_for_target)
               << OptName << (*TC)->getTriple().str();
         }
-        auto TT = SYCLTripleList[I];
-        bool isSpirvAOT = TT.getSubArch() == llvm::Triple::SPIRSubArch_fpga ||
-                          TT.getSubArch() == llvm::Triple::SPIRSubArch_gen ||
-                          TT.getSubArch() == llvm::Triple::SPIRSubArch_x86_64;
         // reflects whether current target is ahead-of-time and can't support
         // runtime setting of specialization constants
         bool isAOT = isNVPTX || isSpirvAOT;
@@ -4021,8 +4083,19 @@ class OffloadingActionBuilder final {
         types::ID PostLinkOutType = isNVPTX || !MultiFileActionDeps
                                         ? types::TY_LLVM_BC
                                         : types::TY_Tempfiletable;
-        auto *PostLinkAction = C.MakeAction<SYCLPostLinkJobAction>(
-            DeviceLinkAction, PostLinkOutType);
+        SYCLPostLinkJobAction *PostLinkDCRAction = nullptr;
+        SYCLPostLinkJobAction *PostLinkAction = nullptr;
+        if (isNVPTX || Args.hasArg(options::OPT_fno_sycl_devicelib)) {
+          PostLinkAction = C.MakeAction<SYCLPostLinkJobAction>(DeviceLinkAction,
+                                                               PostLinkOutType);
+        } else {
+          PostLinkDCRAction = C.MakeAction<SYCLPostLinkJobAction>(
+              DeviceLinkAction, types::TY_LLVM_BC);
+          PostLinkDCRAction->setDeadCodeRemoval(true);
+          PostLinkDCRAction->setRTSetsSpecConstants(false);
+          PostLinkAction = C.MakeAction<SYCLPostLinkJobAction>(
+              PostLinkDCRAction, PostLinkOutType);
+        }
         PostLinkAction->setRTSetsSpecConstants(!isAOT);
 
         if (isNVPTX) {
@@ -4569,7 +4642,7 @@ class OffloadingActionBuilder final {
       return nullptr;
 
     // Let builders add host linking actions.
-    Action* HA;
+    Action *HA = nullptr;
     for (DeviceActionBuilder *SB : SpecializedBuilders) {
       if (!SB->isValid())
         continue;
@@ -4588,7 +4661,6 @@ class OffloadingActionBuilder final {
     for (auto *SB : SpecializedBuilders) {
       if (!SB->isValid())
         continue;
-
       SB->appendLinkDependences(DDeps);
     }
 
diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp
index 064acb6de4b17..9afa87e20f767 100644
--- a/clang/lib/Driver/ToolChains/Clang.cpp
+++ b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -7837,8 +7837,15 @@ void SYCLPostLink::ConstructJob(Compilation &C, const JobAction &JA,
   // OPT_fsycl_device_code_split is not checked as it is an alias to
   // -fsycl-device-code-split=per_source
 
+  auto *SYCLPostLink = llvm::dyn_cast<SYCLPostLinkJobAction>(&JA);
+  bool InDeadCodeRemoval = false;
+  if (SYCLPostLink && SYCLPostLink->getDeadCodeRemoval()) {
+    addArgs(CmdArgs, TCArgs, {"--dead-code-removal"});
+    InDeadCodeRemoval = true;
+  }
+
   // Turn on Dead Parameter Elimination Optimization with early optimizations
-  if (!getToolChain().getTriple().isNVPTX() &&
+  if (!getToolChain().getTriple().isNVPTX() && !InDeadCodeRemoval &&
       TCArgs.hasFlag(options::OPT_fsycl_dead_args_optimization,
                      options::OPT_fno_sycl_dead_args_optimization, false))
     addArgs(CmdArgs, TCArgs, {"-emit-param-info"});
@@ -7854,7 +7861,6 @@ void SYCLPostLink::ConstructJob(Compilation &C, const JobAction &JA,
     addArgs(CmdArgs, TCArgs, {"-symbols"});
   }
   // specialization constants processing is mandatory
-  auto *SYCLPostLink = llvm::dyn_cast<SYCLPostLinkJobAction>(&JA);
   if (SYCLPostLink && SYCLPostLink->getRTSetsSpecConstants())
     addArgs(CmdArgs, TCArgs, {"-spec-const=rt"});
   else
diff --git a/clang/test/Driver/sycl-offload-intelfpga.cpp b/clang/test/Driver/sycl-offload-intelfpga.cpp
index de589affbc5e7..81ded9b647d17 100644
--- a/clang/test/Driver/sycl-offload-intelfpga.cpp
+++ b/clang/test/Driver/sycl-offload-intelfpga.cpp
@@ -21,11 +21,11 @@
 
 /// -fintelfpga -fsycl-link tests
 // RUN:  touch %t.o
-// RUN:  %clangxx -### -target x86_64-unknown-linux-gnu -fsycl -fintelfpga -fsycl-link %t.o -o libfoo.a 2>&1 \
+// RUN:  %clangxx -### -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fintelfpga -fsycl-link %t.o -o libfoo.a 2>&1 \
 // RUN:  | FileCheck -check-prefixes=CHK-FPGA-LINK,CHK-FPGA-EARLY %s
-// RUN:  %clangxx -### -target x86_64-unknown-linux-gnu -fsycl -fintelfpga -fsycl-link=early %t.o -o libfoo.a 2>&1 \
+// RUN:  %clangxx -### -target x86_64-unknown-linux-gnu -fsycl  -fno-sycl-devicelib -fintelfpga -fsycl-link=early %t.o -o libfoo.a 2>&1 \
 // RUN:  | FileCheck -check-prefixes=CHK-FPGA-LINK,CHK-FPGA-EARLY %s
-// RUN:  %clangxx -### -target x86_64-unknown-linux-gnu -fsycl -fintelfpga -fsycl-link=image %t.o -o libfoo.a 2>&1 \
+// RUN:  %clangxx -### -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fintelfpga  -fsycl-link=image %t.o -o libfoo.a 2>&1 \
 // RUN:  | FileCheck -check-prefixes=CHK-FPGA-LINK,CHK-FPGA-IMAGE %s
 // CHK-FPGA-LINK-NOT: clang-offload-bundler{{.*}} "-check-section"
 // CHK-FPGA-LINK: clang-offload-bundler{{.*}} "-type=o" "-targets=sycl-spir64_fpga-unknown-unknown-sycldevice" "-inputs=[[INPUT:.+\.o]]" "-outputs=[[OUTPUT1:.+\.o]]" "-unbundle"
@@ -50,9 +50,9 @@
 
 /// -fintelfpga -fsycl-link clang-cl specific
 // RUN:  touch %t.obj
-// RUN:  %clang_cl -### -fsycl -fintelfpga -fsycl-link %t.obj -Folibfoo.lib 2>&1 \
+// RUN:  %clang_cl -### -fsycl -fintelfpga -fno-sycl-devicelib -fsycl-link %t.obj -Folibfoo.lib 2>&1 \
 // RUN:  | FileCheck -check-prefixes=CHK-FPGA-LINK-WIN %s
-// RUN:  %clang_cl -### -fsycl -fintelfpga -fsycl-link %t.obj -o libfoo.lib 2>&1 \
+// RUN:  %clang_cl -### -fsycl -fintelfpga -fno-sycl-devicelib -fsycl-link %t.obj -o libfoo.lib 2>&1 \
 // RUN:  | FileCheck -check-prefixes=CHK-FPGA-LINK-WIN %s
 // CHK-FPGA-LINK-WIN: clang-offload-bundler{{.*}} "-type=o" "-targets=sycl-spir64_fpga-unknown-unknown-sycldevice{{.*}}" "-inputs=[[INPUT:.+\.obj]]" "-outputs=[[OUTPUT1:.+\.obj]]" "-unbundle"
 // CHK-FPGA-LINK-WIN-NOT: clang-offload-bundler{{.*}}
@@ -185,9 +185,9 @@
 
 /// -fintelfpga -fsycl-link from source
 // RUN: touch %t.cpp
-// RUN: %clangxx -### -target x86_64-unknown-linux-gnu -fsycl -fintelfpga -fsycl-link=early %t.cpp -ccc-print-phases 2>&1 \
+// RUN: %clangxx -### -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fintelfpga -fsycl-link=early %t.cpp -ccc-print-phases 2>&1 \
 // RUN:  | FileCheck -check-prefixes=CHK-FPGA-LINK-SRC,CHK-FPGA-LINK-SRC-DEFAULT %s
-// RUN: %clang_cl -### -target x86_64-unknown-linux-gnu -fsycl -fintelfpga -fsycl-link=early %t.cpp -ccc-print-phases 2>&1 \
+// RUN: %clang_cl -### -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fintelfpga -fsycl-link=early %t.cpp -ccc-print-phases 2>&1 \
 // RUN:  | FileCheck -check-prefixes=CHK-FPGA-LINK-SRC,CHK-FPGA-LINK-SRC-CL %s
 // CHK-FPGA-LINK-SRC: 0: input, "[[INPUT:.+\.cpp]]", c++, (host-sycl)
 // CHK-FPGA-LINK-SRC: 1: preprocessor, {0}, c++-cpp-output, (host-sycl)
@@ -275,9 +275,9 @@
 
 /// -fintelfpga dependency file use from object phases test
 // RUN: touch %t-1.o
-// RUN: %clangxx -fsycl -fintelfpga -ccc-print-phases -### %t-1.o 2>&1 \
+// RUN: %clangxx -fsycl -fno-sycl-devicelib -fintelfpga -ccc-print-phases -### %t-1.o 2>&1 \
 // RUN:  | FileCheck -check-prefix=CHK-FPGA-DEP-FILES-OBJ-PHASES -DINPUT=%t-1.o %s
-// RUN: %clang_cl -fsycl -fintelfpga -ccc-print-phases -### %t-1.o 2>&1 \
+// RUN: %clang_cl -fsycl -fno-sycl-devicelib -fintelfpga -ccc-print-phases -### %t-1.o 2>&1 \
 // RUN:  | FileCheck -check-prefix=CHK-FPGA-DEP-FILES-OBJ-PHASES -DINPUT=%t-1.o %s
 // CHK-FPGA-DEP-FILES-OBJ-PHASES: 0: input, "[[INPUT]]", object, (host-sycl)
 // CHK-FPGA-DEP-FILES-OBJ-PHASES: 1: clang-offload-unbundler, {0}, object, (host-sycl)
@@ -348,7 +348,7 @@
 // RUN:  llc -filetype=obj -o %t-aoco_cl.o %t-aoco_cl.bc
 // RUN:  llvm-ar crv %t_aoco.a %t.o %t2.o %t-aoco.o
 // RUN:  llvm-ar crv %t_aoco_cl.a %t.o %t2_cl.o %t-aoco_cl.o
-// RUN:  %clangxx -target x86_64-unknown-linux-gnu -fsycl -fintelfpga -foffload-static-lib=%t_aoco.a %s -### -ccc-print-phases 2>&1 \
+// RUN:  %clangxx -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fintelfpga -foffload-static-lib=%t_aoco.a %s -### -ccc-print-phases 2>&1 \
 // RUN:  | FileCheck -check-prefix=CHK-FPGA-AOCO-PHASES %s
 // CHK-FPGA-AOCO-PHASES: 0: input, "[[INPUTA:.+\.a]]", object, (host-sycl)
 // CHK-FPGA-AOCO-PHASES: 1: input, "[[INPUTCPP:.+\.cpp]]", c++, (host-sycl)
@@ -375,7 +375,7 @@
 // CHK-FPGA-AOCO-PHASES: 22: offload, "host-sycl (x86_64-unknown-linux-gnu)" {10}, "device-sycl (spir64_fpga-unknown-unknown-sycldevice)" {21}, image
 
 /// FPGA AOCO Windows phases check
-// RUN:  %clang_cl -fsycl -fintelfpga -foffload-static-lib=%t_aoco_cl.a %s -### -ccc-print-phases 2>&1 \
+// RUN:  %clang_cl -fsycl -fno-sycl-devicelib -fintelfpga -foffload-static-lib=%t_aoco_cl.a %s -### -ccc-print-phases 2>&1 \
 // RUN:  | FileCheck -check-prefixes=CHK-FPGA-AOCO-PHASES-WIN %s
 // CHK-FPGA-AOCO-PHASES-WIN: 0: input, "{{.*}}", object, (host-sycl)
 // CHK-FPGA-AOCO-PHASES-WIN: 1: input, "[[INPUTSRC:.+\.cpp]]", c++, (host-sycl)
@@ -401,13 +401,13 @@
 // CHK-FPGA-AOCO-PHASES-WIN: 21: offload, "host-sycl (x86_64-pc-windows-msvc)" {10}, "device-sycl (spir64_fpga-unknown-unknown-sycldevice)" {20}, image
 
 /// aoco test, checking tools
-// RUN:  %clangxx -target x86_64-unknown-linux-gnu -fsycl -fintelfpga -foffload-static-lib=%t_aoco.a -### %s 2>&1 \
+// RUN:  %clangxx -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fintelfpga -foffload-static-lib=%t_aoco.a -### %s 2>&1 \
 // RUN:  | FileCheck -check-prefixes=CHK-FPGA-AOCO,CHK-FPGA-AOCO-LIN %s
-// RUN:  %clangxx -target x86_64-unknown-linux-gnu -fsycl -fintelfpga %t_aoco.a -### %s 2>&1 \
+// RUN:  %clangxx -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fintelfpga %t_aoco.a -### %s 2>&1 \
 // RUN:  | FileCheck -check-prefixes=CHK-FPGA-AOCO,CHK-FPGA-AOCO-LIN %s
-// RUN:  %clang_cl -fsycl -fintelfpga -foffload-static-lib=%t_aoco_cl.a -### %s 2>&1 \
+// RUN:  %clang_cl -fsycl -fno-sycl-devicelib -fintelfpga -foffload-static-lib=%t_aoco_cl.a -### %s 2>&1 \
 // RUN:  | FileCheck -check-prefixes=CHK-FPGA-AOCO,CHK-FPGA-AOCO-WIN %s
-// RUN:  %clang_cl -fsycl -fintelfpga %t_aoco_cl.a -### %s 2>&1 \
+// RUN:  %clang_cl -fsycl -fno-sycl-devicelib -fintelfpga %t_aoco_cl.a -### %s 2>&1 \
 // RUN:  | FileCheck -check-prefixes=CHK-FPGA-AOCO,CHK-FPGA-AOCO-WIN %s
 // CHK-FPGA-AOCO-LIN: clang-offload-bundler{{.*}} "-type=ao" "-targets=sycl-fpga_aoco-intel-unknown-sycldevice" "-inputs=[[INPUTLIB:.+\.a]]" "-check-section"
 // CHK-FPGA-AOCO-LIN: clang{{.*}} "-emit-obj" {{.*}} "-o" "[[HOSTOBJ:.+\.o]]"
diff --git a/clang/test/Driver/sycl-offload-static-lib-2.cpp b/clang/test/Driver/sycl-offload-static-lib-2.cpp
index eff45a9d19f30..5a6d9c1955c90 100644
--- a/clang/test/Driver/sycl-offload-static-lib-2.cpp
+++ b/clang/test/Driver/sycl-offload-static-lib-2.cpp
@@ -99,9 +99,9 @@
 /// ###########################################################################
 
 /// test behaviors of static lib with no source/object
-// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -L/dummy/dir %t.a -### 2>&1 \
+// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -L/dummy/dir %t.a -### 2>&1 \
 // RUN:   | FileCheck %s -check-prefix=STATIC_LIB_NOSRC -DINPUTLIB=%t.a
-// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -L/dummy/dir %t.lo -### 2>&1 \
+// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -L/dummy/dir %t.lo -### 2>&1 \
 // RUN:   | FileCheck %s -check-prefix=STATIC_LIB_NOSRC -DINPUTLIB=%t.lo
 // STATIC_LIB_NOSRC: clang-offload-bundler{{.*}} "-type=ao" "-targets=host-x86_64-unknown-linux-gnu" "-inputs=[[INPUTLIB]]" "-check-section"
 // STATIC_LIB_NOSRC: ld{{.*}} "-r" "-o" "[[PARTIALOBJ:.+\.o]]" "{{.*}}crt1.o" {{.*}} "-L/dummy/dir" {{.*}} "[[INPUTLIB]]"
diff --git a/clang/test/Driver/sycl-offload-static-lib.cpp b/clang/test/Driver/sycl-offload-static-lib.cpp
index e32b244162e38..2e557405b4fb1 100644
--- a/clang/test/Driver/sycl-offload-static-lib.cpp
+++ b/clang/test/Driver/sycl-offload-static-lib.cpp
@@ -47,7 +47,7 @@
 
 /// test behaviors of -foffload-static-lib=<lib> from source
 // RUN: touch %t.a
-// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -foffload-static-lib=%t.a -ccc-print-phases %s 2>&1 \
+// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -foffload-static-lib=%t.a -ccc-print-phases %s 2>&1 \
 // RUN:   | FileCheck %s -check-prefix=FOFFLOAD_STATIC_LIB_SRC
 
 // FOFFLOAD_STATIC_LIB_SRC: 0: input, "[[INPUTA:.+\.a]]", object, (host-sycl)
@@ -122,9 +122,9 @@
 /// ###########################################################################
 
 /// test behaviors of -foffload-static-lib with no source/object
-// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -L/dummy/dir -foffload-static-lib=%t.a -### -ccc-print-phases 2>&1 \
+// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -L/dummy/dir -foffload-static-lib=%t.a -### -ccc-print-phases 2>&1 \
 // RUN:   | FileCheck %s -check-prefixes=FOFFLOAD_STATIC_LIB_NOSRC_PHASES,FOFFLOAD_STATIC_LIB_NOSRC_PHASES_1
-// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -L/dummy/dir -foffload-whole-static-lib=%t.a -### -ccc-print-phases 2>&1 \
+// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -L/dummy/dir -foffload-whole-static-lib=%t.a -### -ccc-print-phases 2>&1 \
 // RUN:   | FileCheck %s -check-prefixes=FOFFLOAD_STATIC_LIB_NOSRC_PHASES,FOFFLOAD_STATIC_LIB_NOSRC_PHASES_2
 // FOFFLOAD_STATIC_LIB_NOSRC_PHASES: 0: input, "[[INPUTA:.+\.a]]", object, (host-sycl)
 // FOFFLOAD_STATIC_LIB_NOSRC_PHASES: 1: linker, {0}, image, (host-sycl)
diff --git a/clang/test/Driver/sycl-offload-win.c b/clang/test/Driver/sycl-offload-win.c
index 4d2ab26532a27..35748739b5a67 100644
--- a/clang/test/Driver/sycl-offload-win.c
+++ b/clang/test/Driver/sycl-offload-win.c
@@ -57,9 +57,9 @@
 
 /// Test behaviors of -foffload-static-lib=<lib> from source.
 // RUN: touch %t.lib
-// RUN: %clang --target=x86_64-pc-windows-msvc -fsycl -foffload-static-lib=%t.lib -ccc-print-phases %s 2>&1 \
+// RUN: %clang --target=x86_64-pc-windows-msvc -fsycl -fno-sycl-devicelib -foffload-static-lib=%t.lib -ccc-print-phases %s 2>&1 \
 // RUN:   | FileCheck -DLIB=%t.lib %s -check-prefix=FOFFLOAD_STATIC_LIB_SRC
-// RUN: %clang_cl --target=x86_64-pc-windows-msvc -fsycl -foffload-static-lib=%t.lib -ccc-print-phases %s 2>&1 \
+// RUN: %clang_cl --target=x86_64-pc-windows-msvc -fsycl -fno-sycl-devicelib -foffload-static-lib=%t.lib -ccc-print-phases %s 2>&1 \
 // RUN:   | FileCheck -DLIB=%t.lib %s -check-prefix=FOFFLOAD_STATIC_LIB_SRC
 
 // FOFFLOAD_STATIC_LIB_SRC: 0: input, "[[INPUTLIB:.+\.lib]]", object, (host-sycl)
diff --git a/clang/test/Driver/sycl-offload-with-split.c b/clang/test/Driver/sycl-offload-with-split.c
index 4d01f4d5fb800..c2a866200510b 100644
--- a/clang/test/Driver/sycl-offload-with-split.c
+++ b/clang/test/Driver/sycl-offload-with-split.c
@@ -12,17 +12,17 @@
 /// preprocessor and another one joining the device linking outputs to the host
 /// action.  The same graph should be generated when no -fsycl-targets is used
 /// The same phase graph will be used with -fsycl-use-bitcode
-// RUN:   %clang -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -fsycl-device-code-split -fsycl-targets=spir64-unknown-unknown-sycldevice %s 2>&1 \
+// RUN:   %clang -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-device-code-split -fsycl-targets=spir64-unknown-unknown-sycldevice %s 2>&1 \
 // RUN:   | FileCheck -check-prefixes=CHK-PHASES,CHK-PHASES-DEFAULT-MODE %s
-// RUN:   %clang_cl -ccc-print-phases -fsycl -fsycl-device-code-split=per_source -fsycl-targets=spir64-unknown-unknown-sycldevice %s 2>&1 \
+// RUN:   %clang_cl -ccc-print-phases -fsycl -fno-sycl-devicelib -fsycl-device-code-split=per_source -fsycl-targets=spir64-unknown-unknown-sycldevice %s 2>&1 \
 // RUN:   | FileCheck -check-prefixes=CHK-PHASES,CHK-PHASES-CL-MODE %s
-// RUN:   %clang -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -fsycl-device-code-split=per_source -fno-sycl-use-bitcode %s 2>&1 \
+// RUN:   %clang -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-device-code-split=per_source -fno-sycl-use-bitcode %s 2>&1 \
 // RUN:   | FileCheck -check-prefixes=CHK-PHASES,CHK-PHASES-DEFAULT-MODE %s
-// RUN:   %clang_cl -ccc-print-phases -fsycl -fsycl-device-code-split=per_source -fno-sycl-use-bitcode %s 2>&1 \
+// RUN:   %clang_cl -ccc-print-phases -fsycl -fno-sycl-devicelib -fsycl-device-code-split=per_source -fno-sycl-use-bitcode %s 2>&1 \
 // RUN:   | FileCheck -check-prefixes=CHK-PHASES,CHK-PHASES-CL-MODE %s
-// RUN:   %clang -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -fsycl-device-code-split=per_source -fsycl-use-bitcode %s 2>&1 \
+// RUN:   %clang -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-device-code-split=per_source -fsycl-use-bitcode %s 2>&1 \
 // RUN:   | FileCheck -check-prefixes=CHK-PHASES,CHK-PHASES-DEFAULT-MODE %s
-// RUN:   %clang_cl -ccc-print-phases -fsycl -fsycl-device-code-split=per_source -fsycl-use-bitcode %s 2>&1 \
+// RUN:   %clang_cl -ccc-print-phases -fsycl -fno-sycl-devicelib -fsycl-device-code-split=per_source -fsycl-use-bitcode %s 2>&1 \
 // RUN:   | FileCheck -check-prefixes=CHK-PHASES,CHK-PHASES-CL-MODE %s
 // CHK-PHASES: 0: input, "[[INPUT:.+\.c]]", c, (host-sycl)
 // CHK-PHASES: 1: preprocessor, {0}, cpp-output, (host-sycl)
@@ -49,7 +49,7 @@
 
 /// Check the phases also add a library to make sure it is treated as input by
 /// the device.
-// RUN:   %clang -ccc-print-phases -target x86_64-unknown-linux-gnu -lsomelib -fsycl -fsycl-device-code-split -fsycl-targets=spir64-unknown-unknown-sycldevice %s 2>&1 \
+// RUN:   %clang -ccc-print-phases -target x86_64-unknown-linux-gnu -lsomelib -fsycl -fno-sycl-devicelib -fsycl-device-code-split -fsycl-targets=spir64-unknown-unknown-sycldevice %s 2>&1 \
 // RUN:   | FileCheck -check-prefix=CHK-PHASES-LIB %s
 // CHK-PHASES-LIB: 0: input, "somelib", object, (host-sycl)
 // CHK-PHASES-LIB: 1: input, "[[INPUT:.+\.c]]", c, (host-sycl)
@@ -75,7 +75,7 @@
 
 /// Check the phases when using and multiple source files
 // RUN:   echo " " > %t.c
-// RUN:   %clang -ccc-print-phases -lsomelib -target x86_64-unknown-linux-gnu -fsycl -fsycl-device-code-split -fsycl-targets=spir64-unknown-unknown-sycldevice %s %t.c 2>&1 \
+// RUN:   %clang -ccc-print-phases -lsomelib -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-device-code-split -fsycl-targets=spir64-unknown-unknown-sycldevice %s %t.c 2>&1 \
 // RUN:   | FileCheck -check-prefix=CHK-PHASES-FILES %s
 
 // CHK-PHASES-FILES: 0: input, "somelib", object, (host-sycl)
@@ -112,11 +112,11 @@
 
 /// Check separate compilation with offloading - unbundling actions
 // RUN:   touch %t.o
-// RUN:   %clang -### -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -fsycl-device-code-split -o %t.out -lsomelib -fsycl-targets=spir64-unknown-unknown-sycldevice %t.o 2>&1 \
+// RUN:   %clang -### -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-device-code-split -o %t.out -lsomelib -fsycl-targets=spir64-unknown-unknown-sycldevice %t.o 2>&1 \
 // RUN:   | FileCheck -DINPUT=%t.o -check-prefix=CHK-UBACTIONS %s
 // RUN:   mkdir -p %t_dir
 // RUN:   touch %t_dir/dummy
-// RUN:   %clang -### -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -fsycl-device-code-split -o %t.out -lsomelib -fsycl-targets=spir64-unknown-unknown-sycldevice %t_dir/dummy 2>&1 \
+// RUN:   %clang -### -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-device-code-split -o %t.out -lsomelib -fsycl-targets=spir64-unknown-unknown-sycldevice %t_dir/dummy 2>&1 \
 // RUN:   | FileCheck -DINPUT=%t_dir/dummy -check-prefix=CHK-UBACTIONS %s
 // CHK-UBACTIONS: 0: input, "somelib", object, (host-sycl)
 // CHK-UBACTIONS: 1: input, "[[INPUT]]", object, (host-sycl)
@@ -134,7 +134,7 @@
 
 /// Check separate compilation with offloading - unbundling with source
 // RUN:   touch %t.o
-// RUN:   %clang -### -ccc-print-phases -target x86_64-unknown-linux-gnu -lsomelib -fsycl -fsycl-device-code-split %t.o -fsycl-targets=spir64-unknown-unknown-sycldevice %s 2>&1 \
+// RUN:   %clang -### -ccc-print-phases -target x86_64-unknown-linux-gnu -lsomelib -fsycl -fno-sycl-devicelib -fsycl-device-code-split %t.o -fsycl-targets=spir64-unknown-unknown-sycldevice %s 2>&1 \
 // RUN:   | FileCheck -check-prefix=CHK-UBUACTIONS %s
 // CHK-UBUACTIONS: 0: input, "somelib", object, (host-sycl)
 // CHK-UBUACTIONS: 1: input, "[[INPUT1:.+\.o]]", object, (host-sycl)
@@ -161,11 +161,11 @@
 /// ###########################################################################
 
 /// Ahead of Time compilation for fpga, gen, cpu
-// RUN:   %clang -target x86_64-unknown-linux-gnu -ccc-print-phases -fsycl -fsycl-device-code-split -fsycl-targets=spir64_fpga-unknown-unknown-sycldevice %s 2>&1 \
+// RUN:   %clang -target x86_64-unknown-linux-gnu -ccc-print-phases -fsycl -fno-sycl-devicelib -fsycl-device-code-split -fsycl-targets=spir64_fpga-unknown-unknown-sycldevice %s 2>&1 \
 // RUN:    | FileCheck %s -check-prefixes=CHK-PHASES-AOT,CHK-PHASES-FPGA
-// RUN:   %clang -target x86_64-unknown-linux-gnu -ccc-print-phases -fsycl -fsycl-device-code-split -fsycl-targets=spir64_gen-unknown-unknown-sycldevice %s 2>&1 \
+// RUN:   %clang -target x86_64-unknown-linux-gnu -ccc-print-phases -fsycl -fno-sycl-devicelib -fsycl-device-code-split -fsycl-targets=spir64_gen-unknown-unknown-sycldevice %s 2>&1 \
 // RUN:    | FileCheck %s -check-prefixes=CHK-PHASES-AOT,CHK-PHASES-GEN
-// RUN:   %clang -target x86_64-unknown-linux-gnu -ccc-print-phases -fsycl -fsycl-device-code-split -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %s 2>&1 \
+// RUN:   %clang -target x86_64-unknown-linux-gnu -ccc-print-phases -fsycl -fno-sycl-devicelib -fsycl-device-code-split -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %s 2>&1 \
 // RUN:    | FileCheck %s -check-prefixes=CHK-PHASES-AOT,CHK-PHASES-CPU
 // CHK-PHASES-AOT: 0: input, "[[INPUT:.+\.c]]", c, (host-sycl)
 // CHK-PHASES-AOT: 1: preprocessor, {0}, cpp-output, (host-sycl)
@@ -196,13 +196,13 @@
 /// ###########################################################################
 
 /// Ahead of Time compilation for fpga, gen, cpu - tool invocation
-// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fsycl-device-code-split -fsycl-targets=spir64_fpga-unknown-unknown-sycldevice %s -### 2>&1 \
+// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-device-code-split -fsycl-targets=spir64_fpga-unknown-unknown-sycldevice %s -### 2>&1 \
 // RUN:  | FileCheck %s -check-prefixes=CHK-TOOLS-AOT,CHK-TOOLS-FPGA
-// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fsycl-device-code-split -fintelfpga %s -### 2>&1 \
+// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-device-code-split -fintelfpga %s -### 2>&1 \
 // RUN:  | FileCheck %s -check-prefixes=CHK-TOOLS-AOT,CHK-TOOLS-FPGA
-// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fsycl-device-code-split -fsycl-targets=spir64_gen-unknown-unknown-sycldevice %s -### 2>&1 \
+// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-device-code-split -fsycl-targets=spir64_gen-unknown-unknown-sycldevice %s -### 2>&1 \
 // RUN:  | FileCheck %s -check-prefixes=CHK-TOOLS-AOT,CHK-TOOLS-GEN
-// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fsycl-device-code-split -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %s -### 2>&1 \
+// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-device-code-split -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %s -### 2>&1 \
 // RUN:  | FileCheck %s -check-prefixes=CHK-TOOLS-AOT,CHK-TOOLS-CPU
 // CHK-TOOLS-AOT: clang{{.*}} "-fsycl-is-device" {{.*}} "-o" "[[OUTPUT1:.+\.bc]]"
 // CHK-TOOLS-AOT: llvm-link{{.*}} "[[OUTPUT1]]" "-o" "[[OUTPUT2:.+\.bc]]"
@@ -226,7 +226,7 @@
 /// ###########################################################################
 
 /// offload with multiple targets, including AOT
-// RUN:  %clang -target x86_64-unknown-linux-gnu -fsycl -fsycl-device-code-split -fsycl-targets=spir64-unknown-unknown-sycldevice,spir64_fpga-unknown-unknown-sycldevice,spir64_gen-unknown-unknown-sycldevice -###  -ccc-print-phases %s 2>&1 \
+// RUN:  %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-device-code-split -fsycl-targets=spir64-unknown-unknown-sycldevice,spir64_fpga-unknown-unknown-sycldevice,spir64_gen-unknown-unknown-sycldevice -###  -ccc-print-phases %s 2>&1 \
 // RUN:   | FileCheck -check-prefix=CHK-PHASE-MULTI-TARG %s
 // CHK-PHASE-MULTI-TARG: 0: input, "[[INPUT:.+\.c]]", c, (host-sycl)
 // CHK-PHASE-MULTI-TARG: 1: preprocessor, {0}, cpp-output, (host-sycl)
diff --git a/clang/test/Driver/sycl-offload.c b/clang/test/Driver/sycl-offload.c
index ab6a7c9d90e1b..b249c3bb239a7 100644
--- a/clang/test/Driver/sycl-offload.c
+++ b/clang/test/Driver/sycl-offload.c
@@ -161,17 +161,17 @@
 /// preprocessor and another one joining the device linking outputs to the host
 /// action.  The same graph should be generated when no -fsycl-targets is used
 /// The same phase graph will be used with -fsycl-use-bitcode
-// RUN:   %clang -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=spir64-unknown-unknown-sycldevice %s 2>&1 \
+// RUN:   %clang -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=spir64-unknown-unknown-sycldevice -fno-sycl-devicelib %s 2>&1 \
 // RUN:   | FileCheck -check-prefixes=CHK-PHASES,CHK-PHASES-DEFAULT-MODE %s
-// RUN:   %clang_cl -ccc-print-phases -fsycl -fsycl-targets=spir64-unknown-unknown-sycldevice %s 2>&1 \
+// RUN:   %clang_cl -ccc-print-phases -fsycl -fsycl-targets=spir64-unknown-unknown-sycldevice -fno-sycl-devicelib %s 2>&1 \
 // RUN:   | FileCheck -check-prefixes=CHK-PHASES,CHK-PHASES-CL-MODE %s
-// RUN:   %clang -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-use-bitcode %s 2>&1 \
+// RUN:   %clang -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-use-bitcode -fno-sycl-devicelib %s 2>&1 \
 // RUN:   | FileCheck -check-prefixes=CHK-PHASES,CHK-PHASES-DEFAULT-MODE %s
-// RUN:   %clang_cl -ccc-print-phases -fsycl -fno-sycl-use-bitcode %s 2>&1 \
+// RUN:   %clang_cl -ccc-print-phases -fsycl -fno-sycl-use-bitcode -fno-sycl-devicelib %s 2>&1 \
 // RUN:   | FileCheck -check-prefixes=CHK-PHASES,CHK-PHASES-CL-MODE %s
-// RUN:   %clang -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -fsycl-use-bitcode %s 2>&1 \
+// RUN:   %clang -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -fsycl-use-bitcode -fno-sycl-devicelib %s 2>&1 \
 // RUN:   | FileCheck -check-prefixes=CHK-PHASES,CHK-PHASES-DEFAULT-MODE %s
-// RUN:   %clang_cl -ccc-print-phases -fsycl -fsycl-use-bitcode %s 2>&1 \
+// RUN:   %clang_cl -ccc-print-phases -fsycl -fsycl-use-bitcode -fno-sycl-devicelib %s 2>&1 \
 // RUN:   | FileCheck -check-prefixes=CHK-PHASES,CHK-PHASES-CL-MODE %s
 // CHK-PHASES: 0: input, "[[INPUT:.+\.c]]", c, (host-sycl)
 // CHK-PHASES: 1: preprocessor, {0}, cpp-output, (host-sycl)
@@ -208,7 +208,7 @@
 
 /// Check the phases also add a library to make sure it is treated as input by
 /// the device.
-// RUN:   %clang -ccc-print-phases -target x86_64-unknown-linux-gnu -lsomelib -fsycl -fsycl-targets=spir64-unknown-unknown-sycldevice %s 2>&1 \
+// RUN:   %clang -ccc-print-phases -target x86_64-unknown-linux-gnu -lsomelib -fsycl -fsycl-targets=spir64-unknown-unknown-sycldevice -fno-sycl-devicelib %s 2>&1 \
 // RUN:   | FileCheck -check-prefix=CHK-PHASES-LIB %s
 // CHK-PHASES-LIB: 0: input, "somelib", object, (host-sycl)
 // CHK-PHASES-LIB: 1: input, "[[INPUT:.+\.c]]", c, (host-sycl)
@@ -241,7 +241,7 @@
 
 /// Check the phases when using and multiple source files
 // RUN:   echo " " > %t.c
-// RUN:   %clang -ccc-print-phases -lsomelib -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=spir64-unknown-unknown-sycldevice %s %t.c 2>&1 \
+// RUN:   %clang -ccc-print-phases -lsomelib -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=spir64-unknown-unknown-sycldevice -fno-sycl-devicelib %s %t.c 2>&1 \
 // RUN:   | FileCheck -check-prefix=CHK-PHASES-FILES %s
 
 // CHK-PHASES-FILES: 0: input, "somelib", object, (host-sycl)
@@ -296,11 +296,11 @@
 
 /// Check separate compilation with offloading - unbundling actions
 // RUN:   touch %t.o
-// RUN:   %clang -### -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -o %t.out -lsomelib -fsycl-targets=spir64-unknown-unknown-sycldevice %t.o 2>&1 \
+// RUN:   %clang -### -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -o %t.out -lsomelib -fsycl-targets=spir64-unknown-unknown-sycldevice %t.o 2>&1 \
 // RUN:   | FileCheck -DINPUT=%t.o -check-prefix=CHK-UBACTIONS %s
 // RUN:   mkdir -p %t_dir
 // RUN:   touch %t_dir/dummy
-// RUN:   %clang -### -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -o %t.out -lsomelib -fsycl-targets=spir64-unknown-unknown-sycldevice %t_dir/dummy 2>&1 \
+// RUN:   %clang -### -ccc-print-phases -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -o %t.out -lsomelib -fsycl-targets=spir64-unknown-unknown-sycldevice %t_dir/dummy 2>&1 \
 // RUN:   | FileCheck -DINPUT=%t_dir/dummy -check-prefix=CHK-UBACTIONS %s
 // CHK-UBACTIONS: 0: input, "somelib", object, (host-sycl)
 // CHK-UBACTIONS: 1: input, "[[INPUT]]", object, (host-sycl)
@@ -318,7 +318,7 @@
 
 /// Check separate compilation with offloading - unbundling with source
 // RUN:   touch %t.o
-// RUN:   %clang -### -ccc-print-phases -target x86_64-unknown-linux-gnu -lsomelib -fsycl %t.o -fsycl-targets=spir64-unknown-unknown-sycldevice %s 2>&1 \
+// RUN:   %clang -### -ccc-print-phases -target x86_64-unknown-linux-gnu -lsomelib -fsycl -fno-sycl-devicelib %t.o -fsycl-targets=spir64-unknown-unknown-sycldevice %s 2>&1 \
 // RUN:   | FileCheck -check-prefix=CHK-UBUACTIONS %s
 // CHK-UBUACTIONS: 0: input, "somelib", object, (host-sycl)
 // CHK-UBUACTIONS: 1: input, "[[INPUT1:.+\.o]]", object, (host-sycl)
@@ -508,7 +508,7 @@
 
 /// Check regular offload with an additional AOT binary passed through -fsycl-add-targets (same triple)
 
-// RUN:  %clang -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=spir64-unknown-unknown-sycldevice -fsycl-add-targets=spir64-unknown-unknown-sycldevice:dummy.spv -ccc-print-phases %s 2>&1 \
+// RUN:  %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-targets=spir64-unknown-unknown-sycldevice -fsycl-add-targets=spir64-unknown-unknown-sycldevice:dummy.spv -ccc-print-phases %s 2>&1 \
 // RUN:   | FileCheck -check-prefix=CHK-ADD-TARGETS-REG %s
 // CHK-ADD-TARGETS-REG: 0: input, "[[INPUT:.+\.c]]", c, (host-sycl)
 // CHK-ADD-TARGETS-REG: 1: preprocessor, {0}, cpp-output, (host-sycl)
@@ -534,7 +534,7 @@
 /// ###########################################################################
 
 /// Check regular offload with multiple additional AOT binaries passed through -fsycl-add-targets
-// RUN:  %clang -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=spir64-unknown-unknown-sycldevice -fsycl-add-targets=spir64_fpga-unknown-unknown-sycldevice:dummy.aocx,spir64_gen-unknown-unknown-sycldevice:dummy_Gen9core.bin,spir64_x86_64-unknown-unknown-sycldevice:dummy.ir -ccc-print-phases %s 2>&1 \
+// RUN:  %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-targets=spir64-unknown-unknown-sycldevice -fsycl-add-targets=spir64_fpga-unknown-unknown-sycldevice:dummy.aocx,spir64_gen-unknown-unknown-sycldevice:dummy_Gen9core.bin,spir64_x86_64-unknown-unknown-sycldevice:dummy.ir -ccc-print-phases %s 2>&1 \
 // RUN:   | FileCheck -check-prefix=CHK-ADD-TARGETS-REG-MUL %s
 // CHK-ADD-TARGETS-REG-MUL: 0: input, "[[INPUT:.+\.c]]", c, (host-sycl)
 // CHK-ADD-TARGETS-REG-MUL: 1: preprocessor, {0}, cpp-output, (host-sycl)
@@ -609,11 +609,11 @@
 /// ###########################################################################
 
 /// Ahead of Time compilation for fpga, gen, cpu
-// RUN:   %clang -target x86_64-unknown-linux-gnu -ccc-print-phases -fsycl -fsycl-targets=spir64_fpga-unknown-unknown-sycldevice %s 2>&1 \
+// RUN:   %clang -target x86_64-unknown-linux-gnu -ccc-print-phases -fsycl -fno-sycl-devicelib -fsycl-targets=spir64_fpga-unknown-unknown-sycldevice %s 2>&1 \
 // RUN:    | FileCheck %s -check-prefixes=CHK-PHASES-AOT,CHK-PHASES-FPGA
-// RUN:   %clang -target x86_64-unknown-linux-gnu -ccc-print-phases -fsycl -fsycl-targets=spir64_gen-unknown-unknown-sycldevice %s 2>&1 \
+// RUN:   %clang -target x86_64-unknown-linux-gnu -ccc-print-phases -fsycl -fno-sycl-devicelib -fsycl-targets=spir64_gen-unknown-unknown-sycldevice %s 2>&1 \
 // RUN:    | FileCheck %s -check-prefixes=CHK-PHASES-AOT,CHK-PHASES-GEN
-// RUN:   %clang -target x86_64-unknown-linux-gnu -ccc-print-phases -fsycl -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %s 2>&1 \
+// RUN:   %clang -target x86_64-unknown-linux-gnu -ccc-print-phases -fsycl -fno-sycl-devicelib -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %s 2>&1 \
 // RUN:    | FileCheck %s -check-prefixes=CHK-PHASES-AOT,CHK-PHASES-CPU
 // CHK-PHASES-AOT: 0: input, "[[INPUT:.+\.c]]", c, (host-sycl)
 // CHK-PHASES-AOT: 1: preprocessor, {0}, cpp-output, (host-sycl)
@@ -642,29 +642,29 @@
 /// ###########################################################################
 
 /// Ahead of Time compilation for fpga, gen, cpu - tool invocation
-// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=spir64_fpga-unknown-unknown-sycldevice %s -### 2>&1 \
+// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-targets=spir64_fpga-unknown-unknown-sycldevice %s -### 2>&1 \
 // RUN:  | FileCheck %s -check-prefixes=CHK-TOOLS-AOT,CHK-TOOLS-FPGA,CHK-TOOLS-FPGA-USM-DISABLE
-// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fintelfpga %s -### 2>&1 \
+// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fintelfpga %s -### 2>&1 \
 // RUN:  | FileCheck %s -check-prefixes=CHK-TOOLS-AOT,CHK-TOOLS-FPGA,CHK-TOOLS-FPGA-USM-DISABLE
-// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=spir64_fpga-unknown-unknown-sycldevice -Xshardware %s -### 2>&1 \
+// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-targets=spir64_fpga-unknown-unknown-sycldevice -Xshardware %s -### 2>&1 \
 // RUN:  | FileCheck %s -check-prefixes=CHK-TOOLS-AOT,CHK-TOOLS-FPGA,CHK-TOOLS-FPGA-USM-ENABLE
-// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fintelfpga -Xshardware %s -### 2>&1 \
+// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fintelfpga -Xshardware %s -### 2>&1 \
 // RUN:  | FileCheck %s -check-prefixes=CHK-TOOLS-AOT,CHK-TOOLS-FPGA,CHK-TOOLS-FPGA-USM-ENABLE
-// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=spir64_fpga-unknown-unknown-sycldevice -Xssimulation %s -### 2>&1 \
+// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-targets=spir64_fpga-unknown-unknown-sycldevice -Xssimulation %s -### 2>&1 \
 // RUN:  | FileCheck %s -check-prefixes=CHK-TOOLS-AOT,CHK-TOOLS-FPGA,CHK-TOOLS-FPGA-USM-ENABLE
-// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fintelfpga -Xssimulation %s -### 2>&1 \
+// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fintelfpga -Xssimulation %s -### 2>&1 \
 // RUN:  | FileCheck %s -check-prefixes=CHK-TOOLS-AOT,CHK-TOOLS-FPGA,CHK-TOOLS-FPGA-USM-ENABLE
-// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=spir64_fpga-unknown-unknown-sycldevice -Xsemulator %s -### 2>&1 \
+// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-targets=spir64_fpga-unknown-unknown-sycldevice -Xsemulator %s -### 2>&1 \
 // RUN:  | FileCheck %s -check-prefixes=CHK-TOOLS-AOT,CHK-TOOLS-FPGA,CHK-TOOLS-FPGA-USM-DISABLE
-// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fintelfpga -Xsemulator %s -### 2>&1 \
+// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fintelfpga -Xsemulator %s -### 2>&1 \
 // RUN:  | FileCheck %s -check-prefixes=CHK-TOOLS-AOT,CHK-TOOLS-FPGA,CHK-TOOLS-FPGA-USM-DISABLE
-// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=spir64_gen-unknown-unknown-sycldevice %s -### 2>&1 \
+// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-targets=spir64_gen-unknown-unknown-sycldevice %s -### 2>&1 \
 // RUN:  | FileCheck %s -check-prefixes=CHK-TOOLS-AOT,CHK-TOOLS-GEN
-// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %s -### 2>&1 \
+// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %s -### 2>&1 \
 // RUN:  | FileCheck %s -check-prefixes=CHK-TOOLS-AOT,CHK-TOOLS-CPU
-// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=spir64_gen-unknown-unknown-sycldevice %s -### 2>&1 \
+// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-targets=spir64_gen-unknown-unknown-sycldevice %s -### 2>&1 \
 // RUN:  | FileCheck %s -check-prefixes=CHK-TOOLS-AOT,CHK-TOOLS-GEN
-// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %s -### 2>&1 \
+// RUN: %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %s -### 2>&1 \
 // RUN:  | FileCheck %s -check-prefixes=CHK-TOOLS-AOT,CHK-TOOLS-CPU
 // CHK-TOOLS-AOT: clang{{.*}} "-fsycl-is-device" {{.*}} "-o" "[[OUTPUT1:.+\.bc]]"
 // CHK-TOOLS-AOT: llvm-link{{.*}} "[[OUTPUT1]]" "-o" "[[OUTPUT2:.+\.bc]]"
@@ -774,7 +774,7 @@
 /// ###########################################################################
 
 /// offload with multiple targets, including AOT
-// RUN:  %clang -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=spir64-unknown-unknown-sycldevice,spir64_fpga-unknown-unknown-sycldevice,spir64_gen-unknown-unknown-sycldevice -###  -ccc-print-phases %s 2>&1 \
+// RUN:  %clang -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-devicelib -fsycl-targets=spir64-unknown-unknown-sycldevice,spir64_fpga-unknown-unknown-sycldevice,spir64_gen-unknown-unknown-sycldevice -###  -ccc-print-phases %s 2>&1 \
 // RUN:   | FileCheck -check-prefix=CHK-PHASE-MULTI-TARG %s
 // CHK-PHASE-MULTI-TARG: 0: input, "[[INPUT:.+\.c]]", c, (host-sycl)
 // CHK-PHASE-MULTI-TARG: 1: preprocessor, {0}, cpp-output, (host-sycl)
@@ -813,9 +813,9 @@
 
 /// ###########################################################################
 /// Verify that -save-temps does not crash
-// RUN: %clang -fsycl -target x86_64-unknown-linux-gnu -save-temps %s -### 2>&1
-// RUN: %clang -fsycl -fsycl-targets=spir64-unknown-unknown-sycldevice -target x86_64-unknown-linux-gnu -save-temps %s -### 2>&1
-// RUN: %clangxx -fsycl -fsycl-targets=spir64-unknown-unknown-sycldevice -target x86_64-unknown-linux-gnu -save-temps %s -### 2>&1 \
+// RUN: %clang -fsycl -fno-sycl-devicelib -target x86_64-unknown-linux-gnu -save-temps %s -### 2>&1
+// RUN: %clang -fsycl -fno-sycl-devicelib -fsycl-targets=spir64-unknown-unknown-sycldevice -target x86_64-unknown-linux-gnu -save-temps %s -### 2>&1
+// RUN: %clangxx -fsycl -fno-sycl-devicelib -fsycl-targets=spir64-unknown-unknown-sycldevice -target x86_64-unknown-linux-gnu -save-temps %s -### 2>&1 \
 // RUN: | FileCheck %s --check-prefixes=CHK-FSYCL-SAVE-TEMPS,CHK-FSYCL-SAVE-TEMPS-CONFL
 // CHK-FSYCL-SAVE-TEMPS: clang{{.*}} "-fsycl-is-device"{{.*}} "-o" "[[DEVICE_BASE_NAME:[a-z0-9-]+]].ii"
 // CHK-FSYCL-SAVE-TEMPS: clang{{.*}} "-fsycl-is-device"{{.*}} "-o" "[[DEVICE_BASE_NAME]].bc"{{.*}} "[[DEVICE_BASE_NAME]].ii"
@@ -847,9 +847,9 @@
 /// passing of a library should not trigger the unbundler
 // RUN: touch %t.a
 // RUN: touch %t.lib
-// RUN: %clang -ccc-print-phases -fsycl %t.a %s 2>&1 \
+// RUN: %clang -ccc-print-phases -fsycl -fno-sycl-devicelib %t.a %s 2>&1 \
 // RUN:   | FileCheck -check-prefix=LIB-UNBUNDLE-CHECK %s
-// RUN: %clang_cl -ccc-print-phases -fsycl %t.lib %s 2>&1 \
+// RUN: %clang_cl -ccc-print-phases -fsycl -fno-sycl-devicelib %t.lib %s 2>&1 \
 // RUN:   | FileCheck -check-prefix=LIB-UNBUNDLE-CHECK %s
 // LIB-UNBUNDLE-CHECK-NOT: clang-offload-unbundler
 
diff --git a/llvm/tools/sycl-post-link/sycl-post-link.cpp b/llvm/tools/sycl-post-link/sycl-post-link.cpp
index a7fa663e31f7d..b348c04dadab9 100644
--- a/llvm/tools/sycl-post-link/sycl-post-link.cpp
+++ b/llvm/tools/sycl-post-link/sycl-post-link.cpp
@@ -89,6 +89,10 @@ static cl::opt<bool> OutputAssembly{"S",
                                     cl::desc("Write output as LLVM assembly"),
                                     cl::Hidden, cl::cat(PostLinkCat)};
 
+static cl::opt<bool> DeadCodeRemoval{
+    "dead-code-removal", cl::desc("Remove all dead code in current module"),
+    cl::cat(PostLinkCat)};
+
 enum IRSplitMode {
   SPLIT_PER_TU,    // one module per translation unit
   SPLIT_PER_KERNEL // one module per kernel
@@ -563,6 +567,31 @@ static string_vector saveResultSymbolsLists(string_vector &ResSymbolsLists) {
   }
   return std::move(Res);
 }
+// Remove 'unused' functions in device LLVM IR module, there is one execption
+// that functions with 'reference-indirectly' attribute can't be removed.
+static void RemoveDeadCode(Module &M) {
+  std::vector<Function *> unusedFuncVec;
+  bool isClean = false;
+  while (!isClean) {
+    unusedFuncVec.clear();
+    for (Function &F : M) {
+      if (F.user_empty() && (F.getCallingConv() == CallingConv::SPIR_FUNC) &&
+          !F.getAttributes().hasFnAttribute("referenced-indirectly")) {
+        F.deleteBody();
+        unusedFuncVec.push_back(&F);
+      }
+    }
+
+    if (!unusedFuncVec.empty()) {
+      for (Function *F : unusedFuncVec) {
+        M.getFunctionList().remove(F);
+      }
+      isClean = false;
+    } else {
+      isClean = true;
+    }
+  }
+}
 
 #define CHECK_AND_EXIT(E)                                                      \
   {                                                                            \
@@ -610,9 +639,13 @@ int main(int argc, char **argv) {
       "will produce single output file example_p.bc suitable for SPIRV\n"
       "translation.\n");
 
-  bool DoSplit = SplitMode.getNumOccurrences() > 0;
+  // For DeadCodeRemoval, the input is original IR and output is a 'clean' IR
+  // whose 'unused' functions are removed. DeadCodeRemoval shouldn't work with
+  // any other functionalities such as code split...
+  bool DoSplit = (SplitMode.getNumOccurrences() > 0) && !DeadCodeRemoval;
   bool DoSpecConst = SpecConstLower.getNumOccurrences() > 0;
-  bool DoParamInfo = EmitKernelParamInfo.getNumOccurrences() > 0;
+  bool DoParamInfo =
+      EmitKernelParamInfo.getNumOccurrences() > 0 && !DeadCodeRemoval;
 
   if (!DoSplit && !DoSpecConst && !DoSymGen && !DoParamInfo) {
     errs() << "no actions specified; try --help for usage info\n";
@@ -646,6 +679,11 @@ int main(int argc, char **argv) {
   if (OutputFilename.getNumOccurrences() == 0)
     OutputFilename = (Twine(sys::path::stem(InputFilename)) + ".files").str();
 
+  if (DeadCodeRemoval && IROutputOnly) {
+    RemoveDeadCode(*MPtr);
+    saveModule(*MPtr, OutputFilename);
+    return 0;
+  }
   std::map<StringRef, std::vector<Function *>> GlobalsSet;
 
   if (DoSplit || DoSymGen) {
diff --git a/sycl/test/devicelib/assert-aot.cpp b/sycl/test/devicelib/assert-aot.cpp
index 3867b97039987..811f174b399d7 100644
--- a/sycl/test/devicelib/assert-aot.cpp
+++ b/sycl/test/devicelib/assert-aot.cpp
@@ -1,5 +1,5 @@
 // REQUIRES: opencl-aot, cpu, linux
 
-// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %S/assert.cpp %sycl_libs_dir/libsycl-glibc.o %sycl_libs_dir/libsycl-fallback-cassert.o -o %t.aot.out
+// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %S/assert.cpp -o %t.aot.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.aot.out >%t.aot.msg
 // RUN: FileCheck %S/assert.cpp --input-file %t.aot.msg --check-prefixes=CHECK-MESSAGE
diff --git a/sycl/test/devicelib/assert-windows.cpp b/sycl/test/devicelib/assert-windows.cpp
index fb38520601d98..01db1858e5e67 100644
--- a/sycl/test/devicelib/assert-windows.cpp
+++ b/sycl/test/devicelib/assert-windows.cpp
@@ -4,8 +4,7 @@
 // Disable the test until the fix reaches SYCL test infrastructure.
 // XFAIL: *
 //
-// RUN: %clangxx -fsycl -c %s -o %t.o
-// RUN: %clangxx -fsycl %t.o %sycl_libs_dir/../bin/libsycl-msvc.o -o %t.out
+// RUN: %clangxx -fsycl %s -o %t.out
 //
 // MSVC implementation of assert does not call an unreachable built-in, so the
 // program doesn't terminate when fallback is used.
diff --git a/sycl/test/devicelib/assert.cpp b/sycl/test/devicelib/assert.cpp
index 343d949b74052..e86923ee09f10 100644
--- a/sycl/test/devicelib/assert.cpp
+++ b/sycl/test/devicelib/assert.cpp
@@ -1,6 +1,5 @@
 // REQUIRES: cpu,linux
-// RUN: %clangxx -fsycl -c %s -o %t.o
-// RUN: %clangxx -fsycl %t.o %sycl_libs_dir/libsycl-glibc.o -o %t.out
+// RUN: %clangxx -fsycl %s -o %t.out
 // (see the other RUN lines below; it is a bit complicated)
 //
 // assert() call in device code guarantees nothing: on some devices it behaves
diff --git a/sycl/test/devicelib/cmath-aot.cpp b/sycl/test/devicelib/cmath-aot.cpp
index e58a87d9f51bc..18ac5095bd074 100644
--- a/sycl/test/devicelib/cmath-aot.cpp
+++ b/sycl/test/devicelib/cmath-aot.cpp
@@ -1,14 +1,14 @@
 // REQUIRES: opencl-aot, cpu
 // UNSUPPORTED: windows
 
-// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %S/cmath_test.cpp %sycl_libs_dir/libsycl-cmath.o %sycl_libs_dir/libsycl-fallback-cmath.o -o %t.cmath.out
+// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %S/cmath_test.cpp -o %t.cmath.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.cmath.out
 
-// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %S/cmath_fp64_test.cpp %sycl_libs_dir/libsycl-cmath-fp64.o %sycl_libs_dir/libsycl-fallback-cmath-fp64.o -o %t.cmath.fp64.out
+// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %S/cmath_fp64_test.cpp -o %t.cmath.fp64.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.cmath.fp64.out
 
-// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %S/std_complex_math_test.cpp %sycl_libs_dir/libsycl-complex.o %sycl_libs_dir/libsycl-cmath.o %sycl_libs_dir/libsycl-fallback-complex.o %sycl_libs_dir/libsycl-fallback-cmath.o -o %t.complex.out
+// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %S/std_complex_math_test.cpp -o %t.complex.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.complex.out
 
-// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %S/std_complex_math_fp64_test.cpp %sycl_libs_dir/libsycl-complex-fp64.o %sycl_libs_dir/libsycl-cmath-fp64.o %sycl_libs_dir/libsycl-fallback-complex-fp64.o %sycl_libs_dir/libsycl-fallback-cmath-fp64.o -o %t.complex.fp64.out
+// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %S/std_complex_math_fp64_test.cpp -o %t.complex.fp64.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.complex.fp64.out
diff --git a/sycl/test/devicelib/cmath_fp64_test.cpp b/sycl/test/devicelib/cmath_fp64_test.cpp
index 30954e0eff59f..92b71027088a8 100644
--- a/sycl/test/devicelib/cmath_fp64_test.cpp
+++ b/sycl/test/devicelib/cmath_fp64_test.cpp
@@ -1,6 +1,5 @@
 // UNSUPPORTED: windows
-// RUN: %clangxx -fsycl -c %s -o %t.o
-// RUN: %clangxx -fsycl %t.o %sycl_libs_dir/libsycl-cmath-fp64.o -o %t.out
+// RUN: %clangxx -fsycl %s -o %t.out
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
diff --git a/sycl/test/devicelib/cmath_test.cpp b/sycl/test/devicelib/cmath_test.cpp
index bb2e37345d6fc..ffb4fc039b54b 100644
--- a/sycl/test/devicelib/cmath_test.cpp
+++ b/sycl/test/devicelib/cmath_test.cpp
@@ -1,6 +1,5 @@
 // UNSUPPORTED: windows
-// RUN: %clangxx -fsycl -c %s -o %t.o
-// RUN: %clangxx -fsycl %t.o %sycl_libs_dir/libsycl-cmath.o -o %t.out
+// RUN: %clangxx -fsycl %s -o %t.out
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
diff --git a/sycl/test/devicelib/complex-fpga.cpp b/sycl/test/devicelib/complex-fpga.cpp
new file mode 100644
index 0000000000000..4d9e4db39e84b
--- /dev/null
+++ b/sycl/test/devicelib/complex-fpga.cpp
@@ -0,0 +1,15 @@
+//==----- accelerator.cpp - AOT compilation for fpga devices using aoc  ----==//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===------------------------------------------------------------------------===//
+
+// REQUIRES: aoc, accelerator
+
+// RUN: %clangxx -fsycl -fsycl-targets=spir64_fpga-unknown-unknown-sycldevice %S/std_complex_math_test.cpp -o %t.out
+// RUN: %ACC_RUN_PLACEHOLDER %t.out
+
+// RUN: %clangxx -fsycl -fintelfpga %S/std_complex_math_test.cpp -o %t.out
+// RUN: %ACC_RUN_PLACEHOLDER %t.out
diff --git a/sycl/test/devicelib/math_fp64_test.cpp b/sycl/test/devicelib/math_fp64_test.cpp
index dec0bf7341d92..23c81033e9ea4 100644
--- a/sycl/test/devicelib/math_fp64_test.cpp
+++ b/sycl/test/devicelib/math_fp64_test.cpp
@@ -1,6 +1,5 @@
 // REQUIRES: cpu, linux
-// RUN: %clangxx -fsycl -c %s -o %t.o
-// RUN: %clangxx -fsycl %t.o %sycl_libs_dir/libsycl-cmath-fp64.o -o %t.out
+// RUN: %clangxx -fsycl %s -o %t.out
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
diff --git a/sycl/test/devicelib/math_fp64_windows_test.cpp b/sycl/test/devicelib/math_fp64_windows_test.cpp
index 64d4e000c5ff5..ac728ad4bf8b5 100644
--- a/sycl/test/devicelib/math_fp64_windows_test.cpp
+++ b/sycl/test/devicelib/math_fp64_windows_test.cpp
@@ -1,6 +1,5 @@
 // REQUIRES: cpu, windows
-// RUN: %clangxx -fsycl -c %s -o %t.o
-// RUN: %clangxx -fsycl %t.o %sycl_libs_dir/../bin/libsycl-cmath-fp64.o -o %t.out
+// RUN: %clangxx -fsycl %s -o %t.out
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
diff --git a/sycl/test/devicelib/math_override_test.cpp b/sycl/test/devicelib/math_override_test.cpp
index e634bf77fbd37..b3014f9bea884 100644
--- a/sycl/test/devicelib/math_override_test.cpp
+++ b/sycl/test/devicelib/math_override_test.cpp
@@ -1,6 +1,5 @@
 // UNSUPPORTED: windows
-// RUN: %clangxx -fsycl -c %s -o %t.o
-// RUN: %clangxx -fsycl %t.o %sycl_libs_dir/libsycl-cmath.o -o %t.out
+// RUN: %clangxx -fsycl %s -o %t.out -fno-builtin
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 #include <CL/sycl.hpp>
 #include <iostream>
@@ -16,6 +15,8 @@ constexpr s::access::mode sycl_write = s::access::mode::write;
 SYCL_EXTERNAL
 extern "C" float sinf(float x) { return x + 100.f; }
 
+SYCL_EXTERNAL
+extern "C" float cosf(float x);
 class DeviceTest;
 
 void device_test() {
@@ -37,7 +38,6 @@ void device_test() {
       });
     });
   }
-
   assert(approx_equal_fp(result_sin, 100.f) && approx_equal_fp(result_cos, 1.f));
 }
 
diff --git a/sycl/test/devicelib/math_test.cpp b/sycl/test/devicelib/math_test.cpp
index 1e3885960b2c5..8394874b7b76c 100644
--- a/sycl/test/devicelib/math_test.cpp
+++ b/sycl/test/devicelib/math_test.cpp
@@ -1,6 +1,5 @@
 // REQUIRES: cpu, linux
-// RUN: %clangxx -fsycl -c %s -o %t.o
-// RUN: %clangxx -fsycl %t.o %sycl_libs_dir/libsycl-cmath.o -o %t.out
+// RUN: %clangxx -fsycl %s -o %t.out
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
diff --git a/sycl/test/devicelib/math_windows_test.cpp b/sycl/test/devicelib/math_windows_test.cpp
index fa295c8bff84c..964db574beec5 100644
--- a/sycl/test/devicelib/math_windows_test.cpp
+++ b/sycl/test/devicelib/math_windows_test.cpp
@@ -1,6 +1,5 @@
 // REQUIRES: cpu, windows
-// RUN: %clangxx -fsycl -c %s -o %t.o
-// RUN: %clangxx -fsycl %t.o %sycl_libs_dir/../bin/libsycl-cmath.o -o %t.out
+// RUN: %clangxx -fsycl %s -o %t.out
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
diff --git a/sycl/test/devicelib/std_complex_math_fp64_test.cpp b/sycl/test/devicelib/std_complex_math_fp64_test.cpp
index 40eeadfd04321..673583b39a20d 100644
--- a/sycl/test/devicelib/std_complex_math_fp64_test.cpp
+++ b/sycl/test/devicelib/std_complex_math_fp64_test.cpp
@@ -1,6 +1,5 @@
 // UNSUPPORTED: windows
-// RUN: %clangxx -fsycl -c %s -o %t.o
-// RUN: %clangxx -fsycl %t.o %sycl_libs_dir/libsycl-complex-fp64.o %sycl_libs_dir/libsycl-cmath-fp64.o -o %t.out
+// RUN: %clangxx -fsycl %s -o %t.out
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
diff --git a/sycl/test/devicelib/std_complex_math_test.cpp b/sycl/test/devicelib/std_complex_math_test.cpp
index b2578076243ff..fb6b1feaf1319 100644
--- a/sycl/test/devicelib/std_complex_math_test.cpp
+++ b/sycl/test/devicelib/std_complex_math_test.cpp
@@ -1,6 +1,5 @@
 // UNSUPPORTED: windows
-// RUN: %clangxx -fsycl -c %s -o %t.o
-// RUN: %clangxx -fsycl %t.o %sycl_libs_dir/libsycl-complex.o %sycl_libs_dir/libsycl-cmath.o -o %t.out
+// RUN: %clangxx -fsycl %s -o %t.out
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out