diff --git a/clang/include/clang/Basic/DiagnosticDriverKinds.td b/clang/include/clang/Basic/DiagnosticDriverKinds.td
index 268d73833d81c..c435da92bae46 100644
--- a/clang/include/clang/Basic/DiagnosticDriverKinds.td
+++ b/clang/include/clang/Basic/DiagnosticDriverKinds.td
@@ -352,6 +352,8 @@ def warn_drv_sycl_offload_target_duplicate : Warning<
 def warn_drv_sycl_target_missing : Warning<
   "linked binaries do not contain expected '%0' target; found targets: '%1'">,
   InGroup<SyclTarget>;
+def err_drv_no_rdc_sycl_target_missing : Error<
+  "linked binaries do not contain expected '%0' target; found targets: '%1', this is not supported with '-fno-sycl-rdc'">;
 def err_drv_multiple_target_with_forced_target : Error<
   "multiple target usage with '%0' is not supported with '%1'">;
 def err_drv_failed_to_deduce_target_from_arch : Error<
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index 58ae955f41a99..cdaa48aa5c666 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -11791,6 +11791,9 @@ def err_sycl_restrict : Error<
 def err_sycl_external_global : Error<
   "invalid reference to 'device_global' variable; external 'device_global'"
   " variable must be marked with SYCL_EXTERNAL macro">;
+def err_sycl_external_no_rdc : Error<
+  "invalid %select{declaration|definition}0 of SYCL_EXTERNAL function in non-relocatable "
+   "device code mode">;
 def warn_sycl_kernel_too_big_args : Warning<
   "size of kernel arguments (%0 bytes) may exceed the supported maximum "
   "of %1 bytes on some devices">, InGroup<SyclStrict>, ShowInSystemHeader;
diff --git a/clang/include/clang/Driver/Action.h b/clang/include/clang/Driver/Action.h
index cc27008eebde5..c7916b101bac0 100644
--- a/clang/include/clang/Driver/Action.h
+++ b/clang/include/clang/Driver/Action.h
@@ -14,6 +14,7 @@
 #include "clang/Driver/Util.h"
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/SetVector.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/ADT/iterator_range.h"
@@ -817,7 +818,8 @@ class FileTableTformJobAction : public JobAction {
       REPLACE,
       REPLACE_CELL,
       RENAME,
-      COPY_SINGLE_FILE
+      COPY_SINGLE_FILE,
+      MERGE
     };
 
     Tform() = default;
@@ -855,6 +857,10 @@ class FileTableTformJobAction : public JobAction {
   // output file.
   void addCopySingleFileTform(StringRef ColumnName, int Row);
 
+  // Merges all tables from filename listed at column <ColumnName> into a
+  // single output table.
+  void addMergeTform(StringRef ColumnName);
+
   static bool classof(const Action *A) {
     return A->getKind() == FileTableTformJobClass;
   }
@@ -937,6 +943,14 @@ class ForEachWrappingAction : public Action {
   static bool classof(const Action *A) {
     return A->getKind() == ForEachWrappingClass;
   }
+
+  void addSerialAction(const Action *A) { SerialActions.insert(A); }
+  const llvm::SmallSetVector<const Action *, 2> &getSerialActions() const {
+    return SerialActions;
+  }
+
+private:
+  llvm::SmallSetVector<const Action *, 2> SerialActions;
 };
 
 } // namespace driver
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index 8127d6d2e146c..8d1e21d54b7d9 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -1005,8 +1005,9 @@ def fno_cuda_flush_denormals_to_zero : Flag<["-"], "fno-cuda-flush-denormals-to-
   Alias<fno_gpu_flush_denormals_to_zero>;
 defm gpu_rdc : BoolFOption<"gpu-rdc",
   LangOpts<"GPURelocatableDeviceCode">, DefaultFalse,
-  PosFlag<SetTrue, [CC1Option], "Generate relocatable device code, also known as separate compilation mode">,
-  NegFlag<SetFalse>>;
+  PosFlag<SetTrue, [], "Generate relocatable device code, also known as separate compilation mode">,
+  NegFlag<SetFalse, []>,
+  BothFlags<[CC1Option]>>;
 def : Flag<["-"], "fcuda-rdc">, Alias<fgpu_rdc>;
 def : Flag<["-"], "fno-cuda-rdc">, Alias<fno_gpu_rdc>;
 defm cuda_short_ptr : BoolFOption<"cuda-short-ptr",
@@ -2990,6 +2991,8 @@ def fsycl_max_parallel_jobs_EQ : Joined<["-"], "fsycl-max-parallel-link-jobs=">,
   HelpText<"Experimental feature: Controls the maximum parallelism of actions performed "
   "on SYCL device code post-link, i.e. the generation of SPIR-V device images "
   "or AOT compilation of each device image.">;
+def : Flag<["-"], "fsycl-rdc">, Alias<fgpu_rdc>;
+def : Flag<["-"], "fno-sycl-rdc">, Alias<fno_gpu_rdc>;
 def fsyntax_only : Flag<["-"], "fsyntax-only">,
   Flags<[NoXarchOption,CoreOption,CC1Option,FC1Option,FlangOption]>, Group<Action_Group>,
   HelpText<"Run the preprocessor, parser and semantic analysis stages">;
diff --git a/clang/lib/Driver/Action.cpp b/clang/lib/Driver/Action.cpp
index f9af7a8f3f970..73312dcf28744 100644
--- a/clang/lib/Driver/Action.cpp
+++ b/clang/lib/Driver/Action.cpp
@@ -567,6 +567,10 @@ void FileTableTformJobAction::addCopySingleFileTform(StringRef ColumnName,
       Tform(Tform::COPY_SINGLE_FILE, {ColumnName, std::to_string(Row)}));
 }
 
+void FileTableTformJobAction::addMergeTform(StringRef ColumnName) {
+  Tforms.emplace_back(Tform(Tform::MERGE, {ColumnName}));
+}
+
 void AppendFooterJobAction::anchor() {}
 
 AppendFooterJobAction::AppendFooterJobAction(Action *Input, types::ID Type)
diff --git a/clang/lib/Driver/Driver.cpp b/clang/lib/Driver/Driver.cpp
index e6626b807c107..ec8609bdf51dd 100644
--- a/clang/lib/Driver/Driver.cpp
+++ b/clang/lib/Driver/Driver.cpp
@@ -1095,10 +1095,11 @@ void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
     Diag(clang::diag::err_drv_invalid_argument_to_option)
         << ArgValue << A->getOption().getName();
   };
+  Arg *DeviceCodeSplit =
+      C.getInputArgs().getLastArg(options::OPT_fsycl_device_code_split_EQ);
   checkSingleArgValidity(SYCLLink, {"early", "image"});
-  checkSingleArgValidity(
-      C.getInputArgs().getLastArg(options::OPT_fsycl_device_code_split_EQ),
-      {"per_kernel", "per_source", "auto", "off"});
+  checkSingleArgValidity(DeviceCodeSplit,
+                         {"per_kernel", "per_source", "auto", "off"});
 
   Arg *SYCLForceTarget =
       getArgRequiringSYCLRuntime(options::OPT_fsycl_force_target_EQ);
@@ -5374,6 +5375,7 @@ class OffloadingActionBuilder final {
           // Current list is empty, nothing to process.
           continue;
 
+        ActionList DeviceLibs;
         ActionList DeviceLibObjects;
         ActionList LinkObjects;
         auto TT = TC->getTriple();
@@ -5439,6 +5441,8 @@ class OffloadingActionBuilder final {
         //   n - target is NVPTX/AMDGCN
         //   a - SPIRV AOT compilation is requested
         //   s - device code split requested
+        //   r - relocatable device code is requested
+        //   f - link object output type is TY_Tempfilelist (fat archive)
         //   * - "all other cases"
         //     - no condition means output/input is "always" present
         // First symbol indicates output/input type
@@ -5449,34 +5453,63 @@ class OffloadingActionBuilder final {
         //                   .-----------------.
         //                   |Link(LinkObjects)|
         //                   .-----------------.
-        //                            |
-        //         .--------------------------------------.
-        //         |               PostLink               |
-        //         .--------------------------------------.
-        //                             [+*]            [+]
-        //                               |              |
-        //                      .-----------------.     |
-        //                      | FileTableTform  |     |
-        //                      | (extract "Code")|     |
-        //                      .-----------------.     |
-        //                              [-]             |
-        //           --------------------|              |
-        //          [.]                [-*]             |
-        //   .---------------.  .-------------------.   |
-        //   | finalizeNVPTX  | |  SPIRVTranslator  |   |
-        //   | finalizeAMDGCN | |                   |   |
-        //   .---------------.  .-------------------.   |
-        //          [.]             [-as]      [-!a]    |
-        //           |                |          |      |
-        //           |              [-s]         |      |
-        //           |       .----------------.  |      |
-        //           |       | BackendCompile |  |      |
-        //           |       .----------------.  |      |
-        //           |              [-s]         |      |
-        //           |                |          |      |
-        //           |              [-a]      [-!a]    [+]
-        //           |              .--------------------.
-        //           -----------[-n]|   FileTableTform   |
+        //                ----[-!rf]   [*]
+        //               [-!rf]         |
+        //         .-------------.      |
+        //         | llvm-foreach|      |
+        //         .-------------.      |
+        //               [.]            |
+        //                |             |
+        //                |             |
+        //         .---------------------------------------.
+        //         |               PostLink                |
+        //         .---------------------------------------.
+        //                           [+*]                [+]
+        //                             |                  |
+        //                             |                  |
+        //                             |---------         |
+        //                             |        |         |
+        //                             |        |         |
+        //                             |      [+!rf]      |
+        //                             |  .-------------. |
+        //                             |  | llvm-foreach| |
+        //                             |  .-------------. |
+        //                             |        |         |
+        //                            [+*]    [+!rf]      |
+        //                      .-----------------.       |
+        //                      | FileTableTform  |       |
+        //                      | (extract "Code")|       |
+        //                      .-----------------.       |
+        //                              [-]               |-----------
+        //           --------------------|                           |
+        //           |                   |                           |
+        //           |                   |-----------------          |
+        //           |                   |                |          |
+        //           |                   |               [-!rf]      |
+        //           |                   |         .--------------.  |
+        //           |                   |         |FileTableTform|  |
+        //           |                   |         |   (merge)    |  |
+        //           |                   |         .--------------.  |
+        //           |                   |               [-]         |-------
+        //           |                   |                |          |      |
+        //           |                   |                |    ------|      |
+        //           |                   |        --------|    |            |
+        //          [.]                 [-*]   [-!rf]        [+!rf]         |
+        //   .---------------.  .-------------------. .--------------.      |
+        //   | finalizeNVPTX  | |  SPIRVTranslator  | |FileTableTform|      |
+        //   | finalizeAMDGCN | |                   | |   (merge)    |      |
+        //   .---------------.  .-------------------. . -------------.      |
+        //          [.]             [-as]      [-!a]         |              |
+        //           |                |          |           |              |
+        //           |              [-s]         |           |              |
+        //           |       .----------------.  |           |              |
+        //           |       | BackendCompile |  |           |              |
+        //           |       .----------------.  |     ------|              |
+        //           |              [-s]         |     |                    |
+        //           |                |          |     |                    |
+        //           |              [-a]      [-!a]  [-!rf]                 |
+        //           |              .--------------------.                  |
+        //           -----------[-n]|   FileTableTform   |[+*]--------------|
         //                          |  (replace "Code")  |
         //                          .--------------------.
         //                                      |
@@ -5485,16 +5518,20 @@ class OffloadingActionBuilder final {
         //         |            OffloadWrapper            |
         //         .--------------------------------------.
         //
-        Action *DeviceLinkAction =
-            C.MakeAction<LinkJobAction>(LinkObjects, types::TY_LLVM_BC);
         ActionList FullLinkObjects;
-        bool SYCLDeviceLibLinked = false;
-        FullLinkObjects.push_back(DeviceLinkAction);
+        bool IsRDC = !tools::SYCL::shouldDoPerObjectFileLinking(C);
+        if (IsRDC) {
+          Action *DeviceLinkAction =
+              C.MakeAction<LinkJobAction>(LinkObjects, types::TY_LLVM_BC);
+          FullLinkObjects.push_back(DeviceLinkAction);
+        } else
+          FullLinkObjects = LinkObjects;
 
         // FIXME: Link all wrapper and fallback device libraries as default,
         // When spv online link is supported by all backends, the fallback
         // device libraries are only needed when current toolchain is using
         // AOT compilation.
+        bool SYCLDeviceLibLinked = false;
         if (isSPIR || isNVPTX) {
           bool UseJitLink =
               isSPIR &&
@@ -5502,118 +5539,209 @@ class OffloadingActionBuilder final {
                            options::OPT_fno_sycl_device_lib_jit_link, false);
           bool UseAOTLink = isSPIR && (isSpirvAOT || !UseJitLink);
           SYCLDeviceLibLinked = addSYCLDeviceLibs(
-              TC, FullLinkObjects, UseAOTLink,
+              TC, DeviceLibs, UseAOTLink,
               C.getDefaultToolChain().getTriple().isWindowsMSVCEnvironment());
         }
+        JobAction *LinkSYCLLibs =
+            C.MakeAction<LinkJobAction>(DeviceLibs, types::TY_LLVM_BC);
+        for (Action *FullLinkObject : FullLinkObjects) {
+          if (FullLinkObject->getKind() ==
+              clang::driver::Action::OffloadDepsJobClass)
+            continue;
+          Action *FullDeviceLinkAction = nullptr;
+          ActionList WrapperInputs;
+
+          if (SYCLDeviceLibLinked) {
+            if (IsRDC) {
+              // First object has to be non-DeviceLib for only-needed to be
+              // passed.
+              DeviceLibs.insert(DeviceLibs.begin(), FullLinkObject);
+              FullDeviceLinkAction =
+                  C.MakeAction<LinkJobAction>(DeviceLibs, types::TY_LLVM_BC);
+            } else {
+              FullDeviceLinkAction = FullLinkObject;
+
+              ActionList DeviceCodeAndSYCLLibs{FullDeviceLinkAction, LinkSYCLLibs};
+              JobAction *LinkDeviceCode =
+                  C.MakeAction<LinkJobAction>(DeviceCodeAndSYCLLibs, types::TY_LLVM_BC);
+
+              if (FullDeviceLinkAction->getType() == types::TY_Tempfilelist) {
+                // If our compiler input outputs a temp file list (eg. fat
+                // static archive), we need to link the device code against
+                // each entry in the static archive.
+                auto *ParallelLinkDeviceCode =
+                    C.MakeAction<ForEachWrappingAction>(
+                        cast<JobAction>(FullDeviceLinkAction), LinkDeviceCode);
+                // The SYCL device library action tree should not be
+                // for-eached, it only needs to happen once total. The
+                // for-each action should start linking device code with the
+                // device libraries.
+                std::function<void(const Action *)> traverseActionTree =
+                    [&](const Action *Act) {
+                      ParallelLinkDeviceCode->addSerialAction(Act);
+                      for (const auto &Input : Act->getInputs()) {
+                        traverseActionTree(Input);
+                      }
+                    };
+                traverseActionTree(LinkSYCLLibs);
+                ActionList TformInputs{FullDeviceLinkAction,
+                                       ParallelLinkDeviceCode};
+                auto *ReplaceFilesAction =
+                    C.MakeAction<FileTableTformJobAction>(
+                        TformInputs, types::TY_Tempfilelist,
+                        types::TY_Tempfilelist);
+                ReplaceFilesAction->addReplaceColumnTform(
+                    FileTableTformJobAction::COL_ZERO,
+                    FileTableTformJobAction::COL_ZERO);
+                ReplaceFilesAction->addExtractColumnTform(
+                    FileTableTformJobAction::COL_ZERO, false /*drop titles*/);
+                FullDeviceLinkAction = ReplaceFilesAction;
+              } else {
+                // If our compiler input is singular, just do a single link.
+                FullDeviceLinkAction = LinkDeviceCode;
+              }
+            }
+          } else
+            FullDeviceLinkAction = FullLinkObject;
+
+          // reflects whether current target is ahead-of-time and can't
+          // support runtime setting of specialization constants
+          bool isAOT = isNVPTX || isAMDGCN || isSpirvAOT;
+
+          // post link is not optional - even if not splitting, always need to
+          // process specialization constants
+          types::ID PostLinkOutType = isSPIR ? types::TY_Tempfiletable
+                                             : FullDeviceLinkAction->getType();
+          auto createPostLinkAction = [&]() {
+            // For SPIR-V targets, force TY_Tempfiletable.
+            auto TypedPostLinkAction = C.MakeAction<SYCLPostLinkJobAction>(
+                FullDeviceLinkAction, PostLinkOutType, types::TY_Tempfiletable);
+            TypedPostLinkAction->setRTSetsSpecConstants(!isAOT);
+            return TypedPostLinkAction;
+          };
+          Action *PostLinkAction = createPostLinkAction();
+          bool NoRDCFatStaticArchive =
+              !IsRDC &&
+              FullDeviceLinkAction->getType() == types::TY_Tempfilelist;
+          if (NoRDCFatStaticArchive)
+            PostLinkAction = C.MakeAction<ForEachWrappingAction>(
+                cast<JobAction>(FullDeviceLinkAction),
+                cast<JobAction>(PostLinkAction));
+
+          auto createExtractIRFilesAction = [&]() {
+            auto *TypedExtractIRFilesAction =
+                C.MakeAction<FileTableTformJobAction>(
+                    PostLinkAction,
+                    isSPIR ? types::TY_Tempfilelist : PostLinkAction->getType(),
+                    types::TY_Tempfilelist);
+            // single column w/o title fits TY_Tempfilelist format
+            TypedExtractIRFilesAction->addExtractColumnTform(
+                FileTableTformJobAction::COL_CODE, false /*drop titles*/);
+            return TypedExtractIRFilesAction;
+          };
 
-        Action *FullDeviceLinkAction = nullptr;
-        if (SYCLDeviceLibLinked)
-          FullDeviceLinkAction =
-              C.MakeAction<LinkJobAction>(FullLinkObjects, types::TY_LLVM_BC);
-        else
-          FullDeviceLinkAction = DeviceLinkAction;
-
-        // reflects whether current target is ahead-of-time and can't support
-        // runtime setting of specialization constants
-        bool isAOT = isNVPTX || isAMDGCN || isSpirvAOT;
-
-        ActionList WrapperInputs;
-        // post link is not optional - even if not splitting, always need to
-        // process specialization constants
-
-        types::ID PostLinkOutType =
-            isSPIR ? types::TY_Tempfiletable : FullDeviceLinkAction->getType();
-        // For SPIR-V targets, force TY_Tempfiletable.
-        auto *PostLinkAction = C.MakeAction<SYCLPostLinkJobAction>(
-            FullDeviceLinkAction, PostLinkOutType, types::TY_Tempfiletable);
-        PostLinkAction->setRTSetsSpecConstants(!isAOT);
-
-        auto *ExtractIRFilesAction = C.MakeAction<FileTableTformJobAction>(
-            PostLinkAction,
-            isSPIR ? types::TY_Tempfilelist : PostLinkAction->getType(),
-            types::TY_Tempfilelist);
-        // single column w/o title fits TY_Tempfilelist format
-        ExtractIRFilesAction->addExtractColumnTform(
-            FileTableTformJobAction::COL_CODE, false /*drop titles*/);
-
-        if (isNVPTX || isAMDGCN) {
-          JobAction *FinAction =
-              isNVPTX ? finalizeNVPTXDependences(ExtractIRFilesAction,
-                                                 TC->getTriple())
-                      : finalizeAMDGCNDependences(ExtractIRFilesAction,
-                                                  TC->getTriple());
-          auto *ForEachWrapping = C.MakeAction<ForEachWrappingAction>(
-              ExtractIRFilesAction, FinAction);
-
-          ActionList TformInputs{PostLinkAction, ForEachWrapping};
-          auto *ReplaceFilesAction = C.MakeAction<FileTableTformJobAction>(
-              TformInputs, types::TY_Tempfiletable, types::TY_Tempfiletable);
-          ReplaceFilesAction->addReplaceColumnTform(
-              FileTableTformJobAction::COL_CODE,
-              FileTableTformJobAction::COL_CODE);
-
-          WrapperInputs.push_back(ReplaceFilesAction);
-        } else {
-          // For SPIRV-based targets - translate to SPIRV then optionally
-          // compile ahead-of-time to native architecture
-          Action *BuildCodeAction =
-              (Action *)C.MakeAction<SPIRVTranslatorJobAction>(
-                  ExtractIRFilesAction, types::TY_Tempfilelist);
+          Action *ExtractIRFilesAction = createExtractIRFilesAction();
 
-          // After the Link, wrap the files before the final host link
-          if (isAOT) {
-            types::ID OutType = types::TY_Tempfilelist;
-            if (!DeviceCodeSplit) {
-              OutType = (TT.getSubArch() == llvm::Triple::SPIRSubArch_fpga)
-                            ? FPGAOutType
-                            : types::TY_Image;
+          if (isNVPTX || isAMDGCN) {
+            JobAction *FinAction =
+                isNVPTX ? finalizeNVPTXDependences(ExtractIRFilesAction,
+                                                   TC->getTriple())
+                        : finalizeAMDGCNDependences(ExtractIRFilesAction,
+                                                    TC->getTriple());
+            auto *ForEachWrapping = C.MakeAction<ForEachWrappingAction>(
+                cast<JobAction>(ExtractIRFilesAction), FinAction);
+
+            ActionList TformInputs{PostLinkAction, ForEachWrapping};
+            auto *ReplaceFilesAction = C.MakeAction<FileTableTformJobAction>(
+                TformInputs, types::TY_Tempfiletable, types::TY_Tempfiletable);
+            ReplaceFilesAction->addReplaceColumnTform(
+                FileTableTformJobAction::COL_CODE,
+                FileTableTformJobAction::COL_CODE);
+
+            WrapperInputs.push_back(ReplaceFilesAction);
+          } else {
+            if (NoRDCFatStaticArchive) {
+              ExtractIRFilesAction = C.MakeAction<ForEachWrappingAction>(
+                  cast<JobAction>(FullDeviceLinkAction),
+                  cast<JobAction>(ExtractIRFilesAction));
+
+              auto *MergeAllTablesIntoOne =
+                  C.MakeAction<FileTableTformJobAction>(ExtractIRFilesAction,
+                                                        types::TY_Tempfilelist,
+                                                        types::TY_Tempfilelist);
+              MergeAllTablesIntoOne->addMergeTform(
+                  FileTableTformJobAction::COL_ZERO);
+              ExtractIRFilesAction = MergeAllTablesIntoOne;
+            }
+            // For SPIRV-based targets - translate to SPIRV then optionally
+            // compile ahead-of-time to native architecture
+            Action *BuildCodeAction = C.MakeAction<SPIRVTranslatorJobAction>(
+                ExtractIRFilesAction, types::TY_Tempfilelist);
+
+            // After the Link, wrap the files before the final host link
+            if (isAOT) {
+              types::ID OutType = types::TY_Tempfilelist;
+              if (!DeviceCodeSplit) {
+                OutType = (TT.getSubArch() == llvm::Triple::SPIRSubArch_fpga)
+                              ? FPGAOutType
+                              : types::TY_Image;
+              }
+              // Do the additional Ahead of Time compilation when the specific
+              // triple calls for it (provided a valid subarch).
+              ActionList BEInputs;
+              BEInputs.push_back(BuildCodeAction);
+              auto unbundleAdd = [&](Action *A, types::ID T) {
+                ActionList AL;
+                AL.push_back(A);
+                Action *UnbundleAction =
+                    C.MakeAction<OffloadUnbundlingJobAction>(AL, T);
+                BEInputs.push_back(UnbundleAction);
+              };
+              // Send any known objects/archives through the unbundler to grab
+              // the dependency file associated.  This is only done for
+              // -fintelfpga.
+              for (Action *A : FPGAObjectInputs)
+                unbundleAdd(A, types::TY_FPGA_Dependencies);
+              for (Action *A : FPGAArchiveInputs)
+                unbundleAdd(A, types::TY_FPGA_Dependencies_List);
+              for (const auto &A : DeviceLibObjects)
+                BEInputs.push_back(A);
+              BuildCodeAction =
+                  C.MakeAction<BackendCompileJobAction>(BEInputs, OutType);
+            }
+            if (NoRDCFatStaticArchive) {
+              auto *MergeAllTablesIntoOne =
+                  C.MakeAction<FileTableTformJobAction>(PostLinkAction,
+                                                        types::TY_Tempfilelist,
+                                                        types::TY_Tempfilelist);
+              MergeAllTablesIntoOne->addMergeTform(
+                  FileTableTformJobAction::COL_ZERO);
+              PostLinkAction = MergeAllTablesIntoOne;
             }
-            // Do the additional Ahead of Time compilation when the specific
-            // triple calls for it (provided a valid subarch).
-            ActionList BEInputs;
-            BEInputs.push_back(BuildCodeAction);
-            auto unbundleAdd = [&](Action *A, types::ID T) {
-              ActionList AL;
-              AL.push_back(A);
-              Action *UnbundleAction =
-                  C.MakeAction<OffloadUnbundlingJobAction>(AL, T);
-              BEInputs.push_back(UnbundleAction);
-            };
-            // Send any known objects/archives through the unbundler to grab the
-            // dependency file associated.  This is only done for -fintelfpga.
-            for (Action *A : FPGAObjectInputs)
-              unbundleAdd(A, types::TY_FPGA_Dependencies);
-            for (Action *A : FPGAArchiveInputs)
-              unbundleAdd(A, types::TY_FPGA_Dependencies_List);
-            for (const auto &A : DeviceLibObjects)
-              BEInputs.push_back(A);
-            BuildCodeAction =
-                C.MakeAction<BackendCompileJobAction>(BEInputs, OutType);
+            ActionList TformInputs{PostLinkAction, BuildCodeAction};
+            auto *ReplaceFilesAction = C.MakeAction<FileTableTformJobAction>(
+                TformInputs, types::TY_Tempfiletable, types::TY_Tempfiletable);
+            ReplaceFilesAction->addReplaceColumnTform(
+                FileTableTformJobAction::COL_CODE,
+                FileTableTformJobAction::COL_CODE);
+            WrapperInputs.push_back(ReplaceFilesAction);
           }
-          ActionList TformInputs{PostLinkAction, BuildCodeAction};
-          auto *ReplaceFilesAction = C.MakeAction<FileTableTformJobAction>(
-              TformInputs, types::TY_Tempfiletable, types::TY_Tempfiletable);
-          ReplaceFilesAction->addReplaceColumnTform(
-              FileTableTformJobAction::COL_CODE,
-              FileTableTformJobAction::COL_CODE);
-          WrapperInputs.push_back(ReplaceFilesAction);
-        }
 
-        // After the Link, wrap the files before the final host link
-        auto *DeviceWrappingAction = C.MakeAction<OffloadWrapperJobAction>(
-            WrapperInputs, types::TY_Object);
+          // After the Link, wrap the files before the final host link
+          auto *DeviceWrappingAction = C.MakeAction<OffloadWrapperJobAction>(
+              WrapperInputs, types::TY_Object);
 
-        if (isSpirvAOT) {
-          bool AddBA = (TT.getSubArch() == llvm::Triple::SPIRSubArch_gen &&
-                        BoundArch != nullptr);
-          DA.add(*DeviceWrappingAction, *TC, AddBA ? BoundArch : nullptr,
-                 Action::OFK_SYCL);
-        } else
-          withBoundArchForToolChain(TC, [&](const char *BoundArch) {
-            DA.add(*DeviceWrappingAction, *TC, BoundArch, Action::OFK_SYCL);
-          });
+          if (isSpirvAOT) {
+            bool AddBA = (TT.getSubArch() == llvm::Triple::SPIRSubArch_gen &&
+                          BoundArch != nullptr);
+            DA.add(*DeviceWrappingAction, *TC, AddBA ? BoundArch : nullptr,
+                   Action::OFK_SYCL);
+          } else
+            withBoundArchForToolChain(TC, [&](const char *BoundArch) {
+              DA.add(*DeviceWrappingAction, *TC, BoundArch, Action::OFK_SYCL);
+            });
+        }
       }
-
       for (auto &SAI : SYCLAOTInputs) {
         // Extract binary file name
         std::string FN(SAI.second);
@@ -5799,8 +5927,14 @@ class OffloadingActionBuilder final {
           ArchListStr += Section;
           Cnt++;
         }
-        C.getDriver().Diag(diag::warn_drv_sycl_target_missing)
-            << SectionTriple << ArchListStr;
+        if (tools::SYCL::shouldDoPerObjectFileLinking(C)) {
+          C.getDriver().Diag(diag::err_drv_no_rdc_sycl_target_missing)
+              << SectionTriple << ArchListStr;
+          C.setContainsError();
+        } else {
+          C.getDriver().Diag(diag::warn_drv_sycl_target_missing)
+              << SectionTriple << ArchListStr;
+        }
       }
     }
 
@@ -7990,7 +8124,8 @@ static void CollectForEachInputs(
     InputInfoList &InputInfos, const Action *SourceAction, const ToolChain *TC,
     StringRef BoundArch, Action::OffloadKind TargetDeviceOffloadKind,
     const std::map<std::pair<const Action *, std::string>, InputInfoList>
-        &CachedResults) {
+        &CachedResults,
+    const ForEachWrappingAction *FEA) {
   for (const Action *Input : SourceAction->getInputs()) {
     // Search for the Input, if not in the cache assume actions were collapsed
     // so recurse.
@@ -7998,10 +8133,12 @@ static void CollectForEachInputs(
         {Input,
          GetTriplePlusArchString(TC, BoundArch, TargetDeviceOffloadKind)});
     if (Lookup != CachedResults.end()) {
-      InputInfos.append(Lookup->second);
+      if (!FEA->getSerialActions().count(Input)) {
+        InputInfos.append(Lookup->second);
+      }
     } else {
       CollectForEachInputs(InputInfos, Input, TC, BoundArch,
-                           TargetDeviceOffloadKind, CachedResults);
+                           TargetDeviceOffloadKind, CachedResults, FEA);
     }
   }
 }
@@ -8172,18 +8309,25 @@ InputInfoList Driver::BuildJobsForActionNoCache(
          llvm::make_range(std::make_move_iterator(JobsToWrap.begin()),
                           std::make_move_iterator(JobsToWrap.end()))) {
       const JobAction *SourceAction = cast<JobAction>(&Cmd->getSource());
-
+      if (FEA->getSerialActions().count(SourceAction)) {
+        C.addCommand(std::move(Cmd));
+        continue;
+      }
       ActionResult = CachedResults.at(
           {SourceAction,
            GetTriplePlusArchString(TC, BoundArch, TargetDeviceOffloadKind)}).front();
       InputInfoList InputInfos;
       CollectForEachInputs(InputInfos, SourceAction, TC, BoundArch,
-                           TargetDeviceOffloadKind, CachedResults);
+                           TargetDeviceOffloadKind, CachedResults, FEA);
       const Tool *Creator = &Cmd->getCreator();
+      StringRef ParallelJobs;
+      if (TargetDeviceOffloadKind == Action::OFK_SYCL)
+        ParallelJobs = C.getArgs().getLastArgValue(
+            options::OPT_fsycl_max_parallel_jobs_EQ);
 
       tools::SYCL::constructLLVMForeachCommand(
           C, *SourceAction, std::move(Cmd), InputInfos, ActionResult, Creator,
-          "", types::getTypeTempSuffix(ActionResult.getType()));
+          "", types::getTypeTempSuffix(ActionResult.getType()), ParallelJobs);
     }
     return { ActionResult };
   }
diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp
index bcfc8332714cb..f38a773413553 100644
--- a/clang/lib/Driver/ToolChains/Clang.cpp
+++ b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -4869,7 +4869,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
                     options::OPT_no_offload_new_driver, false));
 
   bool IsRDCMode =
-      Args.hasFlag(options::OPT_fgpu_rdc, options::OPT_fno_gpu_rdc, false);
+      Args.hasFlag(options::OPT_fgpu_rdc, options::OPT_fno_gpu_rdc, IsSYCL);
   bool IsUsingLTO = D.isUsingLTO(IsDeviceOffloadAction);
   auto LTOMode = D.getLTOMode(IsDeviceOffloadAction);
   bool IsFPGASYCLOffloadDevice =
@@ -6993,9 +6993,13 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
                       options::OPT_fno_hip_kernel_arg_name);
   }
 
-  if (IsCuda || IsHIP) {
+  if (IsCuda || IsHIP || IsSYCL) {
     if (IsRDCMode)
       CmdArgs.push_back("-fgpu-rdc");
+    else
+      CmdArgs.push_back("-fno-gpu-rdc");
+  }
+  if (IsCuda || IsHIP) {
     if (Args.hasFlag(options::OPT_fgpu_defer_diag,
                      options::OPT_fno_gpu_defer_diag, false))
       CmdArgs.push_back("-fgpu-defer-diag");
@@ -9842,6 +9846,13 @@ void FileTableTform::ConstructJob(Compilation &C, const JobAction &JA,
       addArgs(CmdArgs, TCArgs, {Arg});
       break;
     }
+    case FileTableTformJobAction::Tform::MERGE: {
+      assert(Tf.TheArgs.size() == 1 && "column name expected");
+      SmallString<128> Arg("-merge=");
+      Arg += Tf.TheArgs[0];
+      addArgs(CmdArgs, TCArgs, {Arg});
+      break;
+    }
     }
   }
 
diff --git a/clang/lib/Driver/ToolChains/SYCL.cpp b/clang/lib/Driver/ToolChains/SYCL.cpp
index 54bafca712b47..be7a92200c1ab 100644
--- a/clang/lib/Driver/ToolChains/SYCL.cpp
+++ b/clang/lib/Driver/ToolChains/SYCL.cpp
@@ -127,6 +127,11 @@ void SYCL::constructLLVMForeachCommand(Compilation &C, const JobAction &JA,
   C.addCommand(std::move(Cmd));
 }
 
+bool SYCL::shouldDoPerObjectFileLinking(const Compilation &C) {
+  return !C.getArgs().hasFlag(options::OPT_fgpu_rdc, options::OPT_fno_gpu_rdc,
+                              /*default=*/true);
+}
+
 // The list should match pre-built SYCL device library files located in
 // compiler package. Once we add or remove any SYCL device library files,
 // the list should be updated accordingly.
@@ -163,12 +168,25 @@ const char *SYCL::Linker::constructLLVMLinkCommand(
   // an actual object/archive.  Take that list and pass those to the linker
   // instead of the original object.
   if (JA.isDeviceOffloading(Action::OFK_SYCL)) {
-    auto isSYCLDeviceLib = [&C, this](const InputInfo &II) {
+    bool IsRDC = !shouldDoPerObjectFileLinking(C);
+    auto isNoRDCDeviceCodeLink = [&](const InputInfo &II) {
+      if (IsRDC)
+        return false;
+      if (II.getType() != clang::driver::types::TY_LLVM_BC)
+        return false;
+      if (InputFiles.size() != 2)
+        return false;
+      return &II == &InputFiles[1];
+    };
+    auto isSYCLDeviceLib = [&](const InputInfo &II) {
       const ToolChain *HostTC = C.getSingleOffloadToolChain<Action::OFK_Host>();
       StringRef LibPostfix = ".o";
       if (HostTC->getTriple().isWindowsMSVCEnvironment() &&
           C.getDriver().IsCLMode())
         LibPostfix = ".obj";
+      else if (isNoRDCDeviceCodeLink(II))
+        LibPostfix = ".bc";
+
       std::string FileName = this->getToolChain().getInputFilename(II);
       StringRef InputFilename = llvm::sys::path::filename(FileName);
       if (this->getToolChain().getTriple().isNVPTX()) {
@@ -183,9 +201,21 @@ const char *SYCL::Linker::constructLLVMLinkCommand(
           !InputFilename.endswith(LibPostfix) || (InputFilename.count('-') < 2))
         return false;
       // Skip the prefix "libsycl-"
-      StringRef PureLibName = InputFilename.substr(LibSyclPrefix.size());
+      std::string PureLibName =
+          InputFilename.substr(LibSyclPrefix.size()).str();
+      if (isNoRDCDeviceCodeLink(II)) {
+        // Skip the final - until the . because we linked all device libs into a
+        // single BC in a previous action so we have a temp file name.
+        auto FinalDashPos = PureLibName.find_last_of('-');
+        auto DotPos = PureLibName.find_last_of('.');
+        assert((FinalDashPos != std::string::npos &&
+                DotPos != std::string::npos) &&
+               "Unexpected filename");
+        PureLibName =
+            PureLibName.substr(0, FinalDashPos) + PureLibName.substr(DotPos);
+      }
       for (const auto &L : SYCLDeviceLibList) {
-        if (PureLibName.startswith(L))
+        if (StringRef(PureLibName).startswith(L))
           return true;
       }
       return false;
@@ -203,8 +233,17 @@ const char *SYCL::Linker::constructLLVMLinkCommand(
     for (const auto &II : InputFiles) {
       std::string FileName = getToolChain().getInputFilename(II);
       if (II.getType() == types::TY_Tempfilelist) {
-        // Pass the unbundled list with '@' to be processed.
-        Libs.push_back(C.getArgs().MakeArgString("@" + FileName));
+        if (IsRDC) {
+          // Pass the unbundled list with '@' to be processed.
+          Libs.push_back(C.getArgs().MakeArgString("@" + FileName));
+        } else {
+          assert(InputFiles.size() == 2 &&
+                 "Unexpected inputs for no-RDC with temp file list");
+          // If we're in no-RDC mode and the input is a temp file list,
+          // we want to link multiple object files each against device libs,
+          // so we should consider this input as an object and not pass '@'.
+          Objs.push_back(C.getArgs().MakeArgString(FileName));
+        }
       } else if (II.getType() == types::TY_Archive && !LinkSYCLDeviceLibs) {
         Libs.push_back(C.getArgs().MakeArgString(FileName));
       } else
diff --git a/clang/lib/Driver/ToolChains/SYCL.h b/clang/lib/Driver/ToolChains/SYCL.h
index fb78d2ab7e270..f5aef6c9e5ea8 100644
--- a/clang/lib/Driver/ToolChains/SYCL.h
+++ b/clang/lib/Driver/ToolChains/SYCL.h
@@ -38,7 +38,7 @@ void constructLLVMForeachCommand(Compilation &C, const JobAction &JA,
                                  const InputInfo &Output, const Tool *T,
                                  StringRef Increment, StringRef Ext = "out",
                                  StringRef ParallelJobs = "");
-
+bool shouldDoPerObjectFileLinking(const Compilation &C);
 // Runs llvm-spirv to convert spirv to bc, llvm-link, which links multiple LLVM
 // bitcode. Converts generated bc back to spirv using llvm-spirv, wraps with
 // offloading information. Finally compiles to object using llc
diff --git a/clang/lib/Frontend/CompilerInvocation.cpp b/clang/lib/Frontend/CompilerInvocation.cpp
index 10bfe7caf0991..cc023f93ece19 100644
--- a/clang/lib/Frontend/CompilerInvocation.cpp
+++ b/clang/lib/Frontend/CompilerInvocation.cpp
@@ -3605,6 +3605,9 @@ void CompilerInvocation::GenerateLangArgs(const LangOptions &Opts,
 
   if (!Opts.RandstructSeed.empty())
     GenerateArg(Args, OPT_frandomize_layout_seed_EQ, Opts.RandstructSeed, SA);
+
+  if (!Opts.GPURelocatableDeviceCode)
+    GenerateArg(Args, OPT_fno_gpu_rdc, SA);
 }
 
 bool CompilerInvocation::ParseLangArgs(LangOptions &Opts, ArgList &Args,
@@ -4200,6 +4203,11 @@ bool CompilerInvocation::ParseLangArgs(LangOptions &Opts, ArgList &Args,
       Diags.Report(diag::err_drv_hlsl_unsupported_target) << T.str();
   }
 
+  // GPURelocatableDeviceCode should be true for SYCL if not specified.
+  if (Args.hasArg(OPT_fsycl_is_device) || Args.hasArg(OPT_fsycl_is_host))
+    Opts.GPURelocatableDeviceCode = Args.hasFlag(
+        options::OPT_fgpu_rdc, options::OPT_fno_gpu_rdc, /*default=*/true);
+
   return Diags.getNumErrors() == NumErrorsBefore;
 }
 
diff --git a/clang/lib/Sema/SemaDecl.cpp b/clang/lib/Sema/SemaDecl.cpp
index 8d74734b024ab..44aff4b6c0dff 100644
--- a/clang/lib/Sema/SemaDecl.cpp
+++ b/clang/lib/Sema/SemaDecl.cpp
@@ -10239,6 +10239,15 @@ Sema::ActOnFunctionDeclarator(Scope *S, Declarator &D, DeclContext *DC,
     }
   }
 
+  if (getLangOpts().SYCLIsDevice && !getLangOpts().GPURelocatableDeviceCode &&
+      NewFD->hasAttr<SYCLDeviceAttr>() &&
+      !getSourceManager().isInSystemHeader(NewFD->getLocation())) {
+    Diag(NewFD->getLocation(), diag::err_sycl_external_no_rdc)
+        << (D.getFunctionDefinitionKind() ==
+            clang::FunctionDefinitionKind::Definition);
+    NewFD->setInvalidDecl();
+  }
+
   if (!getLangOpts().CPlusPlus) {
     // Perform semantic checking on the function declaration.
     if (!NewFD->isInvalidDecl() && NewFD->isMain())
diff --git a/clang/test/Driver/sycl-no-rdc-errors.cpp b/clang/test/Driver/sycl-no-rdc-errors.cpp
new file mode 100644
index 0000000000000..3506333efcc9e
--- /dev/null
+++ b/clang/test/Driver/sycl-no-rdc-errors.cpp
@@ -0,0 +1,12 @@
+/// Tests driver errors for -no-sycl-rdc
+
+// RUN: %clang -target %itanium_abi_triple -c %s -o %t.o
+// RUN: %clang -target spir64_gen -emit-llvm -c %s -o %t
+// RUN: clang-offload-bundler -type=o -targets=host-%itanium_abi_triple,sycl-spir64_gen-unknown-unknown -input=%t -input=%t.o -output=%t.fat.o
+// RUN: not %clang -### -fsycl -fno-sycl-rdc %t.fat.o 2>&1 | FileCheck -check-prefix=CHECK-ARCH %s
+
+// CHECK-ARCH: error: linked binaries do not contain expected 'spir64-unknown-unknown' target; found targets: 'spir64_gen-unknown-unknown', this is not supported with '-fno-sycl-rdc'
+
+// Some code so that we can create a binary out of this file.
+void test_func(void) {
+}
diff --git a/clang/test/Driver/sycl-no-rdc-fat-archive.cpp b/clang/test/Driver/sycl-no-rdc-fat-archive.cpp
new file mode 100644
index 0000000000000..9caac166bccac
--- /dev/null
+++ b/clang/test/Driver/sycl-no-rdc-fat-archive.cpp
@@ -0,0 +1,29 @@
+/// test behaviors of passing a fat static lib with -fno-sycl-rdc
+// Build a fat static lib that will be used for all tests
+// RUN: echo "void foo(void) {}" > %t1.cpp
+// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl %t1.cpp -c -o %t1_bundle.o
+// RUN: llvm-ar cr %t_lib.a %t1_bundle.o
+// RUN: %clang -### -fsycl -fno-sycl-rdc -fsycl-device-code-split=none --sysroot=%S/Inputs/SYCL %t_lib.a 2>&1 -ccc-print-phases | FileCheck %s
+// RUN: %clang -### -fsycl -fno-sycl-rdc -fsycl-device-code-split=auto --sysroot=%S/Inputs/SYCL %t_lib.a 2>&1 -ccc-print-phases | FileCheck %s
+// RUN: %clang -### -fsycl -fno-sycl-rdc -fsycl-device-code-split=per_kernel --sysroot=%S/Inputs/SYCL %t_lib.a 2>&1 -ccc-print-phases | FileCheck %s
+// RUN: %clang -### -fsycl -fno-sycl-rdc -fsycl-device-code-split=per_source --sysroot=%S/Inputs/SYCL %t_lib.a 2>&1 -ccc-print-phases | FileCheck %s
+// CHECK: 2: input, "{{.*}}_lib.a", archive
+// CHECK: 3: clang-offload-unbundler, {2}, tempfilelist
+// CHECK: 4: spirv-to-ir-wrapper, {3}, tempfilelist, (device-sycl)
+// CHECK: 5: input, "{{.*}}libsycl-crt{{.*}}", object
+// CHECK: 6: clang-offload-unbundler, {5}, object
+// CHECK: 7: offload, " (spir64-unknown-unknown)" {6}, object
+// CHECK: 65: linker, {7, {{.*}}}, ir, (device-sycl)
+// CHECK: 66: linker, {4, 65}, ir, (device-sycl)
+// CHECK: 67: foreach, {4, 66}, ir, (device-sycl)
+// CHECK: 68: file-table-tform, {4, 67}, tempfilelist, (device-sycl)
+// CHECK: 69: sycl-post-link, {68}, tempfiletable, (device-sycl)
+// CHECK: 70: foreach, {68, 69}, tempfiletable, (device-sycl)
+// CHECK: 71: file-table-tform, {70}, tempfilelist, (device-sycl)
+// CHECK: 72: file-table-tform, {70}, tempfilelist, (device-sycl)
+// CHECK: 73: foreach, {68, 72}, tempfilelist, (device-sycl)
+// CHECK: 74: file-table-tform, {73}, tempfilelist, (device-sycl)
+// CHECK: 75: llvm-spirv, {74}, tempfilelist, (device-sycl)
+// CHECK: 76: file-table-tform, {71, 75}, tempfiletable, (device-sycl)
+// CHECK: 77: clang-offload-wrapper, {76}, object, (device-sycl)
+// CHECK: 78: offload, "host-sycl (x86_64-unknown-linux-gnu)" {1}, "device-sycl (spir64-unknown-unknown)" {77}, image
diff --git a/clang/test/Driver/sycl-no-rdc.cpp b/clang/test/Driver/sycl-no-rdc.cpp
new file mode 100644
index 0000000000000..22f8135f987fc
--- /dev/null
+++ b/clang/test/Driver/sycl-no-rdc.cpp
@@ -0,0 +1,31 @@
+/// Tests for -fno-sycl-rdc
+// RUN: touch %t1.cpp
+// RUN: touch %t2.cpp
+// RUN: %clang -### -fsycl -fno-sycl-rdc --sysroot=%S/Inputs/SYCL %t1.cpp %t2.cpp 2>&1 -ccc-print-phases | FileCheck %s
+
+// CHECK: 3: input, "{{.*}}1.cpp", c++, (device-sycl)
+// CHECK: 4: preprocessor, {3}, c++-cpp-output, (device-sycl)
+// CHECK: 5: compiler, {4}, ir, (device-sycl)
+// CHECK: 13: input, "{{.*}}2.cpp", c++, (device-sycl)
+// CHECK: 14: preprocessor, {13}, c++-cpp-output, (device-sycl)
+// CHECK: 15: compiler, {14}, ir, (device-sycl)
+
+// CHECK: 21: input, {{.*}}libsycl-crt{{.*}}, object
+// CHECK: 22: clang-offload-unbundler, {21}, object
+// CHECK: 23: offload, " (spir64-unknown-unknown)" {22}, object
+// CHECK: 81: linker, {23, {{.*}}}, ir, (device-sycl)
+// CHECK: 82: linker, {5, 81}, ir, (device-sycl)
+// CHECK: 83: sycl-post-link, {82}, tempfiletable, (device-sycl)
+// CHECK: 84: file-table-tform, {83}, tempfilelist, (device-sycl)
+// CHECK: 85: llvm-spirv, {84}, tempfilelist, (device-sycl)
+// CHECK: 86: file-table-tform, {83, 85}, tempfiletable, (device-sycl)
+// CHECK: 87: clang-offload-wrapper, {86}, object, (device-sycl)
+
+// CHECK: 88: linker, {15, 81}, ir, (device-sycl)
+// CHECK: 89: sycl-post-link, {88}, tempfiletable, (device-sycl)
+// CHECK: 90: file-table-tform, {89}, tempfilelist, (device-sycl)
+// CHECK: 91: llvm-spirv, {90}, tempfilelist, (device-sycl)
+// CHECK: 92: file-table-tform, {89, 91}, tempfiletable, (device-sycl)
+// CHECK: 93: clang-offload-wrapper, {92}, object, (device-sycl)
+
+// CHECK: 94: offload, "host-sycl (x86_64-unknown-linux-gnu)" {{{.*}}}, "device-sycl (spir64-unknown-unknown)" {87}, "device-sycl (spir64-unknown-unknown)" {93}, image
diff --git a/clang/test/SemaSYCL/sycl-no-rdc.cpp b/clang/test/SemaSYCL/sycl-no-rdc.cpp
new file mode 100644
index 0000000000000..b3466ee8fa396
--- /dev/null
+++ b/clang/test/SemaSYCL/sycl-no-rdc.cpp
@@ -0,0 +1,24 @@
+// RUN: %clang_cc1 -fsycl-is-device -verify -fsyntax-only -fno-gpu-rdc -internal-isystem %S/Inputs %s
+
+// Check that declarations of SYCL_EXTERNAL functions throw an error if -fno-gpu-rdc is passed
+#include "sycl.hpp"
+
+// expected-error@+1{{invalid declaration of SYCL_EXTERNAL function in non-relocatable device code mode}}
+SYCL_EXTERNAL void syclExternalDecl();
+
+// expected-error@+1{{invalid definition of SYCL_EXTERNAL function in non-relocatable device code mode}}
+SYCL_EXTERNAL void syclExternalDefn() {}
+
+using namespace sycl;
+queue q;
+
+void kernel_wrapper() {
+  q.submit([&](handler &h) {
+    h.single_task([=] {
+    });
+  });
+}
+
+int main() {
+  kernel_wrapper();
+}
diff --git a/llvm/include/llvm/Support/SimpleTable.h b/llvm/include/llvm/Support/SimpleTable.h
index e91f288a6bc26..1881f080736c4 100644
--- a/llvm/include/llvm/Support/SimpleTable.h
+++ b/llvm/include/llvm/Support/SimpleTable.h
@@ -132,6 +132,9 @@ class SimpleTable {
   Row &operator[](int I) { return Rows[I]; }
   const Row &operator[](int I) const { return Rows[I]; }
 
+  // Merge another table into this table
+  Error merge(const SimpleTable &Other);
+
 private:
   Error addColumnName(StringRef ColName);
   void rebuildName2NumMapping();
diff --git a/llvm/lib/Support/SimpleTable.cpp b/llvm/lib/Support/SimpleTable.cpp
index 41aeb57b15d28..b75f65d3bf6d6 100644
--- a/llvm/lib/Support/SimpleTable.cpp
+++ b/llvm/lib/Support/SimpleTable.cpp
@@ -251,5 +251,14 @@ Expected<SimpleTable::UPtrTy> SimpleTable::read(const Twine &FileName,
   return read(MemBuf->get(), ColSep);
 }
 
+Error SimpleTable::merge(const SimpleTable &Other) {
+  if (getNumColumns() != Other.getNumColumns())
+    return makeError("different number of columns");
+  if (ColumnNames != Other.ColumnNames)
+    return makeError("different column names");
+  Rows.insert(Rows.end(), Other.Rows.begin(), Other.Rows.end());
+  return Error::success();
+}
+
 } // namespace util
 } // namespace llvm
diff --git a/llvm/test/tools/file-table-tform/Inputs/a1.txt b/llvm/test/tools/file-table-tform/Inputs/a1.txt
new file mode 100644
index 0000000000000..a26424e50e980
--- /dev/null
+++ b/llvm/test/tools/file-table-tform/Inputs/a1.txt
@@ -0,0 +1,3 @@
+[11|12]
+44|55
+66|77
diff --git a/llvm/test/tools/file-table-tform/Inputs/merge-gold.txt b/llvm/test/tools/file-table-tform/Inputs/merge-gold.txt
new file mode 100644
index 0000000000000..e81ea3c87ecfe
--- /dev/null
+++ b/llvm/test/tools/file-table-tform/Inputs/merge-gold.txt
@@ -0,0 +1,5 @@
+[11|12]
+00|11
+22|33
+44|55
+66|77
diff --git a/llvm/test/tools/file-table-tform/Inputs/merge-input.txt b/llvm/test/tools/file-table-tform/Inputs/merge-input.txt
new file mode 100644
index 0000000000000..67866fe6602ea
--- /dev/null
+++ b/llvm/test/tools/file-table-tform/Inputs/merge-input.txt
@@ -0,0 +1,3 @@
+[One|Two]
+0|a0.txt
+1|a1.txt
diff --git a/llvm/test/tools/file-table-tform/file-table-tform-merge.test b/llvm/test/tools/file-table-tform/file-table-tform-merge.test
new file mode 100644
index 0000000000000..f81f34b2d33c9
--- /dev/null
+++ b/llvm/test/tools/file-table-tform/file-table-tform-merge.test
@@ -0,0 +1,6 @@
+-- Merge two tables listed in merge-input.txt
+RUN: cp %S/Inputs/a{0,1}.txt .
+RUN: file-table-tform --merge=Two %S/Inputs/merge-input.txt -o merge-output.txt
+
+-- Verify result
+RUN: diff %S/Inputs/merge-gold.txt merge-output.txt
diff --git a/llvm/tools/file-table-tform/file-table-tform.cpp b/llvm/tools/file-table-tform/file-table-tform.cpp
index 1ded228953a00..c1add76ad1616 100644
--- a/llvm/tools/file-table-tform/file-table-tform.cpp
+++ b/llvm/tools/file-table-tform/file-table-tform.cpp
@@ -82,6 +82,7 @@ static constexpr char OPT_REPLACE_CELL[] = "replace_cell";
 static constexpr char OPT_RENAME[] = "rename";
 static constexpr char OPT_EXTRACT[] = "extract";
 static constexpr char OPT_COPY_SINGLE_FILE[] = "copy_single_file";
+static constexpr char OPT_MERGE[] = "merge";
 
 static cl::list<std::string> TformReplace{
     OPT_REPLACE, cl::ZeroOrMore, cl::desc("replace a column"),
@@ -107,6 +108,10 @@ static cl::list<std::string> TformCopySingleFile{
     cl::value_desc("<column name or ordinal>,<row id ordinal>"),
     cl::cat(FileTableTformCat)};
 
+static cl::list<std::string> TformMerge{
+    OPT_MERGE, cl::Optional, cl::desc("merge all input tables"),
+    cl::value_desc("<column name or ordinal>"), cl::cat(FileTableTformCat)};
+
 static cl::opt<bool> DropTitles{"drop_titles", cl::Optional,
                                 cl::desc("drop column titles"),
                                 cl::cat(FileTableTformCat)};
@@ -172,7 +177,8 @@ struct TformCmd {
             .Case(OPT_RENAME, [&](TformCmd *Cmd) { return Error::success(); })
             .Case(OPT_EXTRACT, [&](TformCmd *Cmd) { return Error::success(); })
             .Case(OPT_COPY_SINGLE_FILE,
-                  [&](TformCmd *Cmd) { return Error::success(); });
+                  [&](TformCmd *Cmd) { return Error::success(); })
+            .Case(OPT_MERGE, [&](TformCmd *Cmd) { return Error::success(); });
     return F(this);
   }
 
@@ -234,15 +240,24 @@ struct TformCmd {
                             std::back_inserter(Args));
                   return Error::success();
                 })
-            .Case(OPT_COPY_SINGLE_FILE, [&](TformCmd *Cmd) -> Error {
-              // argument is <column name>,<row index>
+            .Case(OPT_COPY_SINGLE_FILE,
+                  [&](TformCmd *Cmd) -> Error {
+                    // argument is <column name>,<row index>
+                    if (Arg.empty())
+                      return makeUserError("empty argument in " +
+                                           Twine(OPT_COPY_SINGLE_FILE));
+                    Arg.split(Args, ',');
+                    if (Args.size() != 2 || Args[0].empty() || Args[1].empty())
+                      return makeUserError("invalid argument in " +
+                                           Twine(OPT_COPY_SINGLE_FILE));
+                    return Error::success();
+                  })
+            .Case(OPT_MERGE, [&](TformCmd *Cmd) -> Error {
               if (Arg.empty())
-                return makeUserError("empty argument in " +
-                                     Twine(OPT_COPY_SINGLE_FILE));
+                return makeUserError("empty argument in " + Twine(OPT_MERGE));
               Arg.split(Args, ',');
-              if (Args.size() != 2 || Args[0].empty() || Args[1].empty())
-                return makeUserError("invalid argument in " +
-                                     Twine(OPT_COPY_SINGLE_FILE));
+              if (Args.size() != 1 || Args[0].empty())
+                return makeUserError("invalid argument in " + Twine(OPT_MERGE));
               return Error::success();
             });
     return F(this);
@@ -286,23 +301,63 @@ struct TformCmd {
                     Error Res = Table.peelColumns(Args);
                     return Res ? std::move(Res) : std::move(Error::success());
                   })
-            .Case(OPT_COPY_SINGLE_FILE, [&](TformCmd *Cmd) -> Error {
-              // argument is <column name>,<row index>
-              assert(Args.size() == 2);
-              const int Row = std::stoi(Args[1].str());
-              if (Row >= Table.getNumRows())
-                return makeUserError("row index is out of bounds");
+            .Case(OPT_COPY_SINGLE_FILE,
+                  [&](TformCmd *Cmd) -> Error {
+                    // argument is <column name>,<row index>
+                    assert(Args.size() == 2);
+                    const int Row = std::stoi(Args[1].str());
+                    if (Row >= Table.getNumRows())
+                      return makeUserError("row index is out of bounds");
 
-              // Copy the file from the only remaining row at specified
-              // column
-              StringRef FileToCopy = Table[Row].getCell(Args[0], "");
+                    // Copy the file from the only remaining row at specified
+                    // column
+                    StringRef FileToCopy = Table[Row].getCell(Args[0], "");
 
-              if (FileToCopy.empty())
-                return makeUserError("no file found in specified column");
+                    if (FileToCopy.empty())
+                      return makeUserError("no file found in specified column");
 
+                    std::error_code EC = sys::fs::copy_file(FileToCopy, Output);
+                    return EC ? createFileError(Output, EC)
+                              : std::move(Error::success());
+                  })
+            .Case(OPT_MERGE, [&](TformCmd *Cmd) -> Error {
+              StringRef FileToCopy = Table[0].getCell(Args[0]);
+              if (!llvm::sys::fs::exists(FileToCopy))
+                return makeUserError("first file not found");
               std::error_code EC = sys::fs::copy_file(FileToCopy, Output);
-              return EC ? createFileError(Output, EC)
-                        : std::move(Error::success());
+              if (EC)
+                return createFileError(Output, EC);
+
+              Expected<util::SimpleTable::UPtrTy> OutputTable =
+                  util::SimpleTable::read(Output);
+              if (!OutputTable)
+                return OutputTable.takeError();
+
+              // The first table will be the table merged into, so start merging
+              // from the second table.
+              for (int Row = 1; Row < Table.getNumRows(); Row++) {
+                // Each cell in the input table is a path
+                // to another table.
+                Expected<util::SimpleTable::UPtrTy> Table1 =
+                    util::SimpleTable::read(Table[Row].getCell(Args[0]));
+                if (!Table1)
+                  return Table1.takeError();
+
+                Error Err = (*OutputTable)->merge(*Table1.get());
+                if (Err)
+                  return Err;
+              }
+              raw_fd_ostream Out{Output, EC, sys::fs::OpenFlags::OF_None};
+
+              if (EC)
+                return createFileError(Output, EC);
+
+              (*OutputTable)->write(Out, (*OutputTable)->getNumColumns() > 1);
+
+              if (Out.has_error())
+                return createFileError(Output, Out.error());
+              Out.close();
+              return Error::success();
             });
     return F(this);
   }
@@ -344,9 +399,10 @@ int main(int argc, char **argv) {
   // yet, as an order across all command line options-commands needs to be
   // established first to properly map inputs to commands.
 
-  auto Lists = {std::addressof(TformReplace), std::addressof(TformReplaceCell),
-                std::addressof(TformRename), std::addressof(TformExtract),
-                std::addressof(TformCopySingleFile)};
+  auto Lists = {
+      std::addressof(TformReplace),        std::addressof(TformReplaceCell),
+      std::addressof(TformRename),         std::addressof(TformExtract),
+      std::addressof(TformCopySingleFile), std::addressof(TformMerge)};
 
   for (const auto *L : Lists) {
     for (auto It = L->begin(); It != L->end(); It++) {
@@ -369,12 +425,17 @@ int main(int argc, char **argv) {
 
   // ensure that if copy_single_file is specified, it must be the last tform
   bool HasCopySingleFileTform = false;
+  bool HasMerge = false;
   for (auto &P : Cmds) {
     if (HasCopySingleFileTform) {
       CHECK_AND_EXIT(
           makeUserError("copy_single_file must be the last transformation"));
     }
+    if (HasMerge) {
+      CHECK_AND_EXIT(makeUserError("merge must be the last transformation"));
+    }
     HasCopySingleFileTform = P.second->Kind == OPT_COPY_SINGLE_FILE;
+    HasMerge = P.second->Kind == OPT_MERGE;
   }
 
   for (auto &P : Cmds) {
@@ -397,9 +458,9 @@ int main(int argc, char **argv) {
     CHECK_AND_EXIT(std::move(Res));
   }
 
-  // If copy_single_file was specified the output file is generated by the
-  // corresponding transformation.
-  if (!HasCopySingleFileTform) {
+  // If copy_single_file or merge was specified the output file is generated by
+  // the corresponding transformation.
+  if (!HasCopySingleFileTform && !HasMerge) {
     // Write the transformed table to file
     std::error_code EC;
     raw_fd_ostream Out{Output, EC, sys::fs::OpenFlags::OF_None};