diff --git a/clang/lib/Driver/Driver.cpp b/clang/lib/Driver/Driver.cpp
index 475c1fddb036..00463e7dfac9 100644
--- a/clang/lib/Driver/Driver.cpp
+++ b/clang/lib/Driver/Driver.cpp
@@ -5016,14 +5016,19 @@ class OffloadingActionBuilder final {
                llvm::zip(SYCLDeviceActions, SYCLTargetInfoList)) {
             Action *&A = std::get<0>(TargetActionInfo);
             auto &TargetInfo = std::get<1>(TargetActionInfo);
-            A = C.getDriver().ConstructPhaseAction(C, Args, CurPhase, A,
-                                                   AssociatedOffloadKind);
-            if (SYCLDeviceOnly)
+            Action *PreprocAction = C.getDriver().ConstructPhaseAction(
+                C, Args, CurPhase, A, AssociatedOffloadKind);
+            if (SYCLDeviceOnly) {
+              A = PreprocAction;
               continue;
+            }
             // Add an additional compile action to generate the integration
-            // header.
+            // header. This action compiles the source file instead of the
+            // generated preprocessed file to allow for control of the
+            // diagnostics that could come from the system headers.
             Action *CompileAction =
                 C.MakeAction<CompileJobAction>(A, types::TY_Nothing);
+            A = PreprocAction;
             DA.add(*CompileAction, *TargetInfo.TC, TargetInfo.BoundArch,
                    Action::OFK_SYCL);
           }
@@ -7974,8 +7979,15 @@ Action *Driver::BuildOffloadingActions(Compilation &C,
               if (isa<PreprocessJobAction>(A)) {
                 PackagerActions.push_back(OA);
                 A->setCannotBeCollapsedWithNextDependentAction();
-                Action *CompileAction =
-                    C.MakeAction<CompileJobAction>(A, types::TY_Nothing);
+                // The input to the compilation job is the preprocessed job.
+                // Take that input action (it should be one input) which is
+                // the source file and compile that file to generate the
+                // integration header/footer.
+                ActionList PreprocInputs = A->getInputs();
+                assert(PreprocInputs.size() == 1 &&
+                       "Single input size to preprocess action expected.");
+                Action *CompileAction = C.MakeAction<CompileJobAction>(
+                    PreprocInputs.front(), types::TY_Nothing);
                 DDeps.add(*CompileAction, *TC, BoundArch, Action::OFK_SYCL);
               }
             });
diff --git a/clang/test/Driver/sycl-offload-nvptx.cpp b/clang/test/Driver/sycl-offload-nvptx.cpp
index d2dcbada56de..0d69ce94dd95 100644
--- a/clang/test/Driver/sycl-offload-nvptx.cpp
+++ b/clang/test/Driver/sycl-offload-nvptx.cpp
@@ -106,7 +106,7 @@
 // RUN: | FileCheck -check-prefix=CHK-PREPROC %s
 // CHK-PREPROC: 1: preprocessor, {0}, c++-cpp-output, (device-sycl, sm_[[CUDA_VERSION:[0-9.]+]])
 // CHK-PREPROC: 2: offload, "device-sycl (nvptx64-nvidia-cuda:sm_[[CUDA_VERSION]])" {1}, c++-cpp-output
-// CHK-PREPROC: 4: compiler, {1}, none, (device-sycl, sm_[[CUDA_VERSION]])
+// CHK-PREPROC: 4: compiler, {0}, none, (device-sycl, sm_[[CUDA_VERSION]])
 //
 // RUN: not %clangxx -### -std=c++11 -target x86_64-unknown-linux-gnu -fsycl \
 // RUN: -fsycl-targets=nvptx64-nvidia-cuda --cuda-path=%S/Inputs/no/CUDA/path/here \
diff --git a/clang/test/Driver/sycl-preprocess-old-model.cpp b/clang/test/Driver/sycl-preprocess-old-model.cpp
index 08c93a154d3e..efe968b7b298 100644
--- a/clang/test/Driver/sycl-preprocess-old-model.cpp
+++ b/clang/test/Driver/sycl-preprocess-old-model.cpp
@@ -22,7 +22,7 @@
 // PREPROC_PHASES: 1: preprocessor, {0}, c++-cpp-output, (device-sycl)
 // PREPROC_PHASES: 2: offload, "device-sycl (spir64-unknown-unknown)" {1}, c++-cpp-output
 // PREPROC_PHASES: 3: input, "[[INPUT]]", c++, (host-sycl)
-// PREPROC_PHASES: 4: compiler, {1}, none, (device-sycl)
+// PREPROC_PHASES: 4: compiler, {0}, none, (device-sycl)
 // PREPROC_PHASES: 5: offload, "host-sycl (x86_64-unknown-linux-gnu)" {3}, "device-sycl (spir64-unknown-unknown)" {4}, c++
 // PREPROC_PHASES: 6: preprocessor, {5}, c++-cpp-output, (host-sycl)
 // PREPROC_PHASES: 7: clang-offload-bundler, {2, 6}, c++-cpp-output, (host-sycl)
diff --git a/clang/test/Driver/sycl-preprocess.cpp b/clang/test/Driver/sycl-preprocess.cpp
index 91fd801e586d..41f037e246d7 100644
--- a/clang/test/Driver/sycl-preprocess.cpp
+++ b/clang/test/Driver/sycl-preprocess.cpp
@@ -24,7 +24,7 @@
 // PREPROC_PHASES: 1: preprocessor, {0}, c++-cpp-output, (host-sycl)
 // PREPROC_PHASES: 2: input, "[[INPUT]]", c++, (device-sycl)
 // PREPROC_PHASES: 3: preprocessor, {2}, c++-cpp-output, (device-sycl)
-// PREPROC_PHASES: 4: compiler, {3}, none, (device-sycl)
+// PREPROC_PHASES: 4: compiler, {2}, none, (device-sycl)
 // PREPROC_PHASES: 5: offload, "device-sycl (spir64-unknown-unknown)" {3}, c++-cpp-output
 // PREPROC_PHASES: 6: llvm-offload-binary, {5, 1}, c++-cpp-output
 // PREPROC_PHASES: 7: offload, "host-sycl ([[TARGET]])" {1}, "device-sycl (spir64-unknown-unknown)" {3}, "device-sycl (spir64-unknown-unknown)" {4}, " ([[TARGET]])" {6}, c++-cpp-output