Index: clang/include/clang/Driver/Options.td
===================================================================
--- clang/include/clang/Driver/Options.td
+++ clang/include/clang/Driver/Options.td
@@ -697,6 +697,8 @@
   HelpText<"Emit Clang AST files for source inputs">;
 def emit_llvm : Flag<["-"], "emit-llvm">, Flags<[CC1Option]>, Group<Action_Group>,
   HelpText<"Use the LLVM representation for assembler and object files">;
+def emit_pch : Flag<["-"], "emit-pch">, Flags<[CC1Option]>, Group<Action_Group>,
+  HelpText<"Generate pre-compiled header file">;
 def emit_interface_stubs : Flag<["-"], "emit-interface-stubs">, Flags<[CC1Option]>, Group<Action_Group>,
   HelpText<"Generate Interface Stub Files.">;
 def emit_merged_ifs : Flag<["-"], "emit-merged-ifs">,
@@ -4113,8 +4115,6 @@
   HelpText<"Generate pre-compiled module file from a C++ module interface">;
 def emit_header_module : Flag<["-"], "emit-header-module">,
   HelpText<"Generate pre-compiled module file from a set of header files">;
-def emit_pch : Flag<["-"], "emit-pch">,
-  HelpText<"Generate pre-compiled header file">;
 def emit_llvm_bc : Flag<["-"], "emit-llvm-bc">,
   HelpText<"Build ASTs then convert to LLVM, emit .bc file">;
 def emit_llvm_only : Flag<["-"], "emit-llvm-only">,
Index: clang/include/clang/Driver/Types.def
===================================================================
--- clang/include/clang/Driver/Types.def
+++ clang/include/clang/Driver/Types.def
@@ -42,8 +42,8 @@
 TYPE("cuda",                     CUDA,         PP_CUDA,         "cu",     phases::Preprocess, phases::Compile, phases::Backend, phases::Assemble, phases::Link)
 TYPE("cuda",                     CUDA_DEVICE,  PP_CUDA,         "cu",     phases::Preprocess, phases::Compile, phases::Backend, phases::Assemble, phases::Link)
 TYPE("hip-cpp-output",           PP_HIP,       INVALID,         "cui",    phases::Compile, phases::Backend, phases::Assemble, phases::Link)
-TYPE("hip",                      HIP,          PP_HIP,          "cu",     phases::Preprocess, phases::Compile, phases::Backend, phases::Assemble, phases::Link)
-TYPE("hip",                      HIP_DEVICE,   PP_HIP,          "cu",     phases::Preprocess, phases::Compile, phases::Backend, phases::Assemble, phases::Link)
+TYPE("hip",                      HIP,          PP_HIP,          "hip",    phases::Preprocess, phases::Compile, phases::Backend, phases::Assemble, phases::Link)
+TYPE("hip",                      HIP_DEVICE,   PP_HIP,          "hip",    phases::Preprocess, phases::Compile, phases::Backend, phases::Assemble, phases::Link)
 TYPE("objective-c-cpp-output",   PP_ObjC,      INVALID,         "mi",     phases::Compile, phases::Backend, phases::Assemble, phases::Link)
 TYPE("objc-cpp-output",          PP_ObjC_Alias, INVALID,        "mi",     phases::Compile, phases::Backend, phases::Assemble, phases::Link)
 TYPE("objective-c",              ObjC,         PP_ObjC,         "m",      phases::Preprocess, phases::Compile, phases::Backend, phases::Assemble, phases::Link)
@@ -66,6 +66,8 @@
 TYPE("objective-c++-header",     ObjCXXHeader, PP_ObjCXXHeader, "h",      phases::Preprocess, phases::Precompile)
 TYPE("c++-module",               CXXModule,    PP_CXXModule,    "cppm",   phases::Preprocess, phases::Precompile, phases::Compile, phases::Backend, phases::Assemble, phases::Link)
 TYPE("c++-module-cpp-output",    PP_CXXModule, INVALID,         "iim",    phases::Precompile, phases::Compile, phases::Backend, phases::Assemble, phases::Link)
+TYPE("hip-header-cpp-output",    PP_HIPHeader, INVALID,         "hipii",  phases::Precompile)
+TYPE("hip-header",               HIPHeader,    PP_HIPHeader,    "hiphh",  phases::Preprocess, phases::Precompile)
 
 // Other languages.
 TYPE("ada",                      Ada,          INVALID,         nullptr,  phases::Compile, phases::Backend, phases::Assemble, phases::Link)
Index: clang/lib/Driver/Driver.cpp
===================================================================
--- clang/lib/Driver/Driver.cpp
+++ clang/lib/Driver/Driver.cpp
@@ -292,7 +292,8 @@
     FinalPhase = phases::Preprocess;
 
   // --precompile only runs up to precompilation.
-  } else if ((PhaseArg = DAL.getLastArg(options::OPT__precompile))) {
+  } else if ((PhaseArg = DAL.getLastArg(options::OPT__precompile)) ||
+             (PhaseArg = DAL.getLastArg(options::OPT_emit_pch))) {
     FinalPhase = phases::Precompile;
 
   // -{fsyntax-only,-analyze,emit-ast} only run up to the compiler.
@@ -2394,6 +2395,7 @@
     bool CompileDeviceOnly = false;
     bool EmitLLVM = false;
     bool EmitAsm = false;
+    bool EmitPCH = false;
 
     /// ID to identify each device compilation. For CUDA it is simply the
     /// GPU arch string. For HIP it is either the GPU arch string or GPU
@@ -2445,7 +2447,8 @@
         // If the host input is not CUDA or HIP, we don't need to bother about
         // this input.
         if (IA->getType() != types::TY_CUDA &&
-            IA->getType() != types::TY_HIP) {
+            IA->getType() != types::TY_HIP &&
+            IA->getType() != types::TY_HIPHeader) {
           // The builder will ignore this input.
           IsActive = false;
           return ABRT_Inactive;
@@ -2458,8 +2461,11 @@
           return ABRT_Success;
 
         // Replicate inputs for each GPU architecture.
-        auto Ty = IA->getType() == types::TY_HIP ? types::TY_HIP_DEVICE
-                                                 : types::TY_CUDA_DEVICE;
+        auto Ty = IA->getType();
+        if(Ty != types::TY_HIPHeader)
+          Ty = Ty == types::TY_HIP ?
+              types::TY_HIP_DEVICE : types::TY_CUDA_DEVICE;
+
         for (unsigned I = 0, E = GpuArchList.size(); I != E; ++I) {
           CudaDeviceActions.push_back(
               C.MakeAction<InputAction>(IA->getInputArg(), Ty));
@@ -2584,6 +2590,7 @@
                               options::OPT_cuda_device_only);
       EmitLLVM = Args.getLastArg(options::OPT_emit_llvm);
       EmitAsm = Args.getLastArg(options::OPT_S);
+      EmitPCH = Args.getLastArg(options::OPT_emit_pch);
 
       // Collect all cuda_gpu_arch parameters, removing duplicates.
       std::set<StringRef> GpuArchs;
@@ -2806,6 +2813,24 @@
       if (CudaDeviceActions.empty())
         return ABRT_Success;
 
+      // In EmitPCH mode, construct Phase Actions up to Precompile,
+      // and ignore phases coming afterwards.
+      // Otherwise, ignore the Precompile phase.
+      if (EmitPCH) {
+        if (CurPhase == phases::Precompile) {
+          for (Action *&A : CudaDeviceActions) {
+            A = C.getDriver().ConstructPhaseAction(
+                C, Args, phases::Precompile, A, AssociatedOffloadKind);
+          }
+          return CompileDeviceOnly ? ABRT_Ignore_Host : ABRT_Success;
+        }
+        if (CurPhase >= phases::Compile)
+          return CompileDeviceOnly ? ABRT_Ignore_Host : ABRT_Success;
+      } else {
+        if (CurPhase == phases::Precompile)
+          return ABRT_Success;
+      }
+
       assert(((CurPhase == phases::Link && Relocatable) ||
               CudaDeviceActions.size() == GpuArchList.size()) &&
              "Expecting one action per GPU architecture.");
Index: clang/lib/Driver/ToolChains/Clang.cpp
===================================================================
--- clang/lib/Driver/ToolChains/Clang.cpp
+++ clang/lib/Driver/ToolChains/Clang.cpp
@@ -6318,6 +6318,7 @@
 
   // Disable warnings for clang -E -emit-llvm foo.c
   Args.ClaimAllArgs(options::OPT_emit_llvm);
+  Args.ClaimAllArgs(options::OPT_emit_pch);
 }
 
 Clang::Clang(const ToolChain &TC)
Index: clang/lib/Driver/Types.cpp
===================================================================
--- clang/lib/Driver/Types.cpp
+++ clang/lib/Driver/Types.cpp
@@ -103,7 +103,7 @@
       TY_Plist,         TY_RewrittenObjC, TY_RewrittenLegacyObjC,
       TY_Remap,         TY_PCH,           TY_Object,
       TY_Image,         TY_dSYM,          TY_Dependencies,
-      TY_CUDA_FATBIN,   TY_HIP_FATBIN};
+      TY_CUDA_FATBIN,   TY_HIP_FATBIN,    TY_PP_HIPHeader};
   return !llvm::is_contained(kStaticLangageTypes, Id);
 }
 
@@ -141,6 +141,7 @@
   case TY_CXXHeader: case TY_PP_CXXHeader:
   case TY_ObjCXXHeader: case TY_PP_ObjCXXHeader:
   case TY_CXXModule: case TY_PP_CXXModule:
+  case TY_HIPHeader: case TY_PP_HIPHeader:
   case TY_AST: case TY_ModuleFile: case TY_PCH:
   case TY_LLVM_IR: case TY_LLVM_BC:
     return true;
@@ -170,10 +171,9 @@
   case TY_CXXHeader: case TY_PP_CXXHeader:
   case TY_ObjCXXHeader: case TY_PP_ObjCXXHeader:
   case TY_CXXModule: case TY_PP_CXXModule:
+  case TY_HIPHeader: case TY_PP_HIPHeader:
   case TY_CUDA: case TY_PP_CUDA: case TY_CUDA_DEVICE:
-  case TY_HIP:
-  case TY_PP_HIP:
-  case TY_HIP_DEVICE:
+  case TY_HIP: case TY_PP_HIP: case TY_HIP_DEVICE:
     return true;
   }
 }
@@ -211,6 +211,7 @@
   case TY_HIP:
   case TY_PP_HIP:
   case TY_HIP_DEVICE:
+  case TY_HIPHeader: case TY_PP_HIPHeader:
     return true;
   }
 }
@@ -289,6 +290,7 @@
            .Case("c++m", TY_CXXModule)
            .Case("cppm", TY_CXXModule)
            .Case("cxxm", TY_CXXModule)
+           .Case("hipp", TY_HIPHeader)
            .Default(TY_INVALID);
 }
 
@@ -333,7 +335,8 @@
 
   // --precompile only runs up to precompilation.
   // This is a clang extension and is not compatible with GCC.
-  else if (DAL.getLastArg(options::OPT__precompile))
+  else if (DAL.getLastArg(options::OPT__precompile) ||
+           DAL.getLastArg(options::OPT_emit_pch))
     LastPhase = phases::Precompile;
 
   // -{fsyntax-only,-analyze,emit-ast} only run up to the compiler.
@@ -395,5 +398,7 @@
     return types::TY_ObjCXXHeader;
   case types::TY_CL:
     return types::TY_CLHeader;
+  case types::TY_HIP:
+    return types::TY_HIPHeader;
   }
 }
Index: clang/test/Driver/hip-binding.hipp
===================================================================
--- /dev/null
+++ clang/test/Driver/hip-binding.hipp
@@ -0,0 +1,21 @@
+// REQUIRES: clang-driver
+// REQUIRES: x86-registered-target
+// REQUIRES: amdgpu-registered-target
+
+// Host only
+// RUN: %clang -ccc-print-bindings -target x86_64-linux-gnu \
+// RUN:   -x hip-header --cuda-host-only %s \
+// RUN:   -emit-pch 2>&1 | FileCheck -check-prefix=PCHH %s
+// PCHH: # "x86_64-unknown-linux-gnu" - "clang", inputs: ["{{.*}}hip-binding.hipp"],
+// PCHH-SAME: output: "{{.*}}gch"
+//
+//
+// Device only
+// RUN: %clang -ccc-print-bindings -target x86_64-linux-gnu \
+// RUN:   -x hip-header --cuda-device-only \
+// RUN:   --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -emit-pch 2>&1 | FileCheck -check-prefix=PCHD %s
+// PCHD: # "amdgcn-amd-amdhsa" - "clang", inputs: ["[[IN:.*hip-binding.hipp]]"],
+// PCHD-SAME: output: "{{.*}}gfx803.gch"
+// PCHD: # "amdgcn-amd-amdhsa" - "clang", inputs: ["[[IN]]"],
+// PCHD-SAME: output: "{{.*}}gfx900.gch"
Index: clang/test/Driver/hip-device-compile.hipp
===================================================================
--- /dev/null
+++ clang/test/Driver/hip-device-compile.hipp
@@ -0,0 +1,153 @@
+// REQUIRES: clang-driver
+// REQUIRES: x86-registered-target
+// REQUIRES: amdgpu-registered-target
+
+// Host only -emit-pch and -include-pch in compilation
+// RUN: %clang -emit-pch --cuda-host-only -### \
+// RUN:   -target x86_64-linux-gnu -x hip-header \
+// RUN:   -nogpulib -o a.hipp.pch \
+// RUN:   %S/Inputs/hip_multiple_inputs/a.hipp \
+// RUN: 2>&1 | FileCheck -check-prefixes=HOSTP %s
+//
+// RUN: %clang -x hip --cuda-host-only -### \
+// RUN:   -target x86_64-linux-gnu -include-pch a.hipp.pch \
+// RUN:   -nogpulib -c -emit-llvm -o b.bc \
+// RUN:   %S/Inputs/hip_multiple_inputs/b.hip \
+// RUN: 2>&1 | FileCheck -check-prefixes=HOSTC %s
+//
+// HOSTP: {{".*clang.*"}} "-cc1" "-triple" "x86_64-unknown-linux-gnu"
+// HOSTP-SAME: "-aux-triple" "amdgcn-amd-amdhsa"
+// HOSTP-SAME: "-emit-pch"
+// HOSTP-SAME: "-main-file-name" "a.hipp"
+// HOSTP-SAME: "-o" "a.hipp.pch"
+// HOSTP-SAME: {{".*a.hipp"}}
+//
+// HOSTC: {{".*clang.*"}} "-cc1" "-triple" "x86_64-unknown-linux-gnu"
+// HOSTC-SAME: "-aux-triple" "amdgcn-amd-amdhsa"
+// HOSTC-SAME: "-emit-llvm-bc"
+// HOSTC-SAME: "-main-file-name" "b.hip"
+// HOSTC-SAME: "-o" "b.bc"
+// HOSTC-SAME: {{".*b.hip"}}
+
+// Single Device -emit-pch and -include-pch in compilation
+// RUN: %clang -emit-pch --cuda-device-only -### \
+// RUN:   -target x86_64-linux-gnu -o a.hipp.pch -x hip-header \
+// RUN:   --cuda-gpu-arch=gfx900 -nogpulib \
+// RUN:   %S/Inputs/hip_multiple_inputs/a.hipp \
+// RUN: 2>&1 | FileCheck -check-prefixes=DEV1P %s
+//
+// RUN: %clang -x hip --cuda-device-only -### \
+// RUN:   -target x86_64-linux-gnu -include-pch a.hipp.pch \
+// RUN:   -c -emit-llvm -o b.bc \
+// RUN:   --cuda-gpu-arch=gfx900 -nogpulib \
+// RUN:   %S/Inputs/hip_multiple_inputs/b.hip \
+// RUN: 2>&1 | FileCheck -check-prefixes=DEV1C %s
+//
+// DEV1P: {{".*clang.*"}} "-cc1" "-triple" "amdgcn-amd-amdhsa"
+// DEV1P-SAME: "-aux-triple" "x86_64-unknown-linux-gnu"
+// DEV1P-SAME: "-emit-pch"
+// DEV1P-SAME: "-main-file-name" "a.hipp"
+// DEV1P-SAME: "-fcuda-is-device"
+// DEV1P-SAME: "-target-cpu" "gfx900"
+// DEV1P-SAME: "-o" "a.hipp.pch"
+// DEV1P-SAME: {{".*a.hipp"}}
+//
+// DEV1P-NOT: {{"*.llvm-link"}}
+// DEV1P-NOT: {{".*opt"}}
+// DEV1P-NOT: {{".*llc"}}
+// DEV1P-NOT: {{".*lld.*"}}
+// DEV1P-NOT: {{".*clang-offload-bundler"}}
+// DEV1P-NOT: {{".*ld.*"}}
+//
+// DEV1C: {{".*clang.*"}} "-cc1" "-triple" "amdgcn-amd-amdhsa"
+// DEV1C-SAME: "-aux-triple" "x86_64-unknown-linux-gnu"
+// DEV1C-SAME: "-emit-llvm-bc"
+// DEV1C-SAME: "-main-file-name" "b.hip"
+// DEV1C-SAME: "-fcuda-is-device"
+// DEV1C-SAME: "-target-cpu" "gfx900"
+// DEV1C-SAME: "-include-pch" "a.hipp.pch"
+// DEV1C-SAME: "-o" "b.bc"
+// DEV1C-SAME: {{".*b.hip"}}
+
+// Multi Device -emit-pch and -include-pch in compilation
+// RUN: %clang -emit-pch --cuda-device-only -### \
+// RUN:   -target x86_64-linux-gnu -x hip-header -nogpulib \
+// RUN:   --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 \
+// RUN:   %S/Inputs/hip_multiple_inputs/a.hipp \
+// RUN: 2>&1 | FileCheck -check-prefixes=DEV2P %s
+//
+// RUN: %clang -x hip --cuda-device-only -### \
+// RUN:   -target x86_64-linux-gnu -c -emit-llvm \
+// RUN:   -include-pch a.hipp-hip-amdgcn-amd-amdhsa-gfx803.gch \
+// RUN:   --cuda-gpu-arch=gfx803 -nogpulib -o a.bc \
+// RUN:   %S/Inputs/hip_multiple_inputs/a.cu \
+// RUN: 2>&1 | FileCheck -check-prefixes=DEV2C8 %s
+//
+// RUN: %clang -x hip --cuda-device-only -### \
+// RUN:   -target x86_64-linux-gnu -c -emit-llvm \
+// RUN:   -include-pch a.hipp-hip-amdgcn-amd-amdhsa-gfx900.gch \
+// RUN:   --cuda-gpu-arch=gfx900 -nogpulib -o b.bc \
+// RUN:   %S/Inputs/hip_multiple_inputs/b.hip \
+// RUN: 2>&1 | FileCheck -check-prefixes=DEV2C9 %s
+//
+// DEV2P: {{".*clang.*"}} "-cc1" "-triple" "amdgcn-amd-amdhsa"
+// DEV2P-SAME: "-aux-triple" "x86_64-unknown-linux-gnu"
+// DEV2P-SAME: "-emit-pch"
+// DEV2P-SAME: "-main-file-name" "a.hipp"
+// DEV2P-SAME: "-fcuda-is-device"
+// DEV2P-SAME: "-target-cpu" "gfx803"
+// DEV2P-SAME: "-o" "{{.*gfx803.gch}}"
+// DEV2P-SAME: {{".*a.hipp"}}
+//
+// DEV2P: {{".*clang.*"}} "-cc1" "-triple" "amdgcn-amd-amdhsa"
+// DEV2P-SAME: "-aux-triple" "x86_64-unknown-linux-gnu"
+// DEV2P-SAME: "-emit-pch"
+// DEV2P-SAME: "-main-file-name" "a.hipp"
+// DEV2P-SAME: "-fcuda-is-device"
+// DEV2P-SAME: "-target-cpu" "gfx900"
+// DEV2P-SAME: "-o" "{{.*gfx900.gch}}"
+// DEV2P-SAME: {{".*a.hipp"}}
+//
+// DEV2P-NOT: {{"*.llvm-link"}}
+// DEV2P-NOT: {{".*opt"}}
+// DEV2P-NOT: {{".*llc"}}
+// DEV2P-NOT: {{".*lld.*"}}
+// DEV2P-NOT: {{".*clang-offload-bundler"}}
+// DEV2P-NOT: {{".*ld.*"}}
+//
+// DEV2C8: {{".*clang.*"}} "-cc1" "-triple" "amdgcn-amd-amdhsa"
+// DEV2C8-SAME: "-aux-triple" "x86_64-unknown-linux-gnu"
+// DEV2C8-SAME: "-emit-llvm-bc"
+// DEV2C8-SAME: "-main-file-name" "a.cu"
+// DEV2C8-SAME: "-fcuda-is-device"
+// DEV2C8-SAME: "-target-cpu" "gfx803"
+// DEV2C8-SAME: "-include-pch" "a.hipp-hip-amdgcn-amd-amdhsa-gfx803.gch"
+// DEV2C8-SAME: "-o" "a.bc"
+// DEV2C8-SAME: {{".*a.cu"}}
+//
+// DEV2C9: {{".*clang.*"}} "-cc1" "-triple" "amdgcn-amd-amdhsa"
+// DEV2C9-SAME: "-aux-triple" "x86_64-unknown-linux-gnu"
+// DEV2C9-SAME: "-emit-llvm-bc"
+// DEV2C9-SAME: "-main-file-name" "b.hip"
+// DEV2C9-SAME: "-fcuda-is-device"
+// DEV2C9-SAME: "-target-cpu" "gfx900"
+// DEV2C9-SAME: "-include-pch" "a.hipp-hip-amdgcn-amd-amdhsa-gfx900.gch"
+// DEV2C9-SAME: "-o" "b.bc"
+// DEV2C9-SAME: {{".*b.hip"}}
+
+// Compiling with the wrong GPU variant of PCH should fail
+// RUN: %clang -emit-pch --cuda-device-only -v \
+// RUN:   -target x86_64-linux-gnu -x hip-header -nogpulib \
+// RUN:   --cuda-gpu-arch=gfx900 -o a-900.hipp.pch \
+// RUN:   %S/Inputs/hip_multiple_inputs/a.hipp \
+// RUN: 2>&1 | FileCheck -check-prefixes=CREATE %s
+// CREATE: {{.*}} -o a-900.hipp.pch
+//
+// RUN: not %clang -x hip --cuda-device-only \
+// RUN:   -target x86_64-linux-gnu -c -emit-llvm \
+// RUN:   -include-pch a-900.hipp.pch \
+// RUN:   --cuda-gpu-arch=gfx803 -nogpulib -o b.bc \
+// RUN:   %S/Inputs/hip_multiple_inputs/b.hip \
+// RUN: 2>&1 | FileCheck -check-prefixes=ERROR --implicit-check-not=error: %s
+//
+// ERROR: error: PCH file was compiled for the target CPU 'gfx900' but the current translation unit is being compiled for target 'gfx803'
Index: clang/test/Driver/hip-phases.hipp
===================================================================
--- /dev/null
+++ clang/test/Driver/hip-phases.hipp
@@ -0,0 +1,41 @@
+// REQUIRES: clang-driver
+// REQUIRES: x86-registered-target
+// REQUIRES: amdgpu-registered-target
+//
+//
+// Test single gpu architecture up to the precompile phase in device-only mode.
+//
+// RUN: %clang -x hip-header -target x86_64-unknown-linux-gnu -ccc-print-phases \
+// RUN: --cuda-gpu-arch=gfx803 %s --cuda-device-only -emit-pch 2>&1 \
+// RUN: | FileCheck -check-prefixes=PCH %s
+//
+// PCH-DAG: [[P0:[0-9]+]]: input, "{{.*}}hip-phases.hipp", [[T:hip-header]], (device-hip, [[ARCH:gfx803]])
+// PCH-DAG: [[P1:[0-9]+]]: preprocessor, {[[P0]]}, [[T]]-cpp-output, (device-hip, [[ARCH]])
+// PCH-DAG: [[P2:[0-9]+]]: precompiler, {[[P1]]}, precompiled-header, (device-hip, [[ARCH]])
+// PCH-NOT: host
+//
+// Test two gpu architecture up to the precompile phase in device-only mode.
+//
+// RUN: %clang -x hip-header -target x86_64-unknown-linux-gnu \
+// RUN: -ccc-print-phases --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN: --cuda-device-only -emit-pch 2>&1 \
+// RUN: | FileCheck -check-prefixes=PCH2 %s
+//
+// PCH2-DAG: [[P0:[0-9]+]]: input, "{{.*}}hip-phases.hipp", [[T:hip-header]], (device-hip, [[ARCH:gfx803]])
+// PCH2-DAG: [[P1:[0-9]+]]: preprocessor, {[[P0]]}, [[T]]-cpp-output, (device-hip, [[ARCH]])
+// PCH2-DAG: [[P2:[0-9]+]]: precompiler, {[[P1]]}, precompiled-header, (device-hip, [[ARCH]])
+// PCH2-DAG: [[P4:[0-9]+]]: input, "{{.*}}hip-phases.hipp", [[T]], (device-hip, [[ARCH:gfx900]])
+// PCH2-DAG: [[P5:[0-9]+]]: preprocessor, {[[P4]]}, [[T]]-cpp-output, (device-hip, [[ARCH]])
+// PCH2-DAG: [[P6:[0-9]+]]: precompiler, {[[P5]]}, precompiled-header, (device-hip, [[ARCH]])
+// PCH2-NOT: host
+//
+// Test the precompile phase in host-only mode.
+//
+// RUN: %clang -x hip-header -target x86_64-unknown-linux-gnu \
+// RUN: -ccc-print-phases %s --cuda-host-only \
+// RUN: -emit-pch 2>&1 | FileCheck -check-prefixes=PCHH %s
+//
+// PCHH-DAG: [[P0:[0-9]+]]: input, "{{.*}}hip-phases.hipp", [[T:hip-header]], (host-hip)
+// PCHH-DAG: [[P1:[0-9]+]]: preprocessor, {[[P0]]}, [[T]]-cpp-output, (host-hip)
+// PCHH-DAG: [[P2:[0-9]+]]: precompiler, {[[P1]]}, precompiled-header, (host-hip)
+// PCHH-NOT: device