[CUDA] Disable LTO for device-side compilations.

[tra]

Summary

This fixes host-side LTO during CUDA compilation. Before, LTO
pipeline construction was clashing with CUDA pipeline construction.

At the moment there's no point doing LTO on device side as each
device-side TU is a complete program. We will need to figure out
compilation pipeline construction for the device-side LTO when we
have working support for multi-TU device-side CUDA compilation.

[yroux]

Hi Artem,

[tra]

The test failure should be fixed by r328213.

cfe/trunk/include/clang/Driver/Driver.h

public:
   ///
   /// \return Whether any compilation should be built for this
   /// invocation.
   bool HandleImmediateArgs(const Compilation &C);
 
   /// ConstructAction - Construct the appropriate action to do for
   /// \p Phase on the \p Input, taking in to account arguments
   /// like -fsyntax-only or --analyze.
-  Action *ConstructPhaseAction(Compilation &C, const llvm::opt::ArgList &Args,
-                               phases::ID Phase, Action *Input) const;
+  Action *ConstructPhaseAction(
+      Compilation &C, const llvm::opt::ArgList &Args, phases::ID Phase,
+      Action *Input,
+      Action::OffloadKind TargetDeviceOffloadKind = Action::OFK_None) const;
 
   /// BuildJobsForAction - Construct the jobs to perform for the action \p A and
   /// return an InputInfo for the result of running \p A.  Will only construct
   /// jobs for a given (Action, ToolChain, BoundArch, DeviceKind) tuple once.
   InputInfo
   BuildJobsForAction(Compilation &C, const Action *A, const ToolChain *TC,
                      StringRef BoundArch, bool AtTopLevel, bool MultipleArchs,
                      const char *LinkingOutput,

cfe/trunk/lib/Driver/Driver.cpp

getDeviceDependences(OffloadAction::DeviceDependences &DA,
             // Skip the phases that were already dealt with.
             if (Ph < CurPhase)
               continue;
             // We have to be consistent with the host final phase.
             if (Ph > FinalPhase)
               break;
 
             CudaDeviceActions[I] = C.getDriver().ConstructPhaseAction(
-                C, Args, Ph, CudaDeviceActions[I]);
+                C, Args, Ph, CudaDeviceActions[I], Action::OFK_Cuda);
 
             if (Ph == phases::Assemble)
               break;
           }
 
           // If we didn't reach the assemble phase, we can't generate the fat
           // binary. We don't need to generate the fat binary if we are not in
           // device-only mode.
void Driver::BuildActions(Compilation &C, DerivedArgList &Args,
   Args.ClaimAllArgs(options::OPT_cl_ignored_Group);
 
   // Claim --cuda-host-only and --cuda-compile-host-device, which may be passed
   // to non-CUDA compilations and should not trigger warnings there.
   Args.ClaimAllArgs(options::OPT_cuda_host_only);
   Args.ClaimAllArgs(options::OPT_cuda_compile_host_device);
 }
 
-Action *Driver::ConstructPhaseAction(Compilation &C, const ArgList &Args,
-                                     phases::ID Phase, Action *Input) const {
+Action *Driver::ConstructPhaseAction(
+    Compilation &C, const ArgList &Args, phases::ID Phase, Action *Input,
+    Action::OffloadKind TargetDeviceOffloadKind) const {
   llvm::PrettyStackTraceString CrashInfo("Constructing phase actions");
 
   // Some types skip the assembler phase (e.g., llvm-bc), but we can't
   // encode this in the steps because the intermediate type depends on
   // arguments. Just special case here.
   if (Phase == phases::Assemble && Input->getType() != types::TY_PP_Asm)
     return Input;
 
if (Args.hasArg(options::OPT_emit_ast))
       return C.MakeAction<CompileJobAction>(Input, types::TY_AST);
     if (Args.hasArg(options::OPT_module_file_info))
       return C.MakeAction<CompileJobAction>(Input, types::TY_ModuleFile);
     if (Args.hasArg(options::OPT_verify_pch))
       return C.MakeAction<VerifyPCHJobAction>(Input, types::TY_Nothing);
     return C.MakeAction<CompileJobAction>(Input, types::TY_LLVM_BC);
   }
   case phases::Backend: {
-    if (isUsingLTO()) {
+    if (isUsingLTO() && TargetDeviceOffloadKind == Action::OFK_None) {
       types::ID Output =
           Args.hasArg(options::OPT_S) ? types::TY_LTO_IR : types::TY_LTO_BC;
       return C.MakeAction<BackendJobAction>(Input, Output);
     }
     if (Args.hasArg(options::OPT_emit_llvm)) {
       types::ID Output =
           Args.hasArg(options::OPT_S) ? types::TY_LLVM_IR : types::TY_LLVM_BC;
       return C.MakeAction<BackendJobAction>(Input, Output);

cfe/trunk/lib/Driver/ToolChains/Clang.cpp

if (isa<AnalyzeJobAction>(JA)) {
 
     // Preserve use-list order by default when emitting bitcode, so that
     // loading the bitcode up in 'opt' or 'llc' and running passes gives the
     // same result as running passes here.  For LTO, we don't need to preserve
     // the use-list order, since serialization to bitcode is part of the flow.
     if (JA.getType() == types::TY_LLVM_BC)
       CmdArgs.push_back("-emit-llvm-uselists");
 
-    if (D.isUsingLTO()) {
+    // Device-side jobs do not support LTO.
+    bool isDeviceOffloadAction = !(JA.isDeviceOffloading(Action::OFK_None) ||
+                                   JA.isDeviceOffloading(Action::OFK_Host));
+
+    if (D.isUsingLTO() && !isDeviceOffloadAction) {
       Args.AddLastArg(CmdArgs, options::OPT_flto, options::OPT_flto_EQ);
 
       // The Darwin and PS4 linkers currently use the legacy LTO API, which
       // does not support LTO unit features (CFI, whole program vtable opt)
       // under ThinLTO.
       if (!(RawTriple.isOSDarwin() || RawTriple.isPS4()) ||
           D.getLTOMode() == LTOK_Full)
         CmdArgs.push_back("-flto-unit");

cfe/trunk/test/Driver/lto.cu

+// -flto causes a switch to llvm-bc object files.
+// RUN: %clangxx -nocudainc -nocudalib -ccc-print-phases -c %s -flto 2> %t
+// RUN: FileCheck -check-prefix=CHECK-COMPILE-ACTIONS < %t %s
+//
+// CHECK-COMPILE-ACTIONS: 2: compiler, {1}, ir, (host-cuda)
+// CHECK-COMPILE-ACTIONS-NOT: lto-bc
+// CHECK-COMPILE-ACTIONS: 12: backend, {11}, lto-bc, (host-cuda)
+
+// RUN: %clangxx -nocudainc -nocudalib -ccc-print-phases %s -flto 2> %t
+// RUN: FileCheck -check-prefix=CHECK-COMPILELINK-ACTIONS < %t %s
+//
+// CHECK-COMPILELINK-ACTIONS: 0: input, "{{.*}}lto.cu", cuda, (host-cuda)
+// CHECK-COMPILELINK-ACTIONS: 1: preprocessor, {0}, cuda-cpp-output
+// CHECK-COMPILELINK-ACTIONS: 2: compiler, {1}, ir, (host-cuda)
+// CHECK-COMPILELINK-ACTIONS: 3: input, "{{.*}}lto.cu", cuda, (device-cuda, sm_20)
+// CHECK-COMPILELINK-ACTIONS: 4: preprocessor, {3}, cuda-cpp-output, (device-cuda, sm_20)
+// CHECK-COMPILELINK-ACTIONS: 5: compiler, {4}, ir, (device-cuda, sm_20)
+// CHECK-COMPILELINK-ACTIONS: 6: backend, {5}, assembler, (device-cuda, sm_20)
+// CHECK-COMPILELINK-ACTIONS: 7: assembler, {6}, object, (device-cuda, sm_20)
+// CHECK-COMPILELINK-ACTIONS: 8: offload, "device-cuda (nvptx64-nvidia-cuda:sm_20)" {7}, object
+// CHECK-COMPILELINK-ACTIONS: 9: offload, "device-cuda (nvptx64-nvidia-cuda:sm_20)" {6}, assembler
+// CHECK-COMPILELINK-ACTIONS: 10: linker, {8, 9}, cuda-fatbin, (device-cuda)
+// CHECK-COMPILELINK-ACTIONS: 11: offload, "host-cuda {{.*}}" {2}, "device-cuda{{.*}}" {10}, ir
+// CHECK-COMPILELINK-ACTIONS: 12: backend, {11}, lto-bc, (host-cuda)
+// CHECK-COMPILELINK-ACTIONS: 13: linker, {12}, image, (host-cuda)
+
+// llvm-bc and llvm-ll outputs need to match regular suffixes
+// (unfortunately).
+// RUN: %clangxx %s -nocudainc -nocudalib -flto -save-temps -### 2> %t
+// RUN: FileCheck -check-prefix=CHECK-COMPILELINK-SUFFIXES < %t %s
+//
+// CHECK-COMPILELINK-SUFFIXES: "-o" "[[CPP:.*lto-host.*\.cui]]" "-x" "cuda" "{{.*}}lto.cu"
+// CHECK-COMPILELINK-SUFFIXES: "-o" "[[BC:.*lto-host.*\.bc]]" {{.*}}[[CPP]]"
+// CHECK-COMPILELINK-SUFFIXES: "-o" "[[OBJ:.*lto-host.*\.o]]" {{.*}}[[BC]]"
+// CHECK-COMPILELINK-SUFFIXES: "{{.*}}a.{{(out|exe)}}" {{.*}}[[OBJ]]"
+
+// RUN: %clangxx %s -nocudainc -nocudalib -flto -S -### 2> %t
+// RUN: FileCheck -check-prefix=CHECK-COMPILE-SUFFIXES < %t %s
+//
+// CHECK-COMPILE-SUFFIXES: "-o" "{{.*}}lto.s" "-x" "cuda" "{{.*}}lto.cu"
+
+// RUN: not %clangxx -nocudainc -nocudalib %s -emit-llvm 2>&1 \
+// RUN:    | FileCheck --check-prefix=LLVM-LINK %s
+// LLVM-LINK: -emit-llvm cannot be used when linking
+
+// -flto should cause link using gold plugin
+// RUN: %clangxx -nocudainc -nocudalib \
+// RUN:          -target x86_64-unknown-linux -### %s -flto 2> %t
+// RUN: FileCheck -check-prefix=CHECK-LINK-LTO-ACTION < %t %s
+//
+// CHECK-LINK-LTO-ACTION: "-plugin" "{{.*}}{{[/\\]}}LLVMgold.{{dll|dylib|so}}"
+
+// -flto=full should cause link using gold plugin
+// RUN: %clangxx -nocudainc -nocudalib \
+// RUN:          -target x86_64-unknown-linux -### %s -flto=full 2> %t
+// RUN: FileCheck -check-prefix=CHECK-LINK-FULL-ACTION < %t %s
+//
+// CHECK-LINK-FULL-ACTION: "-plugin" "{{.*}}{{[/\\]}}LLVMgold.{{dll|dylib|so}}"
+
+// Check that subsequent -fno-lto takes precedence
+// RUN: %clangxx -nocudainc -nocudalib \
+// RUN:          -target x86_64-unknown-linux -### %s -flto=full -fno-lto 2> %t
+// RUN: FileCheck -check-prefix=CHECK-LINK-NOLTO-ACTION < %t %s
+//
+// CHECK-LINK-NOLTO-ACTION-NOT: "-plugin" "{{.*}}{{[/\\]}}LLVMgold.{{dll|dylib|so}}"
+
+// -flto passes along an explicit debugger tuning argument.
+// RUN: %clangxx -nocudainc -nocudalib \
+// RUN:          -target x86_64-unknown-linux -### %s -flto -glldb 2> %t
+// RUN: FileCheck -check-prefix=CHECK-TUNING-LLDB < %t %s
+// RUN: %clangxx -nocudainc -nocudalib \
+// RUN:          -target x86_64-unknown-linux -### %s -flto -g 2> %t
+// RUN: FileCheck -check-prefix=CHECK-NO-TUNING < %t %s
+//
+// CHECK-TUNING-LLDB:   "-plugin-opt=-debugger-tune=lldb"
+// CHECK-NO-TUNING-NOT: "-plugin-opt=-debugger-tune

cfe/trunk/test/Driver/thinlto.cu

+// -flto=thin causes a switch to llvm-bc object files.
+// RUN: %clangxx -ccc-print-phases -nocudainc -nocudalib -c %s -flto=thin 2> %t
+// RUN: FileCheck -check-prefix=CHECK-COMPILE-ACTIONS < %t %s
+//
+// CHECK-COMPILE-ACTIONS: 2: compiler, {1}, ir, (host-cuda)
+// CHECK-COMPILE-ACTIONS-NOT: lto-bc
+// CHECK-COMPILE-ACTIONS: 12: backend, {11}, lto-bc, (host-cuda)
+
+// RUN: %clangxx -ccc-print-phases -nocudainc -nocudalib %s -flto=thin 2> %t
+// RUN: FileCheck -check-prefix=CHECK-COMPILELINK-ACTIONS < %t %s
+//
+// CHECK-COMPILELINK-ACTIONS: 0: input, "{{.*}}thinlto.cu", cuda, (host-cuda)
+// CHECK-COMPILELINK-ACTIONS: 1: preprocessor, {0}, cuda-cpp-output
+// CHECK-COMPILELINK-ACTIONS: 2: compiler, {1}, ir, (host-cuda)
+// CHECK-COMPILELINK-ACTIONS: 3: input, "{{.*}}thinlto.cu", cuda, (device-cuda, sm_20)
+// CHECK-COMPILELINK-ACTIONS: 4: preprocessor, {3}, cuda-cpp-output, (device-cuda, sm_20)
+// CHECK-COMPILELINK-ACTIONS: 5: compiler, {4}, ir, (device-cuda, sm_20)
+// CHECK-COMPILELINK-ACTIONS: 6: backend, {5}, assembler, (device-cuda, sm_20)
+// CHECK-COMPILELINK-ACTIONS: 7: assembler, {6}, object, (device-cuda, sm_20)
+// CHECK-COMPILELINK-ACTIONS: 8: offload, "device-cuda (nvptx64-nvidia-cuda:sm_20)" {7}, object
+// CHECK-COMPILELINK-ACTIONS: 9: offload, "device-cuda (nvptx64-nvidia-cuda:sm_20)" {6}, assembler
+// CHECK-COMPILELINK-ACTIONS: 10: linker, {8, 9}, cuda-fatbin, (device-cuda)
+// CHECK-COMPILELINK-ACTIONS: 11: offload, "host-cuda {{.*}}" {2}, "device-cuda{{.*}}" {10}, ir
+// CHECK-COMPILELINK-ACTIONS: 12: backend, {11}, lto-bc, (host-cuda)
+// CHECK-COMPILELINK-ACTIONS: 13: linker, {12}, image, (host-cuda)
+
+// -flto=thin should cause link using gold plugin with thinlto option,
+// also confirm that it takes precedence over earlier -fno-lto and -flto=full.
+// RUN: %clangxx -nocudainc -nocudalib \
+// RUN:        -target x86_64-unknown-linux -### %s -flto=full -fno-lto -flto=thin 2> %t
+// RUN: FileCheck -check-prefix=CHECK-LINK-THIN-ACTION < %t %s
+//
+// CHECK-LINK-THIN-ACTION: "-plugin" "{{.*}}{{[/\\]}}LLVMgold.{{dll|dylib|so}}"
+// CHECK-LINK-THIN-ACTION: "-plugin-opt=thinlto"
+
+// Check that subsequent -flto=full takes precedence
+// RUN: %clangxx -nocudainc -nocudalib \
+// RUN:        -target x86_64-unknown-linux -### %s -flto=thin -flto=full 2> %t
+// RUN: FileCheck -check-prefix=CHECK-LINK-FULL-ACTION < %t %s
+//
+// CHECK-LINK-FULL-ACTION: "-plugin" "{{.*}}{{[/\\]}}LLVMgold.{{dll|dylib|so}}"
+// CHECK-LINK-FULL-ACTION-NOT: "-plugin-opt=thinlto"
+
+// Check that subsequent -fno-lto takes precedence
+// RUN: %clangxx -nocudainc -nocudalib \
+// RUN:        -target x86_64-unknown-linux -### %s -flto=thin -fno-lto 2> %t
+// RUN: FileCheck -check-prefix=CHECK-LINK-NOLTO-ACTION < %t %s
+//
+// CHECK-LINK-NOLTO-ACTION-NOT: "-plugin" "{{.*}}{{[/\\]}}LLVMgold.{{dll|dylib|so}}"
+// CHECK-LINK-NOLTO-ACTION-NOT: "-plugin-opt=thinlto"