[clang] [llvm] [Offload][CUDA] Allow CUDA kernels to use LLVM/Offload (PR #94549)

Johannes Doerfert via cfe-commits cfe-commits at lists.llvm.org
Sun Aug 11 17:12:06 PDT 2024


https://github.com/jdoerfert updated https://github.com/llvm/llvm-project/pull/94549

>From a9f228a344d928c2c7a0ffd70605a0665b2b9a07 Mon Sep 17 00:00:00 2001
From: Johannes Doerfert <johannes at jdoerfert.de>
Date: Mon, 3 Jun 2024 19:52:12 -0700
Subject: [PATCH] [Offload][CUDA] Allow CUDA kernels to use LLVM/Offload
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Through the new `-foffload-via-llvm` flag, CUDA kernels can now be
lowered to the LLVM/Offload API. On the Clang side, this is simply done
by using the OpenMP offload toolchain and emitting calls to `llvm*`
functions to orchestrate the kernel launch rather than `cuda*`
functions. These `llvm*` functions are implemented on top of the
existing LLVM/Offload API.

As we are about to redefine the Offload API, this wil help us in the
design process as a second offload language.

We do not support any CUDA APIs yet, however, we could:
  https://www.osti.gov/servlets/purl/1892137

For proper host execution we need to resurrect/rebase
  https://tianshilei.me/wp-content/uploads/2021/12/llpp-2021.pdf
(which was designed for debugging).

```
❯❯❯ cat test.cu
extern "C" {
void *llvm_omp_target_alloc_shared(size_t Size, int DeviceNum);
void llvm_omp_target_free_shared(void *DevicePtr, int DeviceNum);
}

__global__ void square(int *A) { *A = 42; }

int main(int argc, char **argv) {
  int DevNo = 0;
  int *Ptr = reinterpret_cast<int *>(llvm_omp_target_alloc_shared(4, DevNo));
  *Ptr = 7;
  printf("Ptr %p, *Ptr %i\n", Ptr, *Ptr);
  square<<<1, 1>>>(Ptr);
  printf("Ptr %p, *Ptr %i\n", Ptr, *Ptr);
  llvm_omp_target_free_shared(Ptr, DevNo);
}

❯❯❯ clang++ test.cu -O3 -o test123 -foffload-via-llvm --offload-arch=native

❯❯❯ llvm-objdump --offloading test123

test123:        file format elf64-x86-64

OFFLOADING IMAGE [0]:
kind            elf
arch            gfx90a
triple          amdgcn-amd-amdhsa
producer        openmp

❯❯❯ LIBOMPTARGET_INFO=16 ./test123
Ptr 0x155448ac8000, *Ptr 7
Ptr 0x155448ac8000, *Ptr 42
```
---
 clang/include/clang/Basic/LangOptions.def     |  1 +
 clang/include/clang/Driver/Options.td         |  6 ++
 clang/lib/CodeGen/CGCUDANV.cpp                | 97 ++++++++++++++++---
 clang/lib/CodeGen/CodeGenFunction.cpp         |  3 +-
 clang/lib/Driver/Driver.cpp                   | 19 ++--
 clang/lib/Driver/ToolChains/Clang.cpp         | 31 +++++-
 clang/lib/Driver/ToolChains/CommonArgs.cpp    |  7 +-
 clang/lib/Driver/ToolChains/Cuda.cpp          | 27 +++---
 clang/lib/Headers/CMakeLists.txt              | 18 +++-
 .../llvm_offload_wrappers/__llvm_offload.h    | 31 ++++++
 .../__llvm_offload_device.h                   | 10 ++
 .../__llvm_offload_host.h                     | 15 +++
 .../__clang_openmp_device_functions.h         |  9 +-
 clang/lib/Sema/SemaCUDA.cpp                   |  3 +
 clang/test/CodeGenCUDA/Inputs/cuda.h          |  5 +
 clang/test/CodeGenCUDA/offload_via_llvm.cu    | 90 +++++++++++++++++
 clang/test/Driver/cuda-via-liboffload.cu      | 23 +++++
 offload/include/Shared/APITypes.h             |  5 +-
 offload/include/omptarget.h                   |  2 +-
 .../common/src/PluginInterface.cpp            | 14 ++-
 offload/src/CMakeLists.txt                    |  1 +
 offload/src/KernelLanguage/API.cpp            | 74 ++++++++++++++
 offload/src/exports                           |  3 +
 offload/test/lit.cfg                          |  2 +-
 offload/test/offloading/CUDA/basic_launch.cu  | 30 ++++++
 .../CUDA/basic_launch_blocks_and_threads.cu   | 32 ++++++
 .../offloading/CUDA/basic_launch_multi_arg.cu | 41 ++++++++
 offload/test/offloading/CUDA/kernel_tu.cu.inc |  1 +
 offload/test/offloading/CUDA/launch_tu.cu     | 32 ++++++
 29 files changed, 578 insertions(+), 54 deletions(-)
 create mode 100644 clang/lib/Headers/llvm_offload_wrappers/__llvm_offload.h
 create mode 100644 clang/lib/Headers/llvm_offload_wrappers/__llvm_offload_device.h
 create mode 100644 clang/lib/Headers/llvm_offload_wrappers/__llvm_offload_host.h
 create mode 100644 clang/test/CodeGenCUDA/offload_via_llvm.cu
 create mode 100644 clang/test/Driver/cuda-via-liboffload.cu
 create mode 100644 offload/src/KernelLanguage/API.cpp
 create mode 100644 offload/test/offloading/CUDA/basic_launch.cu
 create mode 100644 offload/test/offloading/CUDA/basic_launch_blocks_and_threads.cu
 create mode 100644 offload/test/offloading/CUDA/basic_launch_multi_arg.cu
 create mode 100644 offload/test/offloading/CUDA/kernel_tu.cu.inc
 create mode 100644 offload/test/offloading/CUDA/launch_tu.cu

diff --git a/clang/include/clang/Basic/LangOptions.def b/clang/include/clang/Basic/LangOptions.def
index 54e689a7a42213..ca9c00a1473bbd 100644
--- a/clang/include/clang/Basic/LangOptions.def
+++ b/clang/include/clang/Basic/LangOptions.def
@@ -298,6 +298,7 @@ LANGOPT(GPUMaxThreadsPerBlock, 32, 1024, "default max threads per block for kern
 LANGOPT(GPUDeferDiag, 1, 0, "defer host/device related diagnostic messages for CUDA/HIP")
 LANGOPT(GPUExcludeWrongSideOverloads, 1, 0, "always exclude wrong side overloads in overloading resolution for CUDA/HIP")
 LANGOPT(OffloadingNewDriver, 1, 0, "use the new driver for generating offloading code.")
+LANGOPT(OffloadViaLLVM, 1, 0, "target LLVM/Offload as portable offloading runtime.")
 
 LANGOPT(SYCLIsDevice      , 1, 0, "Generate code for SYCL device")
 LANGOPT(SYCLIsHost        , 1, 0, "SYCL host compilation")
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index e196c3dc5cb3be..e212c6dd80467a 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -1302,6 +1302,12 @@ def no_offload_compress : Flag<["--"], "no-offload-compress">;
 def offload_compression_level_EQ : Joined<["--"], "offload-compression-level=">,
   Flags<[HelpHidden]>,
   HelpText<"Compression level for offload device binaries (HIP only)">;
+
+defm offload_via_llvm : BoolFOption<"offload-via-llvm",
+  LangOpts<"OffloadViaLLVM">, DefaultFalse,
+  PosFlag<SetTrue, [], [ClangOption, CC1Option], "Use">,
+  NegFlag<SetFalse, [], [ClangOption], "Don't use">,
+  BothFlags<[], [ClangOption], " LLVM/Offload as portable offloading runtime.">>;
 }
 
 // CUDA options
diff --git a/clang/lib/CodeGen/CGCUDANV.cpp b/clang/lib/CodeGen/CGCUDANV.cpp
index 1f5fb630a47038..59c5927717933d 100644
--- a/clang/lib/CodeGen/CGCUDANV.cpp
+++ b/clang/lib/CodeGen/CGCUDANV.cpp
@@ -15,10 +15,12 @@
 #include "CGCXXABI.h"
 #include "CodeGenFunction.h"
 #include "CodeGenModule.h"
+#include "clang/AST/CharUnits.h"
 #include "clang/AST/Decl.h"
 #include "clang/Basic/Cuda.h"
 #include "clang/CodeGen/CodeGenABITypes.h"
 #include "clang/CodeGen/ConstantInitBuilder.h"
+#include "llvm/ADT/StringRef.h"
 #include "llvm/Frontend/Offloading/Utility.h"
 #include "llvm/IR/BasicBlock.h"
 #include "llvm/IR/Constants.h"
@@ -36,6 +38,11 @@ constexpr unsigned HIPFatMagic = 0x48495046; // "HIPF"
 
 class CGNVCUDARuntime : public CGCUDARuntime {
 
+  /// The prefix used for function calls and section names (CUDA, HIP, LLVM)
+  StringRef Prefix;
+  /// TODO: We should transition the OpenMP section to LLVM/Offload
+  StringRef SectionPrefix;
+
 private:
   llvm::IntegerType *IntTy, *SizeTy;
   llvm::Type *VoidTy;
@@ -132,6 +139,9 @@ class CGNVCUDARuntime : public CGCUDARuntime {
     return DummyFunc;
   }
 
+  Address prepareKernelArgs(CodeGenFunction &CGF, FunctionArgList &Args);
+  Address prepareKernelArgsLLVMOffload(CodeGenFunction &CGF,
+                                       FunctionArgList &Args);
   void emitDeviceStubBodyLegacy(CodeGenFunction &CGF, FunctionArgList &Args);
   void emitDeviceStubBodyNew(CodeGenFunction &CGF, FunctionArgList &Args);
   std::string getDeviceSideName(const NamedDecl *ND) override;
@@ -191,15 +201,11 @@ class CGNVCUDARuntime : public CGCUDARuntime {
 } // end anonymous namespace
 
 std::string CGNVCUDARuntime::addPrefixToName(StringRef FuncName) const {
-  if (CGM.getLangOpts().HIP)
-    return ((Twine("hip") + Twine(FuncName)).str());
-  return ((Twine("cuda") + Twine(FuncName)).str());
+  return (Prefix + FuncName).str();
 }
 std::string
 CGNVCUDARuntime::addUnderscoredPrefixToName(StringRef FuncName) const {
-  if (CGM.getLangOpts().HIP)
-    return ((Twine("__hip") + Twine(FuncName)).str());
-  return ((Twine("__cuda") + Twine(FuncName)).str());
+  return ("__" + Prefix + FuncName).str();
 }
 
 static std::unique_ptr<MangleContext> InitDeviceMC(CodeGenModule &CGM) {
@@ -227,6 +233,14 @@ CGNVCUDARuntime::CGNVCUDARuntime(CodeGenModule &CGM)
   SizeTy = CGM.SizeTy;
   VoidTy = CGM.VoidTy;
   PtrTy = CGM.UnqualPtrTy;
+
+  if (CGM.getLangOpts().OffloadViaLLVM) {
+    Prefix = "llvm";
+    SectionPrefix = "omp";
+  } else if (CGM.getLangOpts().HIP)
+    SectionPrefix = Prefix = "hip";
+  else
+    SectionPrefix = Prefix = "cuda";
 }
 
 llvm::FunctionCallee CGNVCUDARuntime::getSetupArgumentFn() const {
@@ -305,18 +319,58 @@ void CGNVCUDARuntime::emitDeviceStub(CodeGenFunction &CGF,
   }
   if (CudaFeatureEnabled(CGM.getTarget().getSDKVersion(),
                          CudaFeature::CUDA_USES_NEW_LAUNCH) ||
-      (CGF.getLangOpts().HIP && CGF.getLangOpts().HIPUseNewLaunchAPI))
+      (CGF.getLangOpts().HIP && CGF.getLangOpts().HIPUseNewLaunchAPI) ||
+      (CGF.getLangOpts().OffloadViaLLVM))
     emitDeviceStubBodyNew(CGF, Args);
   else
     emitDeviceStubBodyLegacy(CGF, Args);
 }
 
-// CUDA 9.0+ uses new way to launch kernels. Parameters are packed in a local
-// array and kernels are launched using cudaLaunchKernel().
-void CGNVCUDARuntime::emitDeviceStubBodyNew(CodeGenFunction &CGF,
-                                            FunctionArgList &Args) {
-  // Build the shadow stack entry at the very start of the function.
+/// CUDA passes the arguments with a level of indirection. For example, a
+/// (void*, short, void*) is passed as {void **, short *, void **} to the launch
+/// function. For the LLVM/offload launch we flatten the arguments into the
+/// struct directly. In addition, we include the size of the arguments, thus
+/// pass {sizeof({void *, short, void *}), ptr to {void *, short, void *},
+/// nullptr}. The last nullptr needs to be initialized to an array of pointers
+/// pointing to the arguments if we want to offload to the host.
+Address CGNVCUDARuntime::prepareKernelArgsLLVMOffload(CodeGenFunction &CGF,
+                                                      FunctionArgList &Args) {
+  SmallVector<llvm::Type *> ArgTypes, KernelLaunchParamsTypes;
+  for (auto &Arg : Args)
+    ArgTypes.push_back(CGF.ConvertTypeForMem(Arg->getType()));
+  llvm::StructType *KernelArgsTy = llvm::StructType::create(ArgTypes);
+
+  auto *Int64Ty = CGF.Builder.getInt64Ty();
+  KernelLaunchParamsTypes.push_back(Int64Ty);
+  KernelLaunchParamsTypes.push_back(PtrTy);
+  KernelLaunchParamsTypes.push_back(PtrTy);
+
+  llvm::StructType *KernelLaunchParamsTy =
+      llvm::StructType::create(KernelLaunchParamsTypes);
+  Address KernelArgs = CGF.CreateTempAllocaWithoutCast(
+      KernelArgsTy, CharUnits::fromQuantity(16), "kernel_args");
+  Address KernelLaunchParams = CGF.CreateTempAllocaWithoutCast(
+      KernelLaunchParamsTy, CharUnits::fromQuantity(16),
+      "kernel_launch_params");
+
+  auto KernelArgsSize = CGM.getDataLayout().getTypeAllocSize(KernelArgsTy);
+  CGF.Builder.CreateStore(llvm::ConstantInt::get(Int64Ty, KernelArgsSize),
+                          CGF.Builder.CreateStructGEP(KernelLaunchParams, 0));
+  CGF.Builder.CreateStore(KernelArgs.emitRawPointer(CGF),
+                          CGF.Builder.CreateStructGEP(KernelLaunchParams, 1));
+  CGF.Builder.CreateStore(llvm::Constant::getNullValue(PtrTy),
+                          CGF.Builder.CreateStructGEP(KernelLaunchParams, 2));
+
+  for (unsigned i = 0; i < Args.size(); ++i) {
+    auto *ArgVal = CGF.Builder.CreateLoad(CGF.GetAddrOfLocalVar(Args[i]));
+    CGF.Builder.CreateStore(ArgVal, CGF.Builder.CreateStructGEP(KernelArgs, i));
+  }
 
+  return KernelLaunchParams;
+}
+
+Address CGNVCUDARuntime::prepareKernelArgs(CodeGenFunction &CGF,
+                                           FunctionArgList &Args) {
   // Calculate amount of space we will need for all arguments.  If we have no
   // args, allocate a single pointer so we still have a valid pointer to the
   // argument array that we can pass to runtime, even if it will be unused.
@@ -331,6 +385,17 @@ void CGNVCUDARuntime::emitDeviceStubBodyNew(CodeGenFunction &CGF,
         VoidVarPtr, CGF.Builder.CreateConstGEP1_32(
                         PtrTy, KernelArgs.emitRawPointer(CGF), i));
   }
+  return KernelArgs;
+}
+
+// CUDA 9.0+ uses new way to launch kernels. Parameters are packed in a local
+// array and kernels are launched using cudaLaunchKernel().
+void CGNVCUDARuntime::emitDeviceStubBodyNew(CodeGenFunction &CGF,
+                                            FunctionArgList &Args) {
+  // Build the shadow stack entry at the very start of the function.
+  Address KernelArgs = CGF.getLangOpts().OffloadViaLLVM
+                           ? prepareKernelArgsLLVMOffload(CGF, Args)
+                           : prepareKernelArgs(CGF, Args);
 
   llvm::BasicBlock *EndBlock = CGF.createBasicBlock("setup.end");
 
@@ -1129,8 +1194,9 @@ void CGNVCUDARuntime::transformManagedVars() {
 // registered. The linker will provide a pointer to this section so we can
 // register the symbols with the linked device image.
 void CGNVCUDARuntime::createOffloadingEntries() {
-  StringRef Section = CGM.getLangOpts().HIP ? "hip_offloading_entries"
-                                            : "cuda_offloading_entries";
+  SmallVector<char, 32> Out;
+  StringRef Section = (SectionPrefix + "_offloading_entries").toStringRef(Out);
+
   llvm::Module &M = CGM.getModule();
   for (KernelInfo &I : EmittedKernels)
     llvm::offloading::emitOffloadingEntry(
@@ -1199,7 +1265,8 @@ llvm::Function *CGNVCUDARuntime::finalizeModule() {
     }
     return nullptr;
   }
-  if (CGM.getLangOpts().OffloadingNewDriver && RelocatableDeviceCode)
+  if (CGM.getLangOpts().OffloadViaLLVM ||
+      (CGM.getLangOpts().OffloadingNewDriver && RelocatableDeviceCode))
     createOffloadingEntries();
   else
     return makeModuleCtorFunction();
diff --git a/clang/lib/CodeGen/CodeGenFunction.cpp b/clang/lib/CodeGen/CodeGenFunction.cpp
index 2b2e23f1e5d7fb..7adc5a48b7c47b 100644
--- a/clang/lib/CodeGen/CodeGenFunction.cpp
+++ b/clang/lib/CodeGen/CodeGenFunction.cpp
@@ -1013,7 +1013,8 @@ void CodeGenFunction::StartFunction(GlobalDecl GD, QualType RetTy,
   }
 
   if (FD && (getLangOpts().OpenCL ||
-             (getLangOpts().HIP && getLangOpts().CUDAIsDevice))) {
+             ((getLangOpts().HIP || getLangOpts().OffloadViaLLVM) &&
+              getLangOpts().CUDAIsDevice))) {
     // Add metadata for a kernel function.
     EmitKernelMetadata(FD, Fn);
   }
diff --git a/clang/lib/Driver/Driver.cpp b/clang/lib/Driver/Driver.cpp
index f4e909b79389bc..e12416e51f8d24 100644
--- a/clang/lib/Driver/Driver.cpp
+++ b/clang/lib/Driver/Driver.cpp
@@ -786,11 +786,13 @@ void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
                    }) ||
       C.getInputArgs().hasArg(options::OPT_hip_link) ||
       C.getInputArgs().hasArg(options::OPT_hipstdpar);
+  bool UseLLVMOffload = C.getInputArgs().hasArg(
+      options::OPT_foffload_via_llvm, options::OPT_fno_offload_via_llvm, false);
   if (IsCuda && IsHIP) {
     Diag(clang::diag::err_drv_mix_cuda_hip);
     return;
   }
-  if (IsCuda) {
+  if (IsCuda && !UseLLVMOffload) {
     const ToolChain *HostTC = C.getSingleOffloadToolChain<Action::OFK_Host>();
     const llvm::Triple &HostTriple = HostTC->getTriple();
     auto OFK = Action::OFK_Cuda;
@@ -812,7 +814,7 @@ void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
         CudaInstallation.WarnIfUnsupportedVersion();
     }
     C.addOffloadDeviceToolChain(CudaTC.get(), OFK);
-  } else if (IsHIP) {
+  } else if (IsHIP && !UseLLVMOffload) {
     if (auto *OMPTargetArg =
             C.getInputArgs().getLastArg(options::OPT_fopenmp_targets_EQ)) {
       Diag(clang::diag::err_drv_unsupported_opt_for_language_mode)
@@ -836,10 +838,11 @@ void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
   // We need to generate an OpenMP toolchain if the user specified targets with
   // the -fopenmp-targets option or used --offload-arch with OpenMP enabled.
   bool IsOpenMPOffloading =
-      C.getInputArgs().hasFlag(options::OPT_fopenmp, options::OPT_fopenmp_EQ,
-                               options::OPT_fno_openmp, false) &&
-      (C.getInputArgs().hasArg(options::OPT_fopenmp_targets_EQ) ||
-       C.getInputArgs().hasArg(options::OPT_offload_arch_EQ));
+      ((IsCuda || IsHIP) && UseLLVMOffload) ||
+      (C.getInputArgs().hasFlag(options::OPT_fopenmp, options::OPT_fopenmp_EQ,
+                                options::OPT_fno_openmp, false) &&
+       (C.getInputArgs().hasArg(options::OPT_fopenmp_targets_EQ) ||
+        C.getInputArgs().hasArg(options::OPT_offload_arch_EQ)));
   if (IsOpenMPOffloading) {
     // We expect that -fopenmp-targets is always used in conjunction with the
     // option -fopenmp specifying a valid runtime with offloading support, i.e.
@@ -867,7 +870,7 @@ void Driver::CreateOffloadingDeviceToolChains(Compilation &C,
       for (StringRef T : OpenMPTargets->getValues())
         OpenMPTriples.insert(T);
     } else if (C.getInputArgs().hasArg(options::OPT_offload_arch_EQ) &&
-               !IsHIP && !IsCuda) {
+               ((!IsHIP && !IsCuda) || UseLLVMOffload)) {
       const ToolChain *HostTC = C.getSingleOffloadToolChain<Action::OFK_Host>();
       auto AMDTriple = getHIPOffloadTargetTriple(*this, C.getInputArgs());
       auto NVPTXTriple = getNVIDIAOffloadTargetTriple(*this, C.getInputArgs(),
@@ -4152,6 +4155,8 @@ void Driver::BuildActions(Compilation &C, DerivedArgList &Args,
 
   bool UseNewOffloadingDriver =
       C.isOffloadingHostKind(Action::OFK_OpenMP) ||
+      Args.hasFlag(options::OPT_foffload_via_llvm,
+                   options::OPT_fno_offload_via_llvm, false) ||
       Args.hasFlag(options::OPT_offload_new_driver,
                    options::OPT_no_offload_new_driver, false);
 
diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp
index c698d38b80e578..96aa930ea28612 100644
--- a/clang/lib/Driver/ToolChains/Clang.cpp
+++ b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -1095,6 +1095,18 @@ void Clang::AddPreprocessingOptions(Compilation &C, const JobAction &JA,
     CmdArgs.push_back("__clang_openmp_device_functions.h");
   }
 
+  if (Args.hasArg(options::OPT_foffload_via_llvm)) {
+    // Add llvm_wrappers/* to our system include path.  This lets us wrap
+    // standard library headers and other headers.
+    SmallString<128> P(D.ResourceDir);
+    llvm::sys::path::append(P, "include", "llvm_offload_wrappers");
+    CmdArgs.append({"-internal-isystem", Args.MakeArgString(P), "-include"});
+    if (JA.isDeviceOffloading(Action::OFK_OpenMP))
+      CmdArgs.push_back("__llvm_offload_device.h");
+    else
+      CmdArgs.push_back("__llvm_offload_host.h");
+  }
+
   // Add -i* options, and automatically translate to
   // -include-pch/-include-pth for transparent PCH support. It's
   // wonky, but we include looking for .gch so we can support seamless
@@ -6665,6 +6677,8 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   // device offloading action other than OpenMP.
   if (Args.hasFlag(options::OPT_fopenmp, options::OPT_fopenmp_EQ,
                    options::OPT_fno_openmp, false) &&
+      !Args.hasFlag(options::OPT_foffload_via_llvm,
+                    options::OPT_fno_offload_via_llvm, false) &&
       (JA.isDeviceOffloading(Action::OFK_None) ||
        JA.isDeviceOffloading(Action::OFK_OpenMP))) {
     switch (D.getOpenMPRuntime(Args)) {
@@ -6742,11 +6756,16 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
     Args.addOptOutFlag(CmdArgs, options::OPT_fopenmp_extensions,
                        options::OPT_fno_openmp_extensions);
   }
-
-  // Forward the new driver to change offloading code generation.
-  if (Args.hasFlag(options::OPT_offload_new_driver,
-                   options::OPT_no_offload_new_driver, false))
+  // Forward the offload runtime change to code generation, liboffload implies
+  // new driver. Otherwise, check if we should forward the new driver to change
+  // offloading code generation.
+  if (Args.hasFlag(options::OPT_foffload_via_llvm,
+                   options::OPT_fno_offload_via_llvm, false)) {
+    CmdArgs.append({"--offload-new-driver", "-foffload-via-llvm"});
+  } else if (Args.hasFlag(options::OPT_offload_new_driver,
+                          options::OPT_no_offload_new_driver, false)) {
     CmdArgs.push_back("--offload-new-driver");
+  }
 
   SanitizeArgs.addArgs(TC, Args, CmdArgs, InputType);
 
@@ -7778,6 +7797,10 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   // so that only the relevant declarations are emitted.
   if (IsOpenMPDevice) {
     CmdArgs.push_back("-fopenmp-is-target-device");
+    // If we are offloading cuda/hip via llvm, it's also "cuda device code".
+    if (Args.hasArg(options::OPT_foffload_via_llvm))
+      CmdArgs.push_back("-fcuda-is-device");
+
     if (OpenMPDeviceInput) {
       CmdArgs.push_back("-fopenmp-host-ir-file-path");
       CmdArgs.push_back(Args.MakeArgString(OpenMPDeviceInput->getFilename()));
diff --git a/clang/lib/Driver/ToolChains/CommonArgs.cpp b/clang/lib/Driver/ToolChains/CommonArgs.cpp
index 6e9744607d9ebc..1cba3e1220264a 100644
--- a/clang/lib/Driver/ToolChains/CommonArgs.cpp
+++ b/clang/lib/Driver/ToolChains/CommonArgs.cpp
@@ -1199,8 +1199,13 @@ bool tools::addOpenMPRuntime(const Compilation &C, ArgStringList &CmdArgs,
                              bool ForceStaticHostRuntime, bool IsOffloadingHost,
                              bool GompNeedsRT) {
   if (!Args.hasFlag(options::OPT_fopenmp, options::OPT_fopenmp_EQ,
-                    options::OPT_fno_openmp, false))
+                    options::OPT_fno_openmp, false)) {
+    // We need libomptarget (liboffload) if it's the choosen offloading runtime.
+    if (Args.hasFlag(options::OPT_foffload_via_llvm,
+                     options::OPT_fno_offload_via_llvm, false))
+      CmdArgs.push_back("-lomptarget");
     return false;
+  }
 
   Driver::OpenMPRuntimeKind RTKind = TC.getDriver().getOpenMPRuntime(Args);
 
diff --git a/clang/lib/Driver/ToolChains/Cuda.cpp b/clang/lib/Driver/ToolChains/Cuda.cpp
index 67a427b9d5ceee..3f9885b196ec55 100644
--- a/clang/lib/Driver/ToolChains/Cuda.cpp
+++ b/clang/lib/Driver/ToolChains/Cuda.cpp
@@ -839,17 +839,15 @@ void CudaToolChain::addClangTargetOptions(
           DeviceOffloadingKind == Action::OFK_Cuda) &&
          "Only OpenMP or CUDA offloading kinds are supported for NVIDIA GPUs.");
 
-  if (DeviceOffloadingKind == Action::OFK_Cuda) {
-    CC1Args.append(
-        {"-fcuda-is-device", "-mllvm", "-enable-memcpyopt-without-libcalls"});
-
-    // Unsized function arguments used for variadics were introduced in CUDA-9.0
-    // We still do not support generating code that actually uses variadic
-    // arguments yet, but we do need to allow parsing them as recent CUDA
-    // headers rely on that. https://github.com/llvm/llvm-project/issues/58410
-    if (CudaInstallation.version() >= CudaVersion::CUDA_90)
-      CC1Args.push_back("-fcuda-allow-variadic-functions");
-  }
+  CC1Args.append(
+      {"-fcuda-is-device", "-mllvm", "-enable-memcpyopt-without-libcalls"});
+
+  // Unsized function arguments used for variadics were introduced in CUDA-9.0
+  // We still do not support generating code that actually uses variadic
+  // arguments yet, but we do need to allow parsing them as recent CUDA
+  // headers rely on that. https://github.com/llvm/llvm-project/issues/58410
+  if (CudaInstallation.version() >= CudaVersion::CUDA_90)
+    CC1Args.push_back("-fcuda-allow-variadic-functions");
 
   if (DriverArgs.hasArg(options::OPT_nogpulib))
     return;
@@ -867,6 +865,13 @@ void CudaToolChain::addClangTargetOptions(
   CC1Args.push_back("-mlink-builtin-bitcode");
   CC1Args.push_back(DriverArgs.MakeArgString(LibDeviceFile));
 
+  // For now, we don't use any Offload/OpenMP device runtime when we offload
+  // CUDA via LLVM/Offload. We should split the Offload/OpenMP device runtime
+  // and include the "generic" (or CUDA-specific) parts.
+  if (DriverArgs.hasFlag(options::OPT_foffload_via_llvm,
+                         options::OPT_fno_offload_via_llvm, false))
+    return;
+
   clang::CudaVersion CudaInstallationVersion = CudaInstallation.version();
 
   if (DriverArgs.hasFlag(options::OPT_fcuda_short_ptr,
diff --git a/clang/lib/Headers/CMakeLists.txt b/clang/lib/Headers/CMakeLists.txt
index b61aeca6bbc910..598bc556e8330a 100644
--- a/clang/lib/Headers/CMakeLists.txt
+++ b/clang/lib/Headers/CMakeLists.txt
@@ -333,6 +333,12 @@ set(openmp_wrapper_files
   openmp_wrappers/new
 )
 
+set(llvm_offload_wrapper_files
+  llvm_offload_wrappers/__llvm_offload.h
+  llvm_offload_wrappers/__llvm_offload_host.h
+  llvm_offload_wrappers/__llvm_offload_device.h
+)
+
 set(llvm_libc_wrapper_files
   llvm_libc_wrappers/assert.h
   llvm_libc_wrappers/stdio.h
@@ -383,7 +389,7 @@ endfunction(clang_generate_header)
 # Copy header files from the source directory to the build directory
 foreach( f ${files} ${cuda_wrapper_files} ${cuda_wrapper_bits_files}
            ${ppc_wrapper_files} ${openmp_wrapper_files} ${zos_wrapper_files} ${hlsl_files}
-           ${llvm_libc_wrapper_files})
+	   ${llvm_libc_wrapper_files} ${llvm_offload_wrapper_files})
   copy_header_to_output_dir(${CMAKE_CURRENT_SOURCE_DIR} ${f})
 endforeach( f )
 
@@ -509,6 +515,7 @@ add_header_target("hlsl-resource-headers" ${hlsl_files})
 add_header_target("opencl-resource-headers" ${opencl_files})
 add_header_target("llvm-libc-resource-headers" ${llvm_libc_wrapper_files})
 add_header_target("openmp-resource-headers" ${openmp_wrapper_files})
+add_header_target("llvm-offload-resource-headers" ${llvm_libc_wrapper_files})
 add_header_target("windows-resource-headers" ${windows_only_files})
 add_header_target("utility-resource-headers" ${utility_files})
 
@@ -550,6 +557,11 @@ install(
   DESTINATION ${header_install_dir}/openmp_wrappers
   COMPONENT clang-resource-headers)
 
+install(
+  FILES ${llvm_offload_wrapper_files}
+  DESTINATION ${header_install_dir}/llvm_offload_wrappers
+  COMPONENT clang-resource-headers)
+
 install(
   FILES ${zos_wrapper_files}
   DESTINATION ${header_install_dir}/zos_wrappers
@@ -712,8 +724,8 @@ install(
   COMPONENT openmp-resource-headers)
 
 install(
-  FILES ${openmp_wrapper_files}
-  DESTINATION ${header_install_dir}/openmp_wrappers
+  FILES ${llvm_offload_wrapper_files}
+  DESTINATION ${header_install_dir}/llvm_offload_wrappers
   EXCLUDE_FROM_ALL
   COMPONENT openmp-resource-headers)
 
diff --git a/clang/lib/Headers/llvm_offload_wrappers/__llvm_offload.h b/clang/lib/Headers/llvm_offload_wrappers/__llvm_offload.h
new file mode 100644
index 00000000000000..2898898904e299
--- /dev/null
+++ b/clang/lib/Headers/llvm_offload_wrappers/__llvm_offload.h
@@ -0,0 +1,31 @@
+/*===------ LLVM/Offload helpers for kernel languages (CUDA/HIP) -*- c++ -*-===
+ *
+ * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+ * See https://llvm.org/LICENSE.txt for license information.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ *
+ *===-----------------------------------------------------------------------===
+ */
+
+#include <stddef.h>
+
+#define __host__ __attribute__((host))
+#define __device__ __attribute__((device))
+#define __global__ __attribute__((global))
+#define __shared__ __attribute__((shared))
+#define __constant__ __attribute__((constant))
+#define __managed__ __attribute__((managed))
+
+extern "C" {
+
+typedef struct dim3 {
+  dim3() {}
+  dim3(unsigned x) : x(x) {}
+  unsigned x = 0, y = 0, z = 0;
+} dim3;
+
+// TODO: For some reason the CUDA device compilation requires this declaration
+// to be present on the device while it is only used on the host.
+unsigned __llvmPushCallConfiguration(dim3 gridDim, dim3 blockDim,
+                                     size_t sharedMem = 0, void *stream = 0);
+}
diff --git a/clang/lib/Headers/llvm_offload_wrappers/__llvm_offload_device.h b/clang/lib/Headers/llvm_offload_wrappers/__llvm_offload_device.h
new file mode 100644
index 00000000000000..1a813b331515b2
--- /dev/null
+++ b/clang/lib/Headers/llvm_offload_wrappers/__llvm_offload_device.h
@@ -0,0 +1,10 @@
+/*===------ LLVM/Offload helpers for kernel languages (CUDA/HIP) -*- c++ -*-===
+ *
+ * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+ * See https://llvm.org/LICENSE.txt for license information.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ *
+ *===-----------------------------------------------------------------------===
+ */
+
+#include "__llvm_offload.h"
diff --git a/clang/lib/Headers/llvm_offload_wrappers/__llvm_offload_host.h b/clang/lib/Headers/llvm_offload_wrappers/__llvm_offload_host.h
new file mode 100644
index 00000000000000..160289d169b55e
--- /dev/null
+++ b/clang/lib/Headers/llvm_offload_wrappers/__llvm_offload_host.h
@@ -0,0 +1,15 @@
+/*===------ LLVM/Offload helpers for kernel languages (CUDA/HIP) -*- c++ -*-===
+ *
+ * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+ * See https://llvm.org/LICENSE.txt for license information.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ *
+ *===-----------------------------------------------------------------------===
+ */
+
+#include "__llvm_offload.h"
+
+extern "C" {
+unsigned llvmLaunchKernel(const void *func, dim3 gridDim, dim3 blockDim,
+                          void **args, size_t sharedMem = 0, void *stream = 0);
+}
diff --git a/clang/lib/Headers/openmp_wrappers/__clang_openmp_device_functions.h b/clang/lib/Headers/openmp_wrappers/__clang_openmp_device_functions.h
index d5b6846b034885..3e354c63efc668 100644
--- a/clang/lib/Headers/openmp_wrappers/__clang_openmp_device_functions.h
+++ b/clang/lib/Headers/openmp_wrappers/__clang_openmp_device_functions.h
@@ -10,17 +10,15 @@
 #ifndef __CLANG_OPENMP_DEVICE_FUNCTIONS_H__
 #define __CLANG_OPENMP_DEVICE_FUNCTIONS_H__
 
-#ifndef _OPENMP
-#error "This file is for OpenMP compilation only."
-#endif
-
 #ifdef __cplusplus
 extern "C" {
 #endif
 
+#ifdef __NVPTX__
 #pragma omp begin declare variant match(                                       \
     device = {arch(nvptx, nvptx64)}, implementation = {extension(match_any)})
 
+#pragma push_macro("__CUDA__")
 #define __CUDA__
 #define __OPENMP_NVPTX__
 
@@ -31,9 +29,10 @@ extern "C" {
 #include <__clang_cuda_device_functions.h>
 
 #undef __OPENMP_NVPTX__
-#undef __CUDA__
+#pragma pop_macro("__CUDA__")
 
 #pragma omp end declare variant
+#endif
 
 #ifdef __AMDGCN__
 #pragma omp begin declare variant match(device = {arch(amdgcn)})
diff --git a/clang/lib/Sema/SemaCUDA.cpp b/clang/lib/Sema/SemaCUDA.cpp
index 580b9872c6a1de..ec37c0df56c671 100644
--- a/clang/lib/Sema/SemaCUDA.cpp
+++ b/clang/lib/Sema/SemaCUDA.cpp
@@ -1068,6 +1068,9 @@ void SemaCUDA::inheritTargetAttrs(FunctionDecl *FD,
 }
 
 std::string SemaCUDA::getConfigureFuncName() const {
+  if (getLangOpts().OffloadViaLLVM)
+    return "__llvmPushCallConfiguration";
+
   if (getLangOpts().HIP)
     return getLangOpts().HIPUseNewLaunchAPI ? "__hipPushCallConfiguration"
                                             : "hipConfigureCall";
diff --git a/clang/test/CodeGenCUDA/Inputs/cuda.h b/clang/test/CodeGenCUDA/Inputs/cuda.h
index 8df425b77ac823..dc760500e65d41 100644
--- a/clang/test/CodeGenCUDA/Inputs/cuda.h
+++ b/clang/test/CodeGenCUDA/Inputs/cuda.h
@@ -46,6 +46,11 @@ extern "C" hipError_t hipLaunchKernel_spt(const void *func, dim3 gridDim,
                                       size_t sharedMem,
                                       hipStream_t stream);
 #endif // __HIP_API_PER_THREAD_DEFAULT_STREAM__
+#elif __OFFLOAD_VIA_LLVM__
+extern "C" unsigned __llvmPushCallConfiguration(dim3 gridDim, dim3 blockDim,
+                                     size_t sharedMem = 0, void *stream = 0);
+extern "C" unsigned llvmLaunchKernel(const void *func, dim3 gridDim, dim3 blockDim,
+                          void **args, size_t sharedMem = 0, void *stream = 0);
 #else
 typedef struct cudaStream *cudaStream_t;
 typedef enum cudaError {} cudaError_t;
diff --git a/clang/test/CodeGenCUDA/offload_via_llvm.cu b/clang/test/CodeGenCUDA/offload_via_llvm.cu
new file mode 100644
index 00000000000000..434eba99c1795d
--- /dev/null
+++ b/clang/test/CodeGenCUDA/offload_via_llvm.cu
@@ -0,0 +1,90 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 5
+// RUN: %clang_cc1 %s -triple nvptx-unknown-unknown -foffload-via-llvm -emit-llvm -o - | FileCheck %s --check-prefix=HST
+// RUN: %clang_cc1 %s -triple nvptx-unknown-unknown -fcuda-is-device -foffload-via-llvm -emit-llvm -o - | FileCheck %s --check-prefix=DEV
+
+// Check that we generate LLVM/Offload calls, including the KERNEL_LAUNCH_PARAMS argument.
+
+#define __OFFLOAD_VIA_LLVM__ 1
+#include "Inputs/cuda.h"
+
+// HST-LABEL: define dso_local void @_Z18__device_stub__fooisPvS_(
+// HST-SAME: i32 noundef [[TMP0:%.*]], i16 noundef signext [[TMP1:%.*]], ptr noundef [[TMP2:%.*]], ptr noundef [[TMP3:%.*]]) #[[ATTR0:[0-9]+]] {
+// HST-NEXT:  [[ENTRY:.*:]]
+// HST-NEXT:    [[DOTADDR:%.*]] = alloca i32, align 4
+// HST-NEXT:    [[DOTADDR1:%.*]] = alloca i16, align 2
+// HST-NEXT:    [[DOTADDR2:%.*]] = alloca ptr, align 4
+// HST-NEXT:    [[DOTADDR3:%.*]] = alloca ptr, align 4
+// HST-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[TMP0]], align 16
+// HST-NEXT:    [[KERNEL_LAUNCH_PARAMS:%.*]] = alloca [[TMP1]], align 16
+// HST-NEXT:    [[GRID_DIM:%.*]] = alloca [[STRUCT_DIM3:%.*]], align 8
+// HST-NEXT:    [[BLOCK_DIM:%.*]] = alloca [[STRUCT_DIM3]], align 8
+// HST-NEXT:    [[SHMEM_SIZE:%.*]] = alloca i32, align 4
+// HST-NEXT:    [[STREAM:%.*]] = alloca ptr, align 4
+// HST-NEXT:    store i32 [[TMP0]], ptr [[DOTADDR]], align 4
+// HST-NEXT:    store i16 [[TMP1]], ptr [[DOTADDR1]], align 2
+// HST-NEXT:    store ptr [[TMP2]], ptr [[DOTADDR2]], align 4
+// HST-NEXT:    store ptr [[TMP3]], ptr [[DOTADDR3]], align 4
+// HST-NEXT:    [[TMP4:%.*]] = getelementptr inbounds nuw [[TMP1]], ptr [[KERNEL_LAUNCH_PARAMS]], i32 0, i32 0
+// HST-NEXT:    store i64 16, ptr [[TMP4]], align 16
+// HST-NEXT:    [[TMP5:%.*]] = getelementptr inbounds nuw [[TMP1]], ptr [[KERNEL_LAUNCH_PARAMS]], i32 0, i32 1
+// HST-NEXT:    store ptr [[KERNEL_ARGS]], ptr [[TMP5]], align 8
+// HST-NEXT:    [[TMP6:%.*]] = getelementptr inbounds nuw [[TMP1]], ptr [[KERNEL_LAUNCH_PARAMS]], i32 0, i32 2
+// HST-NEXT:    store ptr null, ptr [[TMP6]], align 4
+// HST-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTADDR]], align 4
+// HST-NEXT:    [[TMP8:%.*]] = getelementptr inbounds nuw [[TMP0]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// HST-NEXT:    store i32 [[TMP7]], ptr [[TMP8]], align 16
+// HST-NEXT:    [[TMP9:%.*]] = load i16, ptr [[DOTADDR1]], align 2
+// HST-NEXT:    [[TMP10:%.*]] = getelementptr inbounds nuw [[TMP0]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// HST-NEXT:    store i16 [[TMP9]], ptr [[TMP10]], align 4
+// HST-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[DOTADDR2]], align 4
+// HST-NEXT:    [[TMP12:%.*]] = getelementptr inbounds nuw [[TMP0]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// HST-NEXT:    store ptr [[TMP11]], ptr [[TMP12]], align 8
+// HST-NEXT:    [[TMP13:%.*]] = load ptr, ptr [[DOTADDR3]], align 4
+// HST-NEXT:    [[TMP14:%.*]] = getelementptr inbounds nuw [[TMP0]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// HST-NEXT:    store ptr [[TMP13]], ptr [[TMP14]], align 4
+// HST-NEXT:    [[TMP15:%.*]] = call i32 @__llvmPopCallConfiguration(ptr [[GRID_DIM]], ptr [[BLOCK_DIM]], ptr [[SHMEM_SIZE]], ptr [[STREAM]])
+// HST-NEXT:    [[TMP16:%.*]] = load i32, ptr [[SHMEM_SIZE]], align 4
+// HST-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[STREAM]], align 4
+// HST-NEXT:    [[CALL:%.*]] = call noundef i32 @llvmLaunchKernel(ptr noundef @_Z18__device_stub__fooisPvS_, ptr noundef byval([[STRUCT_DIM3]]) align 4 [[GRID_DIM]], ptr noundef byval([[STRUCT_DIM3]]) align 4 [[BLOCK_DIM]], ptr noundef [[KERNEL_LAUNCH_PARAMS]], i32 noundef [[TMP16]], ptr noundef [[TMP17]])
+// HST-NEXT:    br label %[[SETUP_END:.*]]
+// HST:       [[SETUP_END]]:
+// HST-NEXT:    ret void
+//
+// DEV-LABEL: define dso_local void @_Z3fooisPvS_(
+// DEV-SAME: i32 noundef [[TMP0:%.*]], i16 noundef signext [[TMP1:%.*]], ptr noundef [[TMP2:%.*]], ptr noundef [[TMP3:%.*]]) #[[ATTR0:[0-9]+]] {
+// DEV-NEXT:  [[ENTRY:.*:]]
+// DEV-NEXT:    [[DOTADDR:%.*]] = alloca i32, align 4
+// DEV-NEXT:    [[DOTADDR1:%.*]] = alloca i16, align 2
+// DEV-NEXT:    [[DOTADDR2:%.*]] = alloca ptr, align 4
+// DEV-NEXT:    [[DOTADDR3:%.*]] = alloca ptr, align 4
+// DEV-NEXT:    store i32 [[TMP0]], ptr [[DOTADDR]], align 4
+// DEV-NEXT:    store i16 [[TMP1]], ptr [[DOTADDR1]], align 2
+// DEV-NEXT:    store ptr [[TMP2]], ptr [[DOTADDR2]], align 4
+// DEV-NEXT:    store ptr [[TMP3]], ptr [[DOTADDR3]], align 4
+// DEV-NEXT:    ret void
+//
+__global__ void foo(int, short, void *, void *) {}
+
+// HST-LABEL: define dso_local void @_Z5test1Pv(
+// HST-SAME: ptr noundef [[PTR:%.*]]) #[[ATTR1:[0-9]+]] {
+// HST-NEXT:  [[ENTRY:.*:]]
+// HST-NEXT:    [[PTR_ADDR:%.*]] = alloca ptr, align 4
+// HST-NEXT:    [[AGG_TMP:%.*]] = alloca [[STRUCT_DIM3:%.*]], align 4
+// HST-NEXT:    [[AGG_TMP1:%.*]] = alloca [[STRUCT_DIM3]], align 4
+// HST-NEXT:    store ptr [[PTR]], ptr [[PTR_ADDR]], align 4
+// HST-NEXT:    call void @_ZN4dim3C1Ejjj(ptr noundef nonnull align 4 dereferenceable(12) [[AGG_TMP]], i32 noundef 3, i32 noundef 1, i32 noundef 1)
+// HST-NEXT:    call void @_ZN4dim3C1Ejjj(ptr noundef nonnull align 4 dereferenceable(12) [[AGG_TMP1]], i32 noundef 7, i32 noundef 1, i32 noundef 1)
+// HST-NEXT:    [[CALL:%.*]] = call i32 @__llvmPushCallConfiguration(ptr noundef byval([[STRUCT_DIM3]]) align 4 [[AGG_TMP]], ptr noundef byval([[STRUCT_DIM3]]) align 4 [[AGG_TMP1]], i32 noundef 0, ptr noundef null)
+// HST-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[CALL]], 0
+// HST-NEXT:    br i1 [[TOBOOL]], label %[[KCALL_END:.*]], label %[[KCALL_CONFIGOK:.*]]
+// HST:       [[KCALL_CONFIGOK]]:
+// HST-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[PTR_ADDR]], align 4
+// HST-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[PTR_ADDR]], align 4
+// HST-NEXT:    call void @_Z18__device_stub__fooisPvS_(i32 noundef 13, i16 noundef signext 1, ptr noundef [[TMP0]], ptr noundef [[TMP1]]) #[[ATTR3:[0-9]+]]
+// HST-NEXT:    br label %[[KCALL_END]]
+// HST:       [[KCALL_END]]:
+// HST-NEXT:    ret void
+//
+void test1(void *Ptr) {
+  foo<<<3, 7>>>(13, 1, Ptr, Ptr);
+}
diff --git a/clang/test/Driver/cuda-via-liboffload.cu b/clang/test/Driver/cuda-via-liboffload.cu
new file mode 100644
index 00000000000000..68dc963e906b20
--- /dev/null
+++ b/clang/test/Driver/cuda-via-liboffload.cu
@@ -0,0 +1,23 @@
+// RUN: %clang -### -target x86_64-linux-gnu -foffload-via-llvm -ccc-print-bindings \
+// RUN:        --offload-arch=sm_35 --offload-arch=sm_70 %s 2>&1 \
+// RUN: | FileCheck -check-prefix BINDINGS %s
+
+//      BINDINGS: "x86_64-unknown-linux-gnu" - "clang", inputs: ["[[INPUT:.+]]"], output: "[[HOST_BC:.+]]"
+// BINDINGS-NEXT: "nvptx64-nvidia-cuda" - "clang", inputs: ["[[INPUT]]", "[[HOST_BC]]"], output: "[[PTX_SM_35:.+]]"
+// BINDINGS-NEXT: "nvptx64-nvidia-cuda" - "NVPTX::Assembler", inputs: ["[[PTX_SM_35]]"], output: "[[CUBIN_SM_35:.+]]"
+// BINDINGS-NEXT: "nvptx64-nvidia-cuda" - "clang", inputs: ["[[INPUT]]", "[[HOST_BC]]"], output: "[[PTX_SM_70:.+]]"
+// BINDINGS-NEXT: "nvptx64-nvidia-cuda" - "NVPTX::Assembler", inputs: ["[[PTX_SM_70:.+]]"], output: "[[CUBIN_SM_70:.+]]"
+// BINDINGS-NEXT: "x86_64-unknown-linux-gnu" - "Offload::Packager", inputs: ["[[CUBIN_SM_35]]", "[[CUBIN_SM_70]]"], output: "[[BINARY:.+]]"
+// BINDINGS-NEXT: "x86_64-unknown-linux-gnu" - "clang", inputs: ["[[HOST_BC]]", "[[BINARY]]"], output: "[[HOST_OBJ:.+]]"
+// BINDINGS-NEXT: "x86_64-unknown-linux-gnu" - "Offload::Linker", inputs: ["[[HOST_OBJ]]"], output: "a.out"
+
+// RUN: %clang -### -target x86_64-linux-gnu -foffload-via-llvm -ccc-print-bindings \
+// RUN:        --offload-arch=sm_35 --offload-arch=sm_70 %s 2>&1 \
+// RUN: | FileCheck -check-prefix BINDINGS-DEVICE %s
+
+// BINDINGS-DEVICE: # "nvptx64-nvidia-cuda" - "clang", inputs: ["[[INPUT:.+]]"], output: "[[PTX:.+]]"
+// BINDINGS-DEVICE: # "nvptx64-nvidia-cuda" - "NVPTX::Assembler", inputs: ["[[PTX]]"], output: "[[CUBIN:.+]]"
+
+// RUN: %clang -### -target x86_64-linux-gnu -ccc-print-bindings --offload-link -foffload-via-llvm %s 2>&1 | FileCheck -check-prefix DEVICE-LINK %s
+
+// DEVICE-LINK: "x86_64-unknown-linux-gnu" - "Offload::Linker", inputs: ["[[INPUT:.+]]"], output: "a.out"
diff --git a/offload/include/Shared/APITypes.h b/offload/include/Shared/APITypes.h
index 5b22bbaac144fe..4c1f7712249a3a 100644
--- a/offload/include/Shared/APITypes.h
+++ b/offload/include/Shared/APITypes.h
@@ -102,8 +102,9 @@ struct KernelArgsTy {
       0; // Tripcount for the teams / distribute loop, 0 otherwise.
   struct {
     uint64_t NoWait : 1; // Was this kernel spawned with a `nowait` clause.
-    uint64_t Unused : 63;
-  } Flags = {0, 0};
+    uint64_t IsCUDA : 1; // Was this kernel spawned via CUDA.
+    uint64_t Unused : 62;
+  } Flags = {0, 0, 0};
   // The number of teams (for x,y,z dimension).
   uint32_t NumTeams[3] = {0, 0, 0};
    // The number of threads (for x,y,z dimension).
diff --git a/offload/include/omptarget.h b/offload/include/omptarget.h
index 323dee41630f2f..2b6445e9fbe550 100644
--- a/offload/include/omptarget.h
+++ b/offload/include/omptarget.h
@@ -107,7 +107,7 @@ enum TargetAllocTy : int32_t {
 
 inline KernelArgsTy CTorDTorKernelArgs = {1,       0,       nullptr,   nullptr,
 	     nullptr, nullptr, nullptr,   nullptr,
-	     0,      {0,0},       {1, 0, 0}, {1, 0, 0}, 0};
+	     0,      {0,0,0},       {1, 0, 0}, {1, 0, 0}, 0};
 
 struct DeviceTy;
 
diff --git a/offload/plugins-nextgen/common/src/PluginInterface.cpp b/offload/plugins-nextgen/common/src/PluginInterface.cpp
index c3ecbcc62f71f1..84d946507ea74a 100644
--- a/offload/plugins-nextgen/common/src/PluginInterface.cpp
+++ b/offload/plugins-nextgen/common/src/PluginInterface.cpp
@@ -552,9 +552,17 @@ Error GenericKernelTy::launch(GenericDeviceTy &GenericDevice, void **ArgPtrs,
   if (!KernelLaunchEnvOrErr)
     return KernelLaunchEnvOrErr.takeError();
 
-  KernelLaunchParamsTy LaunchParams =
-      prepareArgs(GenericDevice, ArgPtrs, ArgOffsets, KernelArgs.NumArgs, Args,
-                  Ptrs, *KernelLaunchEnvOrErr);
+  KernelLaunchParamsTy LaunchParams;
+
+  // Kernel languages don't use indirection.
+  if (KernelArgs.Flags.IsCUDA) {
+    LaunchParams =
+        *reinterpret_cast<KernelLaunchParamsTy *>(KernelArgs.ArgPtrs);
+  } else {
+    LaunchParams =
+        prepareArgs(GenericDevice, ArgPtrs, ArgOffsets, KernelArgs.NumArgs,
+                    Args, Ptrs, *KernelLaunchEnvOrErr);
+  }
 
   uint32_t NumThreads = getNumThreads(GenericDevice, KernelArgs.ThreadLimit);
   uint64_t NumBlocks =
diff --git a/offload/src/CMakeLists.txt b/offload/src/CMakeLists.txt
index 344069b6fcdcf6..c5f5d902fad14c 100644
--- a/offload/src/CMakeLists.txt
+++ b/offload/src/CMakeLists.txt
@@ -22,6 +22,7 @@ add_llvm_library(omptarget
   OpenMP/InteropAPI.cpp
   OpenMP/OMPT/Callback.cpp
 
+  KernelLanguage/API.cpp
 
   ADDITIONAL_HEADER_DIRS
   ${LIBOMPTARGET_INCLUDE_DIR}
diff --git a/offload/src/KernelLanguage/API.cpp b/offload/src/KernelLanguage/API.cpp
new file mode 100644
index 00000000000000..360435a6a8361f
--- /dev/null
+++ b/offload/src/KernelLanguage/API.cpp
@@ -0,0 +1,74 @@
+//===------ API.cpp - Kernel Language (CUDA/HIP) entry points ----- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+//===----------------------------------------------------------------------===//
+
+#include "Shared/APITypes.h"
+
+#include <cstdio>
+
+struct dim3 {
+  unsigned x = 0, y = 0, z = 0;
+};
+
+struct __omp_kernel_t {
+  dim3 __grid_size;
+  dim3 __block_size;
+  size_t __shared_memory;
+
+  void *__stream;
+};
+
+static __omp_kernel_t __current_kernel = {};
+#pragma omp threadprivate(__current_kernel);
+
+extern "C" {
+
+// TODO: There is little reason we need to keep these names or the way calls are
+// issued. For now we do to avoid modifying Clang's CUDA codegen. Unclear when
+// we actually need to push/pop configurations.
+unsigned __llvmPushCallConfiguration(dim3 __grid_size, dim3 __block_size,
+                                     size_t __shared_memory, void *__stream) {
+  __omp_kernel_t &__kernel = __current_kernel;
+  __kernel.__grid_size = __grid_size;
+  __kernel.__block_size = __block_size;
+  __kernel.__shared_memory = __shared_memory;
+  __kernel.__stream = __stream;
+  return 0;
+}
+
+unsigned __llvmPopCallConfiguration(dim3 *__grid_size, dim3 *__block_size,
+                                    size_t *__shared_memory, void *__stream) {
+  __omp_kernel_t &__kernel = __current_kernel;
+  *__grid_size = __kernel.__grid_size;
+  *__block_size = __kernel.__block_size;
+  *__shared_memory = __kernel.__shared_memory;
+  *((void **)__stream) = __kernel.__stream;
+  return 0;
+}
+
+int __tgt_target_kernel(void *Loc, int64_t DeviceId, int32_t NumTeams,
+                        int32_t ThreadLimit, const void *HostPtr,
+                        KernelArgsTy *Args);
+
+unsigned llvmLaunchKernel(const void *func, dim3 gridDim, dim3 blockDim,
+                          void *args, size_t sharedMem, void *stream) {
+  KernelArgsTy Args = {};
+  Args.DynCGroupMem = sharedMem;
+  Args.NumTeams[0] = gridDim.x;
+  Args.NumTeams[1] = gridDim.y;
+  Args.NumTeams[2] = gridDim.z;
+  Args.ThreadLimit[0] = blockDim.x;
+  Args.ThreadLimit[1] = blockDim.y;
+  Args.ThreadLimit[2] = blockDim.z;
+  Args.ArgPtrs = reinterpret_cast<void **>(args);
+  Args.Flags.IsCUDA = true;
+  int rv = __tgt_target_kernel(nullptr, 0, gridDim.x, blockDim.x, func, &Args);
+  return rv;
+}
+}
diff --git a/offload/src/exports b/offload/src/exports
index f95544ec8329c8..7bdc7d2a531bb3 100644
--- a/offload/src/exports
+++ b/offload/src/exports
@@ -71,6 +71,9 @@ VERS1.0 {
     __tgt_interop_use;
     __tgt_interop_destroy;
     ompt_libomptarget_connect;
+    __llvmPushCallConfiguration;
+    __llvmPopCallConfiguration;
+    llvmLaunchKernel;
   local:
     *;
 };
diff --git a/offload/test/lit.cfg b/offload/test/lit.cfg
index 6d6c77db80e16c..b4fc7d3b333b35 100644
--- a/offload/test/lit.cfg
+++ b/offload/test/lit.cfg
@@ -66,7 +66,7 @@ def evaluate_bool_env(env):
 config.name = 'libomptarget :: ' + config.libomptarget_current_target
 
 # suffixes: A list of file extensions to treat as test files.
-config.suffixes = ['.c', '.cpp', '.cc', '.f90']
+config.suffixes = ['.c', '.cpp', '.cc', '.f90', '.cu']
 
 # excludes: A list of directories to exclude from the testuites.
 config.excludes = ['Inputs']
diff --git a/offload/test/offloading/CUDA/basic_launch.cu b/offload/test/offloading/CUDA/basic_launch.cu
new file mode 100644
index 00000000000000..dd14a0b56b5b68
--- /dev/null
+++ b/offload/test/offloading/CUDA/basic_launch.cu
@@ -0,0 +1,30 @@
+// RUN: %clang++ -foffload-via-llvm --offload-arch=native %s -o %t
+// RUN: %t | %fcheck-generic
+// RUN: %clang++ -foffload-via-llvm --offload-arch=native %s -o %t -fopenmp
+// RUN: %t | %fcheck-generic
+
+// UNSUPPORTED: aarch64-unknown-linux-gnu
+// UNSUPPORTED: aarch64-unknown-linux-gnu-LTO
+// UNSUPPORTED: x86_64-pc-linux-gnu
+// UNSUPPORTED: x86_64-pc-linux-gnu-LTO
+
+#include <stdio.h>
+
+extern "C" {
+void *llvm_omp_target_alloc_shared(size_t Size, int DeviceNum);
+void llvm_omp_target_free_shared(void *DevicePtr, int DeviceNum);
+}
+
+__global__ void square(int *A) { *A = 42; }
+
+int main(int argc, char **argv) {
+  int DevNo = 0;
+  int *Ptr = reinterpret_cast<int *>(llvm_omp_target_alloc_shared(4, DevNo));
+  *Ptr = 7;
+  printf("Ptr %p, *Ptr: %i\n", Ptr, *Ptr);
+  // CHECK: Ptr [[Ptr:0x.*]], *Ptr: 7
+  square<<<1, 1>>>(Ptr);
+  printf("Ptr %p, *Ptr: %i\n", Ptr, *Ptr);
+  // CHECK: Ptr [[Ptr]], *Ptr: 42
+  llvm_omp_target_free_shared(Ptr, DevNo);
+}
diff --git a/offload/test/offloading/CUDA/basic_launch_blocks_and_threads.cu b/offload/test/offloading/CUDA/basic_launch_blocks_and_threads.cu
new file mode 100644
index 00000000000000..615cae6f7b2339
--- /dev/null
+++ b/offload/test/offloading/CUDA/basic_launch_blocks_and_threads.cu
@@ -0,0 +1,32 @@
+// RUN: %clang++ -foffload-via-llvm --offload-arch=native %s -o %t
+// RUN: %t | %fcheck-generic
+// RUN: %clang++ -foffload-via-llvm --offload-arch=native %s -o %t -fopenmp
+// RUN: %t | %fcheck-generic
+
+// UNSUPPORTED: aarch64-unknown-linux-gnu
+// UNSUPPORTED: aarch64-unknown-linux-gnu-LTO
+// UNSUPPORTED: x86_64-pc-linux-gnu
+// UNSUPPORTED: x86_64-pc-linux-gnu-LTO
+
+#include <stdio.h>
+
+extern "C" {
+void *llvm_omp_target_alloc_shared(size_t Size, int DeviceNum);
+void llvm_omp_target_free_shared(void *DevicePtr, int DeviceNum);
+}
+
+__global__ void square(int *A) {
+  __scoped_atomic_fetch_add(A, 1, __ATOMIC_SEQ_CST, __MEMORY_SCOPE_DEVICE);
+}
+
+int main(int argc, char **argv) {
+  int DevNo = 0;
+  int *Ptr = reinterpret_cast<int *>(llvm_omp_target_alloc_shared(4, DevNo));
+  *Ptr = 0;
+  printf("Ptr %p, *Ptr: %i\n", Ptr, *Ptr);
+  // CHECK: Ptr [[Ptr:0x.*]], *Ptr: 0
+  square<<<7, 6>>>(Ptr);
+  printf("Ptr %p, *Ptr: %i\n", Ptr, *Ptr);
+  // CHECK: Ptr [[Ptr]], *Ptr: 42
+  llvm_omp_target_free_shared(Ptr, DevNo);
+}
diff --git a/offload/test/offloading/CUDA/basic_launch_multi_arg.cu b/offload/test/offloading/CUDA/basic_launch_multi_arg.cu
new file mode 100644
index 00000000000000..f95f1dbacc79c4
--- /dev/null
+++ b/offload/test/offloading/CUDA/basic_launch_multi_arg.cu
@@ -0,0 +1,41 @@
+// RUN: %clang++ -foffload-via-llvm --offload-arch=native %s -o %t
+// RUN: %t | %fcheck-generic
+// RUN: %clang++ -foffload-via-llvm --offload-arch=native %s -o %t -fopenmp
+// RUN: %t | %fcheck-generic
+
+// UNSUPPORTED: aarch64-unknown-linux-gnu
+// UNSUPPORTED: aarch64-unknown-linux-gnu-LTO
+// UNSUPPORTED: x86_64-pc-linux-gnu
+// UNSUPPORTED: x86_64-pc-linux-gnu-LTO
+
+#include <stdio.h>
+
+extern "C" {
+void *llvm_omp_target_alloc_shared(size_t Size, int DeviceNum);
+void llvm_omp_target_free_shared(void *DevicePtr, int DeviceNum);
+}
+
+__global__ void square(int *Dst, short Q, int *Src, short P) {
+  *Dst = (Src[0] + Src[1]) * (Q + P);
+  Src[0] = Q;
+  Src[1] = P;
+}
+
+int main(int argc, char **argv) {
+  int DevNo = 0;
+  int *Ptr = reinterpret_cast<int *>(llvm_omp_target_alloc_shared(4, DevNo));
+  int *Src = reinterpret_cast<int *>(llvm_omp_target_alloc_shared(8, DevNo));
+  *Ptr = 7;
+  Src[0] = -2;
+  Src[1] = 8;
+  printf("Ptr %p, *Ptr: %i\n", Ptr, *Ptr);
+  // CHECK: Ptr [[Ptr:0x.*]], *Ptr: 7
+  printf("Src: %i : %i\n", Src[0], Src[1]);
+  // CHECK: Src: -2 : 8
+  square<<<1, 1>>>(Ptr, 3, Src, 4);
+  printf("Ptr %p, *Ptr: %i\n", Ptr, *Ptr);
+  // CHECK: Ptr [[Ptr]], *Ptr: 42
+  printf("Src: %i : %i\n", Src[0], Src[1]);
+  // CHECK: Src: 3 : 4
+  llvm_omp_target_free_shared(Ptr, DevNo);
+}
diff --git a/offload/test/offloading/CUDA/kernel_tu.cu.inc b/offload/test/offloading/CUDA/kernel_tu.cu.inc
new file mode 100644
index 00000000000000..d7d28a109dfc5a
--- /dev/null
+++ b/offload/test/offloading/CUDA/kernel_tu.cu.inc
@@ -0,0 +1 @@
+__global__ void square(int *A) { *A = 42; }
diff --git a/offload/test/offloading/CUDA/launch_tu.cu b/offload/test/offloading/CUDA/launch_tu.cu
new file mode 100644
index 00000000000000..c45f40b115b56b
--- /dev/null
+++ b/offload/test/offloading/CUDA/launch_tu.cu
@@ -0,0 +1,32 @@
+// clang-format off
+// RUN: %clang++ -foffload-via-llvm --offload-arch=native %s -o %t.launch_tu.o -c
+// RUN: %clang++ -foffload-via-llvm --offload-arch=native -x cuda %S/kernel_tu.cu.inc -o %t.kernel_tu.o -c
+// RUN: %clang++ -foffload-via-llvm --offload-arch=native %t.launch_tu.o %t.kernel_tu.o -o %t
+// RUN: %t | %fcheck-generic
+// clang-format on
+
+// UNSUPPORTED: aarch64-unknown-linux-gnu
+// UNSUPPORTED: aarch64-unknown-linux-gnu-LTO
+// UNSUPPORTED: x86_64-pc-linux-gnu
+// UNSUPPORTED: x86_64-pc-linux-gnu-LTO
+
+#include <stdio.h>
+
+extern "C" {
+void *llvm_omp_target_alloc_shared(size_t Size, int DeviceNum);
+void llvm_omp_target_free_shared(void *DevicePtr, int DeviceNum);
+}
+
+extern __global__ void square(int *A);
+
+int main(int argc, char **argv) {
+  int DevNo = 0;
+  int *Ptr = reinterpret_cast<int *>(llvm_omp_target_alloc_shared(4, DevNo));
+  *Ptr = 7;
+  printf("Ptr %p, *Ptr: %i\n", Ptr, *Ptr);
+  // CHECK: Ptr [[Ptr:0x.*]], *Ptr: 7
+  square<<<1, 1>>>(Ptr);
+  printf("Ptr %p, *Ptr: %i\n", Ptr, *Ptr);
+  // CHECK: Ptr [[Ptr]], *Ptr: 42
+  llvm_omp_target_free_shared(Ptr, DevNo);
+}



More information about the cfe-commits mailing list