[Clang][OpenMP] Require CUDA 9.2+ for OpenMP offloading on NVPTX target

In current implementation of `deviceRTLs`, we're using some functions
that are CUDA version dependent (if CUDA_VERSION < 9, it is one; otheriwse, it
is another one). As a result, we have to compile one bitcode library for each
CUDA version supported. A worse problem is forward compatibility. If a new CUDA
version is released, we have to update CMake file as well.

CUDA 9.2 has been released for three years. Instead of using various weird tricks
to make `deviceRTLs` work with different CUDA versions and still have forward
compatibility, we can simply drop support for CUDA 9.1 or lower version. It has at
least two benifits:
- We don't need to generate bitcode libraries for each CUDA version;
- Clang driver doesn't need to search for the bitcode lib based on CUDA version.

We can claim that starting from LLVM 12, OpenMP offloading on NVPTX target requires
CUDA 9.2+.

Reviewed By: jdoerfert, JonChesterfield

Differential Revision: https://reviews.llvm.org/D97003

clang/include/clang/Basic/DiagnosticDriverKinds.td

@@ -263,6 +263,7 @@ def err_drv_expecting_fopenmp_with_fopenmp_targets : Error<
 def err_drv_omp_offload_target_missingbcruntime : Error<
   "No library '%0' found in the default clang lib directory or in LIBRARY_PATH. Please use --libomptarget-%1-bc-path to specify %1 bitcode library.">;
 def err_drv_omp_offload_target_bcruntime_not_found : Error<"Bitcode library '%0' does not exist.">;
+def err_drv_omp_offload_target_cuda_version_not_support : Error<"NVPTX target requires CUDA 9.2 or above. CUDA %0 is detected.">;
 def warn_drv_omp_offload_target_duplicate : Warning<
   "The OpenMP offloading target '%0' is similar to target '%1' already specified - will be ignored.">,
   InGroup<OpenMPTarget>;

clang/lib/Driver/ToolChains/Cuda.cpp

@@ -710,13 +710,14 @@ void CudaToolChain::addClangTargetOptions(
   CC1Args.push_back("-mlink-builtin-bitcode");
   CC1Args.push_back(DriverArgs.MakeArgString(LibDeviceFile));
 
+  clang::CudaVersion CudaInstallationVersion = CudaInstallation.version();
   std::string CudaVersionStr;
 
   // New CUDA versions often introduce new instructions that are only supported
   // by new PTX version, so we need to raise PTX level to enable them in NVPTX
   // back-end.
   const char *PtxFeature = nullptr;
-  switch (CudaInstallation.version()) {
+  switch (CudaInstallationVersion) {
 #define CASE_CUDA_VERSION(CUDA_VER, PTX_VER)                                   \
   case CudaVersion::CUDA_##CUDA_VER:                                           \
     CudaVersionStr = #CUDA_VER;                                                \
@@ -743,12 +744,19 @@ void CudaToolChain::addClangTargetOptions(
                          options::OPT_fno_cuda_short_ptr, false))
     CC1Args.append({"-mllvm", "--nvptx-short-ptr"});
 
-  if (CudaInstallation.version() >= CudaVersion::UNKNOWN)
-    CC1Args.push_back(DriverArgs.MakeArgString(
-        Twine("-target-sdk-version=") +
-        CudaVersionToString(CudaInstallation.version())));
+  if (CudaInstallationVersion >= CudaVersion::UNKNOWN)
+    CC1Args.push_back(
+        DriverArgs.MakeArgString(Twine("-target-sdk-version=") +
+                                 CudaVersionToString(CudaInstallationVersion)));
 
   if (DeviceOffloadingKind == Action::OFK_OpenMP) {
+    if (CudaInstallationVersion < CudaVersion::CUDA_92) {
+      getDriver().Diag(
+          diag::err_drv_omp_offload_target_cuda_version_not_support)
+          << CudaVersionToString(CudaInstallationVersion);
+      return;
+    }
+
     std::string BitcodeSuffix =
         "nvptx-cuda_" + CudaVersionStr + "-" + GpuArch.str();
     addOpenMPDeviceRTL(getDriver(), DriverArgs, CC1Args, BitcodeSuffix,

clang/test/Driver/openmp-offload-gpu.c

@@ -154,17 +154,17 @@
 /// Check that the runtime bitcode library is part of the compile line. Create a bogus
 /// bitcode library and add it to the LIBRARY_PATH.
 // RUN:   env LIBRARY_PATH=%S/Inputs/libomptarget %clang -### -fopenmp=libomp -fopenmp-targets=nvptx64-nvidia-cuda \
-// RUN:   -Xopenmp-target -march=sm_35 --cuda-path=%S/Inputs/CUDA_80/usr/local/cuda \
+// RUN:   -Xopenmp-target -march=sm_35 --cuda-path=%S/Inputs/CUDA_102/usr/local/cuda \
 // RUN:   -fopenmp-relocatable-target -save-temps -no-canonical-prefixes %s 2>&1 \
 // RUN:   | FileCheck -check-prefix=CHK-BCLIB %s
 /// The user can override default detection using --libomptarget-nvptx-bc-path=.
 // RUN:   %clang -### -fopenmp=libomp -fopenmp-targets=nvptx64-nvidia-cuda \
 // RUN:   --libomptarget-nvptx-bc-path=%S/Inputs/libomptarget/libomptarget-nvptx-test.bc \
-// RUN:   -Xopenmp-target -march=sm_35 --cuda-path=%S/Inputs/CUDA_80/usr/local/cuda \
+// RUN:   -Xopenmp-target -march=sm_35 --cuda-path=%S/Inputs/CUDA_102/usr/local/cuda \
 // RUN:   -fopenmp-relocatable-target -save-temps -no-canonical-prefixes %s 2>&1 \
 // RUN:   | FileCheck -check-prefix=CHK-BCLIB-USER %s
 
-// CHK-BCLIB: clang{{.*}}-triple{{.*}}nvptx64-nvidia-cuda{{.*}}-mlink-builtin-bitcode{{.*}}libomptarget-nvptx-cuda_80-sm_35.bc
+// CHK-BCLIB: clang{{.*}}-triple{{.*}}nvptx64-nvidia-cuda{{.*}}-mlink-builtin-bitcode{{.*}}libomptarget-nvptx-cuda_102-sm_35.bc
 // CHK-BCLIB-USER: clang{{.*}}-triple{{.*}}nvptx64-nvidia-cuda{{.*}}-mlink-builtin-bitcode{{.*}}libomptarget-nvptx-test.bc
 // CHK-BCLIB-NOT: {{error:|warning:}}
 
@@ -173,23 +173,33 @@
 /// Check that the warning is thrown when the libomptarget bitcode library is not found.
 /// Libomptarget requires sm_35 or newer so an sm_35 bitcode library should never exist.
 // RUN:   %clang -### -fopenmp=libomp -fopenmp-targets=nvptx64-nvidia-cuda \
-// RUN:   -Xopenmp-target -march=sm_35 --cuda-path=%S/Inputs/CUDA_80/usr/local/cuda \
+// RUN:   -Xopenmp-target -march=sm_35 --cuda-path=%S/Inputs/CUDA_102/usr/local/cuda \
 // RUN:   -fopenmp-relocatable-target -save-temps -no-canonical-prefixes %s 2>&1 \
 // RUN:   | FileCheck -check-prefix=CHK-BCLIB-WARN %s
 
-// CHK-BCLIB-WARN: No library 'libomptarget-nvptx-cuda_80-sm_35.bc' found in the default clang lib directory or in LIBRARY_PATH. Please use --libomptarget-nvptx-bc-path to specify nvptx bitcode library.
+// CHK-BCLIB-WARN: No library 'libomptarget-nvptx-cuda_102-sm_35.bc' found in the default clang lib directory or in LIBRARY_PATH. Please use --libomptarget-nvptx-bc-path to specify nvptx bitcode library.
 
 /// ###########################################################################
 
 /// Check that the error is thrown when the libomptarget bitcode library does not exist.
 // RUN:   %clang -### -fopenmp=libomp -fopenmp-targets=nvptx64-nvidia-cuda \
-// RUN:   -Xopenmp-target -march=sm_35 --cuda-path=%S/Inputs/CUDA_80/usr/local/cuda \
+// RUN:   -Xopenmp-target -march=sm_35 --cuda-path=%S/Inputs/CUDA_102/usr/local/cuda \
 // RUN:   --libomptarget-nvptx-bc-path=not-exist.bc \
 // RUN:   -fopenmp-relocatable-target -save-temps -no-canonical-prefixes %s 2>&1 \
 // RUN:   | FileCheck -check-prefix=CHK-BCLIB-ERROR %s
 
 // CHK-BCLIB-ERROR: Bitcode library 'not-exist.bc' does not exist.
 
+/// ###########################################################################
+
+/// Check that the error is thrown when CUDA 9.1 or lower version is used.
+// RUN:   %clang -### -fopenmp=libomp -fopenmp-targets=nvptx64-nvidia-cuda \
+// RUN:   -Xopenmp-target -march=sm_35 --cuda-path=%S/Inputs/CUDA_90/usr/local/cuda \
+// RUN:   -fopenmp-relocatable-target -save-temps -no-canonical-prefixes %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-CUDA-VERSION-ERROR %s
+
+// CHK-CUDA-VERSION-ERROR: NVPTX target requires CUDA 9.2 or above. CUDA 9.0 is detected.
+
 /// Check that debug info is emitted in dwarf-2
 // RUN:   %clang -### -no-canonical-prefixes -fopenmp=libomp -fopenmp-targets=nvptx64-nvidia-cuda -Xopenmp-target -march=sm_60 %s -g -O1 --no-cuda-noopt-device-debug 2>&1 \
 // RUN:   | FileCheck -check-prefix=DEBUG_DIRECTIVES %s