summaryrefslogtreecommitdiffstats
path: root/lib/Driver/ToolChains/Cuda.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'lib/Driver/ToolChains/Cuda.cpp')
-rw-r--r--lib/Driver/ToolChains/Cuda.cpp53
1 files changed, 34 insertions, 19 deletions
diff --git a/lib/Driver/ToolChains/Cuda.cpp b/lib/Driver/ToolChains/Cuda.cpp
index 57b8d4340e..96f8c513bb 100644
--- a/lib/Driver/ToolChains/Cuda.cpp
+++ b/lib/Driver/ToolChains/Cuda.cpp
@@ -1,9 +1,8 @@
//===--- Cuda.cpp - Cuda Tool and ToolChain Implementations -----*- C++ -*-===//
//
-// The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
+// 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
//
//===----------------------------------------------------------------------===//
@@ -61,6 +60,8 @@ static CudaVersion ParseCudaVersionFile(llvm::StringRef V) {
return CudaVersion::CUDA_92;
if (Major == 10 && Minor == 0)
return CudaVersion::CUDA_100;
+ if (Major == 10 && Minor == 1)
+ return CudaVersion::CUDA_101;
return CudaVersion::UNKNOWN;
}
@@ -453,7 +454,8 @@ void NVPTX::Linker::ConstructJob(Compilation &C, const JobAction &JA,
assert(TC.getTriple().isNVPTX() && "Wrong platform");
ArgStringList CmdArgs;
- CmdArgs.push_back("--cuda");
+ if (TC.CudaInstallation.version() <= CudaVersion::CUDA_100)
+ CmdArgs.push_back("--cuda");
CmdArgs.push_back(TC.getTriple().isArch64Bit() ? "-64" : "-32");
CmdArgs.push_back(Args.MakeArgString("--create"));
CmdArgs.push_back(Args.MakeArgString(Output.getFilename()));
@@ -643,28 +645,41 @@ void CudaToolChain::addClangTargetOptions(
CC1Args.push_back("-mlink-builtin-bitcode");
CC1Args.push_back(DriverArgs.MakeArgString(LibDeviceFile));
- // Libdevice in CUDA-7.0 requires PTX version that's more recent than LLVM
- // defaults to. Use PTX4.2 by default, which is the PTX version that came with
- // CUDA-7.0.
- const char *PtxFeature = "+ptx42";
- // TODO(tra): CUDA-10+ needs PTX 6.3 to support new features. However that
- // requires fair amount of work on LLVM side. We'll keep using PTX 6.1 until
- // all prerequisites are in place.
- if (CudaInstallation.version() >= CudaVersion::CUDA_91) {
- // CUDA-9.1 uses new instructions that are only available in PTX6.1+
- PtxFeature = "+ptx61";
- } else if (CudaInstallation.version() >= CudaVersion::CUDA_90) {
- // CUDA-9.0 uses new instructions that are only available in PTX6.0+
- PtxFeature = "+ptx60";
+ // 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()) {
+ case CudaVersion::CUDA_101:
+ PtxFeature = "+ptx64";
+ break;
+ case CudaVersion::CUDA_100:
+ PtxFeature = "+ptx63";
+ break;
+ case CudaVersion::CUDA_92:
+ PtxFeature = "+ptx61";
+ break;
+ case CudaVersion::CUDA_91:
+ PtxFeature = "+ptx61";
+ break;
+ case CudaVersion::CUDA_90:
+ PtxFeature = "+ptx60";
+ break;
+ default:
+ PtxFeature = "+ptx42";
}
CC1Args.append({"-target-feature", PtxFeature});
if (DriverArgs.hasFlag(options::OPT_fcuda_short_ptr,
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 (DeviceOffloadingKind == Action::OFK_OpenMP) {
SmallVector<StringRef, 8> LibraryPaths;
-
if (const Arg *A = DriverArgs.getLastArg(options::OPT_libomptarget_nvptx_path_EQ))
LibraryPaths.push_back(A->getValue());