diff options
author | Raul Tambre <raul@tambre.ee> | 2020-09-05 16:40:02 (GMT) |
---|---|---|
committer | Brad King <brad.king@kitware.com> | 2020-09-24 19:19:54 (GMT) |
commit | c63fe018353cf6afb30980c4cac7493be7cd0a82 (patch) | |
tree | 68d2daf0cd8ab91a9feaa49392607c6cfecd2ac4 /Source/cmMakefileLibraryTargetGenerator.h | |
parent | c98ec731f90eb0180c89108b7d2e42263b66d1ed (diff) | |
download | CMake-c63fe018353cf6afb30980c4cac7493be7cd0a82.zip CMake-c63fe018353cf6afb30980c4cac7493be7cd0a82.tar.gz CMake-c63fe018353cf6afb30980c4cac7493be7cd0a82.tar.bz2 |
CUDA: Clang separable compilation
For NVCC the compiler takes care of device linking when passed the "-dlink"
flag.
Clang doesn't support such magic and requires the buildsystem to do the work
that NVCC does behind the scenes.
The implementation is based on Bazel's device linking documentation:
https://github.com/tensorflow/tensorflow/blob/7cabcdf073abad8c46e9dda62bb8fa4682d2061e/third_party/nccl/build_defs.bzl.tpl#L259
Closes: #20726
Diffstat (limited to 'Source/cmMakefileLibraryTargetGenerator.h')
-rw-r--r-- | Source/cmMakefileLibraryTargetGenerator.h | 5 |
1 files changed, 5 insertions, 0 deletions
diff --git a/Source/cmMakefileLibraryTargetGenerator.h b/Source/cmMakefileLibraryTargetGenerator.h index 6a38e18..cc989e7 100644 --- a/Source/cmMakefileLibraryTargetGenerator.h +++ b/Source/cmMakefileLibraryTargetGenerator.h @@ -5,6 +5,7 @@ #include "cmConfigure.h" // IWYU pragma: keep #include <string> +#include <vector> #include "cmMakefileTargetGenerator.h" @@ -27,6 +28,10 @@ protected: void WriteModuleLibraryRules(bool relink); void WriteDeviceLibraryRules(const std::string& linkRule, bool relink); + void WriteNvidiaDeviceLibraryRules(const std::string& linkRuleVar, + bool relink, + std::vector<std::string>& commands, + const std::string& targetOutput); void WriteLibraryRules(const std::string& linkRule, const std::string& extraFlags, bool relink); // MacOSX Framework support methods |