[CUDA] Disable LTO for device-side compilations.
authorArtem Belevich <tra@google.com>
Wed, 21 Mar 2018 22:22:59 +0000 (22:22 +0000)
committerArtem Belevich <tra@google.com>
Wed, 21 Mar 2018 22:22:59 +0000 (22:22 +0000)
commitecb178bb356fc058a51bbda0d083068b5d83ba2b
tree8f1e67971ee6bb3575f461000f388531c7a92a8e
parent1dce44e8e82c09822e843ccd1df3c6c5de7839f4
[CUDA] Disable LTO for device-side compilations.

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.

Differential Revision: https://reviews.llvm.org/D44691

llvm-svn: 328161
clang/include/clang/Driver/Driver.h
clang/lib/Driver/Driver.cpp
clang/lib/Driver/ToolChains/Clang.cpp
clang/test/Driver/lto.cu [new file with mode: 0644]
clang/test/Driver/thinlto.cu [new file with mode: 0644]