experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Windows CUDA build: explicitly specify system include paths and library search paths.

Browse Source 
NVCC doesn't respect the INCLUDE and LIB env vars from MSVC (which are set by Bazel). This change ensures NVCC uses the same include and library paths chosen by Bazel.

PiperOrigin-RevId: 357959501
Change-Id: I5a7461d75b4255e59c7c7f5750da55e04d620c83
This commit is contained in:
A. Unique TensorFlower 2021-02-17 09:00:15 -08:00 committed by TensorFlower Gardener
parent 15b8b79364
commit 4bbf4dd025
1 changed files with 7 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
third_party/gpus/crosstool/windows/msvc_wrapper_for_nvcc.py.tpl vendored
View File

@ -153,6 +153,13 @@ def InvokeNvcc(argv, log=False):
]
_, argv = GetOptionValue(argv, '--no-cuda-include-ptx')
# nvcc doesn't respect the INCLUDE and LIB env vars from MSVC,
# so we explicity specify the system include paths and library search paths.
if 'INCLUDE' in os.environ:
nvccopts += [('--system-include="%s"' % p) for p in os.environ['INCLUDE'].split(";")]
if 'LIB' in os.environ:
nvccopts += [('--library-path="%s"' % p) for p in os.environ['LIB'].split(";")]
nvccopts += nvcc_compiler_options
nvccopts += undefines
nvccopts += defines