OSSForks/tensorflow
1
0
Fork 0
mirror of https://github.com/zebrajr/tensorflow.git synced 2025-12-06 12:20:11 +01:00
Code Issues Actions 5 Packages Projects Releases Wiki Activity

Adding "-fcuda-flush-denormals-to-zero" as a default hipcc option

Browse Source 
Prior to ROCm 3.8, hipcc (hipclang) flushed denormal values to zero by default. Starting with ROCm 3.8 that is no longer true, denormal values are kept as is.

TF expects denormals to be flushed to zero. This is enforced on the CUDA side by explicitly passing the "-fcuda-flush-denormals-to-zero" (see tensorflow.bzl). This commit does the same for the ROCm side.

Also removing the no_rocm tag from the corresponding unit test - //tensorflow/python/kernel_tests:denormal_test_gpu
This commit is contained in:
Deven Desai 2020-09-26 00:21:19 +00:00
parent df9ab6f480
commit 5dbd0a2046
1 changed files with 1 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
third_party/gpus/crosstool/clang/bin/crosstool_wrapper_driver_rocm.tpl vendored
View File

@ -186,6 +186,7 @@ def InvokeHipcc(argv, log=False):
# of link time. This allows the default host compiler (gcc) be used as the # of link time. This allows the default host compiler (gcc) be used as the
# linker for TensorFlow on ROCm platform. # linker for TensorFlow on ROCm platform.
hipccopts += ' -fno-gpu-rdc ' hipccopts += ' -fno-gpu-rdc '
hipccopts += ' -fcuda-flush-denormals-to-zero '
hipccopts += undefines hipccopts += undefines
hipccopts += defines hipccopts += defines
hipccopts += std_options hipccopts += std_options