OSSForks/pytorch
1
0
Fork 0
mirror of https://github.com/zebrajr/pytorch.git synced 2025-12-06 12:20:52 +01:00
Code Issues Actions 38 Packages Projects Releases Wiki Activity

Don't build CUDA with debug info by default. (#102617)

Browse Source 
Fixes https://github.com/pytorch/pytorch/issues/102594

Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/102617
Approved by: https://github.com/malfet
This commit is contained in:
Edward Z. Yang 2023-07-05 17:08:16 +00:00 committed by PyTorch MergeBot
parent 0cee4e3c32
commit 3dc4adc7a6
2 changed files with 7 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
CMakeLists.txt
View File

@ -340,6 +340,7 @@ option(HAVE_SOVERSION "Whether to add SOVERSION to the shared objects" OFF)
option(BUILD_LIBTORCH_CPU_WITH_DEBUG "Enable RelWithDebInfo for libtorch_cpu target only" OFF)
cmake_dependent_option(USE_CCACHE "Attempt using CCache to wrap the compilation" ON "UNIX" OFF)
option(WERROR "Build with -Werror supported by the compiler" OFF)
option(DEBUG_CUDA "When compiling DEBUG, also attempt to compile CUDA with debug flags (may cause nvcc to OOM)" OFF)
option(USE_COREML_DELEGATE "Use the CoreML backend through delegate APIs" OFF)
option(USE_PER_OPERATOR_HEADERS "Whether ATen should generate separate headers for each operator" ON)
cmake_dependent_option(
@ -722,7 +723,7 @@ cmake_dependent_option(
"Whether to build the flash_attention kernel for scaled dot product attention" ON
"USE_CUDA AND NOT ROCM AND NOT CMAKE_CUDA_COMPILER_VERSION VERSION_LESS 11.6" OFF)
if(NOT MSVC)
if(DEBUG_CUDA)
string(APPEND CMAKE_CUDA_FLAGS_DEBUG " -lineinfo")
string(APPEND CMAKE_CUDA_FLAGS_RELWITHDEBINFO " -lineinfo")
# CUDA-12.1 crashes when trying to compile with --source-in-ptx

5
setup.py
View File

@ -25,6 +25,11 @@
#
# Environment variables for feature toggles:
#
# DEBUG_CUDA=1
# if used in conjunction with DEBUG or REL_WITH_DEB_INFO, will also
# build CUDA kernels with -lineinfo --source-in-ptx. Note that
# on CUDA 12 this may cause nvcc to OOM, so this is disabled by default.
#
# USE_CUDNN=0
# disables the cuDNN build
#