Add autotools config files for NVIDIA compilers, always add -fPIC flag for nvfortran. #3642
Conversation
lrknox
requested review from
derobins,
byrnHDF,
fortnern,
jhendersonHDF,
qkoziol,
vchoi-hdfgroup,
bmribler,
glennsong09,
mattjala and
brtnfld
as code owners
lrknox
added
Merge - To 1.12
Priority - 1. High 🔼
lrknox
changed the title
…a_compiler_config Adjust nvhpc compiler flags.
| @@ -57,6 +57,15 @@ if (CMAKE_Fortran_COMPILER_ID STREQUAL "NAG") | |||
| list (APPEND HDF5_CMAKE_Fortran_FLAGS "-ieee=full") | |||
| endif () | |||
|
|
|||
| if (CMAKE_Fortran_COMPILER_ID STREQUAL "NVHPC") | |||
| set (CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} -fPIC") | |||
There was a problem hiding this comment.
This might not be needed as CMake should handle this flag.
| @@ -49,6 +49,15 @@ if (CMAKE_CXX_COMPILER_ID STREQUAL SunPro AND CMAKE_CXX_COMPILER_LOADED) | |||
| endif () | |||
| endif () | |||
|
|
|||
| if (CMAKE_CXX_COMPILER_ID STREQUAL "NVHPC") | |||
| set (CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -Minform=warn") | |||
There was a problem hiding this comment.
c++11 flag should have a CMake option available.
There was a problem hiding this comment.
if (CMAKE_CXX_COMPILER_ID STREQUAL "NVHPC" AND CMAKE_CXX_COMPILER_LOADED)
if (NOT DEFINED CMAKE_CXX${CMAKE_CXX_STANDARD}_STANDARD_COMPILE_OPTION)
if (NOT CMAKE_CXX_STANDARD OR CMAKE_CXX_STANDARD EQUAL 11)
set (CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${CMAKE_C11_STANDARD_COMPILE_OPTION}")
endif ()
endif ()
endif ()
|
linux autotools nvhpc test fails when disk space is filled. I tried it twice with the same results: "H5Tdbg.c", line 365: warning: statement is unreachable [code_is_unreachable] I don't know what we can do about this. Otherwise I think the PR is ready for final review. |
|
Replaced with #3686 |
Note that nvcc compiler doesn't match compiler and version processing in nvidia-flags because the output of "nvcc -V" doesn't provide version information on the line beginning with "nvcc". The only version information in the output is the Cuda version. As it is now, nvcc will not get the -c99 flag from nvida-flags. We could add different sed lines to get the compiler_vendor and compiler_version information with the Cuda version, but I'm told we should use nvc and not nvcc, so nvcc is not supported.