Setting `CLANG_DEFAULT_PIE_ON_LINUX` when building clang

[swahtz]

It seems that the behaviour for whether or not -fPIE is on by default in Linux changed in clang 14 to be controlled by a flag when building clang:

https://github.com/llvm/llvm-project/commit/1042de90589c74c6998c127d1ee1b33ff3efb8cc

We were having a bit of an issue switching to the 2024 OS CI images and building with clang++ on the OpenVDB project (when moving from 2022 clang11 to the 2024 image):

https://github.com/AcademySoftwareFoundation/openvdb/pull/1926

Shall we set CLANG_DEFAULT_PIE_ON_LINUX to match the behaviour of gcc?

[jfpanisset]

As a point of comparison, it seems that gcc 11.2.1 which is installed in the aswf-docker containers from gcc-toolset is not configured with either --enable-host-pie or --enable-default-pie :

Configured with: ../configure --enable-bootstrap --enable-languages=c,c++,fortran,lto --prefix=/opt/rh/gcc-toolset-11/root/usr --mandir=/opt/rh/gcc-toolset-11/root/usr/share/man --infodir=/opt/rh/gcc-toolset-11/root/usr/share/info --with-bugurl=https://bugs.rockylinux.org/ --enable-shared --enable-threads=posix --enable-checking=release --enable-multilib --with-system-zlib --enable-__cxa_atexit --disable-libunwind-exceptions --enable-gnu-unique-object --enable-linker-build-id --with-gcc-major-version-only --with-linker-hash-style=gnu --enable-plugin --enable-initfini-array --with-isl=/builddir/build/BUILD/gcc-11.2.1-20220127/obj-x86_64-redhat-linux/isl-install --enable-offload-targets=nvptx-none --without-cuda-driver --enable-gnu-indirect-function --enable-cet --with-tune=generic --with-arch_32=x86-64 --build=x86_64-redhat-linux

The 2022 images were available with multiple Clang versions, from 10-14. Looking at .github/workflows/nanovdb.yml the image is specified as aswf/ci-openvdb:2022 or aswf/ci-openvdb:2024, which according to:

https://hub.docker.com/r/aswf/ci-openvdb/tags

has the 2022 tag defaulting to clang 10.0.1: this is confusing since clang 10 is well before clang added the ability to build clang with PIE enabled by default. The 2024 tag is defaulting to 2024-clang17, where clang 17.0.6 is not built with that flag.

For 2022/clang 10.0.1:

[root@a1865748f473 aswf]# llvm-config --cflags
-I/usr/local/include  -D_GNU_SOURCE -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS
[root@a1865748f473 aswf]# llvm-config --cxxflags
-I/usr/local/include -std=c++14   -fno-exceptions -fno-rtti -D_GNU_SOURCE -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS
[root@a1865748f473 aswf]# llvm-config --ldflags
-L/usr/local/lib

For 2022/clang 17.0.6:

[root@4f19be8f8c79 aswf]# llvm-config --cflags
-I/usr/local/include  -D_GNU_SOURCE -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS
[root@4f19be8f8c79 aswf]# llvm-config --cxxflags
-I/usr/local/include -std=c++17   -fno-exceptions -funwind-tables -fno-rtti -D_GNU_SOURCE -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS
[root@4f19be8f8c79 aswf]# llvm-config --ldflags
-L/usr/local/lib64

Is there a change between CUDA 11.4 (in the 2022 images) and CUDA 12.6 (in latest 2024 images) which now requires compiling with PIE which did not before?