Clang

clang and LLVM are defacto standard tool-chains of the whole modern world. Apple uses clang and it is developing its Swift with it, Google uses clang in NDK (and deprecated gcc), FreeBSD uses clang as a default compiler, and the standard builds of the Chrome browser are done with clang++. LLVM's odegeration back-end is used in GHC Haskell compiler, Rust compiler, Ocaml, Julia and so on.

clang++ comes with libc++ - a clean, rewritten from scratch C++11 standard library and STL implementation. It also supports native atomics, OpenMP 4.x via libomp.so and all the nice things, like native atomics, etc. with its libcompiler_rt.

Here is how to recompile a project with clang++ and libc++ without using or linking with libstdc++. Notice the -stdlib=libc++ flags. Usually the resulting binaries are ~20% faster (when you have re-built libc++ itself with the same optimizations -mtune=native).

This configuration has been tested with whole LLVM toolchain itself and with shit like NodeJS.

export CC=clang
export CXX=clang++
export LD=ld.lld

export FC=gfortran
export FFLAGS="-m64 -march=native -mtune=native -O3 -fPIC -fopenmp"

export CFLAGS="-Wformat -Wformat-security -Werror=format-security \
    -U_FORTIFY_SOURCE -D_FORTIFY_SOURCE=2 \
    -fstack-protector-strong --param ssp-buffer-size=4 \
    -Wstrict-aliasing=2 -fno-strict-aliasing \
    -D_FILE_OFFSET_BITS=64 -D_REENTRANT \
    -ffunction-sections -fdata-sections \
    -m64 -march=native -mtune=native -O3 \
    -fPIC \
    -pthread \
    -fopenmp=libomp \
    -I/usr/local/include"
export LDFLAGS="-L/usr/local/lib \
    -rtlib=compiler-rt \
    -fPIC \
    -pthread \
    -lm \
    -fopenmp=libomp \
    -Wl,--gc-sections -Wl,--as-needed \
    -Wl,-z,relro -Wl,-z,now -Wl,-z,noexecstack"
export CXXFLAGS="-Wformat -Wformat-security -Werror=format-security -Wstrict-aliasing=2 \
    -fno-strict-aliasing \
    -fstack-protector-strong --param ssp-buffer-size=4 \
    -U_FORTIFY_SOURCE -D_FORTIFY_SOURCE=2 \
    -D_FILE_OFFSET_BITS=64 -D_REENTRANT \
    -ffunction-sections -fdata-sections \
    -m64 -march=native -mtune=native -O3 \
    -fPIC \
    -pthread \
    -lm \
    -fopenmp=libomp \
    -std=c++1z -stdlib=libc++ \
    -fno-rtti -fno-exceptions \
    -I/usr/local/include"
export CXXLDFLAGS="-L/usr/local/lib \
    -rtlib=compiler-rt -stdlib=libc++ \
    -fPIC \
    -pthread \
    -fopenmp=libomp \
    -Wl,--gc-sections -Wl,--as-needed \
    -Wl,-z,relro -Wl,-z,now -Wl,-z,noexecstack"

clang supports LTO. Notice that LTO has problems with object files produced outside a compiler, like assembly objects. You cannot build ffmpeg with LTO.

clang and GNU binutils supports -fPIE but this is incompatible with static libraries and you cannot compile Emacs with it.