When trying to use libc++ with cpu2017 xalan benchmark, seeing following performance issue
machine: amd znver5
compiler used: llvm 21
options used: -O3 -march=znver5 -flto
single copy runtime with libc++ : 143
single copy runtime with libstdc++ : 131
There seem to be 9% difference
perf report shows that more time is spent in std::find function. when libstdc++ is used, the function is vectorized and a scalar variant is used with libc++.