clang -O0 is significantly slower than tcc. So the problem is clearly not (just) the optimisation passes.