Improve math performance

-ffast-math not only make the math fast, it is unsafe
-fno-math-errno makes single-instruction math operations not set ERRNO
-fsingle-precision-constant causes floating-point constants to be loaded in single precision even when this is not exact.
But it break strict IEEE compliance

cmake/compile-flags.cmake

@@ -20,6 +20,9 @@ add_compile_flags(BOTH
         -ffunction-sections
         -fdata-sections
         -fstrict-volatile-bitfields
+        -ffast-math
+        -fno-math-errno
+        -fsingle-precision-constant
         -O2
         -ggdb
         )