Allow optimization and use fesetround(), fegetround() #300

Job	Run time
coding-style	25s
host-win (x86_64)	1m 2s
host-win (armv7)	55s
host-win-msvc	58s
host-win (aarch64)	48s
host-x86 (x86_64, g++-10)	2m 5s
host-x86 (x86_64, clang++-11)	26s
host-arm (armv7, none, none, g++-10)	3m 18s
host-arm (armv7, none, none, clang++-11)	3m 45s
host-arm (aarch64, none, none, g++-10)	2m 44s
host-arm (aarch64, none, none, clang++-11)	3m 32s
host-arm (aarch64, crypto+crc, none, g++-10)	2m 51s
host-arm (aarch64, crypto+crc, none, clang++-11)	3m 35s
host-arm (armv7, none, -mcpu=cortex-a32 -mfpu=neon-fp-armv8, g++-10)	3m 20s
host-arm (armv7, none, -mcpu=cortex-a32 -mfpu=neon-fp-armv8, clang++-11)	3m 30s
test-win-msvc	52s
	34m 6s

Provide feedback