Disabling FTZ and DAZ

According to the preceding analysis, the precision varies when FTZ and DAZ convert denormalized floating-point numbers to 0. To achieve better precision, disable FTZ. For ICC, -ftz is enabled by default at all -O optimization levels except -O0. You can use the compilation option -no-ftz to disable FTZ. For the BiSheng Compiler, FTZ is enabled under -ffast-math or -Ofast. FTZ is also separately enabled under -fflushz or -Mflushz (valid only for Flang). For GCC, FTZ is enabled only under -ffast-math.