AVX512VL adds a number of new packed conversion for unsigned 32-bit/64-bit integer types to float/double. These allow to replace less efficient conversions on xarch --- either multiple instructions or dedicated helper methods --- with single accelerated instructions.
AVX512VL adds a number of new packed conversion for unsigned 32-bit/64-bit integer types to float/double. These allow to replace less efficient conversions on xarch --- either multiple instructions or dedicated helper methods --- with single accelerated instructions.