Use appropriate instruction sets Take advantage of the conversion instructions provided by AVX, such as the "vcvtps2pd" and other related instructions. They can efficiently convert packed single-precision floating-point numbers to double-precision floating-point numbers. If the initial data is 32-bit integers, it may be necessary to perform the conversion step by step using appropriate instructions (for example, first convert integers to single-precision and then to double-precision). Choose the most suitable ones according to the instruction set support of the specific processor.