5.Intel 64 Base Architecture:3

Intel AVX extension introduced 256-bit vector processing capability,AVX employs VEX prefix to extend access to 256-bit registers . the new registers are YMM0 through YMM16 in 64-bit mode,the lower 128-bit are alias to XMM registers .the legacy SSE instructions (without VEX prefix) can not access upper 128-bit registers.

AVX provides 256-bit floating-point arithmetic processing ,including add/substract/multiply/divide... .

AVX also provides 256-bit non-arithmetic instructions,including load/store/shuffle/blend/insert/extract... .

AVX provides a full complement if 128-bit numeric processing instructions which employ VEX prefix encoding ,these vex-encoded instruction generally provide the same functionality over instructions operating on XMM registers which are encoded using SIMD prefixes.

as for non-arithmetic processing ,AVX does the same thing,it provides complementary instructions for SIMD prefixed 128-bit non-arithmetic instructions.

AVX2 follows the same programming model as AVX instruction,but enhancement can be gained by encoding with VEX and extending to 256-bit, there are lots of VEX-only instructions whose mnemonics usually start with a 'v'.

AVX also introduced general purpose instruction set. lzcnt/tzcnt are bit-manipulation instructions without VEX prefix,other instructions are vex-prefixed.

as the summary ,we have the following points:

• legacy SIMD can not access upper 128-bit of YMM registers,and VEX.128 prefixed instructions will zero these bit, and these bits are usually accessed by VEX.256 prefixed instructions.

• memory alignment requirement,explicitly-aligned/explicitly-unaligned/most of the others must be aligned,but for VEX-prefixed ,we have more relaxed alignment requirement.