开发者在Intel软件开发手册（SDM）中发现了一条名为vfmaddrnd231pd的指令，该指令声称支持浮点运算的每指令级舍入模式控制1。然而，这条指令实际上并不存在——GNU汇编器无法识别它，在支持FMA的CPU上执行会触发未定义指令异常1。Clang和ICC编译器同样不支持该指令1。
这条虚幻指令仅出现在Intel 2024版SDM中，在2026版本中已被完全移除1。SDM文档中对该指令的唯一记录为"VEX.DDS.128.66.0F3A.W1 B8 /r /ib"，其中DDS为仅用于文档的标记，在后续版本中已被删除1。在Haswell等真实处理器上尝试执行这条指令会直接触发#UD（未定义指令）异常1。
A developer discovered that Intel's Software Development Manual (SDM) documents an instruction called vfmaddrnd231pd that does not actually exist on real processors 1. The instruction claims to provide per-instruction rounding mode control for floating-point operations, yet no CPU implements it 1.
Testing revealed that when executed on systems with FMA support, such as Haswell processors, the vfmaddrnd231pd instruction triggers an undefined instruction (#UD) exception 1. Major development tools including GNU Assembler, Clang, and ICC compiler do not recognize or support the instruction 1. The instruction appeared only in Intel's 2024 SDM release and has been removed in the 2026 version 1. In the documentation, the instruction was referenced solely through the encoding "VEX.DDS.128.66.0F3A.W1 B8 /r /ib," where DDS is a documentation-only marking that was subsequently eliminated in later manual versions 1.
评论
还没有评论，欢迎留下第一条。