Floating Point

Notes

• 2.3.1.6
  • There are 15 floaing point registers
    • xmm0 - xmm15
  • They support
    • 32, 64 and 128 bit floating point
    • Also SIMD : Single Instruction Multiple Data
  • There are also ymm (256 bits) and zmm (512 bit) registers in
• Chapter 18
  • movss, movsd
    • movss: move a 32 bit float between locations
    • movsd: move a 64 bit float between locations
    • Both can not be memory
    • Neither can be an immediate
    • mem to register clears the top bits
    • reg to reg does not.
  • cvtss2sd cvtsd2ss
    • convert 32 to 64 bit number
    • Convert 64 to 32 bit number
  • cvtss2si/ cvttss2si , cvtsd2si/cvttsd2si
    • cvtss2si: Convert 32 bit float to double round
    • cvttss2si: Convert 32 bit float to double truncate
    • cvtsd2si: Convert 64 bit float to double round
    • cvttsd2si: Convert 64 bit float to double truncate
  • addss, addsd : floating point add
  • subss, subsd : subtract
  • mulss, mulsd
  • divss, divsd
  • sqrtss, sqrtsd
  • ucomiss, ucoisd : compare floating point
  • Calling Convention:
    • Parameters in xmm0 through xmm7
    • Others on the stack
    • rax holds the number of floating point parameters.
  • normalDist.asm
• SIMD floating point instructions:
  • movups: move unaligned paced single precision float values
    • memory or register, but both can not be memory
  • addps: add single precison float
  • simd.asm, which I mostly got from gemini