xvresp

Vector Reciprocal Estimate Single-Precision

xvresp XT,XB

Estimates the reciprocal of single-precision floating-point values in a vector.

Encoding

Binary Layout

0:10

11:15

16:20

21:31

Format XX2-form

Opcode 0xF0000268

Extension VSX

Operands

XT
Target Vector Register
XB
Source Vector Register

Example

xvresp vs1, vs3

Registers Altered

FPSCR FX OX UX ZX VXSNAN

Reference

View in PowerISA v3.1C Specification ↗ OPF Power ISA v3.1C

Description

A single-precision floating-point estimate of the reciprocal of src is placed into word element i of VSR[XT] in single-precision format. Unless the reciprocal of src would be a zero, an infinity, or a QNaN, the estimate has a relative error in precision no greater than one part in 16384 of the reciprocal of src.

Operation

if MSR.VSX=0 then VSX_Unavailable()
ex_flag ←0b0
reset_xflags()
do i = 0 to 3
    src ←bfp_CONVERT_FROM_BFP32(VSR[32×BX+B].word[i])
    rnd ←bfp_ROUND_TO_BFP32(FPSCR.RN,v)
    vresult.word[i] ←bfp32_CONVERT_FROM_BFP(rnd)
    if vxsnan_flag=1 then SetFX(FPSCR.VXSNAN)
    if ox_flag=1 then SetFX(FPSCR.OX)
    if ux_flag=1 then SetFX(FPSCR.UX)
    if zx_flag=1 then SetFX(FPSCR.ZX)
    ex_flag ←ex_flag | (FPSCR.VE & vxsnan_flag) | (FPSCR.OE & ox_flag) | (FPSCR.UE & ux_flag) | (FPSCR.ZE & zx_flag)
end
if ex_flag=0 then VSR[32×TX+T] ←vresult

Programming Note

The xvresp instruction is used to estimate the reciprocal of single-precision floating-point numbers in vector registers. It's important to ensure that the VSX (Vector Scalar Extensions) are enabled, as attempting to use this instruction when they are not will result in an exception. The instruction handles NaNs and infinities by setting appropriate flags in the FPSCR register, but developers should be cautious of potential precision loss due to the estimation process.