NASM 2.05 based x86 Instruction Reference[ch293]
A.5.269 RCPSS: Scalar Single-Precision FP Reciprocal RCPSS xmm1,xmm2/m128 ; F3 0F 53 /r [KATMAI,SSE] RCPSS returns an approximation of the reciprocal of the lower single- precision FP value from xmm2/m32; the upper three fields are passed through from xmm1. The maximum error for this approximation is: |Error| <= 1.5 x 2^-12