-fp-rcp flag along with -enable-unsafe-fp-math flag controls the generatation of X86 rcp instruction generation and 
some derived optimizations possible from generated rsqrt instructions.
We have observed some effects on precision and hence has put these transformations under the mentioned flags. 
Following is the details of the -fp-rcp flag along with its values and enabled optimizations.

  -fp-rcp                           -   Enable rcp ops
    =off                                -   No rcp
    =on                                -   y/x => y * rcp(x)  i.e. Standard
    =fda                               -   Standard, Derive FMA

Generate LLVM IR by
gfortran -S -O1 -o - -fplugin=dragonegg.so <filename> -fplugin-arg-dragonegg-emit-ir
.ll -> .s file generation by
llc -O3 -enable-unsafe-fp-math -fp-rsqrt=off/on/advance/fda <.ll filename> -filetype=asm

Example 1. 
    Source code

      real*4     x
      real*4     y
      real*4     r
      r = y/x

LLVM IR

  %0 = load float* %y, align 4
  %1 = load float* %x, align 4
  %2 = fdiv float %0, %1
  store float %2, float* %r, align 4

-fp-rcp=off

        movq    %rdx, %r14
        movq    %rsi, %r15
        movq    %rdi, %rbx
        vmovss  (%r15), %xmm0
        vdivss  (%rbx), %xmm0, %xmm0
        leaq    8(%rsp), %r12
        vmovss  %xmm0, (%r14)

-fp-rcp=on

        vmovss  (%rbx), %xmm0
        vrcpss  %xmm0, %xmm0, %xmm0
        vmulss  (%r15), %xmm0, %xmm0
        vmovss  %xmm0, (%r14)

Input
------
 x =    3.00000000      y =    7.00000000

Output
---------

without rcp:: r =    2.33333325
with rcp::      r =    2.33329773

Example 2

Source

      real*4     x
      real*4     y
      real*4     z
      real*4     r
      r           = z+y/x


LLVM-IR

  %0 = load float* %z, align 4
  %1 = load float* %y, align 4
  %2 = load float* %x, align 4
  %3 = fdiv float %1, %2
  %4 = fadd float %0, %3
  store float %4, float* %r, align 4

-fp-rcp=off

        movq    %rcx, %r14
        movq    %rdx, %r15
        movq    %rsi, %r12
        movq    %rdi, %rbx
        vmovss  (%r12), %xmm0
        vdivss  (%rbx), %xmm0, %xmm0
        leaq    (%rsp), %r13
        vaddss  (%r15), %xmm0, %xmm0
        vmovss  %xmm0, (%r14)

-fp-rcp=fda

        movq    %rcx, %r14
        movq    %rdx, %r15
        movq    %rsi, %r12
        movq    %rdi, %r13
        vmovss  (%r13), %xmm0
        vrcpss  %xmm0, %xmm0, %xmm0
        vmovss  (%r12), %xmm1
        vfmadd213ss     (%r15), %xmm1, %xmm0
        vmovss  %xmm0, (%r14)

Input
-----

 x =    3.00000000      y =    7.00000000      z =   0.330000013

Output
--------

without rcp:: r =    2.66333318
with rcp+fma:: r =    2.66329765