These intrinsics perform the first of two steps in an iteration of the Newton-Raphson method to converge to a reciprocal or a square root. See VRECPS and VRSQRTS in the Assembler Reference.

float32x2_t vrecps_f32(float32x2_t a, float32x2_t b);   // VRECPS.F32 d0, d0, d0 
float32x4_t vrecpsq_f32(float32x4_t a, float32x4_t b);  // VRECPS.F32 q0, q0, q0 
float32x2_t vrsqrts_f32(float32x2_t a, float32x2_t b);  // VRSQRTS.F32 d0, d0, d0
float32x4_t vrsqrtsq_f32(float32x4_t a, float32x4_t b); // VRSQRTS.F32 q0, q0, q0