xref: /aosp_15_r20/external/XNNPACK/src/f32-vsqrt/gen/neonfma-nr2fma1adj-x8.c (revision 4bdc94577ba0e567308109d787f7fec7b531ce36)

// Auto-generated file. Do not edit!
//   Template: src/f32-vsqrt/neonfma-nr2fma1adj.c.in
//   Generator: tools/xngen
//
// Copyright 2020 Google LLC
//
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree.

#include <assert.h>
#include <math.h>

#include <arm_neon.h>

#include <xnnpack/common.h>
#include <xnnpack/vunary.h>


void xnn_f32_vsqrt_ukernel__neonfma_nr2fma1adj_x8(
    size_t n,
    const float* x,
    float* y,
    const union xnn_f32_sqrt_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS
{
  assert(n != 0);
  assert(n % sizeof(float) == 0);

  const float32x4_t vhalf = vmovq_n_f32(0.5f);
  for (; n >= 8 * sizeof(float); n -= 8 * sizeof(float)) {
    const float32x4_t vx0123 = vld1q_f32(x); x += 4;
    const float32x4_t vx4567 = vld1q_f32(x); x += 4;

    const float32x4_t vrsqrtx0123 = vrsqrteq_f32(vx0123);
    const float32x4_t vrsqrtx4567 = vrsqrteq_f32(vx4567);

    float32x4_t vsqrtx0123 = vmulq_f32(vrsqrtx0123, vx0123);
    float32x4_t vhalfrsqrtx0123 = vmulq_f32(vrsqrtx0123, vhalf);
    float32x4_t vsqrtx4567 = vmulq_f32(vrsqrtx4567, vx4567);
    float32x4_t vhalfrsqrtx4567 = vmulq_f32(vrsqrtx4567, vhalf);

    float32x4_t vresidual0123 = vfmsq_f32(vhalf, vsqrtx0123, vhalfrsqrtx0123);
    float32x4_t vresidual4567 = vfmsq_f32(vhalf, vsqrtx4567, vhalfrsqrtx4567);

    vhalfrsqrtx0123 = vfmaq_f32(vhalfrsqrtx0123, vresidual0123, vhalfrsqrtx0123);
    vsqrtx0123 = vfmaq_f32(vsqrtx0123, vresidual0123, vsqrtx0123);
    vhalfrsqrtx4567 = vfmaq_f32(vhalfrsqrtx4567, vresidual4567, vhalfrsqrtx4567);
    vsqrtx4567 = vfmaq_f32(vsqrtx4567, vresidual4567, vsqrtx4567);

    vresidual0123 = vfmsq_f32(vhalf, vsqrtx0123, vhalfrsqrtx0123);
    vresidual4567 = vfmsq_f32(vhalf, vsqrtx4567, vhalfrsqrtx4567);

    vhalfrsqrtx0123 = vfmaq_f32(vhalfrsqrtx0123, vresidual0123, vhalfrsqrtx0123);
    vsqrtx0123 = vfmaq_f32(vsqrtx0123, vresidual0123, vsqrtx0123);
    vhalfrsqrtx4567 = vfmaq_f32(vhalfrsqrtx4567, vresidual4567, vhalfrsqrtx4567);
    vsqrtx4567 = vfmaq_f32(vsqrtx4567, vresidual4567, vsqrtx4567);

    const float32x4_t vadjustment0123 = vfmsq_f32(vx0123, vsqrtx0123, vsqrtx0123);
    const float32x4_t vadjustment4567 = vfmsq_f32(vx4567, vsqrtx4567, vsqrtx4567);

    const float32x4_t vy0123 = vfmaq_f32(vsqrtx0123, vhalfrsqrtx0123, vadjustment0123);
    const float32x4_t vy4567 = vfmaq_f32(vsqrtx4567, vhalfrsqrtx4567, vadjustment4567);

    vst1q_f32(y, vy0123); y += 4;
    vst1q_f32(y, vy4567); y += 4;
  }
  for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
    const float32x4_t vx = vld1q_f32(x); x += 4;
    const float32x4_t vrsqrtx = vrsqrteq_f32(vx);
    float32x4_t vsqrtx = vmulq_f32(vrsqrtx, vx);
    float32x4_t vhalfrsqrtx = vmulq_f32(vrsqrtx, vhalf);
    float32x4_t vresidual = vfmsq_f32(vhalf, vsqrtx, vhalfrsqrtx);
    vhalfrsqrtx = vfmaq_f32(vhalfrsqrtx, vresidual, vhalfrsqrtx);
    vsqrtx = vfmaq_f32(vsqrtx, vresidual, vsqrtx);
    vresidual = vfmsq_f32(vhalf, vsqrtx, vhalfrsqrtx);
    vhalfrsqrtx = vfmaq_f32(vhalfrsqrtx, vresidual, vhalfrsqrtx);
    vsqrtx = vfmaq_f32(vsqrtx, vresidual, vsqrtx);
    const float32x4_t vadjustment = vfmsq_f32(vx, vsqrtx, vsqrtx);
    const float32x4_t vy = vfmaq_f32(vsqrtx, vhalfrsqrtx, vadjustment);
    vst1q_f32(y, vy); y += 4;
  }
  if XNN_UNLIKELY(n != 0) {
    const float32x4_t vx = vld1q_f32(x);
    const float32x4_t vrsqrtx = vrsqrteq_f32(vx);
    float32x4_t vsqrtx = vmulq_f32(vrsqrtx, vx);
    float32x4_t vhalfrsqrtx = vmulq_f32(vrsqrtx, vhalf);
    float32x4_t vresidual = vfmsq_f32(vhalf, vsqrtx, vhalfrsqrtx);
    vhalfrsqrtx = vfmaq_f32(vhalfrsqrtx, vresidual, vhalfrsqrtx);
    vsqrtx = vfmaq_f32(vsqrtx, vresidual, vsqrtx);
    vresidual = vfmsq_f32(vhalf, vsqrtx, vhalfrsqrtx);
    vhalfrsqrtx = vfmaq_f32(vhalfrsqrtx, vresidual, vhalfrsqrtx);
    vsqrtx = vfmaq_f32(vsqrtx, vresidual, vsqrtx);
    const float32x4_t vadjustment = vfmsq_f32(vx, vsqrtx, vsqrtx);
    const float32x4_t vy = vfmaq_f32(vsqrtx, vhalfrsqrtx, vadjustment);

    float32x2_t vy_lo = vget_low_f32(vy);
    if (n & (2 * sizeof(float))) {
      vst1_f32(y, vy_lo); y += 2;
      vy_lo = vget_high_f32(vy);
    }
    if (n & (1 * sizeof(float))) {
      vst1_lane_f32(y, vy_lo, 0);
    }
  }
}