// Auto-generated file. Do not edit!
//   Template: src/f32-vrnd/neonv8.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 <arm_neon.h>

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


void xnn_f32_vrndd_ukernel__neonv8_x8(
    size_t n,
    const float* x,
    float* y,
    const union xnn_f32_rnd_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_DISABLE_TSAN
{
  assert(n != 0);
  assert(n % sizeof(float) == 0);

  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 vy0123 = vrndmq_f32(vx0123);
    const float32x4_t vy4567 = vrndmq_f32(vx4567);

    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 vy = vrndmq_f32(vx);
    vst1q_f32(y, vy); y += 4;
  }
  if XNN_UNLIKELY(n != 0) {
    const float32x4_t vx = vld1q_f32(x);
    const float32x4_t vy = vrndmq_f32(vx);
    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);
    }
  }
}