// clang-format off

// Auto-generated file. Do not edit!

//   Template: src/f32-dwconv/unipass-scalar.c.in

//   Generator: tools/xngen

//

// Copyright 2019 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 <stddef.h>

#include <stdint.h>

#include "src/xnnpack/common.h"

#include "src/xnnpack/dwconv.h"

#include "src/xnnpack/math.h"

#include "src/xnnpack/microparams.h"

void xnn_f32_dwconv_minmax_ukernel_3p1c__scalar_acc2(

    size_t channels,

    size_t output_width,

    const float** input,

    const float* weights,

    float* output,

    intptr_t input_stride,

    size_t output_increment,

    size_t input_offset,

    size_t input_pixel_stride,

    const float* zero,

    const struct xnn_f32_minmax_params* restrict params)

{

  assert(channels != 0);

  assert(output_width != 0);

  const float vmin = params->scalar.min;

  const float vmax = params->scalar.max;

  do {

    const float* i0 = input[0];

    assert(i0 != NULL);

    if XNN_UNPREDICTABLE(i0 != zero) {

      i0 = (const float*) ((uintptr_t) i0 + input_offset);

    }

    const float* i1 = input[1];

    assert(i1 != NULL);

    if XNN_UNPREDICTABLE(i1 != zero) {

      i1 = (const float*) ((uintptr_t) i1 + input_offset);

    }

    const float* i2 = input[2];

    assert(i2 != NULL);

    if XNN_UNPREDICTABLE(i2 != zero) {

      i2 = (const float*) ((uintptr_t) i2 + input_offset);

    }

    input = (const float**) ((uintptr_t) input + input_stride);

    size_t c = channels;

    const float* w = weights;

    do {

      float vacc0p0 = w[0];

      const float vi0 = *i0++;

      const float vk0 = w[1];

      vacc0p0 = math_muladd_f32(vi0, vk0, vacc0p0);

      const float vi1 = *i1++;

      const float vk1 = w[2];

      float vacc0p1 = vi1 * vk1;

      const float vi2 = *i2++;

      const float vk2 = w[3];

      vacc0p0 = math_muladd_f32(vi2, vk2, vacc0p0);

      w += 4;

      vacc0p0 += vacc0p1;

      float vacc0 = math_max_f32(vacc0p0, vmin);

      vacc0 = math_min_f32(vacc0, vmax);

      *output++ = vacc0;

    } while (--c != 0);

    input_offset += input_pixel_stride;

    output = (float*) ((uintptr_t) output + output_increment);

  } while (--output_width != 0);

}