src/f32-dwconv-chw/3x3p1-scalar-1x1-acc3.c

// 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 <xnnpack/dwconv.h>
#include <xnnpack/math.h>


void xnn_f32_dwconv_chw_ukernel_3x3p1__scalar_1x1_acc3(
    size_t input_height,
    size_t input_width,
    const float* input,
    const float* weights,
    const float* zero,
    float* output,
    uint32_t padding_top,
    const union xnn_f32_chw_params params[restrict XNN_MIN_ELEMENTS(1)])
{
  assert(input_height != 0);
  assert(input_width != 0);
  assert(input_width % sizeof(float) == 0);
  assert(padding_top == 1);

  const float params_min = params->scalar.min;
  const float params_max = params->scalar.max;

  const float vbias = weights[0];
  const float vk00 = weights[1];
  const float vk01 = weights[2];
  const float vk02 = weights[3];
  const float vk10 = weights[4];
  const float vk11 = weights[5];
  const float vk12 = weights[6];
  const float vk20 = weights[7];
  const float vk21 = weights[8];
  const float vk22 = weights[9];

  const float* i0 = zero;
  const float* i1 = input;
  const float* i2 = (const float*) ((uintptr_t) i1 + input_width);

  size_t output_height = input_height;
  do {
    if XNN_UNPREDICTABLE(output_height == 1) {
      i2 = zero;
    }

    float vi0x0 = 0.0f;
    float vi1x0 = 0.0f;
    float vi2x0 = 0.0f;
    float vi0x1 = *i0++;
    float vi1x1 = *i1++;
    float vi2x1 = *i2++;

    size_t w = input_width;
    for (; w > 1 * sizeof(float); w -= 1 * sizeof(float)) {
      const float vi0x2 = *i0++;
      const float vi1x2 = *i1++;
      const float vi2x2 = *i2++;

      float vacc0 = vk00 * vi0x0;
      vi0x0 = vi0x1;
      float vacc1 = vk10 * vi1x0;
      vi1x0 = vi1x1;
      float vacc2 = vk20 * vi2x0;
      vi2x0 = vi2x1;

      vacc0 += vk01 * vi0x1;
      vi0x1 = vi0x2;
      vacc1 += vk11 * vi1x1;
      vi1x1 = vi1x2;
      vacc2 += vk21 * vi2x1;
      vi2x1 = vi2x2;

      vacc0 += vk02 * vi0x2;
      vacc1 += vk12 * vi1x2;
      vacc2 += vk22 * vi2x2;

      float voutput = (vbias + vacc0) + (vacc1 + vacc2);

      voutput = math_max_f32(voutput, params_min);
      voutput = math_min_f32(voutput, params_max);

      *output++ = voutput;
    }
    // Always process the last pixel separately to account for right edge.
    assert(w == 1 * sizeof(float));
    {
      float vacc0 = vk00 * vi0x0;
      float vacc1 = vk10 * vi1x0;
      float vacc2 = vk20 * vi2x0;

      vacc0 += vk01 * vi0x1;
      vacc1 += vk11 * vi1x1;
      vacc2 += vk21 * vi2x1;

      float voutput = (vbias + vacc0) + (vacc1 + vacc2);

      voutput = math_max_f32(voutput, params_min);
      voutput = math_min_f32(voutput, params_max);

      *output++ = voutput;
    }

    i0 = (const float*) ((uintptr_t) i1 - input_width);
  } while (--output_height != 0);
}