src/qc8-igemm/gen/2x4-minmax-fp32-scalar-magic.c

// Auto-generated file. Do not edit!
//   Template: src/qs8-igemm/scalar.c.in
//   Generator: tools/xngen
//
// Copyright 2021 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 <fp16.h>

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


void xnn_qc8_igemm_minmax_fp32_ukernel_2x4__scalar_magic(
    size_t mr,
    size_t nc,
    size_t kc,
    size_t ks,
    const int8_t**restrict a,
    const void*restrict w,
    int8_t*restrict c,
    size_t cm_stride,
    size_t cn_stride,
    size_t a_offset,
    const int8_t* zero,
    const union xnn_qs8_minmax_params params[restrict XNN_MIN_ELEMENTS(1)])
{
  assert(mr != 0);
  assert(mr <= 2);
  assert(nc != 0);
  assert(kc != 0);
  assert(ks != 0);
  assert(ks % (2 * sizeof(void*)) == 0);
  assert(a != NULL);
  assert(w != NULL);
  assert(c != NULL);

  int8_t* c0 = c;
  int8_t* c1 = (int8_t*) ((uintptr_t) c0 + cm_stride);
  if XNN_UNPREDICTABLE(mr != 2) {
    c1 = c0;
  }

  do {
    int32_t vacc0x0 = ((const int32_t*) w)[0];
    int32_t vacc0x1 = ((const int32_t*) w)[1];
    int32_t vacc0x2 = ((const int32_t*) w)[2];
    int32_t vacc0x3 = ((const int32_t*) w)[3];
    int32_t vacc1x0 = vacc0x0;
    int32_t vacc1x1 = vacc0x1;
    int32_t vacc1x2 = vacc0x2;
    int32_t vacc1x3 = vacc0x3;
    w = (const void*) ((const int32_t*) w + 4);

    size_t p = ks;
    do {
      const int8_t* restrict a0 = a[0];
      assert(a0 != NULL);
      if XNN_UNPREDICTABLE(a0 != zero) {
        a0 = (const int8_t*) ((uintptr_t) a0 + a_offset);
      }
      const int8_t* restrict a1 = a[1];
      assert(a1 != NULL);
      if XNN_UNPREDICTABLE(a1 != zero) {
        a1 = (const int8_t*) ((uintptr_t) a1 + a_offset);
      }
      a += 2;

      size_t k = kc;
      do {
        const int32_t va0 = (int32_t) *a0++;
        const int32_t va1 = (int32_t) *a1++;

        const int32_t vb0 = (int32_t) ((const int8_t*) w)[0];
        const int32_t vb1 = (int32_t) ((const int8_t*) w)[1];
        const int32_t vb2 = (int32_t) ((const int8_t*) w)[2];
        const int32_t vb3 = (int32_t) ((const int8_t*) w)[3];
        w = (const void*) ((const int8_t*) w + 4);

        vacc0x0 += va0 * vb0;
        vacc0x1 += va0 * vb1;
        vacc0x2 += va0 * vb2;
        vacc0x3 += va0 * vb3;
        vacc1x0 += va1 * vb0;
        vacc1x1 += va1 * vb1;
        vacc1x2 += va1 * vb2;
        vacc1x3 += va1 * vb3;

        k -= sizeof(int8_t);
      } while (k != 0);
      p -= 2 * sizeof(void*);
    } while (p != 0);

    float vfpacc0x0 = (float) vacc0x0;
    float vfpacc0x1 = (float) vacc0x1;
    float vfpacc0x2 = (float) vacc0x2;
    float vfpacc0x3 = (float) vacc0x3;
    float vfpacc1x0 = (float) vacc1x0;
    float vfpacc1x1 = (float) vacc1x1;
    float vfpacc1x2 = (float) vacc1x2;
    float vfpacc1x3 = (float) vacc1x3;

    const float vscale0 = ((const float*) w)[0];
    vfpacc0x0 *= vscale0;
    vfpacc1x0 *= vscale0;
    const float vscale1 = ((const float*) w)[1];
    vfpacc0x1 *= vscale1;
    vfpacc1x1 *= vscale1;
    const float vscale2 = ((const float*) w)[2];
    vfpacc0x2 *= vscale2;
    vfpacc1x2 *= vscale2;
    const float vscale3 = ((const float*) w)[3];
    vfpacc0x3 *= vscale3;
    vfpacc1x3 *= vscale3;
    w = (const void*) ((const float*) w + 4);

    const float voutput_min_less_zero_point = params->scalar_magic.output_min_less_zero_point;
    vfpacc0x0 = math_max_f32(vfpacc0x0, voutput_min_less_zero_point);
    vfpacc0x1 = math_max_f32(vfpacc0x1, voutput_min_less_zero_point);
    vfpacc0x2 = math_max_f32(vfpacc0x2, voutput_min_less_zero_point);
    vfpacc0x3 = math_max_f32(vfpacc0x3, voutput_min_less_zero_point);
    vfpacc1x0 = math_max_f32(vfpacc1x0, voutput_min_less_zero_point);
    vfpacc1x1 = math_max_f32(vfpacc1x1, voutput_min_less_zero_point);
    vfpacc1x2 = math_max_f32(vfpacc1x2, voutput_min_less_zero_point);
    vfpacc1x3 = math_max_f32(vfpacc1x3, voutput_min_less_zero_point);

    const float voutput_max_less_zero_point = params->scalar_magic.output_max_less_zero_point;
    vfpacc0x0 = math_min_f32(vfpacc0x0, voutput_max_less_zero_point);
    vfpacc0x1 = math_min_f32(vfpacc0x1, voutput_max_less_zero_point);
    vfpacc0x2 = math_min_f32(vfpacc0x2, voutput_max_less_zero_point);
    vfpacc0x3 = math_min_f32(vfpacc0x3, voutput_max_less_zero_point);
    vfpacc1x0 = math_min_f32(vfpacc1x0, voutput_max_less_zero_point);
    vfpacc1x1 = math_min_f32(vfpacc1x1, voutput_max_less_zero_point);
    vfpacc1x2 = math_min_f32(vfpacc1x2, voutput_max_less_zero_point);
    vfpacc1x3 = math_min_f32(vfpacc1x3, voutput_max_less_zero_point);

    const float vmagic_bias = params->scalar_magic.magic_bias;
    vfpacc0x0 += vmagic_bias;
    vfpacc0x1 += vmagic_bias;
    vfpacc0x2 += vmagic_bias;
    vfpacc0x3 += vmagic_bias;
    vfpacc1x0 += vmagic_bias;
    vfpacc1x1 += vmagic_bias;
    vfpacc1x2 += vmagic_bias;
    vfpacc1x3 += vmagic_bias;

    const int32_t vmagic_bias_less_output_zero_point = params->scalar_magic.magic_bias_less_output_zero_point;
    int32_t vout0x0 = (int32_t) fp32_to_bits(vfpacc0x0) - vmagic_bias_less_output_zero_point;
    int32_t vout0x1 = (int32_t) fp32_to_bits(vfpacc0x1) - vmagic_bias_less_output_zero_point;
    int32_t vout0x2 = (int32_t) fp32_to_bits(vfpacc0x2) - vmagic_bias_less_output_zero_point;
    int32_t vout0x3 = (int32_t) fp32_to_bits(vfpacc0x3) - vmagic_bias_less_output_zero_point;
    int32_t vout1x0 = (int32_t) fp32_to_bits(vfpacc1x0) - vmagic_bias_less_output_zero_point;
    int32_t vout1x1 = (int32_t) fp32_to_bits(vfpacc1x1) - vmagic_bias_less_output_zero_point;
    int32_t vout1x2 = (int32_t) fp32_to_bits(vfpacc1x2) - vmagic_bias_less_output_zero_point;
    int32_t vout1x3 = (int32_t) fp32_to_bits(vfpacc1x3) - vmagic_bias_less_output_zero_point;

    if XNN_LIKELY(nc >= 4) {
      c1[0] = (int8_t) vout1x0;
      c1[1] = (int8_t) vout1x1;
      c1[2] = (int8_t) vout1x2;
      c1[3] = (int8_t) vout1x3;
      c0[0] = (int8_t) vout0x0;
      c0[1] = (int8_t) vout0x1;
      c0[2] = (int8_t) vout0x2;
      c0[3] = (int8_t) vout0x3;

      c1 = (int8_t*) ((uintptr_t) c1 + cn_stride);
      c0 = (int8_t*) ((uintptr_t) c0 + cn_stride);

      a = (const int8_t**restrict) ((uintptr_t) a - ks);
      nc -= 4;
    } else {
      if (nc & 2) {
        c1[0] = (int8_t) vout1x0;
        c1[1] = (int8_t) vout1x1;
        vout1x0 = vout1x2;
        c1 += 2;
        c0[0] = (int8_t) vout0x0;
        c0[1] = (int8_t) vout0x1;
        vout0x0 = vout0x2;
        c0 += 2;
      }
      if (nc & 1) {
        c1[0] = (int8_t) vout1x0;
        c0[0] = (int8_t) vout0x0;
      }

      nc = 0;
    }
  } while (nc != 0);
}