xref: /aosp_15_r20/external/XNNPACK/src/f32-igemm/gen/3x8s4-minmax-wasmrelaxedsimd.c (revision 4bdc94577ba0e567308109d787f7fec7b531ce36)

// Auto-generated file. Do not edit!
//   Template: src/f32-igemm/wasmsimd-s4.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 <wasm_simd128.h>

#include <xnnpack/igemm.h>


void xnn_f32_igemm_minmax_ukernel_3x8s4__wasmrelaxedsimd(
    size_t mr,
    size_t nc,
    size_t kc,
    size_t ks,
    const float**restrict a,
    const float*restrict w,
    float*restrict c,
    size_t cm_stride,
    size_t cn_stride,
    size_t a_offset,
    const float* zero,
    const union xnn_f32_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS
{
  assert(mr != 0);
  assert(mr <= 3);
  assert(nc != 0);
  assert(kc != 0);
  assert(kc % sizeof(float) == 0);
  assert(ks != 0);
  assert(ks % (3 * sizeof(void*)) == 0);
  assert(a_offset % sizeof(float) == 0);
  assert(a != NULL);
  assert(w != NULL);
  assert(c != NULL);

  float* c0 = c;
  float* c1 = (float*) ((uintptr_t) c0 + cm_stride);
  if XNN_UNPREDICTABLE(mr < 2) {
    c1 = c0;
  }
  float* c2 = (float*) ((uintptr_t) c1 + cm_stride);
  if XNN_UNPREDICTABLE(mr <= 2) {
    c2 = c1;
  }

  const v128_t vmin = wasm_v128_load64_splat(params->wasmsimd.min);
  const v128_t vmax = wasm_v128_load64_splat(params->wasmsimd.max);
  do {
    v128_t vacc0x0123 = wasm_v128_load(w);
    v128_t vacc0x4567 = wasm_v128_load(w + 4);
    v128_t vacc1x0123 = vacc0x0123;
    v128_t vacc1x4567 = vacc0x4567;
    v128_t vacc2x0123 = vacc0x0123;
    v128_t vacc2x4567 = vacc0x4567;
    w += 8;

    size_t p = ks;
    do {
      const float* restrict a0 = a[0];
      assert(a0 != NULL);
      if XNN_UNPREDICTABLE(a0 != zero) {
        a0 = (const float*) ((uintptr_t) a0 + a_offset);
      }
      const float* restrict a1 = a[1];
      assert(a1 != NULL);
      if XNN_UNPREDICTABLE(a1 != zero) {
        a1 = (const float*) ((uintptr_t) a1 + a_offset);
      }
      const float* restrict a2 = a[2];
      assert(a2 != NULL);
      if XNN_UNPREDICTABLE(a2 != zero) {
        a2 = (const float*) ((uintptr_t) a2 + a_offset);
      }
      a += 3;

      size_t k = kc;
      while (k >= 4 * sizeof(float)) {
        v128_t va0 = wasm_v128_load(a0);
        a0 += 4;
        v128_t va1 = wasm_v128_load(a1);
        a1 += 4;
        v128_t va2 = wasm_v128_load(a2);
        a2 += 4;


        const v128_t vb0123c0 = wasm_v128_load(w + 0);
        const v128_t vb4567c0 = wasm_v128_load(w + 4);

        vacc0x0123 = wasm_f32x4_add(vacc0x0123, wasm_f32x4_mul(va0, vb0123c0));
        vacc1x0123 = wasm_f32x4_add(vacc1x0123, wasm_f32x4_mul(va1, vb0123c0));
        vacc2x0123 = wasm_f32x4_add(vacc2x0123, wasm_f32x4_mul(va2, vb0123c0));
        vacc0x4567 = wasm_f32x4_add(vacc0x4567, wasm_f32x4_mul(va0, vb4567c0));
        vacc1x4567 = wasm_f32x4_add(vacc1x4567, wasm_f32x4_mul(va1, vb4567c0));
        vacc2x4567 = wasm_f32x4_add(vacc2x4567, wasm_f32x4_mul(va2, vb4567c0));

        va0 = wasm_v32x4_shuffle(va0, va0, 1, 2, 3, 0);
        va1 = wasm_v32x4_shuffle(va1, va1, 1, 2, 3, 0);
        va2 = wasm_v32x4_shuffle(va2, va2, 1, 2, 3, 0);

        const v128_t vb0123c1 = wasm_v128_load(w + 8);
        const v128_t vb4567c1 = wasm_v128_load(w + 12);

        vacc0x0123 = wasm_f32x4_add(vacc0x0123, wasm_f32x4_mul(va0, vb0123c1));
        vacc1x0123 = wasm_f32x4_add(vacc1x0123, wasm_f32x4_mul(va1, vb0123c1));
        vacc2x0123 = wasm_f32x4_add(vacc2x0123, wasm_f32x4_mul(va2, vb0123c1));
        vacc0x4567 = wasm_f32x4_add(vacc0x4567, wasm_f32x4_mul(va0, vb4567c1));
        vacc1x4567 = wasm_f32x4_add(vacc1x4567, wasm_f32x4_mul(va1, vb4567c1));
        vacc2x4567 = wasm_f32x4_add(vacc2x4567, wasm_f32x4_mul(va2, vb4567c1));

        va0 = wasm_v32x4_shuffle(va0, va0, 1, 2, 3, 0);
        va1 = wasm_v32x4_shuffle(va1, va1, 1, 2, 3, 0);
        va2 = wasm_v32x4_shuffle(va2, va2, 1, 2, 3, 0);

        const v128_t vb0123c2 = wasm_v128_load(w + 16);
        const v128_t vb4567c2 = wasm_v128_load(w + 20);

        vacc0x0123 = wasm_f32x4_add(vacc0x0123, wasm_f32x4_mul(va0, vb0123c2));
        vacc1x0123 = wasm_f32x4_add(vacc1x0123, wasm_f32x4_mul(va1, vb0123c2));
        vacc2x0123 = wasm_f32x4_add(vacc2x0123, wasm_f32x4_mul(va2, vb0123c2));
        vacc0x4567 = wasm_f32x4_add(vacc0x4567, wasm_f32x4_mul(va0, vb4567c2));
        vacc1x4567 = wasm_f32x4_add(vacc1x4567, wasm_f32x4_mul(va1, vb4567c2));
        vacc2x4567 = wasm_f32x4_add(vacc2x4567, wasm_f32x4_mul(va2, vb4567c2));

        va0 = wasm_v32x4_shuffle(va0, va0, 1, 2, 3, 0);
        va1 = wasm_v32x4_shuffle(va1, va1, 1, 2, 3, 0);
        va2 = wasm_v32x4_shuffle(va2, va2, 1, 2, 3, 0);

        const v128_t vb0123c3 = wasm_v128_load(w + 24);
        const v128_t vb4567c3 = wasm_v128_load(w + 28);

        vacc0x0123 = wasm_f32x4_add(vacc0x0123, wasm_f32x4_mul(va0, vb0123c3));
        vacc1x0123 = wasm_f32x4_add(vacc1x0123, wasm_f32x4_mul(va1, vb0123c3));
        vacc2x0123 = wasm_f32x4_add(vacc2x0123, wasm_f32x4_mul(va2, vb0123c3));
        vacc0x4567 = wasm_f32x4_add(vacc0x4567, wasm_f32x4_mul(va0, vb4567c3));
        vacc1x4567 = wasm_f32x4_add(vacc1x4567, wasm_f32x4_mul(va1, vb4567c3));
        vacc2x4567 = wasm_f32x4_add(vacc2x4567, wasm_f32x4_mul(va2, vb4567c3));


        w += 32;
        k -= 4 * sizeof(float);
      }
      if XNN_UNLIKELY(k != 0) {
        v128_t va0 = wasm_v128_load(a0);
        a0 = (const float*) ((uintptr_t) a0 + k);
        v128_t va1 = wasm_v128_load(a1);
        a1 = (const float*) ((uintptr_t) a1 + k);
        v128_t va2 = wasm_v128_load(a2);
        a2 = (const float*) ((uintptr_t) a2 + k);

        const v128_t vzero = wasm_f32x4_const_splat(0.0f);

        const v128_t vb0123c0 = wasm_v128_load(w + 0);
        const v128_t vb4567c0 = wasm_v128_load(w + 4);

        vacc0x0123 = wasm_f32x4_add(vacc0x0123, wasm_f32x4_mul(wasm_v128_andnot(va0, wasm_f32x4_eq(vb0123c0, vzero)), vb0123c0));
        vacc1x0123 = wasm_f32x4_add(vacc1x0123, wasm_f32x4_mul(wasm_v128_andnot(va1, wasm_f32x4_eq(vb0123c0, vzero)), vb0123c0));
        vacc2x0123 = wasm_f32x4_add(vacc2x0123, wasm_f32x4_mul(wasm_v128_andnot(va2, wasm_f32x4_eq(vb0123c0, vzero)), vb0123c0));
        vacc0x4567 = wasm_f32x4_add(vacc0x4567, wasm_f32x4_mul(wasm_v128_andnot(va0, wasm_f32x4_eq(vb4567c0, vzero)), vb4567c0));
        vacc1x4567 = wasm_f32x4_add(vacc1x4567, wasm_f32x4_mul(wasm_v128_andnot(va1, wasm_f32x4_eq(vb4567c0, vzero)), vb4567c0));
        vacc2x4567 = wasm_f32x4_add(vacc2x4567, wasm_f32x4_mul(wasm_v128_andnot(va2, wasm_f32x4_eq(vb4567c0, vzero)), vb4567c0));

        va0 = wasm_v32x4_shuffle(va0, va0, 1, 2, 3, 0);
        va1 = wasm_v32x4_shuffle(va1, va1, 1, 2, 3, 0);
        va2 = wasm_v32x4_shuffle(va2, va2, 1, 2, 3, 0);

        const v128_t vb0123c1 = wasm_v128_load(w + 8);
        const v128_t vb4567c1 = wasm_v128_load(w + 12);

        vacc0x0123 = wasm_f32x4_add(vacc0x0123, wasm_f32x4_mul(wasm_v128_andnot(va0, wasm_f32x4_eq(vb0123c1, vzero)), vb0123c1));
        vacc1x0123 = wasm_f32x4_add(vacc1x0123, wasm_f32x4_mul(wasm_v128_andnot(va1, wasm_f32x4_eq(vb0123c1, vzero)), vb0123c1));
        vacc2x0123 = wasm_f32x4_add(vacc2x0123, wasm_f32x4_mul(wasm_v128_andnot(va2, wasm_f32x4_eq(vb0123c1, vzero)), vb0123c1));
        vacc0x4567 = wasm_f32x4_add(vacc0x4567, wasm_f32x4_mul(wasm_v128_andnot(va0, wasm_f32x4_eq(vb4567c1, vzero)), vb4567c1));
        vacc1x4567 = wasm_f32x4_add(vacc1x4567, wasm_f32x4_mul(wasm_v128_andnot(va1, wasm_f32x4_eq(vb4567c1, vzero)), vb4567c1));
        vacc2x4567 = wasm_f32x4_add(vacc2x4567, wasm_f32x4_mul(wasm_v128_andnot(va2, wasm_f32x4_eq(vb4567c1, vzero)), vb4567c1));

        va0 = wasm_v32x4_shuffle(va0, va0, 1, 2, 3, 0);
        va1 = wasm_v32x4_shuffle(va1, va1, 1, 2, 3, 0);
        va2 = wasm_v32x4_shuffle(va2, va2, 1, 2, 3, 0);

        const v128_t vb0123c2 = wasm_v128_load(w + 16);
        const v128_t vb4567c2 = wasm_v128_load(w + 20);

        vacc0x0123 = wasm_f32x4_add(vacc0x0123, wasm_f32x4_mul(wasm_v128_andnot(va0, wasm_f32x4_eq(vb0123c2, vzero)), vb0123c2));
        vacc1x0123 = wasm_f32x4_add(vacc1x0123, wasm_f32x4_mul(wasm_v128_andnot(va1, wasm_f32x4_eq(vb0123c2, vzero)), vb0123c2));
        vacc2x0123 = wasm_f32x4_add(vacc2x0123, wasm_f32x4_mul(wasm_v128_andnot(va2, wasm_f32x4_eq(vb0123c2, vzero)), vb0123c2));
        vacc0x4567 = wasm_f32x4_add(vacc0x4567, wasm_f32x4_mul(wasm_v128_andnot(va0, wasm_f32x4_eq(vb4567c2, vzero)), vb4567c2));
        vacc1x4567 = wasm_f32x4_add(vacc1x4567, wasm_f32x4_mul(wasm_v128_andnot(va1, wasm_f32x4_eq(vb4567c2, vzero)), vb4567c2));
        vacc2x4567 = wasm_f32x4_add(vacc2x4567, wasm_f32x4_mul(wasm_v128_andnot(va2, wasm_f32x4_eq(vb4567c2, vzero)), vb4567c2));

        va0 = wasm_v32x4_shuffle(va0, va0, 1, 2, 3, 0);
        va1 = wasm_v32x4_shuffle(va1, va1, 1, 2, 3, 0);
        va2 = wasm_v32x4_shuffle(va2, va2, 1, 2, 3, 0);

        const v128_t vb0123c3 = wasm_v128_load(w + 24);
        const v128_t vb4567c3 = wasm_v128_load(w + 28);

        vacc0x0123 = wasm_f32x4_add(vacc0x0123, wasm_f32x4_mul(wasm_v128_andnot(va0, wasm_f32x4_eq(vb0123c3, vzero)), vb0123c3));
        vacc1x0123 = wasm_f32x4_add(vacc1x0123, wasm_f32x4_mul(wasm_v128_andnot(va1, wasm_f32x4_eq(vb0123c3, vzero)), vb0123c3));
        vacc2x0123 = wasm_f32x4_add(vacc2x0123, wasm_f32x4_mul(wasm_v128_andnot(va2, wasm_f32x4_eq(vb0123c3, vzero)), vb0123c3));
        vacc0x4567 = wasm_f32x4_add(vacc0x4567, wasm_f32x4_mul(wasm_v128_andnot(va0, wasm_f32x4_eq(vb4567c3, vzero)), vb4567c3));
        vacc1x4567 = wasm_f32x4_add(vacc1x4567, wasm_f32x4_mul(wasm_v128_andnot(va1, wasm_f32x4_eq(vb4567c3, vzero)), vb4567c3));
        vacc2x4567 = wasm_f32x4_add(vacc2x4567, wasm_f32x4_mul(wasm_v128_andnot(va2, wasm_f32x4_eq(vb4567c3, vzero)), vb4567c3));


        w += 32;
      }
      p -= 3 * sizeof(void*);
    } while (p != 0);

    vacc0x0123 = __builtin_wasm_relaxed_max_f32x4(vmin, vacc0x0123);
    vacc1x0123 = __builtin_wasm_relaxed_max_f32x4(vmin, vacc1x0123);
    vacc2x0123 = __builtin_wasm_relaxed_max_f32x4(vmin, vacc2x0123);
    vacc0x4567 = __builtin_wasm_relaxed_max_f32x4(vmin, vacc0x4567);
    vacc1x4567 = __builtin_wasm_relaxed_max_f32x4(vmin, vacc1x4567);
    vacc2x4567 = __builtin_wasm_relaxed_max_f32x4(vmin, vacc2x4567);

    vacc0x0123 = __builtin_wasm_relaxed_min_f32x4(vmax, vacc0x0123);
    vacc1x0123 = __builtin_wasm_relaxed_min_f32x4(vmax, vacc1x0123);
    vacc2x0123 = __builtin_wasm_relaxed_min_f32x4(vmax, vacc2x0123);
    vacc0x4567 = __builtin_wasm_relaxed_min_f32x4(vmax, vacc0x4567);
    vacc1x4567 = __builtin_wasm_relaxed_min_f32x4(vmax, vacc1x4567);
    vacc2x4567 = __builtin_wasm_relaxed_min_f32x4(vmax, vacc2x4567);

    if XNN_LIKELY(nc >= 8) {
      wasm_v128_store(c2, vacc2x0123);
      wasm_v128_store(c2 + 4, vacc2x4567);
      c2 = (float*) ((uintptr_t) c2 + cn_stride);
      wasm_v128_store(c1, vacc1x0123);
      wasm_v128_store(c1 + 4, vacc1x4567);
      c1 = (float*) ((uintptr_t) c1 + cn_stride);
      wasm_v128_store(c0, vacc0x0123);
      wasm_v128_store(c0 + 4, vacc0x4567);
      c0 = (float*) ((uintptr_t) c0 + cn_stride);

      a = (const float**restrict) ((uintptr_t) a - ks);
      nc -= 8;
    } else {
      if (nc & 4) {
        wasm_v128_store(c2, vacc2x0123);
        wasm_v128_store(c1, vacc1x0123);
        wasm_v128_store(c0, vacc0x0123);

        vacc2x0123 = vacc2x4567;
        vacc1x0123 = vacc1x4567;
        vacc0x0123 = vacc0x4567;

        c2 += 4;
        c1 += 4;
        c0 += 4;
      }
      if (nc & 2) {
        *((double*) c2) = wasm_f64x2_extract_lane(vacc2x0123, 0);
        *((double*) c1) = wasm_f64x2_extract_lane(vacc1x0123, 0);
        *((double*) c0) = wasm_f64x2_extract_lane(vacc0x0123, 0);

        vacc2x0123 = wasm_v32x4_shuffle(vacc2x0123, vacc2x0123, 2, 3, 2, 3);
        vacc1x0123 = wasm_v32x4_shuffle(vacc1x0123, vacc1x0123, 2, 3, 2, 3);
        vacc0x0123 = wasm_v32x4_shuffle(vacc0x0123, vacc0x0123, 2, 3, 2, 3);

        c2 += 2;
        c1 += 2;
        c0 += 2;
      }
      if (nc & 1) {
        *c2 = wasm_f32x4_extract_lane(vacc2x0123, 0);
        *c1 = wasm_f32x4_extract_lane(vacc1x0123, 0);
        *c0 = wasm_f32x4_extract_lane(vacc0x0123, 0);
      }

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