xref: /aosp_15_r20/external/XNNPACK/src/f32-igemm/gen/1x8s4-relu-wasmrelaxedsimd-fma.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_relu_ukernel_1x8s4__wasmrelaxedsimd_fma(
    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_relu_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS
{
  assert(mr != 0);
  assert(mr <= 1);
  assert(nc != 0);
  assert(kc != 0);
  assert(kc % sizeof(float) == 0);
  assert(ks != 0);
  assert(ks % (1 * sizeof(void*)) == 0);
  assert(a_offset % sizeof(float) == 0);
  assert(a != NULL);
  assert(w != NULL);
  assert(c != NULL);

  float* c0 = c;

  do {
    v128_t vacc0x0123 = wasm_v128_load(w);
    v128_t vacc0x4567 = wasm_v128_load(w + 4);
    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);
      }
      a += 1;

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


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

        vacc0x0123 = __builtin_wasm_fma_f32x4(vacc0x0123, va0, vb0123c0);
        vacc0x4567 = __builtin_wasm_fma_f32x4(vacc0x4567, va0, vb4567c0);

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

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

        vacc0x0123 = __builtin_wasm_fma_f32x4(vacc0x0123, va0, vb0123c1);
        vacc0x4567 = __builtin_wasm_fma_f32x4(vacc0x4567, va0, vb4567c1);

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

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

        vacc0x0123 = __builtin_wasm_fma_f32x4(vacc0x0123, va0, vb0123c2);
        vacc0x4567 = __builtin_wasm_fma_f32x4(vacc0x4567, va0, vb4567c2);

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

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

        vacc0x0123 = __builtin_wasm_fma_f32x4(vacc0x0123, va0, vb0123c3);
        vacc0x4567 = __builtin_wasm_fma_f32x4(vacc0x4567, va0, 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);

        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 = __builtin_wasm_fma_f32x4(vacc0x0123, wasm_v128_andnot(va0, wasm_f32x4_eq(vb0123c0, vzero)), vb0123c0);
        vacc0x4567 = __builtin_wasm_fma_f32x4(vacc0x4567, wasm_v128_andnot(va0, wasm_f32x4_eq(vb4567c0, vzero)), vb4567c0);

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

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

        vacc0x0123 = __builtin_wasm_fma_f32x4(vacc0x0123, wasm_v128_andnot(va0, wasm_f32x4_eq(vb0123c1, vzero)), vb0123c1);
        vacc0x4567 = __builtin_wasm_fma_f32x4(vacc0x4567, wasm_v128_andnot(va0, wasm_f32x4_eq(vb4567c1, vzero)), vb4567c1);

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

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

        vacc0x0123 = __builtin_wasm_fma_f32x4(vacc0x0123, wasm_v128_andnot(va0, wasm_f32x4_eq(vb0123c2, vzero)), vb0123c2);
        vacc0x4567 = __builtin_wasm_fma_f32x4(vacc0x4567, wasm_v128_andnot(va0, wasm_f32x4_eq(vb4567c2, vzero)), vb4567c2);

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

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

        vacc0x0123 = __builtin_wasm_fma_f32x4(vacc0x0123, wasm_v128_andnot(va0, wasm_f32x4_eq(vb0123c3, vzero)), vb0123c3);
        vacc0x4567 = __builtin_wasm_fma_f32x4(vacc0x4567, wasm_v128_andnot(va0, wasm_f32x4_eq(vb4567c3, vzero)), vb4567c3);


        w += 32;
      }
      p -= 1 * sizeof(void*);
    } while (p != 0);

    const v128_t vzero = wasm_i32x4_const_splat(0);
    vacc0x0123 = wasm_i32x4_max(vacc0x0123, vzero);
    vacc0x4567 = wasm_i32x4_max(vacc0x4567, vzero);

    if XNN_LIKELY(nc >= 8) {
      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(c0, vacc0x0123);

        vacc0x0123 = vacc0x4567;

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

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

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

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