File size: 1,628 Bytes
8b7c501 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 |
// Auto-generated file. Do not edit!
// Template: src/f16-vbinary/vopc-fp16arith.c.in
// Generator: tools/xngen
//
// Copyright 2022 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 <arm_fp16.h>
#include <xnnpack/common.h>
#include <xnnpack/intrinsics-polyfill.h>
#include <xnnpack/math.h>
#include <xnnpack/vbinary.h>
void xnn_f16_vmaxc_ukernel__fp16arith_x4(
size_t batch,
const void* restrict input_a,
const void* restrict input_b,
void* restrict output,
const union xnn_f16_default_params params[restrict XNN_MIN_ELEMENTS(1)])
{
assert(batch != 0);
assert(batch % sizeof(float16_t) == 0);
assert(input_a != NULL);
assert(input_b != NULL);
assert(output != NULL);
const float16_t* a = (const float16_t*) input_a;
const float16_t* b = (const float16_t*) input_b;
float16_t* o = (float16_t*) output;
const float16_t vb = *b;
for (; batch >= 4 * sizeof(float16_t); batch -= 4 * sizeof(float16_t)) {
float16_t vacc0 = a[0];
float16_t vacc1 = a[1];
float16_t vacc2 = a[2];
float16_t vacc3 = a[3];
a += 4;
vacc0 = vmaxnmh_f16(vacc0, vb);
vacc1 = vmaxnmh_f16(vacc1, vb);
vacc2 = vmaxnmh_f16(vacc2, vb);
vacc3 = vmaxnmh_f16(vacc3, vb);
o[0] = vacc0;
o[1] = vacc1;
o[2] = vacc2;
o[3] = vacc3;
o += 4;
}
if XNN_UNLIKELY(batch != 0) {
do {
float16_t vacc = *a++;
vacc = vmaxnmh_f16(vacc, vb);
*o++ = vacc;
batch -= sizeof(float16_t);
} while (batch != 0);
}
}
|