File size: 2,071 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 69 70 71 72 73 74 75 |
// Auto-generated file. Do not edit!
// Template: src/f32-f16-vcvt/scalar-fabsf.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 <math.h>
#include <xnnpack/common.h>
#include <xnnpack/math.h>
#include <xnnpack/vcvt.h>
void xnn_f32_f16_vcvt_ukernel__scalar_fabsf_x1(
size_t batch,
const float* input,
void* output,
const union xnn_f32_f16_cvt_params params[restrict XNN_MIN_ELEMENTS(1)])
{
assert(batch != 0);
assert(batch % sizeof(float) == 0);
assert(input != NULL);
assert(output != NULL);
const float vscale_to_inf = params->scalar_fabsf.scale_to_inf;
const uint32_t vexp_bias = params->scalar_fabsf.exp_bias;
const float vscale_to_zero = params->scalar_fabsf.scale_to_zero;
const uint32_t vexpw_max = params->scalar_fabsf.expw_max;
const uint32_t vbias_min = params->scalar_fabsf.bias_min;
const uint16_t vexph_mask = params->scalar_fabsf.exph_mask;
const uint16_t vmanth_mask = params->scalar_fabsf.manth_mask;
const uint16_t vnanh = params->scalar_fabsf.nanh;
uint16_t* o = (uint16_t*) output;
do {
const float vx = *input++;
const float vabsx = fabsf(vx);
uint32_t vsignw = float_as_uint32(vx);
const uint32_t vnonsignw = float_as_uint32(vabsx);
float vf = vabsx * vscale_to_inf;
uint32_t vbias = vnonsignw + vexp_bias;
vsignw ^= vnonsignw;
vf *= vscale_to_zero;
vbias &= vexpw_max;
vbias = math_max_u32(vbias, vbias_min);
vf += uint32_as_float(vbias);
const uint32_t vbits = float_as_uint32(vf);
const uint16_t vexph = (uint16_t) (vbits >> 13) & vexph_mask;
const uint16_t vmanth = (uint16_t) vbits & vmanth_mask;
const uint16_t vsignh = (uint16_t) (vsignw >> 16);
uint16_t vh = vexph + vmanth;
if XNN_UNPREDICTABLE(vnonsignw > vexpw_max) {
vh = vnanh;
}
vh |= vsignh;
*o++ = vh;
batch -= sizeof(float);
} while (batch != 0);
}
|