File size: 2,468 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 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 |
// Auto-generated file. Do not edit!
// Template: src/f16-gemm/avx2-broadcast.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 <immintrin.h>
#include <xnnpack/gemm.h>
#include <xnnpack/intrinsics-polyfill.h>
void xnn_f16_f32acc_gemm_minmax_ukernel_1x8__avx2_broadcast(
size_t mr,
size_t nc,
size_t kc,
const void* restrict a,
size_t a_stride,
const void* restrict w,
void* restrict c,
size_t cm_stride,
size_t cn_stride,
const union xnn_f16_minmax_params params[restrict XNN_MIN_ELEMENTS(1)])
{
assert(mr != 0);
assert(mr <= 1);
assert(nc != 0);
assert(kc != 0);
assert(kc % sizeof(uint16_t) == 0);
assert(a != NULL);
assert(w != NULL);
assert(c != NULL);
const uint16_t* a0 = a;
uint16_t* c0 = c;
do {
__m256 vacc0x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) w));
w = (const uint16_t*) w + 8;
size_t k = kc;
do {
const __m256 va0 = _mm256_cvtph_ps(_mm_set1_epi16((short) *a0));
a0 += 1;
const __m256 vb01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) w));
w = (const uint16_t*) w + 8;
vacc0x01234567 = _mm256_fmadd_ps(va0, vb01234567, vacc0x01234567);
k -= sizeof(uint16_t);
} while (k != 0);
const __m256 vmin = _mm256_load_ps(params->avx.min);
vacc0x01234567 = _mm256_max_ps(vacc0x01234567, vmin);
const __m256 vmax = _mm256_load_ps(params->avx.max);
vacc0x01234567 = _mm256_min_ps(vacc0x01234567, vmax);
if XNN_LIKELY(nc >= 8) {
_mm_storeu_si128((__m128i*) c0, _mm256_cvtps_ph(vacc0x01234567, _MM_FROUND_TO_NEAREST_INT));
c0 = (uint16_t*) ((uintptr_t) c0 + cn_stride);
a0 = (const uint16_t*) ((uintptr_t) a0 - kc);
nc -= 8;
} else {
__m128i vh0x01234567 = _mm256_cvtps_ph(vacc0x01234567, _MM_FROUND_TO_NEAREST_INT);
if (nc & 4) {
_mm_storel_epi64((__m128i*) c0, vh0x01234567);
vh0x01234567 = _mm_unpackhi_epi64(vh0x01234567, vh0x01234567);
c0 += 4;
}
if (nc & 2) {
_mm_storeu_si32(c0, vh0x01234567);
vh0x01234567 = _mm_srli_epi64(vh0x01234567, 32);
c0 += 2;
}
if (nc & 1) {
*c0 = (uint16_t) _mm_extract_epi16(vh0x01234567, 0);
}
nc = 0;
}
} while (nc != 0);
}
|