File size: 6,251 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 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 |
// 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.
$assert NR % 8 == 0
$ABC = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ"
$assert ACCTYPE in ["F16", "F32"]
$ACC_SUFFIX = "_f32acc" if ACCTYPE == "F32" else ""
#include <assert.h>
#include <immintrin.h>
#include <xnnpack/igemm.h>
#include <xnnpack/intrinsics-polyfill.h>
void xnn_f16${ACC_SUFFIX}_igemm_minmax_ukernel_${MR}x${NR}__avx2_broadcast(
size_t mr,
size_t nc,
size_t kc,
size_t ks,
const void** restrict a,
const void* restrict w,
void* restrict c,
size_t cm_stride,
size_t cn_stride,
size_t a_offset,
const void* zero,
const union xnn_f16_minmax_params params[restrict XNN_MIN_ELEMENTS(1)])
{
assert(mr != 0);
assert(mr <= ${MR});
assert(nc != 0);
assert(kc != 0);
assert(kc % sizeof(uint16_t) == 0);
assert(ks != 0);
assert(ks % (${MR} * sizeof(void*)) == 0);
assert(a_offset % sizeof(uint16_t) == 0);
assert(a != NULL);
assert(w != NULL);
assert(c != NULL);
uint16_t* c0 = c;
$for M in range(1, MR):
uint16_t* c${M} = (uint16_t*) ((uintptr_t) c${M-1} + cm_stride);
$if M % 2 == 0:
if XNN_UNPREDICTABLE(mr <= ${M}) {
c${M} = c${M-1};
}
$elif M + 1 == MR:
if XNN_UNPREDICTABLE(mr != ${M+1}) {
c${M} = c${M-1};
}
$else:
if XNN_UNPREDICTABLE(mr < ${M+1}) {
c${M} = c${M-1};
}
do {
__m256 vacc0x${ABC[0:8]} = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) w));
$for N in range(8, NR, 8):
__m256 vacc0x${ABC[N:N+8]} = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) ((const uint16_t*) w + ${N})));
$for M in range(1, MR):
$for N in range(0, NR, 8):
__m256 vacc${M}x${ABC[N:N+8]} = vacc0x${ABC[N:N+8]};
w = (const uint16_t*) w + ${NR};
size_t p = ks;
do {
$for M in range(MR):
const uint16_t* restrict a${M} = (const uint16_t*) a[${M}];
assert(a${M} != NULL);
if XNN_UNPREDICTABLE(a${M} != zero) {
a${M} = (const uint16_t*) ((uintptr_t) a${M} + a_offset);
}
a += ${MR};
size_t k = kc;
do {
const __m256 vb${ABC[0:8]} = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) w));
$for N in range(8, NR, 8):
const __m256 vb${ABC[N:N+8]} = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) ((const uint16_t*) w + ${N})));
w = (const uint16_t*) w + ${NR};
$for M in range(MR):
const __m256 va${M} = _mm256_cvtph_ps(_mm_set1_epi16((short) *a${M}));
a${M} += 1;
$for M in range(MR):
$for N in range(0, NR, 8):
$if ACCTYPE == "F32":
vacc${M}x${ABC[N:N+8]} = _mm256_fmadd_ps(va${M}, vb${ABC[N:N+8]}, vacc${M}x${ABC[N:N+8]});
$else:
vacc${M}x${ABC[N:N+8]} = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(va${M}, vb${ABC[N:N+8]}, vacc${M}x${ABC[N:N+8]}), _MM_FROUND_TO_NEAREST_INT));
k -= sizeof(uint16_t);
} while (k != 0);
p -= ${MR} * sizeof(void*);
} while (p != 0);
const __m256 vmin = _mm256_load_ps(params->avx.min);
$for N in range(0, NR, 8):
$for M in range(MR):
vacc${M}x${ABC[N:N+8]} = _mm256_max_ps(vacc${M}x${ABC[N:N+8]}, vmin);
const __m256 vmax = _mm256_load_ps(params->avx.max);
$for N in range(0, NR, 8):
$for M in range(MR):
vacc${M}x${ABC[N:N+8]} = _mm256_min_ps(vacc${M}x${ABC[N:N+8]}, vmax);
if XNN_LIKELY(nc >= ${NR}) {
$for M in reversed(range(MR)):
_mm_storeu_si128((__m128i*) c${M}, _mm256_cvtps_ph(vacc${M}x${ABC[0:8]}, _MM_FROUND_TO_NEAREST_INT));
$for N in range(8, NR, 8):
_mm_storeu_si128((__m128i*) (c${M} + ${N}), _mm256_cvtps_ph(vacc${M}x${ABC[N:N+8]}, _MM_FROUND_TO_NEAREST_INT));
c${M} = (uint16_t*) ((uintptr_t) c${M} + cn_stride);
a = (const void**restrict) ((uintptr_t) a - ks);
nc -= ${NR};
} else {
$for LOG2N in reversed(range(NR.bit_length())):
$if LOG2N == 3:
$for M in reversed(range(MR)):
__m128i vh${M}x${ABC[0:8]} = _mm256_cvtps_ph(vacc${M}x${ABC[0:8]}, _MM_FROUND_TO_NEAREST_INT);
$if NR != 1 << LOG2N:
if (nc & ${1 << LOG2N}) {
$if LOG2N >= 4:
$for M in reversed(range(MR)):
_mm_storeu_si128((__m128i*) c${M}, _mm256_cvtps_ph(vacc${M}x${ABC[0:8]}, _MM_FROUND_TO_NEAREST_INT));
$for N in range(8, 1 << LOG2N, 8):
_mm_storeu_si128((__m128i*) (c${M} + ${N}), _mm256_cvtps_ph(vacc${M}x${ABC[N:N+8]}, _MM_FROUND_TO_NEAREST_INT));
$for M in reversed(range(MR)):
$for N in range(0, NR - (1 << LOG2N), 8):
vacc${M}x${ABC[N:N+8]} = vacc${M}x${ABC[N + (1 << LOG2N):N + (1 << LOG2N)+8]};
$for M in reversed(range(MR)):
c${M} += ${1 << LOG2N};
$elif LOG2N == 3:
$for M in reversed(range(MR)):
_mm_storeu_si128((__m128i*) c${M}, vh${M}x${ABC[0:8]});
$for M in reversed(range(MR)):
vh${M}x${ABC[0:8]} = _mm256_cvtps_ph(vacc${M}x${ABC[8:16]}, _MM_FROUND_TO_NEAREST_INT);
$for M in reversed(range(MR)):
c${M} += ${1 << LOG2N};
$elif LOG2N == 2:
$for M in reversed(range(MR)):
_mm_storel_epi64((__m128i*) c${M}, vh${M}x${ABC[0:8]});
$for M in reversed(range(MR)):
vh${M}x${ABC[0:8]} = _mm_unpackhi_epi64(vh${M}x${ABC[0:8]}, vh${M}x${ABC[0:8]});
$for M in reversed(range(MR)):
c${M} += 4;
$elif LOG2N == 1:
$for M in reversed(range(MR)):
_mm_storeu_si32(c${M}, vh${M}x${ABC[0:8]});
$for M in reversed(range(MR)):
vh${M}x${ABC[0:8]} = _mm_srli_epi64(vh${M}x${ABC[0:8]}, 32);
$for M in reversed(range(MR)):
c${M} += 2;
$elif LOG2N == 0:
$for M in reversed(range(MR)):
*c${M} = _mm_extract_epi16(vh${M}x${ABC[0:8]}, 0);
}
nc = 0;
}
} while (nc != 0);
}
|