File size: 5,861 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 |
// 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.
$assert ROW_TILE >= 1
$assert ACCUMULATORS >= 1
#include <assert.h>
#include <xnnpack/dwconv.h>
#include <xnnpack/math.h>
void xnn_f32_dwconv2d_chw_ukernel_3x3s2p1__scalar_${ROW_TILE}x1${"_acc%d" % ACCUMULATORS if ACCUMULATORS > 1 else ""}(
size_t input_height,
size_t input_width,
const float* input,
const float* weights,
const float* zero,
float* output,
uint32_t padding_top,
const union xnn_f32_chw_params params[restrict XNN_MIN_ELEMENTS(1)])
{
assert(input_height != 0);
assert(input_width != 0);
assert(input_width % sizeof(float) == 0);
assert(padding_top >= 0);
assert(padding_top <= 1);
const float vmin = params->scalar.min;
const float vmax = params->scalar.max;
const float vbias = weights[0];
const float vk00 = weights[1];
const float vk01 = weights[2];
const float vk02 = weights[3];
const float vk10 = weights[4];
const float vk11 = weights[5];
const float vk12 = weights[6];
const float vk20 = weights[7];
const float vk21 = weights[8];
const float vk22 = weights[9];
$if ROW_TILE > 1:
const size_t output_width = round_down_po2((input_width + (2 /* padding */ - 3 /* kernel size */ + 2 /* subsampling */) * sizeof(float)) / 2, sizeof(float));
const float* i0 = (const float*) ((uintptr_t) input - ((-padding_top) & input_width));
const float* i1 = (const float*) ((uintptr_t) i0 + input_width);
if XNN_UNPREDICTABLE(padding_top != 0) {
i0 = zero;
}
$for M in range(2, 1 + 2 * ROW_TILE):
const float* i${M} = (const float*) ((uintptr_t) i${M-1} + input_width);
float* o0 = output;
$for M in range(1, ROW_TILE):
float* o${M} = (float*) ((uintptr_t) o${M-1} + output_width);
size_t padded_input_height = input_height + padding_top + 1 /* padding bottom */;
size_t output_height = (padded_input_height - 3 /* kernel size */ + 2 /* subsampling */) / 2;
do {
$for M in range(2, 1 + 2 * ROW_TILE):
if XNN_UNPREDICTABLE(padded_input_height < ${2 + M}) {
i${M} = zero;
$if M % 2 == 1:
o${(M - 1) // 2} = o${(M - 1) // 2 - 1};
}
$for M in range(1 + 2 * ROW_TILE):
float vi${M}x0 = 0.0f;
size_t w = input_width;
for (; w >= 2 * sizeof(float); w -= 2 * sizeof(float)) {
$for M in range(1 + 2 * ROW_TILE):
const float vi${M}x1 = i${M}[0];
$for K in range(3):
$for M in range(ROW_TILE):
$if K == 0:
float vo${M}p0 = vbias + vi${2*M+K}x0 * vk${K}0;
$elif K < ACCUMULATORS:
float vo${M}p${K} = vi${2*M+K}x0 * vk${K}0;
$else:
vo${M}p${K % ACCUMULATORS} += vi${2*M+K}x0 * vk${K}0;
$for M in range(1 + 2 * ROW_TILE):
const float vi${M}x2 = i${M}[1];
i${M} += 2;
$for K in range(3):
$for M in range(ROW_TILE):
$if K + 3 < ACCUMULATORS:
float vo${M}p${K+3} = vi${2*M+K}x1 * vk${K}1;
$else:
vo${M}p${(K+3) % ACCUMULATORS} += vi${2*M+K}x1 * vk${K}1;
$for M in range(1 + 2 * ROW_TILE):
vi${M}x0 = vi${M}x2;
$for K in range(3):
$for M in range(ROW_TILE):
vo${M}p${(K+6) % ACCUMULATORS} += vi${2*M+K}x2 * vk${K}2;
$if ACCUMULATORS > 1:
$ACC_SLICE = 1
$while ACC_SLICE < ACCUMULATORS:
$for A in range(0, ACCUMULATORS, ACC_SLICE * 2):
$if A + ACC_SLICE < ACCUMULATORS:
$for M in range(ROW_TILE):
vo${M}p${A} += vo${M}p${A + ACC_SLICE};
$ACC_SLICE *= 2
$for M in range(ROW_TILE):
float vo${M} = math_max_f32(vo${M}p0, vmin);
$for M in range(ROW_TILE):
vo${M} = math_min_f32(vo${M}, vmax);
$for M in reversed(range(ROW_TILE)):
*o${M}++ = vo${M};
}
// Potentially process the last pixel.
assert(w <= 1 * sizeof(float));
if (w != 0) {
$for M in range(1 + 2 * ROW_TILE):
const float vi${M}x1 = *i${M}++;
$for K in range(3):
$for M in range(ROW_TILE):
$if K == 0:
float vo${M}p0 = vbias + vi${2*M+K}x0 * vk${K}0;
$elif K < ACCUMULATORS:
float vo${M}p${K} = vi${2*M+K}x0 * vk${K}0;
$else:
vo${M}p${K % ACCUMULATORS} += vi${2*M+K}x0 * vk${K}0;
$for K in range(3):
$for M in range(ROW_TILE):
$if K + 3 < ACCUMULATORS:
float vo${M}p${K+3} = vi${2*M+K}x1 * vk${K}1;
$else:
vo${M}p${(K+3) % ACCUMULATORS} += vi${2*M+K}x1 * vk${K}1;
$if ACCUMULATORS > 1:
$ACC_SLICE = 1
$while ACC_SLICE < ACCUMULATORS:
$for A in range(0, ACCUMULATORS, ACC_SLICE * 2):
$if A + ACC_SLICE < ACCUMULATORS:
$for M in range(ROW_TILE):
vo${M}p${A} += vo${M}p${A + ACC_SLICE};
$ACC_SLICE *= 2
$for M in range(ROW_TILE):
float vo${M} = math_max_f32(vo${M}p0, vmin);
$for M in range(ROW_TILE):
vo${M} = math_min_f32(vo${M}, vmax);
$for M in reversed(range(ROW_TILE)):
*o${M}++ = vo${M};
}
i0 = (const float*) ((uintptr_t) i${2 * ROW_TILE - 1});
i1 = (const float*) ((uintptr_t) i${2 * ROW_TILE});
$for M in range(2, 1 + 2 * ROW_TILE):
i${M} = (const float*) ((uintptr_t) i${M-1} + input_width);
$if ROW_TILE > 1:
o0 = o${ROW_TILE - 1};
$for M in range(1, ROW_TILE):
o${M} = (float*) ((uintptr_t) o${M-1} + output_width);
$if ROW_TILE > 1:
output_height = doz(output_height, ${ROW_TILE});
padded_input_height = doz(padded_input_height, ${ROW_TILE * 2});
$else:
output_height -= 1;
padded_input_height -= 2;
} while (output_height != 0);
}
|