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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268
|
#pragma once
#include <fbgemm/FbgemmConvert.h>
#include "caffe2/core/context.h"
#include "caffe2/core/operator.h"
#include "caffe2/perfkernels/typed_axpy.h"
C10_DECLARE_bool(caffe2_fbgemm_fake_fp16_clamp);
C10_DECLARE_bool(caffe2_fbgemm_fake_fp16_clamp_denorms);
namespace caffe2 {
// A templated class that implements SparseLengths[Sum,WeightedSum,Mean].
template <
class InputTypes, // supported input types, such as TensorTypes<float>
bool USE_WEIGHT = 0, // Whether it is SparseLengthsWeightedSum
bool USE_MEAN = 0, // Whether this is SparseLengthsMean
bool USE_POSITIONAL_WEIGHT = 0,
bool USE_ACC_FP16 = 0, // Whether use fp16 accumulation
bool USE_FP16_FOR_EMBEDDING_ONLY =
0 // Whether use fp16 for embedding entries only
// USE_WEIGHT = 1 and USE_POSITIONAL_WEIGHT = 1
// -> SparseLengthsPositionalWeightedSum
>
class SparseLengthsReductionFakeFp16Op final : public Operator<CPUContext> {
public:
USE_OPERATOR_FUNCTIONS(CPUContext);
template <class... Args>
explicit SparseLengthsReductionFakeFp16Op(Args&&... args)
: Operator<CPUContext>(std::forward<Args>(args)...) {
static_assert(
!(USE_WEIGHT & USE_MEAN), "Cannot both specify weight and mean.");
}
~SparseLengthsReductionFakeFp16Op() noexcept override {}
// Currently, we support float and at::Half inputs for input data type, and
// int32_t and int64_t for the index type.
bool RunOnDevice() override {
return DispatchHelper<InputTypes>::call(this, Input(DATA));
}
template <typename InputType>
bool DoRunWithType() {
return DispatchHelper<TensorTypes2<int32_t, int64_t>, InputType>::call(
this, Input(INDICES));
}
template <typename InputType, typename IndexType>
bool DoRunWithType2() {
auto& dataInput = Input(DATA);
auto& indicesInput = Input(INDICES);
auto& lengthsInput = Input(LENGTHS);
CAFFE_ENFORCE_EQ(1, indicesInput.dim(), "INDICES must be a vector");
CAFFE_ENFORCE_EQ(1, lengthsInput.dim(), "LENGTHS must be a vector");
const int64_t N = dataInput.size(0);
const int D = dataInput.size_from_dim(1);
const int64_t M = lengthsInput.size(0);
const int64_t indices_size = indicesInput.numel();
auto shape = dataInput.sizes().vec();
shape[0] = M;
auto* output = Output(0, shape, at::dtype<float>());
float* out_data = output->template mutable_data<float>();
const InputType* in_data = dataInput.template data<InputType>();
const IndexType* indices = indicesInput.template data<IndexType>();
const int* lengths = lengthsInput.template data<int>();
const float* in_weight = nullptr;
if (USE_WEIGHT) {
// static if
auto& weightInput = Input(WEIGHT);
CAFFE_ENFORCE_EQ(1, weightInput.dim(), "WEIGHT must be a vector");
if (!USE_POSITIONAL_WEIGHT) {
CAFFE_ENFORCE_EQ(
weightInput.numel(),
indices_size,
"Weight should have the same length as indices.");
}
in_weight = weightInput.template data<float>();
}
// Copied from EmbeddingLookupGenericSlow in perfkernels/embedding_lookup.cc
int64_t block_size = D;
int64_t output_size = M;
int64_t index_size = indices_size;
int64_t data_size = N;
const InputType* input = in_data;
const float* weights = in_weight;
bool normalize_by_lengths = USE_MEAN;
float* out = out_data;
int64_t current = 0;
for (const auto m : c10::irange(output_size)) {
memset(out, 0, sizeof(float) * block_size);
if (current + lengths[m] > index_size) {
return false;
}
for (int i = 0; i < lengths[m]; ++i) {
int64_t idx = indices[current];
if (idx < 0 || idx >= data_size) {
return false;
}
float w = 1.f;
if (weights) {
w = weights[USE_POSITIONAL_WEIGHT ? i : current];
if (!USE_FP16_FOR_EMBEDDING_ONLY) {
// Fake fp16 rounding of w
fbgemm::RoundToFloat16(
&w, &w, 1, FLAGS_caffe2_fbgemm_fake_fp16_clamp);
}
}
if (USE_FP16_FOR_EMBEDDING_ONLY) {
std::vector<float> product_rounded(block_size);
if (std::is_same<InputType, at::Half>::value) {
TypedAxpy<InputType, float>(
block_size,
w,
input + block_size * indices[current],
product_rounded.data());
} else {
bool is_float = std::is_same<InputType, float>::value;
assert(is_float);
// Fake fp16 rounding of input
std::vector<float> input_rounded(block_size);
fbgemm::RoundToFloat16(
reinterpret_cast<const float*>(
input + block_size * indices[current]),
input_rounded.data(),
block_size,
FLAGS_caffe2_fbgemm_fake_fp16_clamp,
FLAGS_caffe2_fbgemm_fake_fp16_clamp_denorms);
TypedAxpy<float, float>(
block_size,
w,
reinterpret_cast<const float*>(input_rounded.data()),
product_rounded.data());
}
// Accumulate w x input to output
TypedAxpy<float, float>(
block_size,
1.0,
reinterpret_cast<const float*>(product_rounded.data()),
out);
} else if (USE_ACC_FP16) {
std::vector<float> product_rounded(block_size);
if (std::is_same<InputType, at::Half>::value) {
TypedAxpy<InputType, float>(
block_size,
w,
input + block_size * indices[current],
product_rounded.data());
} else {
bool is_float = std::is_same<InputType, float>::value;
assert(is_float);
// Fake fp16 rounding of input
std::vector<float> input_rounded(block_size);
fbgemm::RoundToFloat16(
reinterpret_cast<const float*>(
input + block_size * indices[current]),
input_rounded.data(),
block_size,
FLAGS_caffe2_fbgemm_fake_fp16_clamp);
TypedAxpy<float, float>(
block_size,
w,
reinterpret_cast<const float*>(input_rounded.data()),
product_rounded.data());
}
// Fake fp16 rounding of w x input
fbgemm::RoundToFloat16(
reinterpret_cast<const float*>(product_rounded.data()),
product_rounded.data(),
block_size,
FLAGS_caffe2_fbgemm_fake_fp16_clamp);
// Accumulate w x input to output
TypedAxpy<float, float>(
block_size,
1.0,
reinterpret_cast<const float*>(product_rounded.data()),
out);
// Fake fp16 rounding of out + w x input
fbgemm::RoundToFloat16(
reinterpret_cast<const float*>(out),
out,
block_size,
FLAGS_caffe2_fbgemm_fake_fp16_clamp);
} else {
if (std::is_same<InputType, at::Half>::value) {
TypedAxpy<InputType, float>(
block_size, w, input + block_size * indices[current], out);
} else {
bool is_float = std::is_same<InputType, float>::value;
assert(is_float);
// Fake fp16 rounding of input
std::vector<float> input_rounded(block_size);
fbgemm::RoundToFloat16(
reinterpret_cast<const float*>(
input + block_size * indices[current]),
input_rounded.data(),
block_size,
FLAGS_caffe2_fbgemm_fake_fp16_clamp);
TypedAxpy<float, float>(
block_size,
w,
reinterpret_cast<const float*>(input_rounded.data()),
out);
}
}
++current;
}
if (normalize_by_lengths && lengths[m]) {
float scale = 1.f / lengths[m];
if (!USE_FP16_FOR_EMBEDDING_ONLY) {
// Fake fp16 rounding of scale and out
fbgemm::RoundToFloat16(
&scale, &scale, 1, FLAGS_caffe2_fbgemm_fake_fp16_clamp);
fbgemm::RoundToFloat16(
reinterpret_cast<const float*>(out),
out,
block_size,
FLAGS_caffe2_fbgemm_fake_fp16_clamp);
}
// hack: context is not really used
math::Scale<float, float, CPUContext>(
block_size, scale, out, out, nullptr);
}
if (!USE_FP16_FOR_EMBEDDING_ONLY) {
// Fake fp16 rounding of out
fbgemm::RoundToFloat16(
reinterpret_cast<const float*>(out),
reinterpret_cast<float*>(out),
block_size,
FLAGS_caffe2_fbgemm_fake_fp16_clamp);
}
out += block_size;
}
return current == index_size;
}
enum {
DATA = 0, // Data input.
WEIGHT = 1, // Weight input used in SparseLengthsWeightedSum
INDICES = 1 + USE_WEIGHT, // 1 in SparseLengths[Sum,Mean] and
// 2 in SparseLengthsWeightedSum
LENGTHS = 2 + USE_WEIGHT, // 2 in SparseLengths[Sum, Mean],
// 3 in SparseLengthsWeightedSum
};
};
} // namespace caffe2
|