wehub-resource-sync
226 lines
7.7 KiB
C++
226 lines
7.7 KiB
C++
// Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
// you may not use this file except in compliance with the License.
|
|
// You may obtain a copy of the License at
|
|
//
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
//
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
// See the License for the specific language governing permissions and
|
|
// limitations under the License.
|
|
|
|
#pragma once
|
|
#include <numeric> // std::iota
|
|
|
|
#include "paddle/phi/common/memory_utils.h"
|
|
#include "paddle/phi/core/kernel_registry.h"
|
|
#include "paddle/phi/core/mixed_vector.h"
|
|
#include "paddle/phi/kernels/funcs/eigen/common.h"
|
|
#include "paddle/phi/kernels/funcs/math_function.h"
|
|
|
|
namespace phi {
|
|
|
|
template <typename Context, typename T>
|
|
struct SequenceExpandFunctor {
|
|
void operator()(const Context& dev_ctx,
|
|
const DenseTensor& x,
|
|
const Vector<size_t>& x_lod, /*expand source lod*/
|
|
const Vector<size_t>& ref_lod, /*expand referenced lod*/
|
|
DenseTensor* out);
|
|
};
|
|
|
|
template <typename Context, typename T>
|
|
struct SequenceExpandGradFunctor {
|
|
void operator()(const Context& dev_ctx,
|
|
const DenseTensor& dout,
|
|
const Vector<size_t>& x_lod, /*expand source lod*/
|
|
const Vector<size_t>& ref_lod, /*expand referenced lod*/
|
|
DenseTensor* dx);
|
|
};
|
|
|
|
template <typename T, typename Context>
|
|
void SequenceExpandKernel(const Context& dev_ctx,
|
|
const DenseTensor& x_in,
|
|
const DenseTensor& y_in,
|
|
int ref_level,
|
|
DenseTensor* out) {
|
|
// From InferShape
|
|
const auto& x_dims = x_in.dims();
|
|
auto out_dims = x_dims;
|
|
auto& x_lod = x_in.lod();
|
|
auto& y_lod = y_in.lod();
|
|
|
|
PADDLE_ENFORCE_LE(x_lod.size(),
|
|
1UL,
|
|
common::errors::InvalidArgument(
|
|
"Level of Input(X)'s lod should not be "
|
|
"greater than 1. But received: lod level %u.",
|
|
x_lod.size()));
|
|
PADDLE_ENFORCE_GT(y_lod.size(),
|
|
0UL,
|
|
common::errors::InvalidArgument(
|
|
"Level of Input(Y)'s lod should be greater than 0. But "
|
|
"received: lod level %u.",
|
|
y_lod.size()));
|
|
PADDLE_ENFORCE_EQ(
|
|
ref_level == -1 ||
|
|
(ref_level >= 0 && ref_level < static_cast<int>(y_lod.size())),
|
|
true,
|
|
common::errors::InvalidArgument(
|
|
"Invalid `ref_level`, which should be either equal to -1 "
|
|
"or in [0, %d), but received `ref_level` = %u.",
|
|
y_lod.size(),
|
|
ref_level));
|
|
|
|
if (ref_level == -1) ref_level = static_cast<int>(y_lod.size() - 1);
|
|
|
|
if (!x_lod.empty()) {
|
|
PADDLE_ENFORCE_EQ(
|
|
x_lod[0].size(),
|
|
y_lod[ref_level].size(),
|
|
common::errors::InvalidArgument(
|
|
"Level number of Input(X)'s lod could be 0. Otherwise "
|
|
"size of Input(X)'s first level lod should be equal to "
|
|
"size of Input(Y)'s referred level lod. But received: "
|
|
"Input(X).lod[0].size() = %u, Input(Y).lod[%d].size() = "
|
|
"%u",
|
|
x_lod[0].size(),
|
|
ref_level,
|
|
y_lod[ref_level].size()));
|
|
} else {
|
|
PADDLE_ENFORCE_EQ(x_dims[0],
|
|
static_cast<int64_t>(y_lod[ref_level].size()) - 1,
|
|
common::errors::InvalidArgument(
|
|
"When Input(X)'s lod is null, the dims[0] of "
|
|
"Input(X) should match the "
|
|
"size of Input(Y)'s referred level lod. But received "
|
|
"Input(X): input rank %u, input shape [%s]; received "
|
|
"Input(Y).lod[%d].size() - 1 = %d.",
|
|
x_dims.size(),
|
|
x_dims,
|
|
ref_level,
|
|
static_cast<int64_t>(y_lod[ref_level].size()) - 1));
|
|
}
|
|
|
|
int64_t out_first_dim = 0;
|
|
if (y_lod[ref_level].size() <= 1) {
|
|
out_first_dim = x_dims[0];
|
|
} else {
|
|
for (size_t i = 1; i < y_lod[ref_level].size(); ++i) {
|
|
int x_seq_len = 1;
|
|
if (x_lod.size() == 1) {
|
|
x_seq_len = static_cast<int>(x_lod[0][i] - x_lod[0][i - 1]);
|
|
}
|
|
out_first_dim += static_cast<int64_t>(
|
|
(y_lod[ref_level][i] - y_lod[ref_level][i - 1]) * x_seq_len);
|
|
}
|
|
}
|
|
out_dims[0] = out_first_dim;
|
|
out->Resize(out_dims);
|
|
|
|
auto* x = &x_in;
|
|
PADDLE_ENFORCE_EQ(
|
|
y_lod.empty(),
|
|
false,
|
|
common::errors::InvalidArgument(
|
|
"Input(Y) DenseTensor of SequenceExpandOp does not contain "
|
|
"LoD information."));
|
|
|
|
if (ref_level == -1) ref_level = y_lod.size() - 1;
|
|
|
|
dev_ctx.template Alloc<T>(out);
|
|
|
|
if (y_lod[ref_level].size() <= 1) {
|
|
Copy(dev_ctx, *x, dev_ctx.GetPlace(), false, out);
|
|
return;
|
|
}
|
|
|
|
// x lod level is at most 1.
|
|
Vector<size_t> out_lod;
|
|
if (x_lod.size() == 1) {
|
|
out_lod.push_back(0);
|
|
int out_offset = 0;
|
|
for (size_t i = 1; i < y_lod[ref_level].size(); ++i) {
|
|
int repeat_num = y_lod[ref_level][i] - y_lod[ref_level][i - 1];
|
|
int x_start = x_lod[0][i - 1];
|
|
int x_end = x_lod[0][i];
|
|
int x_seq_len = x_end - x_start;
|
|
for (int j = 0; j < repeat_num; ++j) {
|
|
out_lod.push_back(out_lod.back() + x_seq_len);
|
|
out_offset++;
|
|
}
|
|
}
|
|
// write lod to out if x has lod
|
|
auto& ref_lod = *out->mutable_lod();
|
|
ref_lod[0] = out_lod;
|
|
}
|
|
Vector<size_t> ref_x_lod;
|
|
if (x->lod().size() == 1) {
|
|
ref_x_lod = x->lod()[0];
|
|
} else {
|
|
// x_lod doesn't has lod, use fake x lod, level = 0
|
|
ref_x_lod.resize(x->dims()[0] + 1);
|
|
std::iota(ref_x_lod.begin(), ref_x_lod.end(), 0);
|
|
}
|
|
SequenceExpandFunctor<Context, T> functor;
|
|
functor(dev_ctx, *x, ref_x_lod, y_lod[ref_level], out);
|
|
}
|
|
|
|
template <typename T, typename Context>
|
|
void SequenceExpandGradKernel(const Context& dev_ctx,
|
|
const DenseTensor& x_in,
|
|
const DenseTensor& y_in,
|
|
const DenseTensor& out_grad,
|
|
int ref_level,
|
|
DenseTensor* x_grad) {
|
|
auto* g_out = &out_grad;
|
|
auto* x = &x_in;
|
|
auto* y = &y_in;
|
|
auto* g_x = x_grad;
|
|
|
|
dev_ctx.template Alloc<T>(g_x);
|
|
g_x->set_lod(x->lod());
|
|
|
|
funcs::SetConstant<Context, T> set_zero;
|
|
set_zero(dev_ctx, g_x, static_cast<T>(0));
|
|
|
|
auto& y_lod = y->lod();
|
|
if (ref_level == -1) ref_level = y_lod.size() - 1;
|
|
// just copy the gradient
|
|
if (y_lod[ref_level].size() <= 1) {
|
|
Copy(dev_ctx, *g_out, dev_ctx.GetPlace(), false, g_x);
|
|
return;
|
|
}
|
|
|
|
Vector<size_t> ref_x_lod;
|
|
Vector<size_t> ref_lod = y_lod[ref_level];
|
|
if (x->lod().size() == 1) {
|
|
ref_x_lod = x->lod()[0];
|
|
} else {
|
|
// x_lod doesn't has lod, use fake x lod, level = 0
|
|
ref_x_lod.resize(x->dims()[0] + 1);
|
|
std::iota(ref_x_lod.begin(), ref_x_lod.end(), 0);
|
|
}
|
|
SequenceExpandGradFunctor<Context, T> functor;
|
|
functor(dev_ctx, *g_out, ref_x_lod, ref_lod, g_x);
|
|
}
|
|
|
|
// for GPU kernel
|
|
inline void GetOutputOffset(const Vector<size_t>& x_lod,
|
|
const Vector<size_t>& ref_lod,
|
|
Vector<size_t>* out_offset) {
|
|
size_t offset = 0;
|
|
int lod_size = static_cast<int>(x_lod.size());
|
|
for (int i = 0; i < static_cast<int>(x_lod.size()); ++i) {
|
|
(*out_offset)[i] = offset;
|
|
if (i < lod_size - 1) {
|
|
offset += (ref_lod[i + 1] - ref_lod[i]) * (x_lod[i + 1] - x_lod[i]);
|
|
}
|
|
}
|
|
}
|
|
|
|
} // namespace phi
|