248 lines
8.8 KiB
C++
248 lines
8.8 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 <algorithm>
|
|
#include <vector>
|
|
#include "paddle/phi/core/tensor_utils.h"
|
|
#include "paddle/phi/kernels/funcs/eigen/common.h"
|
|
#include "paddle/phi/kernels/funcs/eigen/eigen_function.h"
|
|
|
|
#define MAX_RANK_SUPPORTED 8
|
|
|
|
namespace phi {
|
|
|
|
template <typename T, typename Context, int Rank>
|
|
void Expand(const Context& dev_ctx,
|
|
const DenseTensor& x_in,
|
|
const IntArray& shape,
|
|
DenseTensor* out) {
|
|
auto* in0 = &x_in;
|
|
auto in_dims = in0->dims();
|
|
auto expand_times = shape.GetData();
|
|
PADDLE_ENFORCE_EQ(static_cast<size_t>(in_dims.size()),
|
|
expand_times.size(),
|
|
common::errors::InvalidArgument(
|
|
"The number of elements (%d) of 'expand_times' for "
|
|
"Op(expand) must be equal to the number "
|
|
"of dimensions (%d) of the input.",
|
|
expand_times.size(),
|
|
static_cast<size_t>(in_dims.size())));
|
|
auto* out0 = out;
|
|
Eigen::DSizes<int64_t, Rank> bcast_dims;
|
|
for (size_t i = 0; i < expand_times.size(); ++i) {
|
|
bcast_dims[i] = expand_times[i];
|
|
}
|
|
|
|
DDim out_dims(in_dims);
|
|
for (size_t i = 0; i < expand_times.size(); ++i) {
|
|
out_dims[i] *= expand_times[i];
|
|
}
|
|
|
|
out0->Resize(out_dims);
|
|
auto x = EigenTensor<T, Rank>::From(*in0);
|
|
dev_ctx.template Alloc<T>(out0);
|
|
auto y = EigenTensor<T, Rank>::From(*out0);
|
|
auto& place = *dev_ctx.eigen_device();
|
|
funcs::EigenBroadcast<std::decay_t<decltype(place)>, T, Rank>::Eval(
|
|
place, y, x, bcast_dims);
|
|
}
|
|
|
|
template <typename T, typename Context>
|
|
void LegacyExpandKernel(const Context& dev_ctx,
|
|
const DenseTensor& x,
|
|
const IntArray& shape,
|
|
DenseTensor* out) {
|
|
auto rank = x.dims().size();
|
|
PADDLE_ENFORCE_GE(
|
|
rank,
|
|
1,
|
|
common::errors::InvalidArgument(
|
|
"The number of dimensions of the input 'x' for Op(expand) "
|
|
"must be greater than or equal to 1, but the value received is %d.",
|
|
rank));
|
|
PADDLE_ENFORCE_LE(
|
|
rank,
|
|
MAX_RANK_SUPPORTED,
|
|
common::errors::InvalidArgument(
|
|
"The number of dimensions of the input 'x' for Op(expand) "
|
|
"must be less than or equal to %d, but the value received is %d.",
|
|
MAX_RANK_SUPPORTED,
|
|
rank));
|
|
switch (rank) {
|
|
case 1:
|
|
Expand<T, Context, 1>(dev_ctx, x, shape, out);
|
|
break;
|
|
case 2:
|
|
Expand<T, Context, 2>(dev_ctx, x, shape, out);
|
|
break;
|
|
case 3:
|
|
Expand<T, Context, 3>(dev_ctx, x, shape, out);
|
|
break;
|
|
case 4:
|
|
Expand<T, Context, 4>(dev_ctx, x, shape, out);
|
|
break;
|
|
case 5:
|
|
Expand<T, Context, 5>(dev_ctx, x, shape, out);
|
|
break;
|
|
case 6:
|
|
Expand<T, Context, 6>(dev_ctx, x, shape, out);
|
|
break;
|
|
case 7:
|
|
Expand<T, Context, 7>(dev_ctx, x, shape, out);
|
|
break;
|
|
case 8:
|
|
Expand<T, Context, 8>(dev_ctx, x, shape, out);
|
|
break;
|
|
}
|
|
}
|
|
|
|
template <typename T, typename Context, int Dims>
|
|
void ExpandBackward(const Context& dev_ctx,
|
|
const DenseTensor& out_grad_in,
|
|
const std::vector<int>& reshape_dims_vec,
|
|
const std::vector<int>& reduce_dims_vec,
|
|
DenseTensor* in_grad) {
|
|
size_t reshape_size = reshape_dims_vec.size();
|
|
size_t reduce_size = reduce_dims_vec.size();
|
|
PADDLE_ENFORCE_EQ(reshape_size,
|
|
reshape_dims_vec.size(),
|
|
common::errors::InvalidArgument(
|
|
"Inconsistent size between template Dims (%d) and "
|
|
"reshape dimensions (%d).",
|
|
reshape_size,
|
|
reshape_dims_vec.size()));
|
|
PADDLE_ENFORCE_EQ(reduce_size,
|
|
reduce_dims_vec.size(),
|
|
common::errors::InvalidArgument(
|
|
"Inconsistent size between template Dims (%d) and "
|
|
"reduce dimensions (%d).",
|
|
reduce_size,
|
|
reduce_dims_vec.size()));
|
|
auto* in0 = &out_grad_in;
|
|
auto* out0 = in_grad;
|
|
dev_ctx.template Alloc<T>(out0);
|
|
auto x_grad = EigenVector<T>::Flatten(*out0);
|
|
Eigen::DSizes<int64_t, Dims * 2> reshape_dims;
|
|
for (size_t i = 0; i < reshape_size; ++i) {
|
|
reshape_dims[i] = reshape_dims_vec[i];
|
|
}
|
|
Eigen::DSizes<int64_t, Dims> reduce_dims;
|
|
for (size_t i = 0; i < reduce_size; ++i) {
|
|
reduce_dims[i] = reduce_dims_vec[i];
|
|
}
|
|
auto out_grad = EigenVector<T>::Flatten(*in0);
|
|
auto& place = *dev_ctx.eigen_device();
|
|
funcs::EigenBroadcastGrad<std::decay_t<decltype(place)>, T, Dims>::Eval(
|
|
place, x_grad, out_grad, reduce_dims, reshape_dims);
|
|
}
|
|
|
|
template <typename T, typename Context>
|
|
void LegacyExpandGradKernel(const Context& dev_ctx,
|
|
const DenseTensor& x,
|
|
const DenseTensor& out_grad,
|
|
const IntArray& shape,
|
|
DenseTensor* in_grad) {
|
|
auto* in0 = &x;
|
|
// auto& expand_times = dev_ctx.Attr<std::vector<int>>("expand_times");
|
|
auto expand_times = shape.GetData();
|
|
auto x_dims = in0->dims();
|
|
// 1. reshape_dims_vec is the broadcast parameter.
|
|
// 2. reduce_dims_vec is the dimension parameter to compute gradients. For
|
|
// each dimension expanded, the gradients should be summed to original
|
|
// size.
|
|
std::vector<int> reshape_dims_vec;
|
|
std::vector<int> reduce_dims_vec;
|
|
for (size_t i = 0; i < expand_times.size(); ++i) {
|
|
reduce_dims_vec.push_back(reshape_dims_vec.size());
|
|
reshape_dims_vec.push_back(expand_times[i]);
|
|
reshape_dims_vec.push_back(x_dims[i]);
|
|
}
|
|
|
|
int dims = reduce_dims_vec.size();
|
|
|
|
bool just_copy = true;
|
|
for (size_t i = 0; i < expand_times.size(); i++) {
|
|
if (expand_times[i] != 1) {
|
|
just_copy = false;
|
|
break;
|
|
}
|
|
}
|
|
// no need reduce, just copy
|
|
if (just_copy) {
|
|
auto* in0 = &out_grad;
|
|
auto* out0 = in_grad;
|
|
dev_ctx.template Alloc<T>(out0);
|
|
Copy(dev_ctx, *in0, dev_ctx.GetPlace(), false, out0);
|
|
} else {
|
|
PADDLE_ENFORCE_GE(dims,
|
|
1,
|
|
common::errors::InvalidArgument(
|
|
"The number of dimensions of the input "
|
|
"'Out@GRAD' for Op(expand_grad)"
|
|
" must be greater than or equal to 1, but "
|
|
"the value received is %d.",
|
|
dims));
|
|
PADDLE_ENFORCE_LE(dims,
|
|
MAX_RANK_SUPPORTED,
|
|
common::errors::InvalidArgument(
|
|
"The number of dimensions of the input 'Out@GRAD' "
|
|
"for Op(expand_grad) must be less than or equal "
|
|
"to %d, but the value received is %d.",
|
|
MAX_RANK_SUPPORTED,
|
|
dims));
|
|
switch (dims) {
|
|
case 1:
|
|
ExpandBackward<T, Context, 1>(
|
|
dev_ctx, out_grad, reshape_dims_vec, reduce_dims_vec, in_grad);
|
|
break;
|
|
case 2:
|
|
ExpandBackward<T, Context, 2>(
|
|
dev_ctx, out_grad, reshape_dims_vec, reduce_dims_vec, in_grad);
|
|
break;
|
|
case 3:
|
|
ExpandBackward<T, Context, 3>(
|
|
dev_ctx, out_grad, reshape_dims_vec, reduce_dims_vec, in_grad);
|
|
break;
|
|
case 4:
|
|
ExpandBackward<T, Context, 4>(
|
|
dev_ctx, out_grad, reshape_dims_vec, reduce_dims_vec, in_grad);
|
|
break;
|
|
case 5:
|
|
ExpandBackward<T, Context, 5>(
|
|
dev_ctx, out_grad, reshape_dims_vec, reduce_dims_vec, in_grad);
|
|
break;
|
|
case 6:
|
|
ExpandBackward<T, Context, 6>(
|
|
dev_ctx, out_grad, reshape_dims_vec, reduce_dims_vec, in_grad);
|
|
break;
|
|
case 7:
|
|
ExpandBackward<T, Context, 7>(
|
|
dev_ctx, out_grad, reshape_dims_vec, reduce_dims_vec, in_grad);
|
|
break;
|
|
case 8:
|
|
ExpandBackward<T, Context, 8>(
|
|
dev_ctx, out_grad, reshape_dims_vec, reduce_dims_vec, in_grad);
|
|
break;
|
|
default:
|
|
PADDLE_THROW(common::errors::InvalidArgument(
|
|
"Only support tensor with rank being between 1 and %d. But "
|
|
"received tensor's rank = %d.",
|
|
MAX_RANK_SUPPORTED,
|
|
dims));
|
|
}
|
|
}
|
|
}
|
|
} // namespace phi
|