Skip to content

P
Paddle

PaddlePaddle / Paddle
大约 1 年 前同步成功

通知 2299
Star 20931
Fork 5422
P
Paddle
未验证 提交 bf6af816 编写于 作者: Y Yiqun Liu 提交者： GitHub
浏览文件
操作

Implement a common dimension simplifier. (#47981) 

* Implement a common dims simplifier.

* Fix the include position error.

* Reduce the cpu overhead of broadcast computing.
上级 cb087beb
显示空白变更内容
内联 并排
Showing with 293 addition and 234 deletion
paddle/phi/kernels/funcs/broadcast_function.h
浏览文件 @ bf6af816
......@@ -17,6 +17,7 @@ limitations under the License. */
#include "paddle/phi/kernels/funcs/elementwise_base.h"
#if defined(__NVCC__) || defined(__HIPCC__) || defined(__xpu__)
#include "paddle/phi/kernels/funcs/dims_simplifier.h"
namespace kps = phi::kps;
......@@ -27,203 +28,6 @@ namespace funcs {
#if defined(__NVCC__) || defined(__HIPCC__) || defined(__xpu__)
struct DimensionsTransform {
using DimVector = std::vector<int64_t>;
typedef void (*MergeFunctor)(
bool &, std::vector<DimVector> &, DimVector &, int, int);
int64_t N;
int64_t dim_size;
DimVector out_dims;
std::vector<DimVector> in_dims;
private:
// To compensate the lackage of input_tensors` dimension with input
// variable 'axis'.
void InputDimensionsExtend(int N, int axis) {
for (auto &in_dim : in_dims) {
int64_t in_idx = 0;
if (in_dim.size() < dim_size) {
DimVector tmp_dim(dim_size, 1);
for (; in_idx < in_dim.size();) {
if (in_dim[in_idx] == out_dims[axis] || in_dim[in_idx] == 1) {
tmp_dim[axis] = in_dim[in_idx];
in_idx++;
axis++;
} else {
PADDLE_THROW(phi::errors::InvalidArgument(
"The %d-th dimension of input tensor is expected to be equal "
"with the %d-th dimension of output tensor %d or 1, but "
"received %d.",
in_idx + 1,
axis + 1,
out_dims[axis],
in_dim[in_idx]));
}
}
in_dim.resize(dim_size);
std::copy(tmp_dim.begin(), tmp_dim.end(), in_dim.begin());
} else {
for (; in_idx < dim_size;) {
if (in_dim[in_idx] == out_dims[in_idx] || in_dim[in_idx] == 1) {
in_idx++;
} else {
PADDLE_THROW(phi::errors::InvalidArgument(
"The %d-th dimension of input tensor is expected to be equal "
"with the %d-th dimension of output tensor %d or 1, but "
"received %d.",
in_idx + 1,
in_idx + 1,
out_dims[in_idx],
in_dim[in_idx]));
}
}
}
std::reverse(in_dim.begin(), in_dim.end());
}
std::reverse(out_dims.begin(), out_dims.end());
}
// Merge sequential dimension to shrink calculation cost for
// offset computation in CUDA Kernel.
template <typename MergeFunctor>
__inline__ void MergeDimensions(MergeFunctor merge_func, int N) {
auto VectorReorganise = [](DimVector *vec, int l_idx, int m_idx) {
(*vec)[m_idx - 1] = std::accumulate(vec->begin() + l_idx,
vec->begin() + m_idx,
1,
std::multiplies<int64_t>());
vec->erase(vec->begin() + l_idx, vec->begin() + m_idx - 1);
};
int64_t i = 0;
while (i < dim_size) {
int cnt = 0;
int low_idx = i;
bool equal = true;
do {
merge_func(equal, in_dims, out_dims, i, N);
if (equal) {
i++;
cnt++;
} else {
break;
}
} while (i < dim_size);
if (cnt > 1) {
for (auto &in_dim : in_dims) {
VectorReorganise(&in_dim, low_idx, i);
}
VectorReorganise(&out_dims, low_idx, i);
dim_size -= --cnt;
i -= cnt;
} else if (cnt < 1) {
i++;
}
}
}
// To judge whether shape of any input tensors is sequential
// 1-value-dimensions, and metric the length of it.
bool FindSequentialOneDim(int *swap_index) {
int index = 0;
int max_one_length = 0;
for (int j = 0; j < N; ++j) {
int seq_one_length = 0;
bool active_seq = false;
for (int i = 0; i < dim_size; ++i) {
if (!active_seq && in_dims[j][i] == 1) {
seq_one_length = 1;
active_seq = true;
} else if (active_seq) {
if (in_dims[j][i] == 1) {
seq_one_length++;
} else {
active_seq = false;
}
}
}
index = seq_one_length > max_one_length ? j : index;
max_one_length = std::max(seq_one_length, max_one_length);
}
bool has_seq_one = max_one_length > 1;
if (has_seq_one) {
std::swap(in_dims[0], in_dims[index]);
*swap_index = index;
}
return has_seq_one;
}
public:
explicit DimensionsTransform(const std::vector<const DenseTensor *> &ins,
const phi::DDim &dims,
int axis) {
N = std::max(static_cast<int>(ins.size()), 2);
dim_size = dims.size();
out_dims = phi::vectorize<int64_t>(dims);
in_dims.resize(N);
if (ins.size() == 1) {
// when ins.size() = 1, broadcast input to output
in_dims[0] = phi::vectorize<int64_t>(ins[0]->dims());
in_dims[1] = out_dims;
// Add out_dims to in_dims to avoid errors in dims merging
} else {
for (int j = 0; j < N; ++j) {
in_dims[j] = phi::vectorize<int64_t>(ins[j]->dims());
}
}
InputDimensionsExtend(N, axis);
// To Merge the dimensions of input_tensors while the consequtive
// equal-dimensions appears. Example below :
// in_1.shape = [2, 3, 4, 5] in_1.shape = [2, 12, 5]
// in_2.shape = [1, 3, 4, 5] -> in_2.shape = [1, 12, 5]
// in_3.shape = [2, 3, 4, 1] in_3.shape = [2, 12, 1]
auto merge_sequential_dims = [](bool &equal,
std::vector<DimVector> &in_dims,
DimVector &out,
int i,
int num) {
for (int j = 1; j < num; ++j) {
equal &= (in_dims[0][i] == in_dims[j][i]) ? true : false;
}
};
MergeFunctor merge_ptr = merge_sequential_dims;
MergeDimensions<MergeFunctor>(merge_ptr, N);
// To Merge the dimension of input_tensors while the sequential
// 1-value-dimensions appears. Example below :
// in_1.shape = [2, 1, 1, 5] in_1.shape = [2, 1, 5]
// in_2.shape = [2, 3, 4, 5] -> in_2.shape = [1, 12, 5]
// in_3.shape = [2, 3, 4, 1] in_3.shape = [2, 12, 1]
// Caution: Once 1-value-dimensions appears, the corresponding
// shape position of other input tensors must be same with the
// output tensor`s shape, or incorrect merge may occur.
auto merge_sequential_one_dims = [](bool &equal,
std::vector<DimVector> &in_dims,
DimVector &out,
int i,
int num) {
equal = in_dims[0][i] == 1;
if (equal) {
for (int j = 1; j < num; ++j) {
equal &= in_dims[j][i] == out[i];
}
}
};
for (auto i = 0; i < dim_size; ++i) {
int swap_idx = 0;
bool has_seq_one = FindSequentialOneDim(&swap_idx);
if (!has_seq_one) break;
merge_ptr = merge_sequential_one_dims;
MergeDimensions<MergeFunctor>(merge_ptr, N);
std::swap(in_dims[swap_idx], in_dims[0]);
}
}
};
template <typename InT, typename OutT>
int GetVecsize(const std::vector<const DenseTensor *> &ins,
std::vector<DenseTensor *> *outs) {
......@@ -313,7 +117,7 @@ struct BroadcastDataLoader<T, VecSize, Arity, IsBoundary, true> {
#pragma unroll
for (int i = 0; i < phi::DDim::kMaxRank; ++i) {
if (i == configs[0].kDims) break;
if (i == configs[0].rank) break;
auto fast_divmoder = configs[0].divmoders[i].Divmod(idx);
idx = fast_divmoder.val[0];
#pragma unroll
......@@ -1071,7 +875,19 @@ void BroadcastKernelForDifferentVecSize(
#endif
// mergedim and get vec_size
const auto merge_dims = DimensionsTransform(ins, (*outs)[0]->dims(), axis);
const auto dims_simplifier =
BroadcastDimsSimplifier(ins, (*outs)[0]->dims(), axis);
if (VLOG_IS_ON(4)) {
for (size_t i = 0; i < dims_simplifier.in_dims.size(); ++i) {
VLOG(4) << "input i=" << i << ": origin_dims={" << ins[i]->dims()
<< "}, simplied_dims={"
<< phi::make_ddim(dims_simplifier.in_dims[i]) << "}";
}
VLOG(4) << "output: origin_dims={" << (*outs)[0]->dims()
<< "}, simplied_dims={" << phi::make_ddim(dims_simplifier.out_dims)
<< "}";
}
phi::Array<kps::details::BroadcastConfig, kArity> configs;
// get vec_size
......@@ -1081,14 +897,14 @@ void BroadcastKernelForDifferentVecSize(
2,
phi::errors::InvalidArgument(
"XPU only support inputs is 2, but received %d", ins.size()));
configs[0] = kps::details::BroadcastConfig(merge_dims.out_dims,
merge_dims.in_dims[0],
merge_dims.in_dims[1],
merge_dims.dim_size);
configs[1] = kps::details::BroadcastConfig(merge_dims.out_dims,
merge_dims.in_dims[1],
merge_dims.in_dims[0],
merge_dims.dim_size);
configs[0] = kps::details::BroadcastConfig(dims_simplifier.out_dims,
dims_simplifier.in_dims[0],
dims_simplifier.in_dims[1],
dims_simplifier.rank);
configs[1] = kps::details::BroadcastConfig(dims_simplifier.out_dims,
dims_simplifier.in_dims[1],
dims_simplifier.in_dims[0],
dims_simplifier.rank);
auto type = kps::details::OptType::CanNotOptimize;
bool is_optimize = configs[0].cmp_type != type;
int vec_size = is_optimize ? VecSizeL : VecSizeM;
......@@ -1099,8 +915,9 @@ void BroadcastKernelForDifferentVecSize(
// eg: out's shape [3, 45, 1]. then out_dims = {1, 45, 3}
// if (ins[i]->numel() != (*outs)[0]->numel()) {
if (ins[i]->numel()) {
configs[i] = kps::details::BroadcastConfig(
merge_dims.out_dims, merge_dims.in_dims[i], merge_dims.dim_size);
configs[i] = kps::details::BroadcastConfig(dims_simplifier.out_dims,
dims_simplifier.in_dims[i],
dims_simplifier.rank);
}
}
int vec_size = GetVecsize<InT, OutT>(ins, outs);
......
paddle/phi/kernels/funcs/dims_simplifier.h 0 → 100644
浏览文件 @ bf6af816
/* Copyright (c) 2022 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 "paddle/phi/core/ddim.h"
#include "paddle/phi/core/dense_tensor.h"
namespace phi {
namespace funcs {
struct BroadcastDimsSimplifier {
using DimVector = std::vector<int64_t>;
typedef void (*MergeFunctor)(
bool &, std::vector<DimVector> &, DimVector &, int, int);
int64_t N;
int64_t rank;
DimVector out_dims;
std::vector<DimVector> in_dims;
public:
BroadcastDimsSimplifier(const std::vector<const DenseTensor *> &ins,
const phi::DDim &dims,
int axis) {
if (!NeedBroadcast(ins, dims)) {
int64_t numel = phi::product(dims);
rank = 1;
N = ins.size();
out_dims = DimVector{numel};
in_dims.resize(N);
for (int64_t i = 0; i < N; ++i) {
in_dims[i] = DimVector{numel};
}
return;
}
N = std::max(static_cast<int>(ins.size()), 2);
in_dims.resize(N);
rank = dims.size();
out_dims = phi::vectorize<int64_t>(dims);
if (ins.size() == 1) {
// When ins.size() = 1, broadcast input to output.
in_dims[0] = phi::vectorize<int64_t>(ins[0]->dims());
// Add out_dims to in_dims to avoid errors in dims merging.
in_dims[1] = out_dims;
} else {
for (int j = 0; j < N; ++j) {
in_dims[j] = phi::vectorize<int64_t>(ins[j]->dims());
}
}
ExtendInputDimensions(N, axis);
// To Merge the dimensions of input_tensors while the consequtive
// equal-dimensions appears. Example below :
// in_1.shape = [2, 3, 4, 5] in_1.shape = [2, 12, 5]
// in_2.shape = [1, 3, 4, 5] -> in_2.shape = [1, 12, 5]
// in_3.shape = [2, 3, 4, 1] in_3.shape = [2, 12, 1]
auto merge_sequential_dims = [](bool &equal,
std::vector<DimVector> &in_dims,
DimVector &out,
int i,
int num) {
for (int j = 1; j < num; ++j) {
equal &= (in_dims[0][i] == in_dims[j][i]) ? true : false;
}
};
MergeFunctor merge_ptr = merge_sequential_dims;
MergeDimensions<MergeFunctor>(merge_ptr, N);
// To Merge the dimension of input_tensors while the sequential
// 1-value-dimensions appears. Example below :
// in_1.shape = [2, 1, 1, 5] in_1.shape = [2, 1, 5]
// in_2.shape = [2, 3, 4, 5] -> in_2.shape = [1, 12, 5]
// in_3.shape = [2, 3, 4, 1] in_3.shape = [2, 12, 1]
// Caution: Once 1-value-dimensions appears, the corresponding
// shape position of other input tensors must be same with the
// output tensor`s shape, or incorrect merge may occur.
auto merge_sequential_one_dims = [](bool &equal,
std::vector<DimVector> &in_dims,
DimVector &out,
int i,
int num) {
equal = in_dims[0][i] == 1;
if (equal) {
for (int j = 1; j < num; ++j) {
equal &= in_dims[j][i] == out[i];
}
}
};
for (auto i = 0; i < rank; ++i) {
int swap_idx = 0;
bool has_seq_one = FindSequentialOneDim(&swap_idx);
if (!has_seq_one) {
break;
}
merge_ptr = merge_sequential_one_dims;
MergeDimensions<MergeFunctor>(merge_ptr, N);
std::swap(in_dims[swap_idx], in_dims[0]);
}
}
private:
bool NeedBroadcast(const std::vector<const DenseTensor *> &ins,
const phi::DDim &dims) {
bool no_broadcast_flag = true;
for (auto *in : ins) {
no_broadcast_flag &= ins[0]->dims() == in->dims();
}
if (ins.size() > 0) {
no_broadcast_flag &= dims == ins[0]->dims();
}
return !no_broadcast_flag;
}
// To compensate the lackage of input_tensors' dimension with axis.
void ExtendInputDimensions(int N, int axis) {
for (auto &in_dim : in_dims) {
int64_t in_idx = 0;
if (in_dim.size() < rank) {
DimVector tmp_dim(rank, 1);
for (; in_idx < in_dim.size();) {
if (in_dim[in_idx] == out_dims[axis] || in_dim[in_idx] == 1) {
tmp_dim[axis] = in_dim[in_idx];
in_idx++;
axis++;
} else {
PADDLE_THROW(phi::errors::InvalidArgument(
"The %d-th dimension of input tensor is expected to be equal "
"with the %d-th dimension of output tensor %d or 1, but "
"received %d.",
in_idx + 1,
axis + 1,
out_dims[axis],
in_dim[in_idx]));
}
}
in_dim.resize(rank);
std::copy(tmp_dim.begin(), tmp_dim.end(), in_dim.begin());
} else {
for (; in_idx < rank;) {
if (in_dim[in_idx] == out_dims[in_idx] || in_dim[in_idx] == 1) {
in_idx++;
} else {
PADDLE_THROW(phi::errors::InvalidArgument(
"The %d-th dimension of input tensor is expected to be equal "
"with the %d-th dimension of output tensor %d or 1, but "
"received %d.",
in_idx + 1,
in_idx + 1,
out_dims[in_idx],
in_dim[in_idx]));
}
}
}
std::reverse(in_dim.begin(), in_dim.end());
}
std::reverse(out_dims.begin(), out_dims.end());
}
// Merge sequential dimension to shrink calculation cost for
// offset computation in CUDA Kernel.
template <typename MergeFunctor>
__inline__ void MergeDimensions(MergeFunctor merge_func, int N) {
auto VectorReorganise = [](DimVector *vec, int l_idx, int m_idx) {
(*vec)[m_idx - 1] = std::accumulate(vec->begin() + l_idx,
vec->begin() + m_idx,
1,
std::multiplies<int64_t>());
vec->erase(vec->begin() + l_idx, vec->begin() + m_idx - 1);
};
int64_t i = 0;
while (i < rank) {
int cnt = 0;
int low_idx = i;
bool equal = true;
do {
merge_func(equal, in_dims, out_dims, i, N);
if (equal) {
i++;
cnt++;
} else {
break;
}
} while (i < rank);
if (cnt > 1) {
for (auto &in_dim : in_dims) {
VectorReorganise(&in_dim, low_idx, i);
}
VectorReorganise(&out_dims, low_idx, i);
rank -= --cnt;
i -= cnt;
} else if (cnt < 1) {
i++;
}
}
}
// To judge whether shape of any input tensors is sequential
// 1-value-dimensions, and metric the length of it.
bool FindSequentialOneDim(int *swap_index) {
int index = 0;
int max_one_length = 0;
for (int j = 0; j < N; ++j) {
int seq_one_length = 0;
bool active_seq = false;
for (int i = 0; i < rank; ++i) {
if (!active_seq && in_dims[j][i] == 1) {
seq_one_length = 1;
active_seq = true;
} else if (active_seq) {
if (in_dims[j][i] == 1) {
seq_one_length++;
} else {
active_seq = false;
}
}
}
index = seq_one_length > max_one_length ? j : index;
max_one_length = std::max(seq_one_length, max_one_length);
}
bool has_seq_one = max_one_length > 1;
if (has_seq_one) {
std::swap(in_dims[0], in_dims[index]);
*swap_index = index;
}
return has_seq_one;
}
};
} // namespace funcs
} // namespace phi
paddle/phi/kernels/primitive/datamover_primitives.h
浏览文件 @ bf6af816
......@@ -85,33 +85,28 @@ struct FastDivMod {
struct BroadcastConfig {
FastDivMod divmoders[phi::DDim::kMaxRank];
uint32_t strides[phi::DDim::kMaxRank];
int kDims{0};
HOSTDEVICE BroadcastConfig() {}
int rank{0};
HOSTDEVICE BroadcastConfig(const std::vector<int64_t>& out_dims,
// BroadcastConfig should be defined on host used on device.
BroadcastConfig() {}
BroadcastConfig(const std::vector<int64_t>& out_dims,
const std::vector<int64_t>& in_dims,
int dim_size) {
std::vector<uint32_t> strides_in;
std::vector<FastDivMod> divmoders_in;
// for divmoders
divmoders_in.resize(dim_size);
for (int i = 0; i < dim_size; ++i) {
divmoders_in[i] = FastDivMod(out_dims[i]);
divmoders[i] = FastDivMod(out_dims[i]);
}
// for strides
strides_in.resize(dim_size, 1);
for (int i = 0; i < dim_size; ++i) {
strides_in[i] = in_dims[i] == 1 ? 0 : strides_in[i];
strides_in[i] = (i != 0 && strides_in[i] != 0)
strides[i] = in_dims[i] == 1 ? 0 : 1;
strides[i] = (i != 0 && strides[i] != 0)
? std::accumulate(in_dims.begin(),
in_dims.begin() + i,
1,
std::multiplies<int64_t>())
: strides_in[i];
: strides[i];
}
kDims = dim_size;
memcpy(strides, strides_in.data(), kDims * sizeof(uint32_t));
memcpy(divmoders, divmoders_in.data(), kDims * sizeof(FastDivMod));
rank = dim_size;
}
};
......@@ -452,7 +447,7 @@ __device__ __forceinline__ void ReadDataBc(
}
#pragma unroll
for (int i = 0; i < phi::DDim::kMaxRank; ++i) {
if (i >= config.kDims) break;
if (i >= config.rank) break;
auto fast_divmoder = config.divmoders[i].Divmod(index_output);
index_output = fast_divmoder.val[0];
index_src += fast_divmoder.val[1] * config.strides[i];
......@@ -784,7 +779,7 @@ __device__ __forceinline__ void ReadDataBc(
}
#pragma unroll
for (int i = 0; i < phi::DDim::kMaxRank; ++i) {
if (i >= config.kDims) break;
if (i >= config.rank) break;
auto fast_divmoder = config.divmoders[i].Divmod(index_output);
index_output = fast_divmoder.val[0];
index_src += fast_divmoder.val[1] * config.strides[i];
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或