Files
paddlepaddle--paddle/paddle/phi/kernels/xpu/set_value_kernel.cc
T
2026-07-13 12:40:42 +08:00

456 lines
16 KiB
C++

// Copyright (c) 2023 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.
#include "paddle/phi/kernels/set_value_kernel.h"
#include <algorithm>
#include <vector>
#include "paddle/phi/backends/xpu/enforce_xpu.h"
#include "paddle/phi/common/memory_utils.h"
#include "paddle/phi/core/kernel_registry.h"
#include "paddle/phi/core/tensor_utils.h"
#include "paddle/phi/kernels/funcs/slice_utils.h"
#include "paddle/phi/kernels/xpu/elementwise.h"
namespace phi {
// check whether the tensor with dimension of second can assign to the
// tensor with dimension of first
inline void CheckIsDimsMatch(const DDim& first, const DDim& second) {
int ignore_axis1 = 0, ignore_axis2 = 0;
for (; ignore_axis1 < first.size(); ++ignore_axis1) {
if (first[ignore_axis1] != 1) {
break;
}
}
for (; ignore_axis2 < second.size(); ++ignore_axis2) {
if (second[ignore_axis2] != 1) {
break;
}
}
if (second.size() == ignore_axis2) {
// second tensor has only one value
return;
}
if (first.size() - ignore_axis1 >= second.size() - ignore_axis2) {
auto idx1 = first.size() - 1;
auto idx2 = second.size() - 1;
bool is_match = true;
for (; idx2 >= ignore_axis2; idx2--) {
if (first[idx1--] != second[idx2] && second[idx2] != 1) {
is_match = false;
break;
}
}
if (is_match) {
return;
}
}
PADDLE_THROW(errors::InvalidArgument(
"The shape of tensor assigned value must match the shape "
"of target shape: %d, but now shape is %d.",
second.to_str(),
first.to_str()));
}
template <typename T, typename Context, size_t RANK>
void SetValueImpl(const Context& dev_ctx,
const DenseTensor& in,
const T* value_data,
const DDim& value_dims,
const IntArray& starts,
const IntArray& ends,
const IntArray& steps,
const std::vector<int64_t>& axes,
const std::vector<int64_t>& decrease_axes,
const std::vector<int64_t>& none_axes,
DenseTensor* out) {
using XPUType = typename XPUTypeTrait<T>::Type;
auto in_dims = in.dims();
auto new_value_dims = value_dims;
// support for 0-d tensor
if (value_dims.size() == 0) {
new_value_dims = {1};
}
std::vector<int64_t> starts_local = starts.GetData();
std::vector<int64_t> ends_local = ends.GetData();
std::vector<int64_t> steps_local = steps.GetData();
funcs::CheckAndUpdateSliceAttrs(
in_dims, axes, &starts_local, &ends_local, &steps_local);
auto slice_dims = funcs::GetSliceDims(
in_dims, axes, starts_local, ends_local, &steps_local);
auto decrease_slice_dims = funcs::GetDecreasedDims(slice_dims, decrease_axes);
auto slice_dims_for_assign = decrease_slice_dims;
if (!none_axes.empty()) {
std::vector<int64_t> slice_dims_with_none;
size_t none_axes_cur = 0, decrease_axes_cur = 0;
for (int i = 0; i < slice_dims.size(); ++i) {
while (none_axes_cur < none_axes.size() &&
none_axes[none_axes_cur] <= i) {
slice_dims_with_none.push_back(1);
none_axes_cur++;
}
if (decrease_axes_cur < decrease_axes.size() &&
decrease_axes[decrease_axes_cur] == i) {
decrease_axes_cur++;
} else {
slice_dims_with_none.push_back(slice_dims[i]);
}
}
while (none_axes_cur < none_axes.size()) {
slice_dims_with_none.push_back(1);
none_axes_cur++;
}
slice_dims_for_assign = make_ddim(slice_dims_with_none);
}
// Here copy data from input to avoid data loss at PE and Graph level.
// TODO(liym27): Speed up in the future version.
// - Q: Why don't call ShareDataWith to speed up?
// - A: Because it's not supported to ShareDataWith on OP's input and output
// https://github.com/PaddlePaddle/Paddle/wiki/ShareDataWith-and-ShareBufferWith-are-prohibited-in-OP
// - Q: Why don't delete Input, after all, the input and output are the same
// Tensor at program level?
// - A: If deleting Input, the graph will be complex, such as there will
// be two ops points to the output in graph: op1 -> output <- set_value.
// In this case, we have to find a way to handle the running order of
// set_value is what we want.
int r = 0;
out->Resize(in.dims());
dev_ctx.template Alloc<T>(out);
if (in.numel() == 0) {
return;
}
r = xpu::copy(dev_ctx.x_context(),
reinterpret_cast<const XPUType*>(in.data<T>()),
reinterpret_cast<XPUType*>(out->data<T>()),
in.numel());
PADDLE_ENFORCE_XDNN_SUCCESS(r, "copy");
xpu::ctx_guard RAII_GUARD(dev_ctx.x_context());
int64_t slice_numels = common::product(slice_dims);
XPUType* slice_data = RAII_GUARD.alloc_l3_or_gm<XPUType>(slice_numels);
int in_size = in_dims.size();
std::vector<int64_t> starts_indices(in_size, 0);
std::vector<int64_t> ends_indices(in_size, 0);
std::vector<int64_t> strides_indices(in_size, 0);
std::vector<int64_t> flip_axis;
for (size_t i = 0; i < RANK; ++i) {
starts_indices[i] = 0;
ends_indices[i] = slice_dims[i];
strides_indices[i] = 1;
}
for (size_t i = 0; i < axes.size(); i++) {
int64_t axis_index = axes[i];
starts_indices[axis_index] = starts_local[i];
ends_indices[axis_index] = ends_local[i];
strides_indices[axis_index] = steps_local[i];
if (starts_local[i] ==
ends_local[i]) { // slice is empty, data will not be changed
return;
}
}
// Step 2: Set slice tensor
// - Step 2.1 Set slice tensor with value
// NOTE(liym27): [ Why resize slice_tensor here? ]
// A: When do broadcasting on slice_tensor and value, the shape of
// slice_tensor should be decreased dims.
// e.g.
// x[:,0] = value
// x's shape = [3, 4], value's shape = [3]
// We get slice_dims = [3, 1], decrease_slice_dims = [3]
// If do broadcasting on Tensor with shape [3, 1] and [3], the result's
// shape is [3, 3], which cross the border;
// If do broadcasting on Tensor with shape [3] and [3], the result's shape
// is [3], which is right.
funcs::CheckIsDimsMatch(slice_dims_for_assign, new_value_dims);
// do broadcasting
auto f = [](xpu::Context* xpu_ctx,
const XPUType* x,
const XPUType* y, /*unused*/
XPUType* z,
const std::vector<int64_t>& xshape,
const std::vector<int64_t>& zshape) {
return xpu::broadcast<XPUType>(xpu_ctx, x, z, xshape, zshape);
};
XPUElementwise<T, XPUType>(dev_ctx,
value_data,
new_value_dims,
nullptr,
slice_dims_for_assign,
-1,
reinterpret_cast<T*>(slice_data),
f);
// - Step 2.2 If stride < 0, flip the slice_tensor.
// Because strided_slice_view_update does not support the case of stride < 0
// temporarily, the coordinates of starts_indices, ends_indices
// and strides_indices need to be converted.
// This logic may be deleted in the future.
bool need_flip = false;
for (size_t i = 0; i < RANK; ++i) {
if (strides_indices[i] < 0) {
if (!need_flip) {
need_flip = true;
}
flip_axis.push_back(i);
strides_indices[i] = strides_indices[i] * (-1);
ends_indices[i] = starts_indices[i] + 1;
starts_indices[i] =
starts_indices[i] - (slice_dims[i] - 1) * strides_indices[i];
}
}
auto out_shape = vectorize<int64_t>(out->dims());
auto slice_shape = vectorize<int64_t>(slice_dims);
if (need_flip) {
r = xpu::flip(dev_ctx.x_context(),
reinterpret_cast<const XPUType*>(slice_data),
slice_data,
slice_shape,
flip_axis);
PADDLE_ENFORCE_XDNN_SUCCESS(r, "flip");
}
// Step 3: Set out tensor with value
r = xpu::strided_slice_view_update(
dev_ctx.x_context(),
reinterpret_cast<const XPUType*>(slice_data),
reinterpret_cast<XPUType*>(out->data<T>()),
slice_shape,
out_shape,
starts_indices,
ends_indices,
strides_indices);
PADDLE_ENFORCE_XDNN_SUCCESS(r, "strided_slice_view_update");
}
template <typename T, typename Context>
void SetValueKernelImpl(const Context& dev_ctx,
const DenseTensor& x,
const T* value_data,
const DDim& value_dims,
const IntArray& starts,
const IntArray& ends,
const IntArray& steps,
const std::vector<int64_t>& axes,
const std::vector<int64_t>& decrease_axes,
const std::vector<int64_t>& none_axes,
DenseTensor* out) {
// rank是x tensor的维度信息
const int rank = x.dims().size();
switch (rank) {
case 1:
SetValueImpl<T, Context, 1>(dev_ctx,
x,
value_data,
value_dims,
starts,
ends,
steps,
axes,
decrease_axes,
none_axes,
out);
break;
case 2:
SetValueImpl<T, Context, 2>(dev_ctx,
x,
value_data,
value_dims,
starts,
ends,
steps,
axes,
decrease_axes,
none_axes,
out);
break;
case 3:
SetValueImpl<T, Context, 3>(dev_ctx,
x,
value_data,
value_dims,
starts,
ends,
steps,
axes,
decrease_axes,
none_axes,
out);
break;
case 4:
SetValueImpl<T, Context, 4>(dev_ctx,
x,
value_data,
value_dims,
starts,
ends,
steps,
axes,
decrease_axes,
none_axes,
out);
break;
case 5:
SetValueImpl<T, Context, 5>(dev_ctx,
x,
value_data,
value_dims,
starts,
ends,
steps,
axes,
decrease_axes,
none_axes,
out);
break;
case 6:
SetValueImpl<T, Context, 6>(dev_ctx,
x,
value_data,
value_dims,
starts,
ends,
steps,
axes,
decrease_axes,
none_axes,
out);
break;
default:
PADDLE_THROW(errors::InvalidArgument(
"The rank of input should be less than 7, but received %d.", rank));
}
}
template <typename T, typename Context>
void SetTensorValueKernel(const Context& dev_ctx,
const DenseTensor& x,
const DenseTensor& value,
const IntArray& starts,
const IntArray& ends,
const IntArray& steps,
const std::vector<int64_t>& axes,
const std::vector<int64_t>& decrease_axes,
const std::vector<int64_t>& none_axes,
DenseTensor* out) {
if (x.numel() == 0) {
dev_ctx.template Alloc<T>(out);
return;
}
SetValueKernelImpl<T, Context>(dev_ctx,
x,
value.data<T>(),
value.dims(),
starts,
ends,
steps,
axes,
decrease_axes,
none_axes,
out);
}
template <typename T, typename Context>
void SetValueKernel(const Context& dev_ctx,
const DenseTensor& x,
const IntArray& starts,
const IntArray& ends,
const IntArray& steps,
const std::vector<int64_t>& axes,
const std::vector<int64_t>& decrease_axes,
const std::vector<int64_t>& none_axes,
const std::vector<int64_t>& shape,
const std::vector<Scalar>& values,
DenseTensor* out) {
// avoid using vector<T> if T is bool or phi::float16
size_t value_size = sizeof(T);
size_t values_size = values.size();
size_t values_length = values_size * value_size;
std::vector<uint8_t> assign_values(values_length);
uint8_t* value_data_uint8_cpu = assign_values.data();
for (size_t i = 0; i < values_size; i++) {
T value = values[i].to<T>();
memcpy(value_data_uint8_cpu + i * value_size, &value, value_size);
}
using XPUType = typename XPUTypeTrait<T>::Type;
xpu::ctx_guard RAII_GUARD(dev_ctx.x_context());
T* value_data =
reinterpret_cast<T*>(RAII_GUARD.alloc_l3_or_gm<XPUType>(values_size));
memory_utils::Copy(dev_ctx.GetPlace(),
value_data,
CPUPlace(),
value_data_uint8_cpu,
values_length);
auto value_dims = make_ddim(shape);
SetValueKernelImpl<T, Context>(dev_ctx,
x,
value_data,
value_dims,
starts,
ends,
steps,
axes,
decrease_axes,
none_axes,
out);
}
} // namespace phi
PD_REGISTER_KERNEL(set_value,
XPU,
ALL_LAYOUT,
phi::SetValueKernel,
float,
phi::float16,
phi::bfloat16,
int,
int64_t,
bool) {}
PD_REGISTER_KERNEL(set_value_with_tensor,
XPU,
ALL_LAYOUT,
phi::SetTensorValueKernel,
float,
phi::float16,
phi::bfloat16,
int,
int64_t,
bool) {}