Skip to content

【PTen】Add Scalar and ScalarArray in pten #37409

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 5 commits into from
Nov 24, 2021
Merged

【PTen】Add Scalar and ScalarArray in pten #37409

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
a6218c4
add scalar and scalar_array
zyfncg Nov 22, 2021
4df29cf
remove DenseTensor include from Scalar and ScalarArray
zyfncg Nov 22, 2021
81fb5d4
remove inner header from scalar_array
zyfncg Nov 22, 2021
e213512
refactor the method of fill_constant and add some comment
zyfncg Nov 23, 2021
72aa162
Merge branch 'develop' into scalar_and_array
zyfncg Nov 24, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion paddle/fluid/operators/cumsum_op.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -254,7 +254,7 @@ class CumCUDAKernel : public framework::OpKernel<T> {
dim3 transpose_grids((width + tile_size - 1) / tile_size,
(height + tile_size - 1) / tile_size);
auto& dev_ctx = context.template device_context<DeviceContext>();
Tensor tmp;
framework::Tensor tmp;
tmp.Resize(out_dims);
auto* tmp_data = tmp.mutable_data<T>(context.GetPlace());
T* next_in_data = out_data;
Expand Down
3 changes: 1 addition & 2 deletions paddle/fluid/operators/math/tree2col.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,7 +19,6 @@
namespace paddle {
namespace operators {
namespace math {
using Tensor = framework::Tensor;
std::vector<TreeNode> Tree2ColUtil::construct_patch(
size_t root, int max_depth, const std::vector<std::vector<int>> &tr) {
std::stack<TreeNode, std::deque<TreeNode>> stack;
Expand Down Expand Up @@ -51,7 +50,7 @@ std::vector<TreeNode> Tree2ColUtil::construct_patch(
return patch;
}

void Tree2ColUtil::construct_tree(const paddle::Tensor &EdgeSet,
void Tree2ColUtil::construct_tree(const framework::Tensor &EdgeSet,
std::vector<std::vector<int>> *tr,
size_t *node_count) {
auto edge_set_dims = EdgeSet.dims();
Expand Down
4 changes: 1 addition & 3 deletions paddle/fluid/operators/math/tree2col.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,8 +21,6 @@
#include "paddle/fluid/operators/math/math_function.h"

namespace paddle {
using Tensor = framework::Tensor;
using DDim = framework::DDim;
namespace operators {
namespace math {
class TreeNode {
Expand Down Expand Up @@ -64,7 +62,7 @@ class Tree2ColUtil {
static std::vector<TreeNode> construct_patch(
size_t root, int max_depth, const std::vector<std::vector<int>> &tr);

static void construct_tree(const Tensor &EdgeSet,
static void construct_tree(const framework::Tensor &EdgeSet,
std::vector<std::vector<int>> *tr,
size_t *node_count);
};
Expand Down
1 change: 1 addition & 0 deletions paddle/pten/api/all.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,6 +37,7 @@ limitations under the License. */
#include "paddle/pten/common/data_type.h"
#include "paddle/pten/common/layout.h"
#include "paddle/pten/common/scalar.h"
#include "paddle/pten/common/scalar_array.h"

// original custom op headers
#include "paddle/pten/api/ext/dispatch.h"
Expand Down
7 changes: 7 additions & 0 deletions paddle/pten/api/include/creation.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,6 +18,7 @@
#include "paddle/pten/common/backend.h"
#include "paddle/pten/common/data_type.h"
#include "paddle/pten/common/scalar.h"
#include "paddle/pten/common/scalar_array.h"

namespace paddle {
namespace experimental {
Expand All @@ -28,6 +29,12 @@ PD_DLL_DECL Tensor full(const std::vector<int64_t>& shape,
Backend backend = Backend::CPU,
DataLayout layout = DataLayout::NCHW);

PD_DLL_DECL Tensor full_new(const ScalarArray& shape,
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

while not replace original full？

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done, remove original full.

const Scalar& value,
DataType dtype = DataType::FLOAT32,
Backend backend = Backend::CPU,
DataLayout layout = DataLayout::NCHW);

PD_DLL_DECL Tensor full_like(const Tensor& x,
const Scalar& value,
DataType dtype = DataType::UNDEFINED,
Expand Down
40 changes: 38 additions & 2 deletions paddle/pten/api/lib/creation.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,7 +42,43 @@ PD_DLL_DECL Tensor full(const std::vector<int64_t>& shape,
auto kernel_context = pten::KernelContext(dev_ctx);

// 3. Auto data transform
kernel_context.EmplaceBackAttr(value);
kernel_context.EmplaceBackAttr(pten::Scalar(value));

// 4. InferShape
auto out_meta = pten::FullInferShape(shape, dtype, layout);

// 5. Prepare outputs
const auto allocator =
std::make_shared<paddle::experimental::DefaultAllocator>(
pten::TransToFluidPlace(kernel_key.backend()));
auto dense_out = std::make_shared<pten::DenseTensor>(allocator, out_meta);
kernel_context.EmplaceBackOutput(dense_out);
Tensor out;
out.set_impl(dense_out);

// 6. Call kernel
kernel(&kernel_context);

return out;
}

PD_DLL_DECL Tensor full_new(const ScalarArray& shape,
const Scalar& value,
DataType dtype,
Backend backend,
DataLayout layout) {
// 1. Get kernel signature and kernel
pten::KernelKey kernel_key{backend, layout, dtype};
auto kernel = pten::KernelFactory::Instance().SelectKernelOrThrowError(
"fill_constant.new", kernel_key);

// 2. Get Device Context
auto* dev_ctx = GetDeviceContextByBackend(kernel_key.backend());
auto kernel_context = pten::KernelContext(dev_ctx);

// 3. Auto data transform
kernel_context.EmplaceBackAttr(pten::ScalarArray(shape));
kernel_context.EmplaceBackAttr(pten::Scalar(value));

// 4. InferShape
auto out_meta = pten::FullInferShape(shape, dtype, layout);
Expand Down Expand Up @@ -87,7 +123,7 @@ PD_DLL_DECL Tensor full_like(const Tensor& x,

// 3. Auto data transform
auto dense_x = std::dynamic_pointer_cast<pten::DenseTensor>(x.impl());
kernel_context.EmplaceBackAttr(value);
kernel_context.EmplaceBackAttr(pten::Scalar(value));

// 4. InferShape
auto out_meta = FullLikeInferShape(dense_x->meta(), dtype, layout);
Expand Down
5 changes: 5 additions & 0 deletions paddle/pten/api/lib/tensor.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -214,17 +214,22 @@ const T *Tensor::data() const {
template PD_DLL_DECL const float *Tensor::data<float>() const;
template PD_DLL_DECL const double *Tensor::data<double>() const;
template PD_DLL_DECL const int64_t *Tensor::data<int64_t>() const;
template PD_DLL_DECL const uint64_t *Tensor::data<uint64_t>() const;
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why need uint64 and uint32? If there is no current demand, over-design is not recommended

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done, removed.

template PD_DLL_DECL const int32_t *Tensor::data<int32_t>() const;
template PD_DLL_DECL const uint32_t *Tensor::data<uint32_t>() const;
template PD_DLL_DECL const uint8_t *Tensor::data<uint8_t>() const;
template PD_DLL_DECL const int8_t *Tensor::data<int8_t>() const;
template PD_DLL_DECL const int16_t *Tensor::data<int16_t>() const;
template PD_DLL_DECL const uint16_t *Tensor::data<uint16_t>() const;
template PD_DLL_DECL const bool *Tensor::data<bool>() const;
template PD_DLL_DECL const paddle::platform::complex<float>
*Tensor::data<paddle::platform::complex<float>>() const;
template PD_DLL_DECL const paddle::platform::complex<double>
*Tensor::data<paddle::platform::complex<double>>() const;
template PD_DLL_DECL const paddle::platform::float16 *
Tensor::data<paddle::platform::float16>() const;
template PD_DLL_DECL const paddle::platform::bfloat16 *
Tensor::data<paddle::platform::bfloat16>() const;

template <typename T>
T *Tensor::data() {
Expand Down
Loading