Support SparseCooTensorType #62868
Support SparseCooTensorType #62868
Conversation
|
你的PR提交成功，感谢你对开源项目的贡献! |
| std::size_t operator()(const pir::DenseTensorType &obj) const { | ||
| // return | ||
| // pir::DenseTensorTypeStorage::HashValue(std::make_tuple(pir::Type(), | ||
| // pir::DDim(), pir::DataLayout(), pir::LoD(), size_t())); |
There was a problem hiding this comment.
suggest delete
好的
| using ParamKey = std:: | ||
| tuple<DataType, Dim, DataLayout, DenseTensorType, DenseTensorType, bool>; | ||
| SparseCooTensorTypeStorage(DataType dtype, | ||
| Dim dims, |
There was a problem hiding this comment.
和phi体系下数据结构对应，添加meta_dims
paddle/pir/src/core/sparse_type.cc
Outdated
| // See the License for the specific language governing permissions and | ||
| // limitations under the License. | ||
|
|
||
| #include "paddle/pir/include/core/sparse_type.h" |
There was a problem hiding this comment.
直接放在 op_type.h 中吧
好的
| // See the License for the specific language governing permissions and | ||
| // limitations under the License. | ||
|
|
||
| #pragma once |
There was a problem hiding this comment.
同理，直接放在 paddle/fluid/pir/dialect/operator/ir/type_storage.h
There was a problem hiding this comment.
同理，直接放在 paddle/fluid/pir/dialect/operator/ir/type_storage.h
好的
|
|
||
| #pragma once | ||
|
|
||
| #include "paddle/pir/include/core/sparse_type_storage.h" |
There was a problem hiding this comment.
上移到：paddle/fluid/pir/dialect/operator/ir/op_type.h
There was a problem hiding this comment.
上移到：paddle/fluid/pir/dialect/operator/ir/op_type.h
好的
| std::get<2>(key)))); | ||
| // hash DenseTensorType | ||
| hash_value = pir::detail::hash_combine( | ||
| hash_value, std::hash<DenseTensorType>()(std::get<3>(key))); |
There was a problem hiding this comment.
DenseTensorTypeStorage:: HashValue(...)
There was a problem hiding this comment.
DenseTensorTypeStorage:: HashValue(...)
好的
| std::size_t operator()(const pir::DenseTensorType &obj) const { | ||
| // return | ||
| // pir::DenseTensorTypeStorage::HashValue(std::make_tuple(pir::Type(), | ||
| // pir::DDim(), pir::DataLayout(), pir::LoD(), size_t())); |
| Type dtype() const; | ||
| const Dim &dims() const; | ||
| DataLayout data_layout() const; | ||
| DenseTensorType get_indices() const; |
There was a problem hiding this comment.
| DenseTensorType get_indices() const; | |
| DenseTensorType non_zero_indices() const; |
同下
| /// | ||
| /// \brief Declare ParamKey according to parameter type. | ||
| /// | ||
| using Dim = pir::DDim; |
There was a problem hiding this comment.
| using Dim = pir::DDim; | |
| using Dim = common::DDim; |
| /// \brief Declare ParamKey according to parameter type. | ||
| /// | ||
| using Dim = pir::DDim; | ||
| using DataLayout = pir::DataLayout; |
test/cpp/pir/core/type_test.cc
Outdated
| @@ -249,6 +250,39 @@ TEST(type_test, custom_type_dialect) { | |||
| EXPECT_EQ(dialect_integer1, dialect_integer2); | |||
| } | |||
|
|
|||
| TEST(type_test, sparse_dialect) { | |||
There was a problem hiding this comment.
| TEST(type_test, sparse_dialect) { | |
| TEST(type_test, sparse_coo) { |
paddle/pir/src/core/sparse_type.cc
Outdated
|
|
||
| SparseCooTensorType SparseCooTensorType::dyn_cast_impl(Type type) { | ||
| if (type) { | ||
| if (type.type_id() == type_id()) return SparseCooTensorType(type.storage()); |
There was a problem hiding this comment.
if内单行语句也建议用大括号包裹起来，其他地方同
There was a problem hiding this comment.
if内单行语句也建议用大括号包裹起来，其他地方同
ok
| using Type = pir::Type; | ||
| using Dim = SparseCooTensorTypeStorage::Dim; | ||
| using DataLayout = common::DataLayout; | ||
| using DenseTensorType = pir::DenseTensorType; |
There was a problem hiding this comment.
这些是不是相当于在类SparseCooTensorType内取了这些type的别名并且还是public的啊，感觉不太好？比如你可以这样访问SparseCooTensorType::DenseTensorType，不建议这样做
| using Dim = common::DDim; | ||
| using DataLayout = common::DataLayout; | ||
| using DataType = pir::Type; | ||
| using DenseTensorType = pir::DenseTensorType; |
| public: | ||
| using Base::Base; | ||
| using Type = pir::Type; | ||
| using Dim = SparseCooTensorTypeStorage::Dim; |
There was a problem hiding this comment.
| using Dim = SparseCooTensorTypeStorage::Dim; | |
| using Dim = common::Dim; |
test/cpp/pir/core/type_test.cc
Outdated
| ctx->GetOrRegisterDialect<paddle::dialect::OperatorDialect>(); | ||
| pir::Type fp32_dtype = pir::Float32Type::get(ctx); | ||
| common::DDim dims = {4, 4}; | ||
| common::DDim meta_ddims = {4, 1}; |
There was a problem hiding this comment.
meta_ddims -> meta_dims
test/cpp/pir/core/type_test.cc
Outdated
| coalesced); | ||
|
|
||
| paddle::dialect::SparseCooTensorType sparse_coo_tensor_type = | ||
| paddle::dialect::SparseCooTensorType::dyn_cast_impl(pir_type); |
There was a problem hiding this comment.
这里一般验证方法是pir_type.dyn_cast<paddle::dialect::SparseCooTensorType>()，而不用dyn_cast_impl
test/cpp/pir/core/type_test.cc
Outdated
| paddle::dialect::SparseCooTensorType sparse_coo_tensor_type = | ||
| paddle::dialect::SparseCooTensorType::dyn_cast_impl(pir_type); | ||
|
|
||
| EXPECT_EQ(sparse_coo_tensor_type.isa<paddle::dialect::SparseCooTensorType>(), |
There was a problem hiding this comment.
一般先判断isa<> 再进行dyn_cast
test/cpp/pir/core/type_test.cc
Outdated
| paddle::dialect::SparseCooTensorType sparse_coo_tensor_type = | ||
| paddle::dialect::SparseCooTensorType::dyn_cast_impl(pir_type); | ||
|
|
||
| EXPECT_EQ(sparse_coo_tensor_type.isa<paddle::dialect::SparseCooTensorType>(), |
There was a problem hiding this comment.
| EXPECT_EQ(sparse_coo_tensor_type.isa<paddle::dialect::SparseCooTensorType>(), | |
| EXPECT_EQ(pir_type.isa<paddle::dialect::SparseCooTensorType>(), |
| if (type.type_id() == type_id()) { | ||
| return true; | ||
| } | ||
| if (auto wrap_type = type.dyn_cast<pir::WrapTypeInterface>()) { |
There was a problem hiding this comment.
SparseCooTensorType 在定义的时候就没有WrapTypeInterface，不需要这段逻辑
| if (type.type_id() == type_id()) { | ||
| return SparseCooTensorType(type.storage()); | ||
| } | ||
| if (auto wrap_type = type.dyn_cast<pir::WrapTypeInterface>()) { |
|
|
||
| DataType dtype_; | ||
| Dim dims_; | ||
| Dim meta_dims_; |
There was a problem hiding this comment.
meta_dims 与 dims 的关系是什么，在 SparseCooTensor 中，这两个变量的具体作用有重叠么？
| using Type = pir::Type; | ||
| using Dim = common::Dim; | ||
| using DataLayout = common::DataLayout; |
PR types
New features
PR changes
Others
Description
支持 SparceCooTensorType
Other
Pcard-67164