[pten] add concat pten kernel #38955
Conversation
|
Thanks for your contribution! |
| @@ -13,6 +13,8 @@ See the License for the specific language governing permissions and | |||
| limitations under the License. */ | |||
|
|
|||
| #include "paddle/fluid/operators/math/concat_and_split.h" | |||
|
|
|||
| #include "paddle/pten/kernels/cpu/concat_and_split.h" | |||
There was a problem hiding this comment.
建议直接把concat_and_split.cc迁移过来，马上我们也要把它移过来了，可以下个PR再做
| #include "paddle/pten/kernels/funcs/concat_funcs.h" | ||
| namespace pten { | ||
|
|
||
| DenseTensorMeta ConcatInferMeta(const std::vector<DenseTensorMeta>& x_meta, |
There was a problem hiding this comment.
这里后面会改成的返回值作为输入参数指针的形式，和kernel保持一致
|
|
||
| DenseTensorMeta ConcatInferMeta(const std::vector<DenseTensorMeta>& x_meta, | ||
| const Scalar& axis_scalar, | ||
| bool is_runtime) { |
| #include "paddle/pten/core/tensor_meta.h" | ||
| namespace pten { | ||
|
|
||
| // TODO(chentianyu03) use std::vector<DenseTensor> as InferMeta inputs |
There was a problem hiding this comment.
这里后面会新增MetaTensor概念，作为inferMeta的输入
| #include "paddle/fluid/framework/mixed_vector.h" | ||
| #include "paddle/fluid/memory/malloc.h" | ||
| #include "paddle/fluid/operators/math/concat_and_split.h" | ||
| #include "paddle/fluid/platform/cuda_graph_with_memory_pool.h" |
There was a problem hiding this comment.
[TODO] 这里需要梳理下platform下还依赖了哪些组件，是需要我们提前迁移的
| #include <vector> | ||
| #include "gflags/gflags.h" | ||
| #include "paddle/fluid/framework/mixed_vector.h" | ||
| #include "paddle/fluid/memory/malloc.h" |
There was a problem hiding this comment.
这里依赖malloc有点严重，看下我们有替代写法吗
| tmp_dev_ins_data = paddle::memory::Alloc(context, in_num * sizeof(T*)); | ||
| auto* restored = paddle::platform::RestoreHostMemIfCapturingCUDAGraph( | ||
| inputs_data, in_num); | ||
| paddle::memory::Copy(context.GetPlace(), |
There was a problem hiding this comment.
copy_kernle 是对tensor的操作，这里使用的是指针地址
|
|
||
| #include "paddle/pten/kernels/concat_kernel.h" | ||
|
|
||
| #include "paddle/fluid/framework/lod_tensor.h" |
There was a problem hiding this comment.
为什么还需要lod_tensor，这里的头文件确认下必要性
There was a problem hiding this comment.
这里引用lod_tensor.h，是因为使用了AppendLoD 等辅助函数
| @@ -0,0 +1,86 @@ | |||
| /* Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved. | |||
| std::vector<pten::DenseTensor> pt_tensors; | ||
|
|
||
| for(auto & t : tensors) { | ||
| pt_tensors.push_back(*std::dynamic_pointer_cast<pten::DenseTensor>(t.impl())); |
There was a problem hiding this comment.
这里避免使用dynamic_cast，使用is_dense_tensor判断+static cast
There was a problem hiding this comment.
已沟通，先保持现状
| auto in_lod = paddle::framework::ConvertToLengthBasedLoD(x[i].lod()); | ||
| paddle::framework::AppendLoD(out_lod, in_lod); |
There was a problem hiding this comment.
这个kernel文件里引入较多fluid下的代码，建议评估一下迁移难度，如果可以尽量将依赖函数迁移到pten下
MingMingShangTian
dismissed stale reviews from YuanRisheng and chenwhql
via
70c7d9b
PR types
Others
PR changes
Others
Describe
add concat pten kernel