optimization of index_select op backward#32955
Merged
Xreki merged 30 commits intoPaddlePaddle:developfrom Jul 20, 2021
Merged
Conversation
|
Thanks for your contribution! |
|
Sorry to inform you that d121f02's CIs have passed for more than 7 days. To prevent PR conflicts, you need to re-run all CIs manually. |
JamesLim-sy
reviewed
Jun 9, 2021
| @@ -11,47 +11,50 @@ | |||
| // 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. | |||
Contributor
There was a problem hiding this comment.
在Conversation的Comment区域要描述本次PR的目的，PR修改前后性能变化情况等信息。
Xreki
reviewed
Jul 5, 2021
| using Tensor = framework::Tensor; | ||
| using LoDTensor = framework::LoDTensor; | ||
| using DDim = framework::DDim; | ||
|
|
| auto input_dim_size = input_dim.size(); | ||
| auto output_dim = x_grad->dims(); | ||
| std::vector<T> out_vec(x_grad->numel(), 0); | ||
| std::memset(out_data, 0.0, x_grad->numel() * sizeof(T)); |
Contributor
There was a problem hiding this comment.
可以用SetConstant，另外初始化为0的部分，放到L196后的for循环里面，每次初始化一部分，对cache是不是友好些？
JamesLim-sy
reviewed
Jul 5, 2021
| }; | ||
|
|
||
| template <typename T> | ||
| struct IndexSelectAdd< |
Contributor
There was a problem hiding this comment.
这一段仿函数的意义感觉不大，感觉除了浮点之外采用的是下述通用形式。
template <typename platform::cpu_isa_t isa, typename T, class Enable = void>
struct IndexSelectAdd {
void operator()(int n, const T* src, T* dst) {
for (int i = 0; i < n; i++) {
dst[i] += src[i];
}
}
};
JamesLim-sy
reviewed
Jul 14, 2021
| @@ -194,8 +220,8 @@ class IndexSelectGradKernel : public framework::OpKernel<T> { | |||
| if (dim < 0) { | |||
Contributor
There was a problem hiding this comment.
Line212 - Line219 可以改成：
auto *x_grad = ctx.Input<framework::LoDTensor>("X");
auto *index = ctx.Input<framework::LoDTensor>("Index");
auto *out_grad = ctx.Output<framework::LoDTensor>("Out");
| }; | ||
|
|
||
| template <typename T, typename IndexT = int> | ||
| #if ((!defined __NVCC__) && (!defined __HIPCC__)) |
JamesLim-sy
reviewed
Jul 14, 2021
| void operator()(const framework::ExecutionContext& ctx, int slice_size, | ||
| const T* src_pointer, const T* p_pointer, T* dist_pointer) { | ||
| auto blas = math::GetBlas<DeviceContext, T>(ctx); | ||
| blas.VADD(slice_size, src_pointer, p_pointer, dist_pointer); |
Contributor
There was a problem hiding this comment.
使用的blas的时候，可以测一下不同OMP设置情况下的加速比。
Xreki
approved these changes
Jul 20, 2021
| template <typename DeviceContext, typename T, typename IndexT = int> | ||
| void IndexSelectGradInner(const framework::ExecutionContext& context, | ||
| const LoDTensor& out_grad, const LoDTensor& index, | ||
| const LoDTensor* out_grad, const LoDTensor* index, |
Contributor
There was a problem hiding this comment.
不要修改参数的类型，不用修改的输入用const Tensor&类型。
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
PR types
Performance optimization
PR changes
OPs
Describe
Optimization of index_select op backward, the optimized data as follows:
Compared with the origin and pytorch, optimization measures have been improved.