Skip to content
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

[NPU] Support npu kernel for gather op fix bug #31541

Merged
merged 11 commits into from
Mar 12, 2021
Merged

[NPU] Support npu kernel for gather op fix bug #31541

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
896b450
add gather npu op
xymyeah Mar 6, 2021
e63333c
Merge branch 'ascendrc' of https://github.com/PaddlePaddle/Paddle int…
xymyeah Mar 8, 2021
eea89c1
code review done
xymyeah Mar 8, 2021
203e842
update python new line
xymyeah Mar 8, 2021
f6d5f85
precommit
xymyeah Mar 9, 2021
6941eb7
fix review
xymyeah Mar 9, 2021
98b75be
del commit
xymyeah Mar 9, 2021
ddc8be7
update gather_grad
xymyeah Mar 10, 2021
3c00e5e
fix bug
xymyeah Mar 10, 2021
f89e1bc
fix bug
xymyeah Mar 11, 2021
8a514d3
fix bug
xymyeah Mar 12, 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
7 changes: 5 additions & 2 deletions paddle/fluid/operators/gather_op_npu.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,10 +51,13 @@ class GatherGradOpNPUKernel : public framework::OpKernel<T> {
auto *dx = ctx.Output<Tensor>(framework::GradVarName("X"));

// step1: Unsqueeze index
framework::Tensor tmp_tensor(index->type());
const auto index_dims = index->dims();
if (index_dims.size() == 1) {
framework::Tensor tmp_index = UnsqueezeTo(*index, 2);
index = &tmp_index;
tmp_tensor.ShareDataWith(*index);
std::vector<int64_t> new_dim = {index_dims[0], 1};
tmp_tensor.Resize(framework::make_ddim(new_dim));
index = &tmp_tensor;
}

auto stream =
Expand Down
10 changes: 5 additions & 5 deletions paddle/fluid/operators/gather_op_npu_test.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -109,17 +109,17 @@ void CompareGrad(f::Scope* scope, const p::DeviceContext& ctx,
auto dout = scope->Var("DOut");
auto tensor_dout = dout->GetMutable<f::LoDTensor>();

std::vector<int> init_index = {0, 1, 2, 0};
std::vector<int> init_index = {0, 1};
paddle::framework::TensorFromVector<int>(init_index, ctx, tensor_index);
tensor_index->Resize(paddle::framework::make_ddim({2, 2}));
tensor_index->Resize(paddle::framework::make_ddim({2}));

std::vector<T> init_x = {1.0, 1.0, 1.0, 1.0, 1.0, 1.0};
TensorFromVector(init_x, ctx, tensor_x);
tensor_x->Resize(paddle::framework::make_ddim({3, 2}));

std::vector<T> init_dout = {5.0, 10.0};
std::vector<T> init_dout = {5.0, 10.0, 2.0, 3.0};
TensorFromVector(init_dout, ctx, tensor_dout);
tensor_dout->Resize(paddle::framework::make_ddim({2}));
tensor_dout->Resize(paddle::framework::make_ddim({2, 2}));

ctx.Wait();

Expand All @@ -143,7 +143,7 @@ void CompareGrad(f::Scope* scope, const p::DeviceContext& ctx,
uint32_t expected_size = 3 * 2;
EXPECT_EQ((uint32_t)dx_vec.size(), expected_size);

std::vector<T> expected_dx_vec = {0.0, 5.0, 0.0, 0.0, 10.0, 0.0};
std::vector<T> expected_dx_vec = {5.0, 10.0, 2.0, 3.0, 0.0, 0.0};
for (uint32_t i = 0; i < dx_vec.size(); i++) {
VLOG(3) << "dx_vec[i]=" << dx_vec[i];
EXPECT_EQ(dx_vec[i], expected_dx_vec[i]);
Expand Down