Skip to content

Correct tensor copy misuse #10412

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
May 7, 2018
Merged

Correct tensor copy misuse #10412

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
e309f42
fix errors in concat_test
JiayiFeng May 4, 2018
2e61733
fix errors in lod_reset_op
JiayiFeng May 4, 2018
baa9f50
fix errors in multiplex_op
JiayiFeng May 4, 2018
bf99396
fix errors in sequence_slice_op
JiayiFeng May 4, 2018
0c99cd7
fix errors in sequence_padding_test
JiayiFeng May 5, 2018
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
3 changes: 1 addition & 2 deletions paddle/fluid/operators/lod_reset_op.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,8 +46,7 @@ class LoDResetKernel : public framework::OpKernel<T> {
auto* lod = lod_t->data<int>();
if (platform::is_gpu_place(ctx.GetPlace())) {
framework::Tensor lod_cpu;
framework::TensorCopy(*lod_t, platform::CPUPlace(),
ctx.device_context(), &lod_cpu);
framework::TensorCopySync(*lod_t, platform::CPUPlace(), &lod_cpu);
lod = lod_cpu.data<int>();
}
level0 = std::vector<int>(lod, lod + lod_t->numel());
Expand Down
32 changes: 16 additions & 16 deletions paddle/fluid/operators/math/concat_test.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -69,8 +69,8 @@ void testConcat() {
}

if (paddle::platform::is_gpu_place(Place())) {
paddle::framework::TensorCopy(input_a_cpu, Place(), *context, &input_a);
paddle::framework::TensorCopy(input_b_cpu, Place(), *context, &input_b);
paddle::framework::TensorCopySync(input_a_cpu, Place(), &input_a);
paddle::framework::TensorCopySync(input_b_cpu, Place(), &input_b);
}

std::vector<paddle::framework::Tensor> input;
Expand All @@ -86,8 +86,8 @@ void testConcat() {

int* out_ptr;
if (paddle::platform::is_gpu_place(Place())) {
paddle::framework::TensorCopy(out, paddle::platform::CPUPlace(), *context,
&out_cpu);
paddle::framework::TensorCopySync(out, paddle::platform::CPUPlace(),
&out_cpu);
out_ptr = out_cpu.data<int>();
} else {
out_ptr = out.data<int>();
Expand Down Expand Up @@ -142,8 +142,8 @@ void testConcat() {
}

if (paddle::platform::is_gpu_place(Place())) {
paddle::framework::TensorCopy(input_a_cpu, Place(), *context, &input_a);
paddle::framework::TensorCopy(input_b_cpu, Place(), *context, &input_b);
paddle::framework::TensorCopySync(input_a_cpu, Place(), &input_a);
paddle::framework::TensorCopySync(input_b_cpu, Place(), &input_b);
}

input.clear();
Expand All @@ -157,8 +157,8 @@ void testConcat() {
PADDLE_ENFORCE_EQ(input_b.dims(), dim_b);

if (paddle::platform::is_gpu_place(Place())) {
paddle::framework::TensorCopy(out, paddle::platform::CPUPlace(), *context,
&out_cpu);
paddle::framework::TensorCopySync(out, paddle::platform::CPUPlace(),
&out_cpu);
out_ptr = out_cpu.data<int>();
} else {
out_ptr = out.data<int>();
Expand Down Expand Up @@ -215,8 +215,8 @@ void testConcat() {
}

if (paddle::platform::is_gpu_place(Place())) {
paddle::framework::TensorCopy(input_a_cpu, Place(), *context, &input_a);
paddle::framework::TensorCopy(input_b_cpu, Place(), *context, &input_b);
paddle::framework::TensorCopySync(input_a_cpu, Place(), &input_a);
paddle::framework::TensorCopySync(input_b_cpu, Place(), &input_b);
}

input.clear();
Expand All @@ -230,8 +230,8 @@ void testConcat() {
PADDLE_ENFORCE_EQ(input_b.dims(), dim_b);

if (paddle::platform::is_gpu_place(Place())) {
paddle::framework::TensorCopy(out, paddle::platform::CPUPlace(), *context,
&out_cpu);
paddle::framework::TensorCopySync(out, paddle::platform::CPUPlace(),
&out_cpu);
out_ptr = out_cpu.data<int>();
} else {
out_ptr = out.data<int>();
Expand Down Expand Up @@ -290,8 +290,8 @@ void testConcat() {
}

if (paddle::platform::is_gpu_place(Place())) {
paddle::framework::TensorCopy(input_a_cpu, Place(), *context, &input_a);
paddle::framework::TensorCopy(input_b_cpu, Place(), *context, &input_b);
paddle::framework::TensorCopySync(input_a_cpu, Place(), &input_a);
paddle::framework::TensorCopySync(input_b_cpu, Place(), &input_b);
}

input.clear();
Expand All @@ -305,8 +305,8 @@ void testConcat() {
PADDLE_ENFORCE_EQ(input_b.dims(), dim_b);

if (paddle::platform::is_gpu_place(Place())) {
paddle::framework::TensorCopy(out, paddle::platform::CPUPlace(), *context,
&out_cpu);
paddle::framework::TensorCopySync(out, paddle::platform::CPUPlace(),
&out_cpu);
out_ptr = out_cpu.data<int>();
} else {
out_ptr = out.data<int>();
Expand Down
4 changes: 2 additions & 2 deletions paddle/fluid/operators/math/sequence_padding_test.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,7 +41,7 @@ void TestSequencePadding(const paddle::framework::LoD& lod,
if (paddle::platform::is_cpu_place(*place)) {
seq = cpu_seq;
} else {
TensorCopy(cpu_seq, *place, *context, &seq);
TensorCopySync(cpu_seq, *place, &seq);
seq.set_lod(lod);
}

Expand All @@ -64,7 +64,7 @@ void TestSequencePadding(const paddle::framework::LoD& lod,
if (paddle::platform::is_cpu_place(*place)) {
cpu_seq_back = seq_back;
} else {
TensorCopy(seq_back, paddle::platform::CPUPlace(), *context, &cpu_seq_back);
TensorCopySync(seq_back, paddle::platform::CPUPlace(), &cpu_seq_back);
cpu_seq_back.set_lod(lod);
}

Expand Down
4 changes: 2 additions & 2 deletions paddle/fluid/operators/multiplex_op.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,7 +33,7 @@ class MultiplexGPUKernel : public framework::OpKernel<T> {
auto cols = ins[0]->numel() / rows;
// copy index to cpu
Tensor index_t_cpu;
TensorCopy(*ids, platform::CPUPlace(), ctx.device_context(), &index_t_cpu);
TensorCopySync(*ids, platform::CPUPlace(), &index_t_cpu);
auto* index = index_t_cpu.data<int32_t>();
auto stream = ctx.cuda_device_context().stream();
platform::CUDAPlace place = boost::get<platform::CUDAPlace>(ctx.GetPlace());
Expand Down Expand Up @@ -69,7 +69,7 @@ class MultiplexGradGPUKernel : public framework::OpKernel<T> {
auto cols = ins[0]->numel() / rows;
// copy index to cpu
Tensor index_t_cpu;
TensorCopy(*ids, platform::CPUPlace(), ctx.device_context(), &index_t_cpu);
TensorCopySync(*ids, platform::CPUPlace(), &index_t_cpu);
auto* index = index_t_cpu.data<int32_t>();

auto stream = ctx.cuda_device_context().stream();
Expand Down
12 changes: 4 additions & 8 deletions paddle/fluid/operators/sequence_slice_op.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -66,13 +66,11 @@ class SequenceSliceOpKernel : public framework::OpKernel<T> {

if (platform::is_gpu_place(ctx.GetPlace())) {
offset_cpu.mutable_data<T>(offset->dims(), platform::CPUPlace());
framework::TensorCopy(*offset, platform::CPUPlace(), ctx.device_context(),
&offset_cpu);
framework::TensorCopySync(*offset, platform::CPUPlace(), &offset_cpu);
offset_data = offset_cpu.data<int64_t>();

length_cpu.mutable_data<T>(length->dims(), platform::CPUPlace());
framework::TensorCopy(*length, platform::CPUPlace(), ctx.device_context(),
&length_cpu);
framework::TensorCopySync(*length, platform::CPUPlace(), &length_cpu);
length_data = length_cpu.data<int64_t>();
}

Expand Down Expand Up @@ -127,13 +125,11 @@ class SequenceSliceGradOpKernel : public framework::OpKernel<T> {

if (platform::is_gpu_place(ctx.GetPlace())) {
offset_cpu.mutable_data<T>(offset->dims(), platform::CPUPlace());
framework::TensorCopy(*offset, platform::CPUPlace(), ctx.device_context(),
&offset_cpu);
framework::TensorCopySync(*offset, platform::CPUPlace(), &offset_cpu);
offset_data = offset_cpu.data<int64_t>();

length_cpu.mutable_data<T>(length->dims(), platform::CPUPlace());
framework::TensorCopy(*length, platform::CPUPlace(), ctx.device_context(),
&length_cpu);
framework::TensorCopySync(*length, platform::CPUPlace(), &length_cpu);
length_data = length_cpu.data<int64_t>();
}

Expand Down