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

fix bug of elementwise_add_grad, *test=kunlun #44545

Merged
merged 5 commits into from
Jul 27, 2022
Merged

fix bug of elementwise_add_grad, *test=kunlun #44545

Show file tree
Hide file tree
Changes from 4 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
db8ac58
fix bug of elementwise_add_grad, *test=kunlun
helen88 Jul 22, 2022
15df22a
fix bug, *test=kunlun
helen88 Jul 22, 2022
566c3cb
rm pooling_t, *test=kunlun
helen88 Jul 22, 2022
4004935
Merge branch 'develop' into ew_add_grad_zh
helen88 Jul 22, 2022
0fedb3e
fix bug of ew_add_grad when inplace, *test=kunlun
helen88 Jul 25, 2022
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
4 changes: 2 additions & 2 deletions cmake/external/xpu.cmake
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,7 @@ set(XPU_RT_LIB_NAME "libxpurt.so")
if(NOT DEFINED XPU_BASE_URL)
set(XPU_BASE_URL_WITHOUT_DATE
"https://baidu-kunlun-product.cdn.bcebos.com/KL-SDK/klsdk-dev")
set(XPU_BASE_URL "${XPU_BASE_URL_WITHOUT_DATE}/20220719")
set(XPU_BASE_URL "${XPU_BASE_URL_WITHOUT_DATE}/20220722")
else()
set(XPU_BASE_URL "${XPU_BASE_URL}")
endif()
Expand All @@ -19,7 +19,7 @@ endif()
if(NOT DEFINED XPU_XDNN_BASE_URL)
set(XPU_XDNN_BASE_URL_WITHOUT_DATE
"https://klx-sdk-release-public.su.bcebos.com/xdnn/dev")
set(XPU_XDNN_BASE_URL "${XPU_XDNN_BASE_URL_WITHOUT_DATE}/20220719")
set(XPU_XDNN_BASE_URL "${XPU_XDNN_BASE_URL_WITHOUT_DATE}/20220722")
else()
set(XPU_XDNN_BASE_URL "${XPU_XDNN_BASE_URL}")
endif()
Expand Down
21 changes: 5 additions & 16 deletions paddle/fluid/operators/elementwise/elementwise_add_op_xpu.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,15 +52,9 @@ class ElementwiseAddGradXPUKernel : public ElemwiseGradKernel<T> {
ctx.template device_context<paddle::platform::XPUDeviceContext>();

if (dx != nullptr) {
T* dx_data = dx->mutable_data<T>(ctx.GetPlace());
if (dx->dims() == dz_dims) {
if (dx_data != dz_data) {
framework::TensorCopy(
*dz,
ctx.GetPlace(),
ctx.template device_context<platform::DeviceContext>(),
dx);
}
// TensorCopy will call mutable_data of dx.
framework::TensorCopy(*dz, ctx.GetPlace(), dev_ctx, dx);
} else {
// For inplace strategy, dx will be stored in addr of dz, which makes
// the result of dy wrong.
Expand All @@ -71,6 +65,7 @@ class ElementwiseAddGradXPUKernel : public ElemwiseGradKernel<T> {
std::vector<int> reduce_dims = GetReduceDim(dx->dims(), dz_dims, axis);
std::vector<int> dz_vector = phi::vectorize<int>(dz_dims);

T* dx_data = dx->mutable_data<T>(ctx.GetPlace());
int ret =
xpu::reduce_sum<XPUType>(dev_ctx.x_context(),
reinterpret_cast<const XPUType*>(dz_data),
Expand All @@ -82,18 +77,12 @@ class ElementwiseAddGradXPUKernel : public ElemwiseGradKernel<T> {
}

if (dy != nullptr) {
T* dy_data = dy->mutable_data<T>(ctx.GetPlace());
if (dy->dims() == dz_dims) {
if (dy_data != dz_data) {
framework::TensorCopy(
*dz,
ctx.GetPlace(),
ctx.template device_context<platform::DeviceContext>(),
dy);
}
framework::TensorCopy(*dz, ctx.GetPlace(), dev_ctx, dy);
} else {
std::vector<int> reduce_dims = GetReduceDim(dy->dims(), dz_dims, axis);
std::vector<int> dz_vector = phi::vectorize<int>(dz_dims);
T* dy_data = dy->mutable_data<T>(ctx.GetPlace());
int ret =
xpu::reduce_sum<XPUType>(dev_ctx.x_context(),
reinterpret_cast<const XPUType*>(dz_data),
Expand Down
17 changes: 0 additions & 17 deletions paddle/fluid/operators/pool_op_xpu.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,23 +21,6 @@ namespace operators {

using framework::Tensor;

xpu::Pooling_t XPUPoolingType(const std::string& pooltype,
bool exclusive,
bool is_test) {
if (pooltype == "max") {
return xpu::Pooling_t::MAX_WITHOUT_INDEX;
} else if (pooltype == "avg") {
if (exclusive) {
return xpu::Pooling_t::AVG_WITHOUT_PAD;
} else {
return xpu::Pooling_t::AVG_WITH_PAD;
}
} else {
PADDLE_THROW(platform::errors::InvalidArgument(
"Pool op only supports 2D and 3D input."));
}
}

template <typename DeviceContext, typename T>
class PoolXPUKernel : public framework::OpKernel<T> {
using XPUType = typename XPUTypeTrait<T>::Type;
Expand Down