Skip to content

Add is_custom_device to support metax-gpu #24

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

Open
wants to merge 1 commit into
base: paddle
Choose a base branch
from
Open

Add is_custom_device to support metax-gpu #24

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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: 4 additions & 3 deletions cpp/open3d/ml/paddle/PaddleHelper.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,9 +19,10 @@
#include "paddle/phi/core/allocator.h"

// Macros for checking tensor properties
#define CHECK_CUDA(x) \
do { \
PD_CHECK(x.is_gpu(), #x " must be a CUDA tensor"); \
#define CHECK_CUDA(x) \
do { \
PD_CHECK(x.is_gpu() || x.is_custom_device(), \
#x " must be a CUDA tensor"); \
} while (0)

// NOTE: The input Tensor will be preprocessed into a contiguous Tensor within
Expand Down
4 changes: 2 additions & 2 deletions cpp/open3d/ml/paddle/continuous_conv/ContinuousConvOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -97,7 +97,7 @@ std::vector<paddle::Tensor> ContinuousConvForward(
return {out_features}; \
}

if (inp_features.is_gpu()) {
if (inp_features.is_gpu() || inp_features.is_custom_device()) {
Copy link
Collaborator

@BeingGod BeingGod Sep 12, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

对于非cuda兼容的 custom device 设备，不应该走到 CUDA kernel

#ifdef BUILD_CUDA_MODULE
CALL(float, float, float, int32_t, ::ContinuousConvCUDA)
#else
Expand Down Expand Up @@ -189,7 +189,7 @@ std::vector<paddle::Tensor> ContinuousConvBackward(
}

bool dispatch_success = false;
if (inp_features.is_gpu()) {
if (inp_features.is_gpu() || inp_features.is_custom_device()) {
#ifdef BUILD_CUDA_MODULE
CALL(float, float, float, int32_t, CUDA)
#else
Expand Down
4 changes: 2 additions & 2 deletions cpp/open3d/ml/paddle/continuous_conv/ContinuousConvTransposeOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -110,7 +110,7 @@ std::vector<paddle::Tensor> ContinuousConvTransposeForward(
return {out_features}; \
}

if (inp_features.is_gpu()) {
if (inp_features.is_gpu() || inp_features.is_custom_device()) {
#ifdef BUILD_CUDA_MODULE
CALL(float, float, float, int32_t, ::ContinuousConvTransposeCUDA)
#else
Expand Down Expand Up @@ -206,7 +206,7 @@ std::vector<paddle::Tensor> ContinuousConvTransposeBackward(
}

bool dispatch_success = false;
if (inp_features.is_gpu()) {
if (inp_features.is_gpu() || inp_features.is_custom_device()) {
#ifdef BUILD_CUDA_MODULE
CALL(float, float, float, int32_t, CUDA)
#else
Expand Down
2 changes: 1 addition & 1 deletion cpp/open3d/ml/paddle/misc/BuildSpatialHashTableOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -86,7 +86,7 @@ std::vector<paddle::Tensor> BuildSpatialHashTable(
return {hash_table_index, hash_table_cell_splits, \
out_hash_table_splits}; \
}
if (points.is_gpu()) {
if (points.is_gpu() || points.is_custom_device()) {
#ifdef BUILD_CUDA_MODULE
// pass to cuda function
CALL(float, BuildSpatialHashTableCUDA)
Expand Down
2 changes: 1 addition & 1 deletion cpp/open3d/ml/paddle/misc/FixedRadiusSearchOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -116,7 +116,7 @@ std::vector<paddle::Tensor> FixedRadiusSearch(
metric, ignore_query_point, return_distances, neighbors_index, \
neighbors_row_splits, neighbors_distance

if (points.is_gpu()) {
if (points.is_gpu() || points.is_custom_device()) {
#ifdef BUILD_CUDA_MODULE
// pass to cuda function
if (ComparePaddleDtype<float>(point_type)) {
Expand Down
3 changes: 2 additions & 1 deletion cpp/open3d/ml/paddle/misc/InvertNeighborsListOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,7 +46,8 @@ std::vector<paddle::Tensor> InvertNeighborsList(

CHECK_SAME_DEVICE_TYPE(inp_neighbors_index, inp_neighbors_row_splits,
inp_neighbors_attributes);
if (inp_neighbors_index.is_gpu()) {
if (inp_neighbors_index.is_gpu() ||
inp_neighbors_index.is_custom_device()) {
#ifdef BUILD_CUDA_MODULE
// pass to cuda function
CALL(int32_t, uint8_t, InvertNeighborsListCUDA)
Expand Down
2 changes: 1 addition & 1 deletion cpp/open3d/ml/paddle/misc/KnnSearchOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -83,7 +83,7 @@ std::vector<paddle::Tensor> KnnSearch(paddle::Tensor& points,
ignore_query_point, return_distances, neighbors_index, \
neighbors_row_splits, neighbors_distance

if (points.is_gpu()) {
if (points.is_gpu() || points.is_custom_device()) {
PD_CHECK(false, "KnnSearch does not support CUDA");
} else {
if (ComparePaddleDtype<float>(point_type)) {
Expand Down
2 changes: 1 addition & 1 deletion cpp/open3d/ml/paddle/misc/NmsOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,7 @@ std::vector<paddle::Tensor> Nms(paddle::Tensor& boxes,
CHECK_TYPE(scores, phi::DataType::FLOAT32);

std::vector<int64_t> keep_indices_blob;
if (boxes.is_gpu()) {
if (boxes.is_gpu() || boxes.is_custom_device()) {
#ifdef BUILD_CUDA_MODULE
keep_indices_blob = open3d::ml::contrib::NmsCUDAKernel(
boxes.data<float>(), scores.data<float>(), boxes.shape()[0],
Expand Down
2 changes: 1 addition & 1 deletion cpp/open3d/ml/paddle/misc/RadiusSearchOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -85,7 +85,7 @@ std::vector<paddle::Tensor> MultiRadiusSearch(
ignore_query_point, return_distances, normalize_distances, \
neighbors_index, neighbors_row_splits, neighbors_distance

if (points.is_gpu()) {
if (points.is_gpu() || points.is_custom_device()) {
PD_CHECK(false, "MultiRadiusSearch does not support CUDA");
} else {
if (ComparePaddleDtype<float>(point_type)) {
Expand Down
2 changes: 1 addition & 1 deletion cpp/open3d/ml/paddle/misc/RaggedToDenseOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,7 +59,7 @@ std::vector<paddle::Tensor> RaggedToDense(paddle::Tensor& values,
return {fn<value_t>(values, row_splits, out_col_size, default_value)}; \
}

if (values.is_gpu()) {
if (values.is_gpu() || values.is_custom_device()) {
#ifdef BUILD_CUDA_MODULE
// pass to cuda function
CALL(uint8_t, RaggedToDenseCUDA)
Expand Down
2 changes: 1 addition & 1 deletion cpp/open3d/ml/paddle/misc/ReduceSubarraysSumOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,7 +33,7 @@ std::vector<paddle::Tensor> ReduceSubarraysSum(paddle::Tensor& values,

CHECK_SAME_DEVICE_TYPE(values, row_splits);

if (values.is_gpu()) {
if (values.is_gpu() || values.is_custom_device()) {
#ifdef BUILD_CUDA_MODULE
// pass to cuda function
CALL(int32_t, ReduceSubarraysSumCUDA)
Expand Down
4 changes: 2 additions & 2 deletions cpp/open3d/ml/paddle/misc/VoxelPoolingOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -92,7 +92,7 @@ std::vector<paddle::Tensor> VoxelPoolingForward(
}

CHECK_SAME_DEVICE_TYPE(positions, features);
if (positions.is_gpu()) {
if (positions.is_gpu() || positions.is_custom_device()) {
PD_CHECK(false, "VoxelPooling does not support CUDA");
} else {
CALL(float, float, VoxelPoolingCPU)
Expand Down Expand Up @@ -170,7 +170,7 @@ std::vector<paddle::Tensor> VoxelPoolingBackward(
}

CHECK_SAME_DEVICE_TYPE(positions, features);
if (positions.is_gpu()) {
if (positions.is_gpu() || positions.is_custom_device()) {
PD_CHECK(false, "VoxelPooling backward does not support CUDA");
} else {
CALL(float, float, VoxelPoolingGradCPU)
Expand Down
2 changes: 1 addition & 1 deletion cpp/open3d/ml/paddle/misc/VoxelizeOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -58,7 +58,7 @@ std::vector<paddle::Tensor> Voxelize(paddle::Tensor& points,
voxel_batch_splits}; \
}

if (points.is_gpu()) {
if (points.is_gpu() || points.is_custom_device()) {
#ifdef BUILD_CUDA_MODULE
// pass to cuda function
CALL(float, VoxelizeCUDA)
Expand Down
4 changes: 2 additions & 2 deletions cpp/open3d/ml/paddle/sparse_conv/SparseConvOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -74,7 +74,7 @@ std::vector<paddle::Tensor> SparseConvForward(
return {out_features}; \
}

if (inp_features.is_gpu()) {
if (inp_features.is_gpu() || inp_features.is_custom_device()) {
#ifdef BUILD_CUDA_MODULE
CALL(float, float, int32_t, uint8_t, ::SparseConvCUDA)
#else
Expand Down Expand Up @@ -166,7 +166,7 @@ std::vector<paddle::Tensor> SparseConvBackward(
}

bool dispatch_success = false;
if (inp_features.is_gpu()) {
if (inp_features.is_gpu() || inp_features.is_custom_device()) {
#ifdef BUILD_CUDA_MODULE
CALL(float, float, int32_t, uint8_t, CUDA)
#else
Expand Down
4 changes: 2 additions & 2 deletions cpp/open3d/ml/paddle/sparse_conv/SparseConvTransposeOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -87,7 +87,7 @@ std::vector<paddle::Tensor> SparseConvTransposeForward(
return {out_features}; \
}

if (inp_features.is_gpu()) {
if (inp_features.is_gpu() || inp_features.is_custom_device()) {
#ifdef BUILD_CUDA_MODULE
CALL(float, float, int32_t, uint8_t, ::SparseConvTransposeCUDA)
#else
Expand Down Expand Up @@ -176,7 +176,7 @@ std::vector<paddle::Tensor> SparseConvTransposeBackward(
}

bool dispatch_success = false;
if (inp_features.is_gpu()) {
if (inp_features.is_gpu() || inp_features.is_custom_device()) {
#ifdef BUILD_CUDA_MODULE
CALL(float, float, int32_t, uint8_t, CUDA)
#else
Expand Down