saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-29 20:14:01 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

[QNN EP] Improve Slice to support opset 9 (#15186)

Browse source 
### Description
Improve Slice to support Onnx opset9 which has starts, ends & axes in
node attributes.

### Motivation and Context
To unblock some models.
This commit is contained in:
Hector Li 2023-03-24 16:07:06 -07:00 committed by GitHub
parent d834ec895a
commit 5a2e43bdd5
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 35 additions and 12 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

47
onnxruntime/core/providers/qnn/builder/opbuilder/slice_op_builder.cc
View file

@ -13,9 +13,6 @@
namespace onnxruntime {
namespace qnn {
const int SLICE_MIN_INPUT = 3;
const int SLICE_MAX_INPUT = 5;
class SliceOpBuilder : public BaseOpBuilder {
public:
SliceOpBuilder() : BaseOpBuilder("SliceOpBuilder") {}
@ -38,6 +35,10 @@ class SliceOpBuilder : public BaseOpBuilder {
private:
Status ExplictOpCheck(QnnModelWrapper& qnn_model_wrapper, const NodeUnit& node_unit) const;
void GetDataFromAttribute(const NodeUnit& node_unit,
TensorShapeVector& raw_starts,
TensorShapeVector& raw_ends,
TensorShapeVector& raw_axes) const;
typedef struct {
int32_t begin, end, stride;
} Range;
@ -46,19 +47,36 @@ class SliceOpBuilder : public BaseOpBuilder {
Status SliceOpBuilder::ExplictOpCheck(QnnModelWrapper& qnn_model_wrapper, const NodeUnit& node_unit) const {
size_t input_count = node_unit.Inputs().size();
ORT_RETURN_IF_NOT(input_count >= SLICE_MIN_INPUT && input_count <= SLICE_MAX_INPUT,
"For ONNX Slice operation the expected number of inputs is between 3 and 5.");
// Skip the first input. All other input need to be initializer
for (size_t i = 1; i < node_unit.Inputs().size(); i++) {
const auto& next_input = node_unit.Inputs()[i].node_arg.Name();
if (!qnn_model_wrapper.IsInitializerInput(next_input)) {
return ORT_MAKE_STATUS(ONNXRUNTIME, FAIL, "QNN desn't support dynamic slice.");
// Op set 9 only has 1 input with starts, ends, axes attribute
// Op set > 9, starts, ends, axes are from node input
if (input_count > 1) {
// Skip the first input. All other input need to be initializer
for (size_t i = 1; i < input_count; i++) {
const auto& next_input = node_unit.Inputs()[i].node_arg.Name();
if (!qnn_model_wrapper.IsInitializerInput(next_input)) {
return ORT_MAKE_STATUS(ONNXRUNTIME, FAIL, "QNN desn't support dynamic slice.");
}
}
}
return Status::OK();
}
void SliceOpBuilder::GetDataFromAttribute(const NodeUnit& node_unit,
TensorShapeVector& raw_starts,
TensorShapeVector& raw_ends,
TensorShapeVector& raw_axes) const {
NodeAttrHelper node_helper(node_unit);
auto starts = node_helper.Get("starts", std::vector<int64_t>{0});
raw_starts.assign(starts.begin(), starts.end());
auto ends = node_helper.Get("ends", std::vector<int64_t>{0});
raw_ends.assign(ends.begin(), ends.end());
if (node_helper.HasAttr("axes")) {
auto axes = node_helper.Get("axes", std::vector<int64_t>{0});
raw_axes.assign(axes.begin(), axes.end());
}
}
// Note: For ONNX Slice operation the expected number of inputs is between 3 and 5
Status SliceOpBuilder::ProcessInputs(QnnModelWrapper& qnn_model_wrapper,
const NodeUnit& node_unit,
@ -79,7 +97,13 @@ Status SliceOpBuilder::ProcessInputs(QnnModelWrapper& qnn_model_wrapper,
std::vector<uint32_t> input0_shape;
auto inputs = node_unit.Inputs();
for (size_t input_i = 0; input_i < inputs.size(); ++input_i) {
auto input_count = inputs.size();
// Opset 9, only 1 input, starts, ends, axes are in attribute
if (1 == input_count) {
GetDataFromAttribute(node_unit, raw_starts, raw_ends, raw_axes);
}
for (size_t input_i = 0; input_i < input_count; ++input_i) {
auto& input_name = inputs[input_i].node_arg.Name();
if (input_name.empty()) {
// Ignore unspecified/unused optional input
@ -144,7 +168,6 @@ Status SliceOpBuilder::ProcessInputs(QnnModelWrapper& qnn_model_wrapper,
continue;
}
}
ORT_ENFORCE(input_i == 0, "QNN ReluMinMax operator expects only one input. other inputs, starts, ends, axes and steps are expected to be parameters, ie. initializer inputs in ONNX model");
input0_shape = input_shape;
input_names.push_back(input_name);