IntenseWebs/openvino
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[PT FE] Remove warnings suppression in frontend code (#15676)

Browse Source 
* [PT FE] Remove warnings suppression in frontend code

* Fix code style

* Fix linux build

* Fix build

* Fix typo
This commit is contained in:
Maxim Vafin 2023-02-13 22:35:53 +01:00 committed by GitHub
parent 5da21f5649
commit efe3b27f5b
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
8 changed files with 31 additions and 35 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
src/frontends/pytorch/CMakeLists.txt
View File

@ -2,10 +2,4 @@
# SPDX-License-Identifier: Apache-2.0 # SPDX-License-Identifier: Apache-2.0
# #
if(CMAKE_CXX_COMPILER_ID STREQUAL "MSVC")
# 'argument': conversion from 'size_t' to 'int', possible loss of data
ie_add_compiler_flags(/wd4267)
ie_add_compiler_flags(/wd4244)
endif()
add_subdirectory(src) add_subdirectory(src)

2
src/frontends/pytorch/src/op/eye.cpp
View File

@ -22,7 +22,7 @@ OutputVector translate_eye(NodeContext& context) {
// num rows and cols should be integer, but at the moment conversion their data type can be unknown yet // num rows and cols should be integer, but at the moment conversion their data type can be unknown yet
x = context.mark_node(std::make_shared<v0::Convert>(x, element::i64)); x = context.mark_node(std::make_shared<v0::Convert>(x, element::i64));
Output<Node> y; Output<Node> y;
size_t dtype_id; int dtype_id;
auto dtype = element::f32; auto dtype = element::f32;
// aten::eye support only main diagonal // aten::eye support only main diagonal
auto diagonal = context.mark_node(v0::Constant::create(element::i64, Shape{}, {0})); auto diagonal = context.mark_node(v0::Constant::create(element::i64, Shape{}, {0}));

2
src/frontends/pytorch/src/op/repeat_interleave.cpp
View File

@ -26,7 +26,7 @@ OutputVector generate_indices_from_repeats_tensor(const NodeContext& context, co
OutputVector all_indices; OutputVector all_indices;
for (size_t i = 0; i < repeats.size(); i++) { for (size_t i = 0; i < repeats.size(); i++) {
Shape indices_shape{static_cast<size_t>(repeats.at(i))}; Shape indices_shape{static_cast<size_t>(repeats.at(i))};
std::vector<int32_t> indices_vec(repeats.at(i), i); std::vector<int32_t> indices_vec(repeats.at(i), static_cast<int32_t>(i));
auto indices = context.mark_node(v0::Constant::create(element::i32, indices_shape, indices_vec)); auto indices = context.mark_node(v0::Constant::create(element::i32, indices_shape, indices_vec));
all_indices.push_back(indices); all_indices.push_back(indices);
} }

2
src/frontends/pytorch/src/pt_framework_node.hpp
View File

@ -47,7 +47,7 @@ public:
auto op = std::make_shared<PtFrameworkNode>(m_decoder, inputs, get_output_size()); auto op = std::make_shared<PtFrameworkNode>(m_decoder, inputs, get_output_size());
for (size_t body_index = 0; body_index < m_bodies.size(); ++body_index) { for (size_t body_index = 0; body_index < m_bodies.size(); ++body_index) {
op->set_function(body_index, get_function(body_index)->clone()); op->set_function(static_cast<int>(body_index), get_function(static_cast<int>(body_index))->clone());
for (const auto& m_input_descr : m_input_descriptions[body_index]) { for (const auto& m_input_descr : m_input_descriptions[body_index]) {
op->m_input_descriptions[body_index].push_back(m_input_descr->copy()); op->m_input_descriptions[body_index].push_back(m_input_descr->copy());
} }

11
src/frontends/pytorch/src/transforms/append_list_unpack_replacer.cpp
View File

@ -62,11 +62,12 @@ AppendListUnpackReplacer::AppendListUnpackReplacer() {
auto getitem_index_ptr = getitem_node->input_value(1).get_node_shared_ptr(); auto getitem_index_ptr = getitem_node->input_value(1).get_node_shared_ptr();
auto getitem_index_const = std::dynamic_pointer_cast<ov::op::v0::Constant>(getitem_index_ptr); auto getitem_index_const = std::dynamic_pointer_cast<ov::op::v0::Constant>(getitem_index_ptr);
auto index_val = getitem_index_const->cast_vector<int64_t>(); auto index_val = getitem_index_const->cast_vector<int64_t>();
auto index = 0; if (index_val.size() != 1) {
if (index_val[0] >= 0) { return false;
index = index_val[0]; }
} else { auto index = index_val[0];
index = inputs.size() + index_val[0]; if (index_val[0] < 0) {
index = inputs.size() + index;
} }
auto axis_0 = ov::op::v0::Constant::create(element::i64, Shape{}, {0}); auto axis_0 = ov::op::v0::Constant::create(element::i64, Shape{}, {0});
auto split = std::make_shared<ov::op::v1::Split>(inputs[index], axis_0, list_unpack->get_output_size()); auto split = std::make_shared<ov::op::v1::Split>(inputs[index], axis_0, list_unpack->get_output_size());

13
src/frontends/pytorch/src/transforms/aten_getitem_replacer.cpp
View File

@ -93,15 +93,16 @@ AtenGetItemReplacer::AtenGetItemReplacer() {
} else { } else {
auto getitem_index_ptr = getitem->input_value(1).get_node_shared_ptr(); auto getitem_index_ptr = getitem->input_value(1).get_node_shared_ptr();
auto getitem_index_const = std::dynamic_pointer_cast<ov::op::v0::Constant>(getitem_index_ptr); auto getitem_index_const = std::dynamic_pointer_cast<ov::op::v0::Constant>(getitem_index_ptr);
auto index_val = getitem_index_const->cast_vector<int64_t>();
auto split = std::make_shared<ov::op::v1::VariadicSplit>(torch_split->get_input_source_output(0), auto split = std::make_shared<ov::op::v1::VariadicSplit>(torch_split->get_input_source_output(0),
torch_split->get_input_source_output(2), torch_split->get_input_source_output(2),
torch_split->get_input_source_output(1)); torch_split->get_input_source_output(1));
auto index = 0; auto index_val = getitem_index_const->cast_vector<int64_t>();
if (index_val[0] >= 0) { if (index_val.size() != 1) {
index = index_val[0]; return false;
} else { }
index = split->outputs().size() + index_val[0]; auto index = index_val[0];
if (index < 0) {
index = split->outputs().size() + index;
} }
OutputVector res{split->outputs()[index]}; OutputVector res{split->outputs()[index]};
copy_runtime_info({getitem, input_node}, split); copy_runtime_info({getitem, input_node}, split);

26
src/frontends/pytorch/src/utils.cpp
View File

@ -24,7 +24,7 @@ void num_inputs_check(const NodeContext& context, size_t min_inputs, size_t max_
Output<Node> make_optional_bias(const Output<Node>& base_op, Output<Node> make_optional_bias(const Output<Node>& base_op,
const NodeContext& context, const NodeContext& context,
size_t bias_input_idx, int bias_input_idx,
const std::vector<int>& unsqueeze_dims) { const std::vector<int>& unsqueeze_dims) {
using std::make_shared; using std::make_shared;
@ -105,7 +105,7 @@ Output<Node> reshape_kernel_for_group(const NodeContext& context, const Output<N
return make_shared<opset10::Reshape>(kernel, new_kernel_shape, false); return make_shared<opset10::Reshape>(kernel, new_kernel_shape, false);
} }
std::shared_ptr<Node> get_axes_range(const NodeContext& context, size_t input_id) { std::shared_ptr<Node> get_axes_range(const NodeContext& context, int input_id) {
auto x = context.get_input(input_id); auto x = context.get_input(input_id);
auto start = std::make_shared<opset10::Constant>(element::i32, Shape{}, 0); auto start = std::make_shared<opset10::Constant>(element::i32, Shape{}, 0);
auto step = std::make_shared<opset10::Constant>(element::i32, Shape{}, 1); auto step = std::make_shared<opset10::Constant>(element::i32, Shape{}, 1);
@ -122,15 +122,15 @@ std::shared_ptr<Node> numel(const NodeContext& context, const Output<Node>& x) {
}; };
namespace { namespace {
const std::unordered_map<int, element::Type> TORCH_TO_OV_TYPE{{0, element::u8}, const std::unordered_map<int64_t, element::Type> TORCH_TO_OV_TYPE{{0, element::u8},
{1, element::i8}, {1, element::i8},
{2, element::i16}, {2, element::i16},
{3, element::i32}, {3, element::i32},
{4, element::i64}, {4, element::i64},
{5, element::f16}, {5, element::f16},
{6, element::f32}, {6, element::f32},
{7, element::f64}, {7, element::f64},
{11, element::boolean}}; {11, element::boolean}};
const std::unordered_map<std::string, ov::op::PadType> TORCH_AUTO_PAD_TO_OV{{"valid", ov::op::PadType::VALID}, const std::unordered_map<std::string, ov::op::PadType> TORCH_AUTO_PAD_TO_OV{{"valid", ov::op::PadType::VALID},
{"same", ov::op::PadType::SAME_UPPER}}; {"same", ov::op::PadType::SAME_UPPER}};
@ -222,7 +222,7 @@ OutputVector make_framework_node(NodeContext* context) {
} }
// Number of body outputs can be higher then number of pt node outputs, e.g. in case of loop first body output is // Number of body outputs can be higher then number of pt node outputs, e.g. in case of loop first body output is
// condition, we have to skip such outputs. // condition, we have to skip such outputs.
int num_skip_body_outputs = auto num_skip_body_outputs =
num_body_outs > context->get_output_size() ? num_body_outs - context->get_output_size() : 0; num_body_outs > context->get_output_size() ? num_body_outs - context->get_output_size() : 0;
// We need to reduce number of outputs, because some outputs are outputs from body // We need to reduce number of outputs, because some outputs are outputs from body
@ -232,7 +232,7 @@ OutputVector make_framework_node(NodeContext* context) {
context->get_output_size() - num_body_outs + num_skip_body_outputs); context->get_output_size() - num_body_outs + num_skip_body_outputs);
fw_node->set_friendly_name(context->get_op_type()); fw_node->set_friendly_name(context->get_op_type());
for (size_t i = 0; i < bodies.size(); ++i) { for (size_t i = 0; i < bodies.size(); ++i) {
fw_node->set_function(i, bodies[i]); fw_node->set_function(static_cast<int>(i), bodies[i]);
} }
// Connect inputs with external context // Connect inputs with external context

4
src/frontends/pytorch/src/utils.hpp
View File

@ -22,7 +22,7 @@ void num_inputs_check(const NodeContext& context, size_t min_inputs, size_t max_
Output<Node> make_optional_bias(const Output<Node>& base_op, Output<Node> make_optional_bias(const Output<Node>& base_op,
const NodeContext& context, const NodeContext& context,
size_t bias_input_idx, int bias_input_idx,
const std::vector<int>& unsqueeze_dims = {}); const std::vector<int>& unsqueeze_dims = {});
Output<Node> reshape_channelwise(const NodeContext& context, Output<Node> reshape_channelwise(const NodeContext& context,
@ -36,7 +36,7 @@ std::tuple<Output<Node>, Output<Node>> get_shape_rank(const NodeContext& context
Output<Node> reshape_kernel_for_group(const NodeContext& context, const Output<Node>& kernel, int64_t groups); Output<Node> reshape_kernel_for_group(const NodeContext& context, const Output<Node>& kernel, int64_t groups);
std::shared_ptr<Node> get_axes_range(const NodeContext& context, size_t input_id); std::shared_ptr<Node> get_axes_range(const NodeContext& context, int input_id);
std::shared_ptr<Node> numel(const NodeContext& context, const Output<Node>& x); std::shared_ptr<Node> numel(const NodeContext& context, const Output<Node>& x);