Skip to content

Commit

Permalink
Merge branch 'master' into ci/gha/risc-v-sccache
Browse files Browse the repository at this point in the history
  • Loading branch information
@akashchi
akashchi authored Jan 25, 2024
2 parents 221d65c + ff681cc commit c98315b
Show file tree
Hide file tree
Showing 44 changed files with 276 additions and 379 deletions.
16 changes: 7 additions & 9 deletions src/frontends/onnx/frontend/src/op/qlinear_conv.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -7,10 +7,6 @@

#include "op/qlinear_conv.hpp"

#include <cstddef>
#include <memory>
#include <vector>

#include "conv.hpp"
#include "dequantize_linear.hpp"
#include "exceptions.hpp"
Expand All @@ -19,6 +15,8 @@
#include "openvino/op/multiply.hpp"
#include "quantize_linear.hpp"

using namespace ov::op;

OPENVINO_SUPPRESS_DEPRECATED_START
namespace ngraph {
namespace onnx_import {
Expand All @@ -35,22 +33,22 @@ OutputVector qlinear_conv(const Node& node) {
auto w_zero_point = inputs.at(5);
auto y_scale = inputs.at(6);
auto y_zero_point = inputs.at(7);
Output<ngraph::Node> B = inputs.size() > 8 ? inputs.at(8) : std::make_shared<NullNode>()->output(0);
Output<ov::Node> B = inputs.size() > 8 ? inputs.at(8) : std::make_shared<NullNode>()->output(0);

x = set_13::detail::dequantize_linear(x,
x_scale,
std::make_shared<ov::op::v0::Convert>(x_zero_point, element::f32),
std::make_shared<v0::Convert>(x_zero_point, element::f32),
1,
node)[0];
w = set_13::detail::dequantize_linear(w,
w_scale,
std::make_shared<ov::op::v0::Convert>(w_zero_point, element::f32),
std::make_shared<v0::Convert>(w_zero_point, element::f32),
1,
node)[0];

if (!ov::op::util::is_null(B)) {
B = std::make_shared<ov::op::v1::Multiply>(std::make_shared<ov::op::v0::Convert>(B, x_scale.get_element_type()),
std::make_shared<ov::op::v1::Multiply>(x_scale, w_scale))
B = std::make_shared<v1::Multiply>(std::make_shared<v0::Convert>(B, x_scale.get_element_type()),
std::make_shared<v1::Multiply>(x_scale, w_scale))
->output(0);
}

Expand Down
5 changes: 1 addition & 4 deletions src/frontends/onnx/frontend/src/op/qlinear_conv.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,6 @@
#include "openvino/core/deprecated.hpp"
OPENVINO_SUPPRESS_DEPRECATED_START

#include "ngraph/node.hpp"
#include "onnx_import/core/node.hpp"

namespace ngraph {
Expand All @@ -21,14 +20,12 @@ namespace set_1 {
///
/// \param node The ONNX node object representing this operation.
///
/// \return The vector containing Ngraph nodes producing output of ONNX quantizied
/// \return The vector containing OV nodes producing output of ONNX quantizied
/// convolution operation.
OutputVector qlinear_conv(const Node& node);

} // namespace set_1

} // namespace op

} // namespace onnx_import

} // namespace ngraph
Expand Down
10 changes: 4 additions & 6 deletions src/frontends/onnx/frontend/src/op/qlinear_matmul.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -4,16 +4,14 @@

#include "op/qlinear_matmul.hpp"

#include <cstddef>
#include <memory>
#include <vector>

#include "dequantize_linear.hpp"
#include "matmul.hpp"
#include "openvino/op/convert.hpp"
#include "quantize_linear.hpp"
#include "utils/reshape.hpp"

using namespace ov::op;

OPENVINO_SUPPRESS_DEPRECATED_START
namespace ngraph {
namespace onnx_import {
Expand All @@ -34,13 +32,13 @@ OutputVector qlinear_matmul(const Node& node) {
const auto& dequnatize_a =
set_13::detail::dequantize_linear(a,
a_scale,
std::make_shared<ov::op::v0::Convert>(a_zero_point, element::f32),
std::make_shared<v0::Convert>(a_zero_point, element::f32),
1,
node);
const auto& dequnatize_b =
set_13::detail::dequantize_linear(b,
b_scale,
std::make_shared<ov::op::v0::Convert>(b_zero_point, element::f32),
std::make_shared<v0::Convert>(b_zero_point, element::f32),
1,
node);

Expand Down
3 changes: 1 addition & 2 deletions src/frontends/onnx/frontend/src/op/qlinear_matmul.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,6 @@
#include "openvino/core/deprecated.hpp"
OPENVINO_SUPPRESS_DEPRECATED_START

#include "ngraph/node.hpp"
#include "onnx_import/core/node.hpp"

namespace ngraph {
Expand All @@ -18,7 +17,7 @@ namespace set_1 {
///
/// \param node The ONNX node object representing this operation.
///
/// \return The vector containing Ngraph nodes producing output of ONNX quantizied
/// \return The vector containing OV nodes producing output of ONNX quantizied
/// matrix multiplication operation.
OutputVector qlinear_matmul(const Node& node);
} // namespace set_1
Expand Down
1 change: 0 additions & 1 deletion src/frontends/onnx/frontend/src/op/quant_conv.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,6 @@
#include "openvino/core/deprecated.hpp"
OPENVINO_SUPPRESS_DEPRECATED_START

#include "ngraph/node.hpp"
#include "onnx_import/core/node.hpp"

namespace ngraph {
Expand Down
105 changes: 49 additions & 56 deletions src/frontends/onnx/frontend/src/op/quantize_linear.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -4,36 +4,34 @@

#include "op/quantize_linear.hpp"

#include <cstdint>
#include <memory>
#include <numeric>
#include <tuple>

#include "default_opset.hpp"
#include "exceptions.hpp"
#include "ngraph/axis_set.hpp"
#include "ngraph/shape.hpp"
#include "ngraph/type/element_type.hpp"
#include "ngraph/validation_util.hpp"
#include "openvino/core/validation_util.hpp"
#include "openvino/frontend/exception.hpp"
#include "openvino/op/constant.hpp"
#include "openvino/op/convert.hpp"
#include "openvino/op/fake_quantize.hpp"
#include "openvino/op/multiply.hpp"
#include "openvino/op/subtract.hpp"
#include "ov_models/ov_builders/reshape.hpp"
#include "utils/reshape.hpp"

using namespace ov::op;

OPENVINO_SUPPRESS_DEPRECATED_START
namespace ngraph {
namespace onnx_import {
namespace op {
namespace detail {
namespace {
Output<ngraph::Node> get_zero_point(const OutputVector& inputs) {
Output<ov::Node> get_zero_point(const OutputVector& inputs) {
if (inputs.size() > 2) {
return inputs.at(2);
} else {
return std::make_shared<default_opset::Constant>(element::u8, Shape{1}, std::uint8_t(0));
return std::make_shared<v0::Constant>(element::u8, Shape{1}, std::uint8_t(0));
}
}

void validate_zero_point_type(const Node& onnx_node, const Output<ngraph::Node>& y_zero_point) {
void validate_zero_point_type(const Node& onnx_node, const Output<ov::Node>& y_zero_point) {
const auto& y_zero_point_et = y_zero_point.get_element_type();
CHECK_VALID_NODE(
onnx_node,
Expand All @@ -43,49 +41,48 @@ void validate_zero_point_type(const Node& onnx_node, const Output<ngraph::Node>&
"integer type.");
}

Output<ngraph::Node> validate_scale(const Node& onnx_node, const Output<ngraph::Node>& y_scale) {
Output<ov::Node> validate_scale(const Node& onnx_node, const Output<ov::Node>& y_scale) {
const auto& y_scale_et = y_scale.get_element_type();
CHECK_VALID_NODE(onnx_node, y_scale_et.is_static(), "\"y_scale\" input data type must be static.");
if (y_scale_et != element::f32) {
return std::make_shared<default_opset::Convert>(y_scale, element::f32);
return std::make_shared<v0::Convert>(y_scale, element::f32);
}
return y_scale;
}

Output<ngraph::Node> validate_data(const Node& onnx_node, const Output<ngraph::Node>& data) {
Output<ov::Node> validate_data(const Node& onnx_node, const Output<ov::Node>& data) {
const auto& data_et = data.get_element_type();
CHECK_VALID_NODE(onnx_node, data_et.is_static(), "\"x\" input data type must be static.");

if (data_et != element::f32) {
return std::make_shared<default_opset::Convert>(data, element::f32);
return std::make_shared<v0::Convert>(data, element::f32);
}
return data;
}

std::tuple<std::shared_ptr<ngraph::Node>, std::shared_ptr<ngraph::Node>> get_output_bands(
const element::Type& destination_type,
const element::Type& data_type) {
std::shared_ptr<ngraph::Node> output_low;
std::shared_ptr<ngraph::Node> output_high;
std::tuple<std::shared_ptr<ov::Node>, std::shared_ptr<ov::Node>> get_output_bands(const element::Type& destination_type,
const element::Type& data_type) {
std::shared_ptr<ov::Node> output_low;
std::shared_ptr<ov::Node> output_high;

// These values could be used in a ConvertQuantizeDequantize transformation and
// should be aligned
switch (destination_type) {
case element::i8:
output_low = std::make_shared<default_opset::Constant>(data_type, Shape{1}, -128);
output_high = std::make_shared<default_opset::Constant>(data_type, Shape{1}, 127);
output_low = std::make_shared<v0::Constant>(data_type, Shape{1}, -128);
output_high = std::make_shared<v0::Constant>(data_type, Shape{1}, 127);
break;
case element::u8:
output_low = std::make_shared<default_opset::Constant>(data_type, Shape{1}, 0);
output_high = std::make_shared<default_opset::Constant>(data_type, Shape{1}, 255);
output_low = std::make_shared<v0::Constant>(data_type, Shape{1}, 0);
output_high = std::make_shared<v0::Constant>(data_type, Shape{1}, 255);
break;
case element::i16:
output_low = std::make_shared<default_opset::Constant>(data_type, Shape{1}, -32768);
output_high = std::make_shared<default_opset::Constant>(data_type, Shape{1}, 32767);
output_low = std::make_shared<v0::Constant>(data_type, Shape{1}, -32768);
output_high = std::make_shared<v0::Constant>(data_type, Shape{1}, 32767);
break;
case element::u16:
output_low = std::make_shared<default_opset::Constant>(data_type, Shape{1}, 0);
output_high = std::make_shared<default_opset::Constant>(data_type, Shape{1}, 65535);
output_low = std::make_shared<v0::Constant>(data_type, Shape{1}, 0);
output_high = std::make_shared<v0::Constant>(data_type, Shape{1}, 65535);
break;
default:
OPENVINO_THROW("Unsupported element type for QuantizeLinear");
Expand All @@ -95,27 +92,23 @@ std::tuple<std::shared_ptr<ngraph::Node>, std::shared_ptr<ngraph::Node>> get_out
return std::make_tuple(output_low, output_high);
}

std::tuple<std::shared_ptr<ngraph::Node>, std::shared_ptr<ngraph::Node>> get_input_bands(
const Output<ngraph::Node>& y_scale,
const Output<ngraph::Node>& y_zero_point,
const std::shared_ptr<ngraph::Node>& output_low,
const std::shared_ptr<ngraph::Node>& output_high,
std::tuple<std::shared_ptr<ov::Node>, std::shared_ptr<ov::Node>> get_input_bands(
const Output<ov::Node>& y_scale,
const Output<ov::Node>& y_zero_point,
const std::shared_ptr<ov::Node>& output_low,
const std::shared_ptr<ov::Node>& output_high,
const element::Type& data_type) {
std::shared_ptr<ngraph::Node> input_low;
std::shared_ptr<ngraph::Node> input_high;
const auto& zero_point = std::make_shared<default_opset::Convert>(y_zero_point, data_type);
std::shared_ptr<ov::Node> input_low;
std::shared_ptr<ov::Node> input_high;
const auto& zero_point = std::make_shared<v0::Convert>(y_zero_point, data_type);

input_low =
std::make_shared<default_opset::Multiply>(y_scale,
std::make_shared<default_opset::Subtract>(output_low, zero_point));
input_low = std::make_shared<v1::Multiply>(y_scale, std::make_shared<v1::Subtract>(output_low, zero_point));
OPENVINO_SUPPRESS_DEPRECATED_START
if (auto constant = ov::get_constant_from_source(input_low)) {
OPENVINO_SUPPRESS_DEPRECATED_END
input_low = constant;
}
input_high =
std::make_shared<default_opset::Multiply>(y_scale,
std::make_shared<default_opset::Subtract>(output_high, zero_point));
input_high = std::make_shared<v1::Multiply>(y_scale, std::make_shared<v1::Subtract>(output_high, zero_point));
OPENVINO_SUPPRESS_DEPRECATED_START
if (auto constant = ov::get_constant_from_source(input_high)) {
OPENVINO_SUPPRESS_DEPRECATED_END
Expand All @@ -125,25 +118,25 @@ std::tuple<std::shared_ptr<ngraph::Node>, std::shared_ptr<ngraph::Node>> get_inp
return std::make_tuple(input_low, input_high);
}
} // namespace
std::shared_ptr<ngraph::Node> make_fake_quantize(const Output<ngraph::Node>& y_scale,
const Output<ngraph::Node>& y_zero_point,
const Output<ngraph::Node>& data) {
std::shared_ptr<ov::Node> make_fake_quantize(const Output<ov::Node>& y_scale,
const Output<ov::Node>& y_zero_point,
const Output<ov::Node>& data) {
const element::Type& destination_type = y_zero_point.get_element_type();
const element::Type& data_type = data.get_element_type();

std::shared_ptr<ngraph::Node> output_low;
std::shared_ptr<ngraph::Node> output_high;
std::shared_ptr<ov::Node> output_low;
std::shared_ptr<ov::Node> output_high;
std::tie(output_low, output_high) = detail::get_output_bands(destination_type, data_type);

std::shared_ptr<ngraph::Node> input_low;
std::shared_ptr<ngraph::Node> input_high;
std::shared_ptr<ov::Node> input_low;
std::shared_ptr<ov::Node> input_high;
std::tie(input_low, input_high) =
detail::get_input_bands(y_scale, y_zero_point, output_low, output_high, data_type);

const std::size_t levels = static_cast<size_t>(1) << destination_type.bitwidth();

return std::make_shared<default_opset::Convert>(
std::make_shared<default_opset::FakeQuantize>(data, input_low, input_high, output_low, output_high, levels),
return std::make_shared<v0::Convert>(
std::make_shared<v0::FakeQuantize>(data, input_low, input_high, output_low, output_high, levels),
destination_type);
}
} // namespace detail
Expand All @@ -165,9 +158,9 @@ OutputVector quantize_linear(const Node& node) {

namespace set_13 {
namespace {
OutputVector quantize_linear(Output<ngraph::Node> x,
Output<ngraph::Node> y_scale,
Output<ngraph::Node> y_zero_point,
OutputVector quantize_linear(Output<ov::Node> x,
Output<ov::Node> y_scale,
Output<ov::Node> y_zero_point,
int64_t axis,
Node node) {
namespace detail = ngraph::onnx_import::op::detail;
Expand Down
8 changes: 4 additions & 4 deletions src/frontends/onnx/frontend/src/op/quantize_linear.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -7,16 +7,16 @@
#include "openvino/core/deprecated.hpp"
OPENVINO_SUPPRESS_DEPRECATED_START

#include "ngraph/node.hpp"
#include "onnx_import/core/node.hpp"
#include "openvino/core/node.hpp"

namespace ngraph {
namespace onnx_import {
namespace op {
namespace detail {
std::shared_ptr<ngraph::Node> make_fake_quantize(const Output<ngraph::Node>& y_scale,
const Output<ngraph::Node>& y_zero_point,
const Output<ngraph::Node>& data);
std::shared_ptr<ov::Node> make_fake_quantize(const Output<ov::Node>& y_scale,
const Output<ov::Node>& y_zero_point,
const Output<ov::Node>& data);
}
namespace set_1 {
OutputVector quantize_linear(const Node& node);
Expand Down
9 changes: 6 additions & 3 deletions src/frontends/onnx/frontend/src/op/random_normal.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2,12 +2,15 @@
// SPDX-License-Identifier: Apache-2.0
//

#include "op/random_normal.hpp"

#include "exceptions.hpp"
#include "ngraph/shape.hpp"
#include "openvino/frontend/common/random_normal_helper.hpp"
#include "openvino/op/constant.hpp"
#include "utils/common.hpp"

using namespace ov::op;

OPENVINO_SUPPRESS_DEPRECATED_START
namespace ngraph {
namespace onnx_import {
Expand All @@ -23,8 +26,8 @@ OutputVector random_normal(const Node& node) {

const auto mean = node.get_attribute_value<float>("mean", 0.0f);
const auto scale = node.get_attribute_value<float>("scale", 1.0f);
auto scale_node = ov::op::v0::Constant::create(target_type, Shape{1}, {scale});
auto mean_node = ov::op::v0::Constant::create(target_type, Shape{1}, {mean});
auto scale_node = v0::Constant::create(target_type, Shape{1}, {scale});
auto mean_node = v0::Constant::create(target_type, Shape{1}, {mean});

const auto seed = node.get_attribute_value<float>("seed", 0);
const auto shape = node.get_attribute_as_constant<std::vector<int64_t>>("shape");
Expand Down
1 change: 0 additions & 1 deletion src/frontends/onnx/frontend/src/op/random_normal.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,6 @@
#include "openvino/core/deprecated.hpp"
OPENVINO_SUPPRESS_DEPRECATED_START

#include "ngraph/node.hpp"
#include "onnx_import/core/node.hpp"

namespace ngraph {
Expand Down
Loading

0 comments on commit c98315b

Please sign in to comment.