[GPU] add convert precision for converting bool to u8 (#28654)

### Details:
- *Add abs + ceiling + clamp before convert when convert is changed from
bool to u8*

### Tickets:
 - *CVS-160217*

src/plugins/intel_gpu/include/intel_gpu/plugin/transformations_pipeline.hpp

@@ -11,6 +11,7 @@
 
 #include "intel_gpu/runtime/execution_config.hpp"
 #include "intel_gpu/runtime/device.hpp"
+#include "transformations/convert_precision.hpp"
 
 namespace ov::intel_gpu {
 
@@ -21,6 +22,8 @@ class TransformationsPipeline {
     void apply(std::shared_ptr<ov::Model> func);
 
 private:
+    static bool fuse_type_to_convert(const std::shared_ptr<ov::Node>& node, const precisions_map& precisions);
+
     const ExecutionConfig& config;
     std::shared_ptr<RemoteContextImpl> m_context;
     cldnn::device_info device_info;

src/plugins/intel_gpu/src/plugin/transformations_pipeline.cpp

@@ -12,6 +12,7 @@
 #include <tuple>
 #include <vector>
 
+#include "openvino/opsets/opset10.hpp"
 #include "intel_gpu/plugin/transformations_pipeline.hpp"
 #include "intel_gpu/runtime/debug_configuration.hpp"
 #include "intel_gpu/runtime/itt.hpp"
@@ -281,6 +282,49 @@ extern bool query_microkernels_supported(cldnn::engine& e, const cldnn::Executio
 
 namespace ov::intel_gpu {
 
+bool TransformationsPipeline::fuse_type_to_convert(const std::shared_ptr<ov::Node>& node, const precisions_map& precisions) {
+    auto convert = ov::as_type_ptr<ov::opset10::Convert>(node);
+    if (!convert)
+        return false;
+    const auto& from = node->get_output_element_type(0);
+    auto it = precisions.find(from);
+    if (it == precisions.end())
+        return false;
+    const auto& to = it->second;
+
+    if (convert->get_convert_element_type() == ov::element::boolean && to.is_integral_number()) {
+        // For Convert node, converting precision from numerical data types to boolean will lead to mathematical
+        // error, because here the output precision boolean is replaced by u8:
+        //  - floating point value 0.01 is converted to be 1 for boolean, but 0 for u8 - need to insert Ceil.
+        //  - either float or int values should be clipped with the interval [0; 1] to mimic bool cast behavior, i.e.
+        //  0 - is false, 1 - is true
+        //  - to perform clamping correctly an Abs op should be inserted before Clamp
+        // Thus an Abs, Ceil and Clamp nodes should be added before the Convert node for this scenario.
+        ov::pass::NodeRegistry reg;
+        const auto& in_prec = convert->get_input_element_type(0);
+        auto parent_node = convert->input_value(0).get_node_shared_ptr();
+        auto item = precisions.find(in_prec);
+        if (item != precisions.end()) {
+            // Add convert node for unsupported precision, such as FP64 or INT64
+            parent_node = reg.make<ov::opset10::Convert>(parent_node, item->second);
+        }
+        if (in_prec.is_signed()) {
+            parent_node = reg.make<ov::opset10::Abs>(parent_node);
+        }
+        if (in_prec.is_real()) {
+            parent_node = reg.make<ov::opset10::Ceiling>(parent_node);
+        }
+        parent_node = reg.make<ov::opset10::Clamp>(parent_node, 0, 1);
+        const auto new_convert = reg.make<ov::opset10::Convert>(parent_node, to);
+        new_convert->set_friendly_name(convert->get_friendly_name());
+        ov::copy_runtime_info(convert, reg.get());
+        ov::replace_node(convert, new_convert);
+        return true;
+    }
+    convert->set_convert_element_type(to);
+    return true;
+}
+
 void TransformationsPipeline::apply(std::shared_ptr<ov::Model> func) {
     OV_ITT_SCOPED_TASK(itt::domains::intel_gpu_plugin, "TransformationsPipeline::apply");
     using const_node_ptr = const std::shared_ptr<const ov::Node>;
@@ -403,6 +447,7 @@ void TransformationsPipeline::apply(std::shared_ptr<ov::Model> func) {
         const bool keep_precision_sensitive_in_fp32_1 = true;
         const bool convert_input_output_precision = false;
         const bool store_original_precision_as_rt_attribute = true;
+
         manager.register_pass<ov::pass::ConvertPrecision>(fp_convert_precision_map,
                                                           empty_fuse_map,
                                                           keep_precision_sensitive_in_fp32_1,
@@ -514,8 +559,11 @@ void TransformationsPipeline::apply(std::shared_ptr<ov::Model> func) {
 
         manager.register_pass<ov::pass::Validate>();
         const bool keep_precision_sensitive_in_fp32_2 = true;
+
+        // To convert to f16 input to boolean which is converted to u8, add abs + ceiling + clamp before convert.
+        type_to_fuse_map type_to_fuse = {{ov::opset10::Convert::get_type_info_static(), fuse_type_to_convert}};
         manager.register_pass<ov::pass::ConvertPrecision>(int_convert_precision_map,
-                                                          empty_fuse_map,
+                                                          type_to_fuse,
                                                           keep_precision_sensitive_in_fp32_2,
                                                           convert_input_output_precision);
 

src/plugins/intel_gpu/tests/functional/shared_tests_instances/single_layer_tests/conversion.cpp

@@ -4,11 +4,76 @@
 
 #include <vector>
 
-#include "single_op_tests/conversion.hpp"
 #include "common_test_utils/test_constants.hpp"
+#include "common_test_utils/data_utils.hpp"
+#include "common_test_utils/ov_tensor_utils.hpp"
+#include "conversion.hpp"
+
+namespace ov {
+namespace test {
+
+void ConvertToBooleanLayerTest::generate_inputs(const std::vector<ov::Shape>& targetInputStaticShapes) {
+        inputs.clear();
+        const auto& funcInputs = function->inputs();
+
+        auto shape = targetInputStaticShapes.front();
+        auto size = shape_size(shape);
+        auto input_type = funcInputs[0].get_element_type();
+
+        ov::Tensor tensor = ov::Tensor(input_type, shape);
+        const auto first_part_size = size / 2;
+        const auto second_part_size = size - first_part_size;
+
+        // 1). Validate the nearest to zero values (Abs + Ceil)
+        {
+                double start_from = -2;
+                uint32_t range = 4;
+                int32_t resolution = size;
+                if (input_type == ov::element::f32) {
+                auto* rawBlobDataPtr = static_cast<float*>(tensor.data());
+                ov::test::utils::fill_data_random(rawBlobDataPtr, first_part_size, range, start_from, resolution);
+                } else if (input_type == ov::element::f16) {
+                auto* rawBlobDataPtr = static_cast<ov::float16*>(tensor.data());
+                ov::test::utils::fill_data_random(rawBlobDataPtr, first_part_size, range, start_from, resolution);
+                } else {
+                FAIL() << "Generating inputs with precision " << input_type.to_string() << " isn't supported, if output precision is boolean.";
+                }
+        }
+
+        // 2). Validate the values that are more than UINT8_MAX in absolute (Abs + Min)
+        {
+                ov::test::utils::InputGenerateData in_data_neg;
+                double neg_start_from = -1.5 * std::numeric_limits<uint8_t>::max();
+                double pos_start_from = 0.5 * std::numeric_limits<uint8_t>::max();
+                uint32_t range = 256;
+                auto neg_size = second_part_size / 2;
+                auto pos_size = second_part_size - neg_size;
+                int32_t resolution = 1;
+
+                if (input_type == ov::element::f32) {
+                auto* rawBlobDataPtr = static_cast<float*>(tensor.data());
+                ov::test::utils::fill_data_random(rawBlobDataPtr + first_part_size, neg_size, range, neg_start_from, resolution);
+                ov::test::utils::fill_data_random(rawBlobDataPtr + first_part_size + neg_size, pos_size, range, pos_start_from, resolution);
+                } else if (input_type == ov::element::f16) {
+                auto* rawBlobDataPtr = static_cast<ov::float16*>(tensor.data());
+                ov::test::utils::fill_data_random(rawBlobDataPtr + first_part_size, neg_size, range, neg_start_from, resolution);
+                ov::test::utils::fill_data_random(rawBlobDataPtr + first_part_size + neg_size, pos_size, range, pos_start_from, resolution);
+                } else {
+                FAIL() << "Generating inputs with precision " << input_type.to_string() << " isn't supported, if output precision is boolean.";
+                }
+        }
+
+        inputs.insert({funcInputs[0].get_node_shared_ptr(), tensor});
+}
+
+}  // namespace test
+}  // namespace ov
+
 
 namespace {
 using ov::test::ConversionLayerTest;
+using ov::test::ConvertToBooleanLayerTest;
+
 const std::vector<ov::test::utils::ConversionTypes> conversionOpTypes = {
     ov::test::utils::ConversionTypes::CONVERT,
     ov::test::utils::ConversionTypes::CONVERT_LIKE,
@@ -32,4 +97,22 @@ INSTANTIATE_TEST_SUITE_P(smoke_NoReshape, ConversionLayerTest,
                                 ::testing::Values(ov::test::utils::DEVICE_GPU)),
                         ConversionLayerTest::getTestCaseName);
 
+TEST_P(ConvertToBooleanLayerTest, CompareWithRefs) {
+    run();
+};
+
+const std::vector<ov::element::Type> precisions_floating_point = {
+        ov::element::f32,
+        ov::element::f16
+};
+
+INSTANTIATE_TEST_SUITE_P(smoke_NoReshape, ConvertToBooleanLayerTest,
+                        ::testing::Combine(
+                                ::testing::ValuesIn({ov::test::utils::ConversionTypes::CONVERT}),
+                                ::testing::ValuesIn(ov::test::static_shapes_to_test_representation(inShape)),
+                                ::testing::ValuesIn(precisions_floating_point),
+                                ::testing::Values(ov::element::boolean),
+                                ::testing::Values(ov::test::utils::DEVICE_GPU)),
+                        ConvertToBooleanLayerTest::getTestCaseName);
+
 }  // namespace

src/plugins/intel_gpu/tests/functional/shared_tests_instances/single_layer_tests/conversion.hpp

@@ -0,0 +1,22 @@
+// Copyright (C) 2023 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#pragma once
+
+#include "shared_test_classes/base/ov_subgraph.hpp"
+#include "common_test_utils/ov_tensor_utils.hpp"
+#include "gtest/gtest.h"
+#include "single_op_tests/conversion.hpp"
+
+
+namespace ov {
+namespace test {
+
+class ConvertToBooleanLayerTest : public ConversionLayerTest {
+protected:
+    void generate_inputs(const std::vector<ov::Shape>& targetInputStaticShapes) override;
+};
+
+}  // namespace test
+}  // namespace ov