Merge branch 'master' into gn_cleanup

cmake/packaging/debian.cmake

@@ -87,7 +87,7 @@ macro(ov_cpack_settings)
         # - 2022.1.1, 2022.2 do not have debian packages enabled, distributed only as archives
         # - 2022.3 is the first release where Debian updated packages are introduced, others 2022.3.X are LTS
         2022.3.0 2022.3.1 2022.3.2 2022.3.3 2022.3.4 2022.3.5
-        2023.0.0
+        2023.0.0 2023.0.1
         )
 
     #

cmake/packaging/rpm.cmake

@@ -73,7 +73,7 @@ macro(ov_cpack_settings)
         # - 2022.1.1, 2022.2 do not have rpm packages enabled, distributed only as archives
         # - 2022.3 is the first release where RPM updated packages are introduced, others 2022.3.X are LTS
         2022.3.0 2022.3.1 2022.3.2 2022.3.3 2022.3.4 2022.3.5
-        2023.0.0
+        2023.0.0 2023.0.1
         )
 
     find_host_program(rpmlint_PROGRAM NAMES rpmlint DOC "Path to rpmlint")

docs/_static/css/custom.css

@@ -70,6 +70,7 @@ ul#navbar-main-elements li:first-of-type {
 
 ul#navbar-main-elements > li:hover { 
     text-decoration: underline;
+    color: #fff;
 }
 
 

docs/_static/css/homepage_style.css

@@ -3,13 +3,13 @@
 #openvino-documentation > h1 {
     display: none;
 }
-img {
-    cursor: default;
-}
 h1 {
-    font-size: var(--pst-font-size-h2);
-    margin-bottom: 3rem;
+    /*font-size: var(--pst-font-size-h2);*/
+    /*margin-bottom: 3rem;*/
+    display: none!important;
 }
+
+
 #ov-homepage-banner, .openvino-diagram, .ov-homepage-higlight-grid {
     margin-bottom: 90px!important;
 }

docs/home.rst

@@ -1,7 +1,6 @@
-.. OpenVINO Toolkit documentation master file, created by
-   sphinx-quickstart on Wed Jul  7 10:46:56 2021.
-   You can adapt this file completely to your liking, but it should at least
-   contain the root `toctree` directive.
+============================
+OpenVINO 2023.0
+============================
 
 .. meta::
    :google-site-verification: _YqumYQ98cmXUTwtzM_0WIIadtDc6r_TMYGbmGgNvrk
@@ -34,8 +33,6 @@
       :align: center
 
 
-
-
 .. grid:: 2 2 3 3
    :class-container: ov-homepage-higlight-grid
 
@@ -75,11 +72,8 @@
       Reach for performance with post-training and training-time compression with NNCF
 
 
-
-
-
 Feature Overview
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+##############################
 
 .. grid:: 1 2 2 2
    :class-container: ov-homepage-feature-grid
@@ -109,9 +103,6 @@ Feature Overview
 
 
 
-
-
-
 .. toctree::
    :maxdepth: 2
    :hidden:
@@ -122,4 +113,5 @@ Feature Overview
    DOCUMENTATION <documentation>
    MODEL ZOO <model_zoo>
    RESOURCES <resources>
-   RELEASE NOTES <https://software.intel.com/content/www/us/en/develop/articles/openvino-relnotes.html>
+   RELEASE NOTES <release_notes>
+

docs/index.rst

@@ -7,5 +7,5 @@ OpenVINO™ Documentation
    Install <https://www.intel.com/content/www/us/en/developer/tools/openvino-toolkit/download.html> 
    Blog <https://blog.openvino.ai/>
    Forum <https://community.intel.com/t5/Intel-Distribution-of-OpenVINO/bd-p/distribution-openvino-toolkit>
-   Training <https://www.intel.com/content/www/us/en/developer/tools/devcloud/edge/learn/certification.html>
+   Support <https://www.intel.com/content/www/us/en/support/products/96066/software/development-software/openvino-toolkit.html>
    GitHub <https://github.com/openvinotoolkit>

docs/install_guides/installing-openvino-from-archive-windows.md

@@ -75,7 +75,7 @@ Step 1: Download and Install OpenVINO Core Components
       ``C:\Program Files (x86)\Intel`` is the recommended folder. You may also use a different path if desired or if you don't have administrator privileges on your computer.
 
 
-2. Download the `OpenVINO Runtime archive file for Windows <https://storage.openvinotoolkit.org/repositories/openvino/packages/2022.3/windows/>`__ to your local ``Downloads`` folder.
+2. Download the `OpenVINO Runtime archive file for Windows <https://storage.openvinotoolkit.org/repositories/openvino/packages/2023.0/windows/>`__ to your local ``Downloads`` folder.
 
    If you prefer using command-lines, run the following commands in the command prompt window you opened:
 

docs/resources/prerelease_information.md

@@ -17,13 +17,21 @@ Please file a github Issue on these with the label “pre-release” so we can g
    * NOT subject to official support.
    * Subject to change in the future.
    * Introduced to allow early testing and get early feedback from the community.
-
 
-.. dropdown:: OpenVINO Toolkit 2023.0.0.dev20230427
-   :open:
+
+
+.. dropdown:: OpenVINO Toolkit 2023.0.0.dev20230407
    :animate: fade-in-slide-down
    :color: primary
 
+   The first pre-release for OpenVINO 2023.1 focused on fixing bugs and performance issues.
+
+
+
+.. dropdown:: OpenVINO Toolkit 2023.0.0.dev20230407
+   :animate: fade-in-slide-down
+   :color: secondary
+
    Note that a new distribution channel has been introduced for C++ developers: `Conda Forge <https://anaconda.org/conda-forge/openvino>`__ 
    (the 2022.3.0 release is available there now).
 

docs/resources/release_notes.md

@@ -0,0 +1,18 @@
+# Release Notes {#release_notes}
+
+@sphinxdirective
+
+.. raw:: html
+
+   <meta http-equiv="Refresh" content="0; url='https://software.intel.com/content/www/us/en/develop/articles/openvino-relnotes.html'" />
+
+
+.. toctree::
+   :hidden:
+
+   prerelease_information
+
+The official OpenVINO Release Notes are published at `intel.com <https://software.intel.com/content/www/us/en/develop/articles/openvino-relnotes.html>`__
+
+
+@endsphinxdirective

docs/resources/resources.md

@@ -8,7 +8,6 @@
    :hidden:
 
    openvino_docs_performance_benchmarks
-   prerelease_information
 
 .. toctree::
    :maxdepth: 1

src/common/snippets/include/snippets/lowered/loop_manager.hpp

@@ -76,11 +76,19 @@ class LinearIR::LoopManager {
                    LinearIR::constExprIt loop_end_pos,
                    size_t loop_depth, size_t vector_size);
     // Return Loop ID
+    template <typename T>
     size_t mark_loop(LinearIR::constExprIt loop_begin_pos,
                     LinearIR::constExprIt loop_end_pos,
                     size_t work_amount, size_t work_amount_increment, size_t dim_idx,
-                    const std::vector<ExpressionPort>& entries,
-                    const std::vector<ExpressionPort>& exits);
+                    const std::vector<T>& entries,
+                    const std::vector<T>& exits) {
+        const auto loop_info = std::make_shared<LoopManager::LoopInfo>(work_amount, work_amount_increment, dim_idx, entries, exits);
+        const auto loop_id = this->add_loop_info(loop_info);
+        for (auto expr_it = loop_begin_pos; expr_it != loop_end_pos; ++expr_it) {
+            insert_loop_id(*expr_it, loop_id);
+        }
+        return loop_id;
+    }
 
     void fuse_loops(const LinearIR& linear_ir, size_t loop_id_upper, size_t loop_id_lower, bool fuse_into_upper = true);
     void fuse_loops(LinearIR::constExprIt loop_begin_target, LinearIR::constExprIt loop_end_target,
@@ -123,6 +131,8 @@ class LinearIR::LoopManager {
                                 LinearIR::constExprIt& loop_end_pos,
                                 size_t loop_id, bool loop_ops_inserted = false);
 
+    LoopPort get_loop_port_by_expr_port(const ExpressionPort& expr_port, const size_t loop_id);
+
 private:
     static void get_io_loop_ports(LinearIR::constExprIt loop_begin_pos,
                                   LinearIR::constExprIt loop_end_pos,

src/common/snippets/include/snippets/lowered/pass/fuse_loops.hpp

@@ -42,6 +42,12 @@ class FuseLoops : public Pass {
     FuseLoops();
     bool run(LinearIR& linear_ir) override;
 
+    // This method checks that all ports which connect lower and upper loops are incremented.
+    // This helps to avoid fusing for the ports with incompleted data
+    static bool loop_ports_are_compatible(const LinearIR::LoopManagerPtr& loop_manager,
+                                          const size_t loop_lower_id,
+                                          const size_t loop_upper_id);
+
 private:
     static bool can_be_fused(const LinearIR::LoopManager::LoopInfoPtr& loop_current,
                              const LinearIR::LoopManager::LoopInfoPtr& loop_target);

src/common/snippets/include/snippets/lowered/pass/split_loops.hpp

@@ -0,0 +1,46 @@
+// Copyright (C) 2023 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#pragma once
+
+#include "pass.hpp"
+#include "snippets/lowered/loop_manager.hpp"
+
+namespace ov {
+namespace snippets {
+namespace lowered {
+namespace pass {
+
+/**
+ * @interface SplitLoops
+ * @brief If loop_1 has larger increment but the same works amount of loop_2, that follows loop_1, then split loop_2
+ *        into two loops so the outermost of the split loops could be fused with the loop_1 using the pass `FuseLoops`.
+ * Example:
+ *         Loop_1_begin                                Loop_1_begin                                  Loop_1_begin
+ *              ...                                        ...                                           ...
+ *         Loop_1_end (wa = 128, inc = 32)             Loop_1_end (wa = 128, inc = 32)               Split_loop_2_begin
+ *              ...                         Splitting      ...                               Fusing      ...
+ *         Loop_2_begin                         =>     Split_loop_1_begin                      =>    Split_loop_2_end (wa = 32, inc = 1)
+ *              ...                                    Split_loop_2_begin                                ...
+ *         Loop_2_end (wa = 128, inc = 1)                  ...                                       Loop_1_end (wa = 128, inc = 32)
+ *                                                     Split_loop_2_end (wa = 32, inc = 1)
+ *                                                     Split_loop_1_end (wa = 128, inc = 32)
+ * @ingroup snippets
+ */
+
+class SplitLoops : public Pass {
+public:
+    OPENVINO_RTTI("SplitLoops", "Pass")
+    SplitLoops();
+    bool run(LinearIR& linear_ir) override;
+
+private:
+    static bool can_be_split(const LinearIR::LoopManager::LoopInfoPtr& current,
+                             const LinearIR::LoopManager::LoopInfoPtr& target);
+};
+
+} // namespace pass
+} // namespace lowered
+} // namespace snippets
+} // namespace ov

src/common/snippets/include/snippets/op/subgraph.hpp

@@ -104,12 +104,14 @@ class Subgraph : public ov::op::util::SubGraphOp {
                                 ov::pass::Manager& pre_common,
                                 ov::pass::Manager& post_common,
                                 ov::pass::Manager& post_precision,
+                                lowered::pass::PassPipeline& target_lowered_markup_pipeline,
                                 lowered::pass::PassPipeline& target_lowered_pipeline,
                                 const void* compile_params = nullptr);
     snippets::Schedule generate(const BlockedShapeVector& output_shapes, const BlockedShapeVector& input_shapes, const void* compile_params = nullptr);
     snippets::Schedule generate(ov::pass::Manager& pre_common,
                                 ov::pass::Manager& post_common,
                                 ov::pass::Manager& post_precision,
+                                lowered::pass::PassPipeline& target_lowered_markup_pipeline,
                                 lowered::pass::PassPipeline& target_lowered_pipeline,
                                 const void* compile_params = nullptr);
     snippets::Schedule generate(const void* compile_params = nullptr);
@@ -144,7 +146,9 @@ class Subgraph : public ov::op::util::SubGraphOp {
 private:
     void align_element_types(const BlockedShapeVector& outputShapes, const BlockedShapeVector& inputShapes);
     void data_flow_transformations(ov::pass::Manager& pre_common, ov::pass::Manager& post_common, ov::pass::Manager& post_precision);
-    void control_flow_transformations(lowered::LinearIR& linear_ir, lowered::pass::PassPipeline& target_pipeline);
+    void control_flow_transformations(lowered::LinearIR& linear_ir,
+                                      lowered::pass::PassPipeline& target_markup_pipeline,
+                                      lowered::pass::PassPipeline& target_pipeline);
     void init_config();
     // Count of Subgraph virtual ports:
     //  - Potential non-scalar Constants that will be created after some transformations (At the moment it's relevant only for FakeQuantize decomposition)

src/common/snippets/src/lowered/loop_manager.cpp

@@ -113,6 +113,18 @@ void LinearIR::LoopManager::get_loop_bounds(const LinearIR &linear_ir,
     }
 }
 
+LinearIR::LoopManager::LoopPort LinearIR::LoopManager::get_loop_port_by_expr_port(const ExpressionPort& expr_port, const size_t loop_id) {
+    auto get_loop_port = [&](const std::vector<LinearIR::LoopManager::LoopPort>& ports) {
+        auto it = std::find_if(ports.cbegin(), ports.cend(), [&](const LinearIR::LoopManager::LoopPort& p) { return *p.expr_port == expr_port; });
+        if (it == ports.cend())
+            OPENVINO_THROW("Expression has not been found among loop ports. Loop id: " + std::to_string(loop_id));
+        return *it;
+    };
+    const auto& loop_info = get_loop_info(loop_id);
+    return expr_port.get_type() == ExpressionPort::Input ? get_loop_port(loop_info->entry_points)
+                                                         : get_loop_port(loop_info->exit_points);
+}
+
 void LinearIR::LoopManager::get_io_loop_ports(LinearIR::constExprIt loop_begin_pos,
                                               LinearIR::constExprIt loop_end_pos,
                                               std::vector<ExpressionPort> &entries,
@@ -211,18 +223,6 @@ void LinearIR::LoopManager::mark_loop(LinearIR::constExprIt loop_begin_pos,
     }
 }
 
-size_t LinearIR::LoopManager::mark_loop(LinearIR::constExprIt loop_begin_pos,
-                                        LinearIR::constExprIt loop_end_pos,
-                                        size_t work_amount, size_t work_amount_increment, size_t dim_idx,
-                                        const std::vector<ExpressionPort>& entries,
-                                        const std::vector<ExpressionPort>& exits) {
-    const auto loop_info = std::make_shared<LoopManager::LoopInfo>(work_amount, work_amount_increment, dim_idx, entries, exits);
-    const auto loop_id = this->add_loop_info(loop_info);
-    for (auto expr_it = loop_begin_pos; expr_it != loop_end_pos; ++expr_it) {
-        insert_loop_id(*expr_it, loop_id);
-    }
-    return loop_id;
-}
 void LinearIR::LoopManager::fuse_loops(const LinearIR& linear_ir, size_t loop_id_upper, size_t loop_id_lower, bool fuse_into_upper) {
     LinearIR::constExprIt loop_begin_target, loop_end_target;
     get_loop_bounds(linear_ir, fuse_into_upper ? loop_id_lower : loop_id_upper, loop_begin_target, loop_end_target);

src/common/snippets/src/lowered/pass/fuse_loops.cpp

@@ -24,6 +24,23 @@ using LoopInfoPtr = LoopManager::LoopInfoPtr;
 
 FuseLoops::FuseLoops() : Pass() {}
 
+bool FuseLoops::loop_ports_are_compatible(const LinearIR::LoopManagerPtr& loop_manager,
+                                          const size_t loop_lower_id,
+                                          const size_t loop_upper_id) {
+    const auto loop_lower = loop_manager->get_loop_info(loop_lower_id);
+    for (const auto& entry : loop_lower->entry_points) {
+        const auto& src_port = entry.expr_port->get_port_connector_ptr()->get_source();
+        if (is_loop_id_found(src_port.get_expr()->get_loop_ids(), loop_upper_id)) {
+            if (!entry.is_incremented)
+                return false;
+            auto src_loop_port = loop_manager->get_loop_port_by_expr_port(src_port, loop_upper_id);
+            if (!src_loop_port.is_incremented)
+                return false;
+        }
+    }
+    return true;
+}
+
 bool FuseLoops::can_be_fused(const LoopInfoPtr& loop_current, const LoopInfoPtr& loop_target) {
     auto current_work_amount = loop_current->work_amount;
     auto target_work_amount = loop_target->work_amount;
@@ -79,7 +96,7 @@ bool FuseLoops::fuse_upper_into_current(LinearIR& linear_ir, const LinearIR::Loo
                                         LinearIR::constExprIt& current_loop_begin_pos, LinearIR::constExprIt& current_loop_end_pos) {
     const auto& loop_current = loop_manager->get_loop_info(current_loop_id);
     const auto& loop_target = loop_manager->get_loop_info(target_loop_id);
-    if (!can_be_fused(loop_current, loop_target))
+    if (!can_be_fused(loop_current, loop_target) || !loop_ports_are_compatible(loop_manager, current_loop_id, target_loop_id))
         return false;
 
     // We can fuse Loop_up to Loop_down only in cases when other consumers of Loop_up are after Loop_down
@@ -129,7 +146,7 @@ bool FuseLoops::fuse_lower_into_current(LinearIR& linear_ir, const LinearIR::Loo
                                         LinearIR::constExprIt& current_loop_begin_pos, LinearIR::constExprIt& current_loop_end_pos) {
     const auto& loop_current = loop_manager->get_loop_info(current_loop_id);
     const auto& loop_target = loop_manager->get_loop_info(target_loop_id);
-    if (!can_be_fused(loop_current, loop_target))
+    if (!can_be_fused(loop_current, loop_target) || !loop_ports_are_compatible(loop_manager, target_loop_id, current_loop_id))
         return false;
 
     // We can fuse Loop_down to Loop_up only in cases when other parents of Loop_down are before Loop_up

src/common/snippets/src/lowered/pass/init_loops.cpp

@@ -51,10 +51,15 @@ void InitLoops::init_ptr_increments(std::vector<LoopPort>& loop_inputs, std::vec
         const auto& layout = port->get_descriptor_ptr()->get_layout();
         const auto& shape = port->get_descriptor_ptr()->get_shape();
         const auto& dim = *(layout.rbegin() + dim_idx);
+        // Ticket: 113106
+        // WA: the current logic doesn't support the case with transposed output shape for brgemm layer
+        // but for all existing cases planar layout can be used
+        std::vector<size_t> planar(layout.size());
+        std::iota(planar.begin(), planar.end(), 0);
         loop_output.ptr_increment = 0;
         // If relevant dim is not broadcasted, then ptr_increment is the dim stride in the new layout
         if (loop_output.is_incremented && !(shape[dim] == 1 && work_amount != 1)) {
-            loop_output.ptr_increment = get_dim_stride(dim, layout, shape);
+            loop_output.ptr_increment = get_dim_stride(dim, planar, shape);
         }
     }
 }