2.3.14 -> 2.3.15:
  - Fix bugs and revise FPGA workaround codes
  - Handle DSPM size and a number of vISA insts
  - Skip layout conversion of tensors with raw data layout
  - Support Tizen sample apps (Native, Web, and .NET)
  - Verify v2.3.7 models in FPGA
  - Update NPU-DDK documents

2.3.13 -> 2.3.14:
  - Use a global request id managed by a device driver
  - Handle the field for extended metasize in npubinfmt.h
  - Support nnstreamer sub-plugin in Tizen emulator 6.5
  - Support system-wide request profiling using a device driver
  - Support application and request statistics in cuse driver

2.3.12 -> 2.3.13:
  - Add extra field for extened metasize to npubinfmt.h
  - Add extra API to select a NPU scheduler
  - Add apptest to test API interleaving
  - Add nnstreamer filter sub-plugin for srnpu
  - Fix model existence checking
  - Fix text formats reported by trinity-smi

2.3.11 -> 2.3.12:
  - Support NCHW data layout conversion
  - Perform code refactoring of HW input service
  - Add segment size info. for external memory
  - Add github action workflows for CI/CD

2.3.10 -> 2.3.11:
  - Fix bug when setting npumgr param
  - Fix bug when deallocating hwmem (e.g., missing close)
  - Support multiple submission with the same request
  - Add several aging apptests (e.g., repeat iterations)
  - Add data layout for input/output tensors to npubinfmt.h
  - Remove run_internal APIs (invalid scenarios) in libnpuhost.h

2.3.9 -> 2.3.10:
  - Add decoupled APIs for network inferencing
  - Enhance debugging (e.g., logging, API)
  - Update VD I/F sync. with real modules
  - Revise some documents

2.3.8 -> 2.3.9:
  - Fix memcpy bug of getNPUmodel_metadata
  - Add documents for FPGA testing guide
  - Add unittests to increase function coverage
  - Drop Ubuntu 16.04 (Xenial) support

2.3.7 -> 2.3.8:
  - Complete the dummy module of VD NPU manager
  - Add libnpuhost APIs for user convenience
  - Add some unittests to cover RFP checklists
  - Fix minor bugs

2.3.6 -> 2.3.7:
  - Add new profile API, getNPU_profile_opt()
  - Add the opcode of each vISA profile layer
  - Fix minor svace issues
  - Use TRIV v2.3.5 test models

2.3.5 -> 2.3.6:
  - Implement the prototype of VD NPU Manager (dummy)
  - Provide a CUSE-based virtual device node (trinity-cuse)
  - Distinguish debian and emulated envionment
  - Apply clang-format for the whole C++ source codes
  - Add a profile data path to the result of getNPU_profile()

2.3.4 -> 2.3.5:
  - Add start/end cycles of each profiling layer
  - Support NPU emulation mode in Tizen env.

2.3.3 -> 2.3.4:
  - Increase the maximum length of profiling layer name
  - Update the testdata path of unittests (v2.3.4)

2.3.2 -> 2.3.3:
  - Support profiling APIs for requests with any priority
  - Add an apptest program to test the task preemption
  - Fix minor bugs on a few unittest programs

2.3.1 -> 2.3.2:
  - Support the DSP profiling for MDK
  - Increase the overall code coverage
  - Revise the debugging feature

2.3.0 -> 2.3.1:
  - Update npubinfmt.h to store extra model info
  - Support per-layer profiling for MDK
  - Provide hotdoc-generated documents

2.2.10 -> 2.3.0:
  - Support 2-TOPS TRIV2 as the default NPU
  - Support TRIV 2.3.x architecture
  - Make TRIV1 and TRIA devices deprecated

2.2.9 -> 2.2.10:
  - Fix tensor size calculation using dynamic data granularity
  - Increase the code coverage for new APIs

2.2.8 -> 2.2.9:
  - Support profiling APIs in FastModel/FPGA env.
  - Fix some package dependency issues

2.2.7 -> 2.2.8:
  - Implement profiling APIs with the libmrpsim support
  - Fix minor bugs mostly related to NPU emulation

2.2.6 -> 2.2.7:
  - Revise NPU driver APIs to use Trinity API level 2
  - Extend trinity-smi to monitor the status of apps and their tasks
  - Seperate libnpuhost codes from core handler codes

2.2.5 -> 2.2.6:
  - Support 2-TOPS NPU devices for emulation with different command files
  - Add the command line utility (trinity-smi) for monitoring trinity devices
  - Support the Docker envionment (xenial, bionic) for npu-engine users

2.2.4 -> 2.2.5:
  - Revise the utility for trinity-testing applications
  - Replace emulation library from libnpu-engine-emul to libmrpsim
  - Resolve minor issues from static analyzers (i.e., svace, coverity)

2.2.3 -> 2.2.4:
  - Fix minor bugs related to layout conversion

2.2.2 -> 2.2.3:
  - Add a prototype of profiling interface as libnpuhost APIs
  - Add some apptests to check a multi-tasking capability
  - Skip to install dummy (or uncompatible) apptests in non-emulated env.
  - Fix minor bugs related to model unregistration

2.2.1 -> 2.2.2:
  - Fix external segment handling bugs
  - Add npu-engine-utils package
  - Support TRIV2's data manipulation (layout conversion)

2.2.0 -> 2.2.1:
  - Resolve issues reported by static analyzer (svace/coverity)
  - Add driver-assisted memory copy to avoid FPGA memcpy bug
  - Prepare the code release

2.0.5 -> 2.2.0:
  - Revise the user APIs for HW-recurring tasks
  - Sync with the kernel driver's changed ioctl definition
  - Add testcases for external hwmem and HW-recurring tasks

2.0.4 -> 2.0.5:
  - Fix tensor size calculations on TRIV2 layout
  - Add some testcases to run infernces using an async thread

2.0.2 -> 2.0.4:
  - Add ioctl() to get the number of requests for trinity devices
  - Add ioctl() to stop all requests for trinity devices
  - Support log verbose mode for debugging
  - Version sync with npu-engine-emul

2.0.1 -> 2.0.2:
  - Fix minor bugs (e.g., weight offset setting and data size calc)
  - Add user-level APIs to get status (e.g., memory/device status)
  - Support memory status ioctl for trinity devices
  - Support the stop feature to terminate submitted requests

2.0.0 -> 2.0.1:
  - Fully support the TRIV2 device and kernel driver
  - Add missing device functions (e.g., unsetModel)
  - Resolve minor bugs and svace issues

1.2.0 -> 2.0.0:
  - Revise the whole source codes (C++ porting/refactoring)
  - Increase the code coverage (>= 85%)
  - Support a segment table for TRIV2 inferences
  - Sync with the revised NPU kernel drivers

1.1.4 -> 1.2.0:
  - Sync with the NPU kernel driver refactoring
  - Add the version control with the NPU kernel driver

1.1.3 -> 1.1.4:
  - Add APIs to sepecify inference constraints (e.g., timeout/priority)
  - Add an option to control multiple buffering
  - Support a thread pool to support concurrent ioctls
  - Support a HW-recurring input service for runNPU_internalInput()

1.1.2 -> 1.1.3:
  - Fix minor bugs on NPU unittests
  - Add install_rpath for ubuntu users
  - Disable apptests using tvn models (instead, CI will do this)

1.1.1 -> 1.1.2:
  - Fix minor bugs on shamt handling in NPU emulation codes
  - Add an API to get the model metadata

1.1.0 -> 1.1.1:
  - Fix minor bugs on data mainpulation

1.0.1 -> 1.1.0:
  - Remove NPU emulation impl. (moved to AIP/NPU_SystemService_Emulator)
  - Remove testdata generation impl. (moved to AIP/NPU_SystemService_Testdata)
  - Simplify debian/tizen packaging scripts

1.0.0 -> 1.0.1:
  - Fix minor bugs in testdata generation

0.0.7 -> 1.0.0:
  - Support simultaneous request handling from different user apps
  - Support timeout and priority handling for multiple requests
  - Add and verify prebuilt tvn models (TRIV v1) in FastModel
  - Sync with the NPU driver version in NPU_Linux-kernel

0.0.6 -> 0.0.7:
  - Verify test cases on FastModel
  - Fix a few minor bugs

0.0.5 -> 0.0.6:
  - Support new NPU binary format (i.e., version 2)
  - Support non-contiguous multiple input/output tensors
  - Support data manipulation and quantization (only for npubinfmt v2)

0.0.4 -> 0.0.5:
  - Add a prototype for NPU-ASR
  - Fix minor bugs

0.0.3 -> 0.0.4:
  - Change names of sub-packages from libnpuengine to npu-engine

0.0.2 -> 0.0.3:
  - Tarballs containing test data are removed
  - Test data are generated at the build time

0.0.1 -> 0.0.2:
  - NPU Vision functionality verified in the FastModel envionment
  - Support Ubuntu 16.04 packaging & emulation of NPU executions
  - More test cases with single/multi-layer NPU models