2 - Add extra field for extened metasize to npubinfmt.h
3 - Add extra API to select a NPU scheduler
4 - Add apptest to test API interleaving
5 - Add nnstreamer filter sub-plugin for srnpu
6 - Fix model existence checking
7 - Fix text formats reported by trinity-smi
10 - Support NCHW data layout conversion
11 - Perform code refactoring of HW input service
12 - Add segment size info. for external memory
13 - Add github action workflows for CI/CD
16 - Fix bug when setting npumgr param
17 - Fix bug when deallocating hwmem (e.g., missing close)
18 - Support multiple submission with the same request
19 - Add several aging apptests (e.g., repeat iterations)
20 - Add data layout for input/output tensors to npubinfmt.h
21 - Remove run_internal APIs (invalid scenarios) in libnpuhost.h
24 - Add decoupled APIs for network inferencing
25 - Enhance debugging (e.g., logging, API)
26 - Update VD I/F sync. with real modules
27 - Revise some documents
30 - Fix memcpy bug of getNPUmodel_metadata
31 - Add documents for FPGA testing guide
32 - Add unittests to increase function coverage
33 - Drop Ubuntu 16.04 (Xenial) support
36 - Complete the dummy module of VD NPU manager
37 - Add libnpuhost APIs for user convenience
38 - Add some unittests to cover RFP checklists
42 - Add new profile API, getNPU_profile_opt()
43 - Add the opcode of each vISA profile layer
44 - Fix minor svace issues
45 - Use TRIV v2.3.5 test models
48 - Implement the prototype of VD NPU Manager (dummy)
49 - Provide a CUSE-based virtual device node (trinity-cuse)
50 - Distinguish debian and emulated envionment
51 - Apply clang-format for the whole C++ source codes
52 - Add a profile data path to the result of getNPU_profile()
55 - Add start/end cycles of each profiling layer
56 - Support NPU emulation mode in Tizen env.
59 - Increase the maximum length of profiling layer name
60 - Update the testdata path of unittests (v2.3.4)
63 - Support profiling APIs for requests with any priority
64 - Add an apptest program to test the task preemption
65 - Fix minor bugs on a few unittest programs
68 - Support the DSP profiling for MDK
69 - Increase the overall code coverage
70 - Revise the debugging feature
73 - Update npubinfmt.h to store extra model info
74 - Support per-layer profiling for MDK
75 - Provide hotdoc-generated documents
78 - Support 2-TOPS TRIV2 as the default NPU
79 - Support TRIV 2.3.x architecture
80 - Make TRIV1 and TRIA devices deprecated
83 - Fix tensor size calculation using dynamic data granularity
84 - Increase the code coverage for new APIs
87 - Support profiling APIs in FastModel/FPGA env.
88 - Fix some package dependency issues
91 - Implement profiling APIs with the libmrpsim support
92 - Fix minor bugs mostly related to NPU emulation
95 - Revise NPU driver APIs to use Trinity API level 2
96 - Extend trinity-smi to monitor the status of apps and their tasks
97 - Seperate libnpuhost codes from core handler codes
100 - Support 2-TOPS NPU devices for emulation with different command files
101 - Add the command line utility (trinity-smi) for monitoring trinity devices
102 - Support the Docker envionment (xenial, bionic) for npu-engine users
105 - Revise the utility for trinity-testing applications
106 - Replace emulation library from libnpu-engine-emul to libmrpsim
107 - Resolve minor issues from static analyzers (i.e., svace, coverity)
110 - Fix minor bugs related to layout conversion
113 - Add a prototype of profiling interface as libnpuhost APIs
114 - Add some apptests to check a multi-tasking capability
115 - Skip to install dummy (or uncompatible) apptests in non-emulated env.
116 - Fix minor bugs related to model unregistration
119 - Fix external segment handling bugs
120 - Add npu-engine-utils package
121 - Support TRIV2's data manipulation (layout conversion)
124 - Resolve issues reported by static analyzer (svace/coverity)
125 - Add driver-assisted memory copy to avoid FPGA memcpy bug
126 - Prepare the code release
129 - Revise the user APIs for HW-recurring tasks
130 - Sync with the kernel driver's changed ioctl definition
131 - Add testcases for external hwmem and HW-recurring tasks
134 - Fix tensor size calculations on TRIV2 layout
135 - Add some testcases to run infernces using an async thread
138 - Add ioctl() to get the number of requests for trinity devices
139 - Add ioctl() to stop all requests for trinity devices
140 - Support log verbose mode for debugging
141 - Version sync with npu-engine-emul
144 - Fix minor bugs (e.g., weight offset setting and data size calc)
145 - Add user-level APIs to get status (e.g., memory/device status)
146 - Support memory status ioctl for trinity devices
147 - Support the stop feature to terminate submitted requests
150 - Fully support the TRIV2 device and kernel driver
151 - Add missing device functions (e.g., unsetModel)
152 - Resolve minor bugs and svace issues
155 - Revise the whole source codes (C++ porting/refactoring)
156 - Increase the code coverage (>= 85%)
157 - Support a segment table for TRIV2 inferences
158 - Sync with the revised NPU kernel drivers
161 - Sync with the NPU kernel driver refactoring
162 - Add the version control with the NPU kernel driver
165 - Add APIs to sepecify inference constraints (e.g., timeout/priority)
166 - Add an option to control multiple buffering
167 - Support a thread pool to support concurrent ioctls
168 - Support a HW-recurring input service for runNPU_internalInput()
171 - Fix minor bugs on NPU unittests
172 - Add install_rpath for ubuntu users
173 - Disable apptests using tvn models (instead, CI will do this)
176 - Fix minor bugs on shamt handling in NPU emulation codes
177 - Add an API to get the model metadata
180 - Fix minor bugs on data mainpulation
183 - Remove NPU emulation impl. (moved to AIP/NPU_SystemService_Emulator)
184 - Remove testdata generation impl. (moved to AIP/NPU_SystemService_Testdata)
185 - Simplify debian/tizen packaging scripts
188 - Fix minor bugs in testdata generation
191 - Support simultaneous request handling from different user apps
192 - Support timeout and priority handling for multiple requests
193 - Add and verify prebuilt tvn models (TRIV v1) in FastModel
194 - Sync with the NPU driver version in NPU_Linux-kernel
197 - Verify test cases on FastModel
198 - Fix a few minor bugs
201 - Support new NPU binary format (i.e., version 2)
202 - Support non-contiguous multiple input/output tensors
203 - Support data manipulation and quantization (only for npubinfmt v2)
206 - Add a prototype for NPU-ASR
210 - Change names of sub-packages from libnpuengine to npu-engine
213 - Tarballs containing test data are removed
214 - Test data are generated at the build time
217 - NPU Vision functionality verified in the FastModel envionment
218 - Support Ubuntu 16.04 packaging & emulation of NPU executions
219 - More test cases with single/multi-layer NPU models