2 - Fix bug when setting npumgr param
3 - Fix bug when deallocating hwmem (e.g., missing close)
4 - Support multiple submission with the same request
5 - Add several aging apptests (e.g., repeat iterations)
6 - Add data layout for input/output tensors to npubinfmt.h
7 - Remove run_internal APIs (invalid scenarios) in libnpuhost.h
10 - Add decoupled APIs for network inferencing
11 - Enhance debugging (e.g., logging, API)
12 - Update VD I/F sync. with real modules
13 - Revise some documents
16 - Fix memcpy bug of getNPUmodel_metadata
17 - Add documents for FPGA testing guide
18 - Add unittests to increase function coverage
19 - Drop Ubuntu 16.04 (Xenial) support
22 - Complete the dummy module of VD NPU manager
23 - Add libnpuhost APIs for user convenience
24 - Add some unittests to cover RFP checklists
28 - Add new profile API, getNPU_profile_opt()
29 - Add the opcode of each vISA profile layer
30 - Fix minor svace issues
31 - Use TRIV v2.3.5 test models
34 - Implement the prototype of VD NPU Manager (dummy)
35 - Provide a CUSE-based virtual device node (trinity-cuse)
36 - Distinguish debian and emulated envionment
37 - Apply clang-format for the whole C++ source codes
38 - Add a profile data path to the result of getNPU_profile()
41 - Add start/end cycles of each profiling layer
42 - Support NPU emulation mode in Tizen env.
45 - Increase the maximum length of profiling layer name
46 - Update the testdata path of unittests (v2.3.4)
49 - Support profiling APIs for requests with any priority
50 - Add an apptest program to test the task preemption
51 - Fix minor bugs on a few unittest programs
54 - Support the DSP profiling for MDK
55 - Increase the overall code coverage
56 - Revise the debugging feature
59 - Update npubinfmt.h to store extra model info
60 - Support per-layer profiling for MDK
61 - Provide hotdoc-generated documents
64 - Support 2-TOPS TRIV2 as the default NPU
65 - Support TRIV 2.3.x architecture
66 - Make TRIV1 and TRIA devices deprecated
69 - Fix tensor size calculation using dynamic data granularity
70 - Increase the code coverage for new APIs
73 - Support profiling APIs in FastModel/FPGA env.
74 - Fix some package dependency issues
77 - Implement profiling APIs with the libmrpsim support
78 - Fix minor bugs mostly related to NPU emulation
81 - Revise NPU driver APIs to use Trinity API level 2
82 - Extend trinity-smi to monitor the status of apps and their tasks
83 - Seperate libnpuhost codes from core handler codes
86 - Support 2-TOPS NPU devices for emulation with different command files
87 - Add the command line utility (trinity-smi) for monitoring trinity devices
88 - Support the Docker envionment (xenial, bionic) for npu-engine users
91 - Revise the utility for trinity-testing applications
92 - Replace emulation library from libnpu-engine-emul to libmrpsim
93 - Resolve minor issues from static analyzers (i.e., svace, coverity)
96 - Fix minor bugs related to layout conversion
99 - Add a prototype of profiling interface as libnpuhost APIs
100 - Add some apptests to check a multi-tasking capability
101 - Skip to install dummy (or uncompatible) apptests in non-emulated env.
102 - Fix minor bugs related to model unregistration
105 - Fix external segment handling bugs
106 - Add npu-engine-utils package
107 - Support TRIV2's data manipulation (layout conversion)
110 - Resolve issues reported by static analyzer (svace/coverity)
111 - Add driver-assisted memory copy to avoid FPGA memcpy bug
112 - Prepare the code release
115 - Revise the user APIs for HW-recurring tasks
116 - Sync with the kernel driver's changed ioctl definition
117 - Add testcases for external hwmem and HW-recurring tasks
120 - Fix tensor size calculations on TRIV2 layout
121 - Add some testcases to run infernces using an async thread
124 - Add ioctl() to get the number of requests for trinity devices
125 - Add ioctl() to stop all requests for trinity devices
126 - Support log verbose mode for debugging
127 - Version sync with npu-engine-emul
130 - Fix minor bugs (e.g., weight offset setting and data size calc)
131 - Add user-level APIs to get status (e.g., memory/device status)
132 - Support memory status ioctl for trinity devices
133 - Support the stop feature to terminate submitted requests
136 - Fully support the TRIV2 device and kernel driver
137 - Add missing device functions (e.g., unsetModel)
138 - Resolve minor bugs and svace issues
141 - Revise the whole source codes (C++ porting/refactoring)
142 - Increase the code coverage (>= 85%)
143 - Support a segment table for TRIV2 inferences
144 - Sync with the revised NPU kernel drivers
147 - Sync with the NPU kernel driver refactoring
148 - Add the version control with the NPU kernel driver
151 - Add APIs to sepecify inference constraints (e.g., timeout/priority)
152 - Add an option to control multiple buffering
153 - Support a thread pool to support concurrent ioctls
154 - Support a HW-recurring input service for runNPU_internalInput()
157 - Fix minor bugs on NPU unittests
158 - Add install_rpath for ubuntu users
159 - Disable apptests using tvn models (instead, CI will do this)
162 - Fix minor bugs on shamt handling in NPU emulation codes
163 - Add an API to get the model metadata
166 - Fix minor bugs on data mainpulation
169 - Remove NPU emulation impl. (moved to AIP/NPU_SystemService_Emulator)
170 - Remove testdata generation impl. (moved to AIP/NPU_SystemService_Testdata)
171 - Simplify debian/tizen packaging scripts
174 - Fix minor bugs in testdata generation
177 - Support simultaneous request handling from different user apps
178 - Support timeout and priority handling for multiple requests
179 - Add and verify prebuilt tvn models (TRIV v1) in FastModel
180 - Sync with the NPU driver version in NPU_Linux-kernel
183 - Verify test cases on FastModel
184 - Fix a few minor bugs
187 - Support new NPU binary format (i.e., version 2)
188 - Support non-contiguous multiple input/output tensors
189 - Support data manipulation and quantization (only for npubinfmt v2)
192 - Add a prototype for NPU-ASR
196 - Change names of sub-packages from libnpuengine to npu-engine
199 - Tarballs containing test data are removed
200 - Test data are generated at the build time
203 - NPU Vision functionality verified in the FastModel envionment
204 - Support Ubuntu 16.04 packaging & emulation of NPU executions
205 - More test cases with single/multi-layer NPU models