2 - Implement the prototype of VD NPU Manager (dummy)
3 - Provide a CUSE-based virtual device node (trinity-cuse)
4 - Distinguish debian and emulated envionment
5 - Apply clang-format for the whole C++ source codes
6 - Add a profile data path to the result of getNPU_profile()
9 - Add start/end cycles of each profiling layer
10 - Support NPU emulation mode in Tizen env.
13 - Increase the maximum length of profiling layer name
14 - Update the testdata path of unittests (v2.3.4)
17 - Support profiling APIs for requests with any priority
18 - Add an apptest program to test the task preemption
19 - Fix minor bugs on a few unittest programs
22 - Support the DSP profiling for MDK
23 - Increase the overall code coverage
24 - Revise the debugging feature
27 - Update npubinfmt.h to store extra model info
28 - Support per-layer profiling for MDK
29 - Provide hotdoc-generated documents
32 - Support 2-TOPS TRIV2 as the default NPU
33 - Support TRIV 2.3.x architecture
34 - Make TRIV1 and TRIA devices deprecated
37 - Fix tensor size calculation using dynamic data granularity
38 - Increase the code coverage for new APIs
41 - Support profiling APIs in FastModel/FPGA env.
42 - Fix some package dependency issues
45 - Implement profiling APIs with the libmrpsim support
46 - Fix minor bugs mostly related to NPU emulation
49 - Revise NPU driver APIs to use Trinity API level 2
50 - Extend trinity-smi to monitor the status of apps and their tasks
51 - Seperate libnpuhost codes from core handler codes
54 - Support 2-TOPS NPU devices for emulation with different command files
55 - Add the command line utility (trinity-smi) for monitoring trinity devices
56 - Support the Docker envionment (xenial, bionic) for npu-engine users
59 - Revise the utility for trinity-testing applications
60 - Replace emulation library from libnpu-engine-emul to libmrpsim
61 - Resolve minor issues from static analyzers (i.e., svace, coverity)
64 - Fix minor bugs related to layout conversion
67 - Add a prototype of profiling interface as libnpuhost APIs
68 - Add some apptests to check a multi-tasking capability
69 - Skip to install dummy (or uncompatible) apptests in non-emulated env.
70 - Fix minor bugs related to model unregistration
73 - Fix external segment handling bugs
74 - Add npu-engine-utils package
75 - Support TRIV2's data manipulation (layout conversion)
78 - Resolve issues reported by static analyzer (svace/coverity)
79 - Add driver-assisted memory copy to avoid FPGA memcpy bug
80 - Prepare the code release
83 - Revise the user APIs for HW-recurring tasks
84 - Sync with the kernel driver's changed ioctl definition
85 - Add testcases for external hwmem and HW-recurring tasks
88 - Fix tensor size calculations on TRIV2 layout
89 - Add some testcases to run infernces using an async thread
92 - Add ioctl() to get the number of requests for trinity devices
93 - Add ioctl() to stop all requests for trinity devices
94 - Support log verbose mode for debugging
95 - Version sync with npu-engine-emul
98 - Fix minor bugs (e.g., weight offset setting and data size calc)
99 - Add user-level APIs to get status (e.g., memory/device status)
100 - Support memory status ioctl for trinity devices
101 - Support the stop feature to terminate submitted requests
104 - Fully support the TRIV2 device and kernel driver
105 - Add missing device functions (e.g., unsetModel)
106 - Resolve minor bugs and svace issues
109 - Revise the whole source codes (C++ porting/refactoring)
110 - Increase the code coverage (>= 85%)
111 - Support a segment table for TRIV2 inferences
112 - Sync with the revised NPU kernel drivers
115 - Sync with the NPU kernel driver refactoring
116 - Add the version control with the NPU kernel driver
119 - Add APIs to sepecify inference constraints (e.g., timeout/priority)
120 - Add an option to control multiple buffering
121 - Support a thread pool to support concurrent ioctls
122 - Support a HW-recurring input service for runNPU_internalInput()
125 - Fix minor bugs on NPU unittests
126 - Add install_rpath for ubuntu users
127 - Disable apptests using tvn models (instead, CI will do this)
130 - Fix minor bugs on shamt handling in NPU emulation codes
131 - Add an API to get the model metadata
134 - Fix minor bugs on data mainpulation
137 - Remove NPU emulation impl. (moved to AIP/NPU_SystemService_Emulator)
138 - Remove testdata generation impl. (moved to AIP/NPU_SystemService_Testdata)
139 - Simplify debian/tizen packaging scripts
142 - Fix minor bugs in testdata generation
145 - Support simultaneous request handling from different user apps
146 - Support timeout and priority handling for multiple requests
147 - Add and verify prebuilt tvn models (TRIV v1) in FastModel
148 - Sync with the NPU driver version in NPU_Linux-kernel
151 - Verify test cases on FastModel
152 - Fix a few minor bugs
155 - Support new NPU binary format (i.e., version 2)
156 - Support non-contiguous multiple input/output tensors
157 - Support data manipulation and quantization (only for npubinfmt v2)
160 - Add a prototype for NPU-ASR
164 - Change names of sub-packages from libnpuengine to npu-engine
167 - Tarballs containing test data are removed
168 - Test data are generated at the build time
171 - NPU Vision functionality verified in the FastModel envionment
172 - Support Ubuntu 16.04 packaging & emulation of NPU executions
173 - More test cases with single/multi-layer NPU models