2 - Add start/end cycles of each profiling layer
3 - Support NPU emulation mode in Tizen env.
6 - Increase the maximum length of profiling layer name
7 - Update the testdata path of unittests (v2.3.4)
10 - Support profiling APIs for requests with any priority
11 - Add an apptest program to test the task preemption
12 - Fix minor bugs on a few unittest programs
15 - Support the DSP profiling for MDK
16 - Increase the overall code coverage
17 - Revise the debugging feature
20 - Update npubinfmt.h to store extra model info
21 - Support per-layer profiling for MDK
22 - Provide hotdoc-generated documents
25 - Support 2-TOPS TRIV2 as the default NPU
26 - Support TRIV 2.3.x architecture
27 - Make TRIV1 and TRIA devices deprecated
30 - Fix tensor size calculation using dynamic data granularity
31 - Increase the code coverage for new APIs
34 - Support profiling APIs in FastModel/FPGA env.
35 - Fix some package dependency issues
38 - Implement profiling APIs with the libmrpsim support
39 - Fix minor bugs mostly related to NPU emulation
42 - Revise NPU driver APIs to use Trinity API level 2
43 - Extend trinity-smi to monitor the status of apps and their tasks
44 - Seperate libnpuhost codes from core handler codes
47 - Support 2-TOPS NPU devices for emulation with different command files
48 - Add the command line utility (trinity-smi) for monitoring trinity devices
49 - Support the Docker envionment (xenial, bionic) for npu-engine users
52 - Revise the utility for trinity-testing applications
53 - Replace emulation library from libnpu-engine-emul to libmrpsim
54 - Resolve minor issues from static analyzers (i.e., svace, coverity)
57 - Fix minor bugs related to layout conversion
60 - Add a prototype of profiling interface as libnpuhost APIs
61 - Add some apptests to check a multi-tasking capability
62 - Skip to install dummy (or uncompatible) apptests in non-emulated env.
63 - Fix minor bugs related to model unregistration
66 - Fix external segment handling bugs
67 - Add npu-engine-utils package
68 - Support TRIV2's data manipulation (layout conversion)
71 - Resolve issues reported by static analyzer (svace/coverity)
72 - Add driver-assisted memory copy to avoid FPGA memcpy bug
73 - Prepare the code release
76 - Revise the user APIs for HW-recurring tasks
77 - Sync with the kernel driver's changed ioctl definition
78 - Add testcases for external hwmem and HW-recurring tasks
81 - Fix tensor size calculations on TRIV2 layout
82 - Add some testcases to run infernces using an async thread
85 - Add ioctl() to get the number of requests for trinity devices
86 - Add ioctl() to stop all requests for trinity devices
87 - Support log verbose mode for debugging
88 - Version sync with npu-engine-emul
91 - Fix minor bugs (e.g., weight offset setting and data size calc)
92 - Add user-level APIs to get status (e.g., memory/device status)
93 - Support memory status ioctl for trinity devices
94 - Support the stop feature to terminate submitted requests
97 - Fully support the TRIV2 device and kernel driver
98 - Add missing device functions (e.g., unsetModel)
99 - Resolve minor bugs and svace issues
102 - Revise the whole source codes (C++ porting/refactoring)
103 - Increase the code coverage (>= 85%)
104 - Support a segment table for TRIV2 inferences
105 - Sync with the revised NPU kernel drivers
108 - Sync with the NPU kernel driver refactoring
109 - Add the version control with the NPU kernel driver
112 - Add APIs to sepecify inference constraints (e.g., timeout/priority)
113 - Add an option to control multiple buffering
114 - Support a thread pool to support concurrent ioctls
115 - Support a HW-recurring input service for runNPU_internalInput()
118 - Fix minor bugs on NPU unittests
119 - Add install_rpath for ubuntu users
120 - Disable apptests using tvn models (instead, CI will do this)
123 - Fix minor bugs on shamt handling in NPU emulation codes
124 - Add an API to get the model metadata
127 - Fix minor bugs on data mainpulation
130 - Remove NPU emulation impl. (moved to AIP/NPU_SystemService_Emulator)
131 - Remove testdata generation impl. (moved to AIP/NPU_SystemService_Testdata)
132 - Simplify debian/tizen packaging scripts
135 - Fix minor bugs in testdata generation
138 - Support simultaneous request handling from different user apps
139 - Support timeout and priority handling for multiple requests
140 - Add and verify prebuilt tvn models (TRIV v1) in FastModel
141 - Sync with the NPU driver version in NPU_Linux-kernel
144 - Verify test cases on FastModel
145 - Fix a few minor bugs
148 - Support new NPU binary format (i.e., version 2)
149 - Support non-contiguous multiple input/output tensors
150 - Support data manipulation and quantization (only for npubinfmt v2)
153 - Add a prototype for NPU-ASR
157 - Change names of sub-packages from libnpuengine to npu-engine
160 - Tarballs containing test data are removed
161 - Test data are generated at the build time
164 - NPU Vision functionality verified in the FastModel envionment
165 - Support Ubuntu 16.04 packaging & emulation of NPU executions
166 - More test cases with single/multi-layer NPU models