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