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