2 - Fix bugs and revise FPGA workaround codes
3 - Handle DSPM size and a number of vISA insts
4 - Skip layout conversion of tensors with raw data layout
5 - Support Tizen sample apps (Native, Web, and .NET)
6 - Verify v2.3.7 models in FPGA
7 - Update NPU-DDK documents
10 - Use a global request id managed by a device driver
11 - Handle the field for extended metasize in npubinfmt.h
12 - Support nnstreamer sub-plugin in Tizen emulator 6.5
13 - Support system-wide request profiling using a device driver
14 - Support application and request statistics in cuse driver
17 - Add extra field for extened metasize to npubinfmt.h
18 - Add extra API to select a NPU scheduler
19 - Add apptest to test API interleaving
20 - Add nnstreamer filter sub-plugin for srnpu
21 - Fix model existence checking
22 - Fix text formats reported by trinity-smi
25 - Support NCHW data layout conversion
26 - Perform code refactoring of HW input service
27 - Add segment size info. for external memory
28 - Add github action workflows for CI/CD
31 - Fix bug when setting npumgr param
32 - Fix bug when deallocating hwmem (e.g., missing close)
33 - Support multiple submission with the same request
34 - Add several aging apptests (e.g., repeat iterations)
35 - Add data layout for input/output tensors to npubinfmt.h
36 - Remove run_internal APIs (invalid scenarios) in libnpuhost.h
39 - Add decoupled APIs for network inferencing
40 - Enhance debugging (e.g., logging, API)
41 - Update VD I/F sync. with real modules
42 - Revise some documents
45 - Fix memcpy bug of getNPUmodel_metadata
46 - Add documents for FPGA testing guide
47 - Add unittests to increase function coverage
48 - Drop Ubuntu 16.04 (Xenial) support
51 - Complete the dummy module of VD NPU manager
52 - Add libnpuhost APIs for user convenience
53 - Add some unittests to cover RFP checklists
57 - Add new profile API, getNPU_profile_opt()
58 - Add the opcode of each vISA profile layer
59 - Fix minor svace issues
60 - Use TRIV v2.3.5 test models
63 - Implement the prototype of VD NPU Manager (dummy)
64 - Provide a CUSE-based virtual device node (trinity-cuse)
65 - Distinguish debian and emulated envionment
66 - Apply clang-format for the whole C++ source codes
67 - Add a profile data path to the result of getNPU_profile()
70 - Add start/end cycles of each profiling layer
71 - Support NPU emulation mode in Tizen env.
74 - Increase the maximum length of profiling layer name
75 - Update the testdata path of unittests (v2.3.4)
78 - Support profiling APIs for requests with any priority
79 - Add an apptest program to test the task preemption
80 - Fix minor bugs on a few unittest programs
83 - Support the DSP profiling for MDK
84 - Increase the overall code coverage
85 - Revise the debugging feature
88 - Update npubinfmt.h to store extra model info
89 - Support per-layer profiling for MDK
90 - Provide hotdoc-generated documents
93 - Support 2-TOPS TRIV2 as the default NPU
94 - Support TRIV 2.3.x architecture
95 - Make TRIV1 and TRIA devices deprecated
98 - Fix tensor size calculation using dynamic data granularity
99 - Increase the code coverage for new APIs
102 - Support profiling APIs in FastModel/FPGA env.
103 - Fix some package dependency issues
106 - Implement profiling APIs with the libmrpsim support
107 - Fix minor bugs mostly related to NPU emulation
110 - Revise NPU driver APIs to use Trinity API level 2
111 - Extend trinity-smi to monitor the status of apps and their tasks
112 - Seperate libnpuhost codes from core handler codes
115 - Support 2-TOPS NPU devices for emulation with different command files
116 - Add the command line utility (trinity-smi) for monitoring trinity devices
117 - Support the Docker envionment (xenial, bionic) for npu-engine users
120 - Revise the utility for trinity-testing applications
121 - Replace emulation library from libnpu-engine-emul to libmrpsim
122 - Resolve minor issues from static analyzers (i.e., svace, coverity)
125 - Fix minor bugs related to layout conversion
128 - Add a prototype of profiling interface as libnpuhost APIs
129 - Add some apptests to check a multi-tasking capability
130 - Skip to install dummy (or uncompatible) apptests in non-emulated env.
131 - Fix minor bugs related to model unregistration
134 - Fix external segment handling bugs
135 - Add npu-engine-utils package
136 - Support TRIV2's data manipulation (layout conversion)
139 - Resolve issues reported by static analyzer (svace/coverity)
140 - Add driver-assisted memory copy to avoid FPGA memcpy bug
141 - Prepare the code release
144 - Revise the user APIs for HW-recurring tasks
145 - Sync with the kernel driver's changed ioctl definition
146 - Add testcases for external hwmem and HW-recurring tasks
149 - Fix tensor size calculations on TRIV2 layout
150 - Add some testcases to run infernces using an async thread
153 - Add ioctl() to get the number of requests for trinity devices
154 - Add ioctl() to stop all requests for trinity devices
155 - Support log verbose mode for debugging
156 - Version sync with npu-engine-emul
159 - Fix minor bugs (e.g., weight offset setting and data size calc)
160 - Add user-level APIs to get status (e.g., memory/device status)
161 - Support memory status ioctl for trinity devices
162 - Support the stop feature to terminate submitted requests
165 - Fully support the TRIV2 device and kernel driver
166 - Add missing device functions (e.g., unsetModel)
167 - Resolve minor bugs and svace issues
170 - Revise the whole source codes (C++ porting/refactoring)
171 - Increase the code coverage (>= 85%)
172 - Support a segment table for TRIV2 inferences
173 - Sync with the revised NPU kernel drivers
176 - Sync with the NPU kernel driver refactoring
177 - Add the version control with the NPU kernel driver
180 - Add APIs to sepecify inference constraints (e.g., timeout/priority)
181 - Add an option to control multiple buffering
182 - Support a thread pool to support concurrent ioctls
183 - Support a HW-recurring input service for runNPU_internalInput()
186 - Fix minor bugs on NPU unittests
187 - Add install_rpath for ubuntu users
188 - Disable apptests using tvn models (instead, CI will do this)
191 - Fix minor bugs on shamt handling in NPU emulation codes
192 - Add an API to get the model metadata
195 - Fix minor bugs on data mainpulation
198 - Remove NPU emulation impl. (moved to AIP/NPU_SystemService_Emulator)
199 - Remove testdata generation impl. (moved to AIP/NPU_SystemService_Testdata)
200 - Simplify debian/tizen packaging scripts
203 - Fix minor bugs in testdata generation
206 - Support simultaneous request handling from different user apps
207 - Support timeout and priority handling for multiple requests
208 - Add and verify prebuilt tvn models (TRIV v1) in FastModel
209 - Sync with the NPU driver version in NPU_Linux-kernel
212 - Verify test cases on FastModel
213 - Fix a few minor bugs
216 - Support new NPU binary format (i.e., version 2)
217 - Support non-contiguous multiple input/output tensors
218 - Support data manipulation and quantization (only for npubinfmt v2)
221 - Add a prototype for NPU-ASR
225 - Change names of sub-packages from libnpuengine to npu-engine
228 - Tarballs containing test data are removed
229 - Test data are generated at the build time
232 - NPU Vision functionality verified in the FastModel envionment
233 - Support Ubuntu 16.04 packaging & emulation of NPU executions
234 - More test cases with single/multi-layer NPU models