2 - Support Simulator 3.9.1
3 - Backward compatibility for simulator is not supported
6 - Use Apache-2.0 License for now
7 - Use the embedded uapi header temporally
8 - Fix minor bugs (e.g., output callback handling)
9 - Prepare open sourcing for Tizen.org
12 - Support multi tensors for kernel requests
13 - Reset internal callback after submission (sync mode)
14 - Revise reference example code
17 - Fix the dspm size checking to use envvar
18 - Add build macro for cuse driver
19 - Add smack rules for cuse driver in post script
22 - Add PID info to profile data path
23 - Change the dafault data type for test programs
24 - Fix some coverity issues
27 - Support TRIV v2.3.8 models
28 - Update NNStreamer sub-plugin name
29 - Update NPU-DDK documents
32 - Fix double free bug when handling output callback
33 - Fix tensor size calculation for layout NONE (= TRIV2)
34 - Allow mutliple calls of getNPU_profile()
35 - Resolve warning messages from Valgrind analysis
36 - Update action workflows to upload a core dump on failures
37 - Increase code coverage (line & func >= 85%)
38 - Increase the column limit of clang-format for readability
41 - Fix bugs and revise FPGA workaround codes
42 - Handle DSPM size and a number of vISA insts
43 - Skip layout conversion of tensors with raw data layout
44 - Support Tizen sample apps (Native, Web, and .NET)
45 - Verify v2.3.7 models in FPGA
46 - Update NPU-DDK documents
49 - Use a global request id managed by a device driver
50 - Handle the field for extended metasize in npubinfmt.h
51 - Support nnstreamer sub-plugin in Tizen emulator 6.5
52 - Support system-wide request profiling using a device driver
53 - Support application and request statistics in cuse driver
56 - Add extra field for extened metasize to npubinfmt.h
57 - Add extra API to select a NPU scheduler
58 - Add apptest to test API interleaving
59 - Add nnstreamer filter sub-plugin for srnpu
60 - Fix model existence checking
61 - Fix text formats reported by trinity-smi
64 - Support NCHW data layout conversion
65 - Perform code refactoring of HW input service
66 - Add segment size info. for external memory
67 - Add github action workflows for CI/CD
70 - Fix bug when setting npumgr param
71 - Fix bug when deallocating hwmem (e.g., missing close)
72 - Support multiple submission with the same request
73 - Add several aging apptests (e.g., repeat iterations)
74 - Add data layout for input/output tensors to npubinfmt.h
75 - Remove run_internal APIs (invalid scenarios) in libnpuhost.h
78 - Add decoupled APIs for network inferencing
79 - Enhance debugging (e.g., logging, API)
80 - Update VD I/F sync. with real modules
81 - Revise some documents
84 - Fix memcpy bug of getNPUmodel_metadata
85 - Add documents for FPGA testing guide
86 - Add unittests to increase function coverage
87 - Drop Ubuntu 16.04 (Xenial) support
90 - Complete the dummy module of VD NPU manager
91 - Add libnpuhost APIs for user convenience
92 - Add some unittests to cover RFP checklists
96 - Add new profile API, getNPU_profile_opt()
97 - Add the opcode of each vISA profile layer
98 - Fix minor svace issues
99 - Use TRIV v2.3.5 test models
102 - Implement the prototype of VD NPU Manager (dummy)
103 - Provide a CUSE-based virtual device node (trinity-cuse)
104 - Distinguish debian and emulated envionment
105 - Apply clang-format for the whole C++ source codes
106 - Add a profile data path to the result of getNPU_profile()
109 - Add start/end cycles of each profiling layer
110 - Support NPU emulation mode in Tizen env.
113 - Increase the maximum length of profiling layer name
114 - Update the testdata path of unittests (v2.3.4)
117 - Support profiling APIs for requests with any priority
118 - Add an apptest program to test the task preemption
119 - Fix minor bugs on a few unittest programs
122 - Support the DSP profiling for MDK
123 - Increase the overall code coverage
124 - Revise the debugging feature
127 - Update npubinfmt.h to store extra model info
128 - Support per-layer profiling for MDK
129 - Provide hotdoc-generated documents
132 - Support 2-TOPS TRIV2 as the default NPU
133 - Support TRIV 2.3.x architecture
134 - Make TRIV1 and TRIA devices deprecated
137 - Fix tensor size calculation using dynamic data granularity
138 - Increase the code coverage for new APIs
141 - Support profiling APIs in FastModel/FPGA env.
142 - Fix some package dependency issues
145 - Implement profiling APIs with the libmrpsim support
146 - Fix minor bugs mostly related to NPU emulation
149 - Revise NPU driver APIs to use Trinity API level 2
150 - Extend trinity-smi to monitor the status of apps and their tasks
151 - Seperate libnpuhost codes from core handler codes
154 - Support 2-TOPS NPU devices for emulation with different command files
155 - Add the command line utility (trinity-smi) for monitoring trinity devices
156 - Support the Docker envionment (xenial, bionic) for npu-engine users
159 - Revise the utility for trinity-testing applications
160 - Replace emulation library from libnpu-engine-emul to libmrpsim
161 - Resolve minor issues from static analyzers (i.e., svace, coverity)
164 - Fix minor bugs related to layout conversion
167 - Add a prototype of profiling interface as libnpuhost APIs
168 - Add some apptests to check a multi-tasking capability
169 - Skip to install dummy (or uncompatible) apptests in non-emulated env.
170 - Fix minor bugs related to model unregistration
173 - Fix external segment handling bugs
174 - Add npu-engine-utils package
175 - Support TRIV2's data manipulation (layout conversion)
178 - Resolve issues reported by static analyzer (svace/coverity)
179 - Add driver-assisted memory copy to avoid FPGA memcpy bug
180 - Prepare the code release
183 - Revise the user APIs for HW-recurring tasks
184 - Sync with the kernel driver's changed ioctl definition
185 - Add testcases for external hwmem and HW-recurring tasks
188 - Fix tensor size calculations on TRIV2 layout
189 - Add some testcases to run infernces using an async thread
192 - Add ioctl() to get the number of requests for trinity devices
193 - Add ioctl() to stop all requests for trinity devices
194 - Support log verbose mode for debugging
195 - Version sync with npu-engine-emul
198 - Fix minor bugs (e.g., weight offset setting and data size calc)
199 - Add user-level APIs to get status (e.g., memory/device status)
200 - Support memory status ioctl for trinity devices
201 - Support the stop feature to terminate submitted requests
204 - Fully support the TRIV2 device and kernel driver
205 - Add missing device functions (e.g., unsetModel)
206 - Resolve minor bugs and svace issues
209 - Revise the whole source codes (C++ porting/refactoring)
210 - Increase the code coverage (>= 85%)
211 - Support a segment table for TRIV2 inferences
212 - Sync with the revised NPU kernel drivers
215 - Sync with the NPU kernel driver refactoring
216 - Add the version control with the NPU kernel driver
219 - Add APIs to sepecify inference constraints (e.g., timeout/priority)
220 - Add an option to control multiple buffering
221 - Support a thread pool to support concurrent ioctls
222 - Support a HW-recurring input service for runNPU_internalInput()
225 - Fix minor bugs on NPU unittests
226 - Add install_rpath for ubuntu users
227 - Disable apptests using tvn models (instead, CI will do this)
230 - Fix minor bugs on shamt handling in NPU emulation codes
231 - Add an API to get the model metadata
234 - Fix minor bugs on data mainpulation
237 - Remove NPU emulation impl. (moved to AIP/NPU_SystemService_Emulator)
238 - Remove testdata generation impl. (moved to AIP/NPU_SystemService_Testdata)
239 - Simplify debian/tizen packaging scripts
242 - Fix minor bugs in testdata generation
245 - Support simultaneous request handling from different user apps
246 - Support timeout and priority handling for multiple requests
247 - Add and verify prebuilt tvn models (TRIV v1) in FastModel
248 - Sync with the NPU driver version in NPU_Linux-kernel
251 - Verify test cases on FastModel
252 - Fix a few minor bugs
255 - Support new NPU binary format (i.e., version 2)
256 - Support non-contiguous multiple input/output tensors
257 - Support data manipulation and quantization (only for npubinfmt v2)
260 - Add a prototype for NPU-ASR
264 - Change names of sub-packages from libnpuengine to npu-engine
267 - Tarballs containing test data are removed
268 - Test data are generated at the build time
271 - NPU Vision functionality verified in the FastModel envionment
272 - Support Ubuntu 16.04 packaging & emulation of NPU executions
273 - More test cases with single/multi-layer NPU models