2 - Add PID info to profile data path
3 - Change the dafault data type for test programs
4 - Fix some coverity issues
7 - Support TRIV v2.3.8 models
8 - Update NNStreamer sub-plugin name
9 - Update NPU-DDK documents
12 - Fix double free bug when handling output callback
13 - Fix tensor size calculation for layout NONE (= TRIV2)
14 - Allow mutliple calls of getNPU_profile()
15 - Resolve warning messages from Valgrind analysis
16 - Update action workflows to upload a core dump on failures
17 - Increase code coverage (line & func >= 85%)
18 - Increase the column limit of clang-format for readability
21 - Fix bugs and revise FPGA workaround codes
22 - Handle DSPM size and a number of vISA insts
23 - Skip layout conversion of tensors with raw data layout
24 - Support Tizen sample apps (Native, Web, and .NET)
25 - Verify v2.3.7 models in FPGA
26 - Update NPU-DDK documents
29 - Use a global request id managed by a device driver
30 - Handle the field for extended metasize in npubinfmt.h
31 - Support nnstreamer sub-plugin in Tizen emulator 6.5
32 - Support system-wide request profiling using a device driver
33 - Support application and request statistics in cuse driver
36 - Add extra field for extened metasize to npubinfmt.h
37 - Add extra API to select a NPU scheduler
38 - Add apptest to test API interleaving
39 - Add nnstreamer filter sub-plugin for srnpu
40 - Fix model existence checking
41 - Fix text formats reported by trinity-smi
44 - Support NCHW data layout conversion
45 - Perform code refactoring of HW input service
46 - Add segment size info. for external memory
47 - Add github action workflows for CI/CD
50 - Fix bug when setting npumgr param
51 - Fix bug when deallocating hwmem (e.g., missing close)
52 - Support multiple submission with the same request
53 - Add several aging apptests (e.g., repeat iterations)
54 - Add data layout for input/output tensors to npubinfmt.h
55 - Remove run_internal APIs (invalid scenarios) in libnpuhost.h
58 - Add decoupled APIs for network inferencing
59 - Enhance debugging (e.g., logging, API)
60 - Update VD I/F sync. with real modules
61 - Revise some documents
64 - Fix memcpy bug of getNPUmodel_metadata
65 - Add documents for FPGA testing guide
66 - Add unittests to increase function coverage
67 - Drop Ubuntu 16.04 (Xenial) support
70 - Complete the dummy module of VD NPU manager
71 - Add libnpuhost APIs for user convenience
72 - Add some unittests to cover RFP checklists
76 - Add new profile API, getNPU_profile_opt()
77 - Add the opcode of each vISA profile layer
78 - Fix minor svace issues
79 - Use TRIV v2.3.5 test models
82 - Implement the prototype of VD NPU Manager (dummy)
83 - Provide a CUSE-based virtual device node (trinity-cuse)
84 - Distinguish debian and emulated envionment
85 - Apply clang-format for the whole C++ source codes
86 - Add a profile data path to the result of getNPU_profile()
89 - Add start/end cycles of each profiling layer
90 - Support NPU emulation mode in Tizen env.
93 - Increase the maximum length of profiling layer name
94 - Update the testdata path of unittests (v2.3.4)
97 - Support profiling APIs for requests with any priority
98 - Add an apptest program to test the task preemption
99 - Fix minor bugs on a few unittest programs
102 - Support the DSP profiling for MDK
103 - Increase the overall code coverage
104 - Revise the debugging feature
107 - Update npubinfmt.h to store extra model info
108 - Support per-layer profiling for MDK
109 - Provide hotdoc-generated documents
112 - Support 2-TOPS TRIV2 as the default NPU
113 - Support TRIV 2.3.x architecture
114 - Make TRIV1 and TRIA devices deprecated
117 - Fix tensor size calculation using dynamic data granularity
118 - Increase the code coverage for new APIs
121 - Support profiling APIs in FastModel/FPGA env.
122 - Fix some package dependency issues
125 - Implement profiling APIs with the libmrpsim support
126 - Fix minor bugs mostly related to NPU emulation
129 - Revise NPU driver APIs to use Trinity API level 2
130 - Extend trinity-smi to monitor the status of apps and their tasks
131 - Seperate libnpuhost codes from core handler codes
134 - Support 2-TOPS NPU devices for emulation with different command files
135 - Add the command line utility (trinity-smi) for monitoring trinity devices
136 - Support the Docker envionment (xenial, bionic) for npu-engine users
139 - Revise the utility for trinity-testing applications
140 - Replace emulation library from libnpu-engine-emul to libmrpsim
141 - Resolve minor issues from static analyzers (i.e., svace, coverity)
144 - Fix minor bugs related to layout conversion
147 - Add a prototype of profiling interface as libnpuhost APIs
148 - Add some apptests to check a multi-tasking capability
149 - Skip to install dummy (or uncompatible) apptests in non-emulated env.
150 - Fix minor bugs related to model unregistration
153 - Fix external segment handling bugs
154 - Add npu-engine-utils package
155 - Support TRIV2's data manipulation (layout conversion)
158 - Resolve issues reported by static analyzer (svace/coverity)
159 - Add driver-assisted memory copy to avoid FPGA memcpy bug
160 - Prepare the code release
163 - Revise the user APIs for HW-recurring tasks
164 - Sync with the kernel driver's changed ioctl definition
165 - Add testcases for external hwmem and HW-recurring tasks
168 - Fix tensor size calculations on TRIV2 layout
169 - Add some testcases to run infernces using an async thread
172 - Add ioctl() to get the number of requests for trinity devices
173 - Add ioctl() to stop all requests for trinity devices
174 - Support log verbose mode for debugging
175 - Version sync with npu-engine-emul
178 - Fix minor bugs (e.g., weight offset setting and data size calc)
179 - Add user-level APIs to get status (e.g., memory/device status)
180 - Support memory status ioctl for trinity devices
181 - Support the stop feature to terminate submitted requests
184 - Fully support the TRIV2 device and kernel driver
185 - Add missing device functions (e.g., unsetModel)
186 - Resolve minor bugs and svace issues
189 - Revise the whole source codes (C++ porting/refactoring)
190 - Increase the code coverage (>= 85%)
191 - Support a segment table for TRIV2 inferences
192 - Sync with the revised NPU kernel drivers
195 - Sync with the NPU kernel driver refactoring
196 - Add the version control with the NPU kernel driver
199 - Add APIs to sepecify inference constraints (e.g., timeout/priority)
200 - Add an option to control multiple buffering
201 - Support a thread pool to support concurrent ioctls
202 - Support a HW-recurring input service for runNPU_internalInput()
205 - Fix minor bugs on NPU unittests
206 - Add install_rpath for ubuntu users
207 - Disable apptests using tvn models (instead, CI will do this)
210 - Fix minor bugs on shamt handling in NPU emulation codes
211 - Add an API to get the model metadata
214 - Fix minor bugs on data mainpulation
217 - Remove NPU emulation impl. (moved to AIP/NPU_SystemService_Emulator)
218 - Remove testdata generation impl. (moved to AIP/NPU_SystemService_Testdata)
219 - Simplify debian/tizen packaging scripts
222 - Fix minor bugs in testdata generation
225 - Support simultaneous request handling from different user apps
226 - Support timeout and priority handling for multiple requests
227 - Add and verify prebuilt tvn models (TRIV v1) in FastModel
228 - Sync with the NPU driver version in NPU_Linux-kernel
231 - Verify test cases on FastModel
232 - Fix a few minor bugs
235 - Support new NPU binary format (i.e., version 2)
236 - Support non-contiguous multiple input/output tensors
237 - Support data manipulation and quantization (only for npubinfmt v2)
240 - Add a prototype for NPU-ASR
244 - Change names of sub-packages from libnpuengine to npu-engine
247 - Tarballs containing test data are removed
248 - Test data are generated at the build time
251 - NPU Vision functionality verified in the FastModel envionment
252 - Support Ubuntu 16.04 packaging & emulation of NPU executions
253 - More test cases with single/multi-layer NPU models