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