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