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