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