2 - Add new profile API, getNPU_profile_opt()
3 - Add the opcode of each vISA profile layer
4 - Fix minor svace issues
5 - Use TRIV v2.3.5 test models
8 - Implement the prototype of VD NPU Manager (dummy)
9 - Provide a CUSE-based virtual device node (trinity-cuse)
10 - Distinguish debian and emulated envionment
11 - Apply clang-format for the whole C++ source codes
12 - Add a profile data path to the result of getNPU_profile()
15 - Add start/end cycles of each profiling layer
16 - Support NPU emulation mode in Tizen env.
19 - Increase the maximum length of profiling layer name
20 - Update the testdata path of unittests (v2.3.4)
23 - Support profiling APIs for requests with any priority
24 - Add an apptest program to test the task preemption
25 - Fix minor bugs on a few unittest programs
28 - Support the DSP profiling for MDK
29 - Increase the overall code coverage
30 - Revise the debugging feature
33 - Update npubinfmt.h to store extra model info
34 - Support per-layer profiling for MDK
35 - Provide hotdoc-generated documents
38 - Support 2-TOPS TRIV2 as the default NPU
39 - Support TRIV 2.3.x architecture
40 - Make TRIV1 and TRIA devices deprecated
43 - Fix tensor size calculation using dynamic data granularity
44 - Increase the code coverage for new APIs
47 - Support profiling APIs in FastModel/FPGA env.
48 - Fix some package dependency issues
51 - Implement profiling APIs with the libmrpsim support
52 - Fix minor bugs mostly related to NPU emulation
55 - Revise NPU driver APIs to use Trinity API level 2
56 - Extend trinity-smi to monitor the status of apps and their tasks
57 - Seperate libnpuhost codes from core handler codes
60 - Support 2-TOPS NPU devices for emulation with different command files
61 - Add the command line utility (trinity-smi) for monitoring trinity devices
62 - Support the Docker envionment (xenial, bionic) for npu-engine users
65 - Revise the utility for trinity-testing applications
66 - Replace emulation library from libnpu-engine-emul to libmrpsim
67 - Resolve minor issues from static analyzers (i.e., svace, coverity)
70 - Fix minor bugs related to layout conversion
73 - Add a prototype of profiling interface as libnpuhost APIs
74 - Add some apptests to check a multi-tasking capability
75 - Skip to install dummy (or uncompatible) apptests in non-emulated env.
76 - Fix minor bugs related to model unregistration
79 - Fix external segment handling bugs
80 - Add npu-engine-utils package
81 - Support TRIV2's data manipulation (layout conversion)
84 - Resolve issues reported by static analyzer (svace/coverity)
85 - Add driver-assisted memory copy to avoid FPGA memcpy bug
86 - Prepare the code release
89 - Revise the user APIs for HW-recurring tasks
90 - Sync with the kernel driver's changed ioctl definition
91 - Add testcases for external hwmem and HW-recurring tasks
94 - Fix tensor size calculations on TRIV2 layout
95 - Add some testcases to run infernces using an async thread
98 - Add ioctl() to get the number of requests for trinity devices
99 - Add ioctl() to stop all requests for trinity devices
100 - Support log verbose mode for debugging
101 - Version sync with npu-engine-emul
104 - Fix minor bugs (e.g., weight offset setting and data size calc)
105 - Add user-level APIs to get status (e.g., memory/device status)
106 - Support memory status ioctl for trinity devices
107 - Support the stop feature to terminate submitted requests
110 - Fully support the TRIV2 device and kernel driver
111 - Add missing device functions (e.g., unsetModel)
112 - Resolve minor bugs and svace issues
115 - Revise the whole source codes (C++ porting/refactoring)
116 - Increase the code coverage (>= 85%)
117 - Support a segment table for TRIV2 inferences
118 - Sync with the revised NPU kernel drivers
121 - Sync with the NPU kernel driver refactoring
122 - Add the version control with the NPU kernel driver
125 - Add APIs to sepecify inference constraints (e.g., timeout/priority)
126 - Add an option to control multiple buffering
127 - Support a thread pool to support concurrent ioctls
128 - Support a HW-recurring input service for runNPU_internalInput()
131 - Fix minor bugs on NPU unittests
132 - Add install_rpath for ubuntu users
133 - Disable apptests using tvn models (instead, CI will do this)
136 - Fix minor bugs on shamt handling in NPU emulation codes
137 - Add an API to get the model metadata
140 - Fix minor bugs on data mainpulation
143 - Remove NPU emulation impl. (moved to AIP/NPU_SystemService_Emulator)
144 - Remove testdata generation impl. (moved to AIP/NPU_SystemService_Testdata)
145 - Simplify debian/tizen packaging scripts
148 - Fix minor bugs in testdata generation
151 - Support simultaneous request handling from different user apps
152 - Support timeout and priority handling for multiple requests
153 - Add and verify prebuilt tvn models (TRIV v1) in FastModel
154 - Sync with the NPU driver version in NPU_Linux-kernel
157 - Verify test cases on FastModel
158 - Fix a few minor bugs
161 - Support new NPU binary format (i.e., version 2)
162 - Support non-contiguous multiple input/output tensors
163 - Support data manipulation and quantization (only for npubinfmt v2)
166 - Add a prototype for NPU-ASR
170 - Change names of sub-packages from libnpuengine to npu-engine
173 - Tarballs containing test data are removed
174 - Test data are generated at the build time
177 - NPU Vision functionality verified in the FastModel envionment
178 - Support Ubuntu 16.04 packaging & emulation of NPU executions
179 - More test cases with single/multi-layer NPU models