2 - Support 2-TOPS TRIV2 as the default NPU
3 - Support TRIV 2.3.x architecture
4 - Make TRIV1 and TRIA devices deprecated
7 - Fix tensor size calculation using dynamic data granularity
8 - Increase the code coverage for new APIs
11 - Support profiling APIs in FastModel/FPGA env.
12 - Fix some package dependency issues
15 - Implement profiling APIs with the libmrpsim support
16 - Fix minor bugs mostly related to NPU emulation
19 - Revise NPU driver APIs to use Trinity API level 2
20 - Extend trinity-smi to monitor the status of apps and their tasks
21 - Seperate libnpuhost codes from core handler codes
24 - Support 2-TOPS NPU devices for emulation with different command files
25 - Add the command line utility (trinity-smi) for monitoring trinity devices
26 - Support the Docker envionment (xenial, bionic) for npu-engine users
29 - Revise the utility for trinity-testing applications
30 - Replace emulation library from libnpu-engine-emul to libmrpsim
31 - Resolve minor issues from static analyzers (i.e., svace, coverity)
34 - Fix minor bugs related to layout conversion
37 - Add a prototype of profiling interface as libnpuhost APIs
38 - Add some apptests to check a multi-tasking capability
39 - Skip to install dummy (or uncompatible) apptests in non-emulated env.
40 - Fix minor bugs related to model unregistration
43 - Fix external segment handling bugs
44 - Add npu-engine-utils package
45 - Support TRIV2's data manipulation (layout conversion)
48 - Resolve issues reported by static analyzer (svace/coverity)
49 - Add driver-assisted memory copy to avoid FPGA memcpy bug
50 - Prepare the code release
53 - Revise the user APIs for HW-recurring tasks
54 - Sync with the kernel driver's changed ioctl definition
55 - Add testcases for external hwmem and HW-recurring tasks
58 - Fix tensor size calculations on TRIV2 layout
59 - Add some testcases to run infernces using an async thread
62 - Add ioctl() to get the number of requests for trinity devices
63 - Add ioctl() to stop all requests for trinity devices
64 - Support log verbose mode for debugging
65 - Version sync with npu-engine-emul
68 - Fix minor bugs (e.g., weight offset setting and data size calc)
69 - Add user-level APIs to get status (e.g., memory/device status)
70 - Support memory status ioctl for trinity devices
71 - Support the stop feature to terminate submitted requests
74 - Fully support the TRIV2 device and kernel driver
75 - Add missing device functions (e.g., unsetModel)
76 - Resolve minor bugs and svace issues
79 - Revise the whole source codes (C++ porting/refactoring)
80 - Increase the code coverage (>= 85%)
81 - Support a segment table for TRIV2 inferences
82 - Sync with the revised NPU kernel drivers
85 - Sync with the NPU kernel driver refactoring
86 - Add the version control with the NPU kernel driver
89 - Add APIs to sepecify inference constraints (e.g., timeout/priority)
90 - Add an option to control multiple buffering
91 - Support a thread pool to support concurrent ioctls
92 - Support a HW-recurring input service for runNPU_internalInput()
95 - Fix minor bugs on NPU unittests
96 - Add install_rpath for ubuntu users
97 - Disable apptests using tvn models (instead, CI will do this)
100 - Fix minor bugs on shamt handling in NPU emulation codes
101 - Add an API to get the model metadata
104 - Fix minor bugs on data mainpulation
107 - Remove NPU emulation impl. (moved to AIP/NPU_SystemService_Emulator)
108 - Remove testdata generation impl. (moved to AIP/NPU_SystemService_Testdata)
109 - Simplify debian/tizen packaging scripts
112 - Fix minor bugs in testdata generation
115 - Support simultaneous request handling from different user apps
116 - Support timeout and priority handling for multiple requests
117 - Add and verify prebuilt tvn models (TRIV v1) in FastModel
118 - Sync with the NPU driver version in NPU_Linux-kernel
121 - Verify test cases on FastModel
122 - Fix a few minor bugs
125 - Support new NPU binary format (i.e., version 2)
126 - Support non-contiguous multiple input/output tensors
127 - Support data manipulation and quantization (only for npubinfmt v2)
130 - Add a prototype for NPU-ASR
134 - Change names of sub-packages from libnpuengine to npu-engine
137 - Tarballs containing test data are removed
138 - Test data are generated at the build time
141 - NPU Vision functionality verified in the FastModel envionment
142 - Support Ubuntu 16.04 packaging & emulation of NPU executions
143 - More test cases with single/multi-layer NPU models