2 - Support 2-TOPS NPU devices for emulation with different command files
3 - Add the command line utility (trinity-smi) for monitoring trinity devices
4 - Support the Docker envionment (xenial, bionic) for npu-engine users
7 - Revise the utility for trinity-testing applications
8 - Replace emulation library from libnpu-engine-emul to libmrpsim
9 - Resolve minor issues from static analyzers (i.e., svace, coverity)
12 - Fix minor bugs related to layout conversion
15 - Add a prototype of profiling interface as libnpuhost APIs
16 - Add some apptests to check a multi-tasking capability
17 - Skip to install dummy (or uncompatible) apptests in non-emulated env.
18 - Fix minor bugs related to model unregistration
21 - Fix external segment handling bugs
22 - Add npu-engine-utils package
23 - Support TRIV2's data manipulation (layout conversion)
26 - Resolve issues reported by static analyzer (svace/coverity)
27 - Add driver-assisted memory copy to avoid FPGA memcpy bug
28 - Prepare the code release
31 - Revise the user APIs for HW-recurring tasks
32 - Sync with the kernel driver's changed ioctl definition
33 - Add testcases for external hwmem and HW-recurring tasks
36 - Fix tensor size calculations on TRIV2 layout
37 - Add some testcases to run infernces using an async thread
40 - Add ioctl() to get the number of requests for trinity devices
41 - Add ioctl() to stop all requests for trinity devices
42 - Support log verbose mode for debugging
43 - Version sync with npu-engine-emul
46 - Fix minor bugs (e.g., weight offset setting and data size calc)
47 - Add user-level APIs to get status (e.g., memory/device status)
48 - Support memory status ioctl for trinity devices
49 - Support the stop feature to terminate submitted requests
52 - Fully support the TRIV2 device and kernel driver
53 - Add missing device functions (e.g., unsetModel)
54 - Resolve minor bugs and svace issues
57 - Revise the whole source codes (C++ porting/refactoring)
58 - Increase the code coverage (>= 85%)
59 - Support a segment table for TRIV2 inferences
60 - Sync with the revised NPU kernel drivers
63 - Sync with the NPU kernel driver refactoring
64 - Add the version control with the NPU kernel driver
67 - Add APIs to sepecify inference constraints (e.g., timeout/priority)
68 - Add an option to control multiple buffering
69 - Support a thread pool to support concurrent ioctls
70 - Support a HW-recurring input service for runNPU_internalInput()
73 - Fix minor bugs on NPU unittests
74 - Add install_rpath for ubuntu users
75 - Disable apptests using tvn models (instead, CI will do this)
78 - Fix minor bugs on shamt handling in NPU emulation codes
79 - Add an API to get the model metadata
82 - Fix minor bugs on data mainpulation
85 - Remove NPU emulation impl. (moved to AIP/NPU_SystemService_Emulator)
86 - Remove testdata generation impl. (moved to AIP/NPU_SystemService_Testdata)
87 - Simplify debian/tizen packaging scripts
90 - Fix minor bugs in testdata generation
93 - Support simultaneous request handling from different user apps
94 - Support timeout and priority handling for multiple requests
95 - Add and verify prebuilt tvn models (TRIV v1) in FastModel
96 - Sync with the NPU driver version in NPU_Linux-kernel
99 - Verify test cases on FastModel
100 - Fix a few minor bugs
103 - Support new NPU binary format (i.e., version 2)
104 - Support non-contiguous multiple input/output tensors
105 - Support data manipulation and quantization (only for npubinfmt v2)
108 - Add a prototype for NPU-ASR
112 - Change names of sub-packages from libnpuengine to npu-engine
115 - Tarballs containing test data are removed
116 - Test data are generated at the build time
119 - NPU Vision functionality verified in the FastModel envionment
120 - Support Ubuntu 16.04 packaging & emulation of NPU executions
121 - More test cases with single/multi-layer NPU models