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