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