2 - Revise the user APIs for HW-recurring tasks
3 - Sync with the kernel driver's changed ioctl definition
4 - Add testcases for external hwmem and HW-recurring tasks
7 - Fix tensor size calculations on TRIV2 layout
8 - Add some testcases to run infernces using an async thread
11 - Add ioctl() to get the number of requests for trinity devices
12 - Add ioctl() to stop all requests for trinity devices
13 - Support log verbose mode for debugging
14 - Version sync with npu-engine-emul
17 - Fix minor bugs (e.g., weight offset setting and data size calc)
18 - Add user-level APIs to get status (e.g., memory/device status)
19 - Support memory status ioctl for trinity devices
20 - Support the stop feature to terminate submitted requests
23 - Fully support the TRIV2 device and kernel driver
24 - Add missing device functions (e.g., unsetModel)
25 - Resolve minor bugs and svace issues
28 - Revise the whole source codes (C++ porting/refactoring)
29 - Increase the code coverage (>= 85%)
30 - Support a segment table for TRIV2 inferences
31 - Sync with the revised NPU kernel drivers
34 - Sync with the NPU kernel driver refactoring
35 - Add the version control with the NPU kernel driver
38 - Add APIs to sepecify inference constraints (e.g., timeout/priority)
39 - Add an option to control multiple buffering
40 - Support a thread pool to support concurrent ioctls
41 - Support a HW-recurring input service for runNPU_internalInput()
44 - Fix minor bugs on NPU unittests
45 - Add install_rpath for ubuntu users
46 - Disable apptests using tvn models (instead, CI will do this)
49 - Fix minor bugs on shamt handling in NPU emulation codes
50 - Add an API to get the model metadata
53 - Fix minor bugs on data mainpulation
56 - Remove NPU emulation impl. (moved to AIP/NPU_SystemService_Emulator)
57 - Remove testdata generation impl. (moved to AIP/NPU_SystemService_Testdata)
58 - Simplify debian/tizen packaging scripts
61 - Fix minor bugs in testdata generation
64 - Support simultaneous request handling from different user apps
65 - Support timeout and priority handling for multiple requests
66 - Add and verify prebuilt tvn models (TRIV v1) in FastModel
67 - Sync with the NPU driver version in NPU_Linux-kernel
70 - Verify test cases on FastModel
71 - Fix a few minor bugs
74 - Support new NPU binary format (i.e., version 2)
75 - Support non-contiguous multiple input/output tensors
76 - Support data manipulation and quantization (only for npubinfmt v2)
79 - Add a prototype for NPU-ASR
83 - Change names of sub-packages from libnpuengine to npu-engine
86 - Tarballs containing test data are removed
87 - Test data are generated at the build time
90 - NPU Vision functionality verified in the FastModel envionment
91 - Support Ubuntu 16.04 packaging & emulation of NPU executions
92 - More test cases with single/multi-layer NPU models