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