2 - Add a prototype of profiling interface as libnpuhost APIs
3 - Add some apptests to check a multi-tasking capability
4 - Skip to install dummy (or uncompatible) apptests in non-emulated env.
5 - Fix minor bugs related to model unregistration
8 - Fix external segment handling bugs
9 - Add npu-engine-utils package
10 - Support TRIV2's data manipulation (layout conversion)
13 - Resolve issues reported by static analyzer (svace/coverity)
14 - Add driver-assisted memory copy to avoid FPGA memcpy bug
15 - Prepare the code release
18 - Revise the user APIs for HW-recurring tasks
19 - Sync with the kernel driver's changed ioctl definition
20 - Add testcases for external hwmem and HW-recurring tasks
23 - Fix tensor size calculations on TRIV2 layout
24 - Add some testcases to run infernces using an async thread
27 - Add ioctl() to get the number of requests for trinity devices
28 - Add ioctl() to stop all requests for trinity devices
29 - Support log verbose mode for debugging
30 - Version sync with npu-engine-emul
33 - Fix minor bugs (e.g., weight offset setting and data size calc)
34 - Add user-level APIs to get status (e.g., memory/device status)
35 - Support memory status ioctl for trinity devices
36 - Support the stop feature to terminate submitted requests
39 - Fully support the TRIV2 device and kernel driver
40 - Add missing device functions (e.g., unsetModel)
41 - Resolve minor bugs and svace issues
44 - Revise the whole source codes (C++ porting/refactoring)
45 - Increase the code coverage (>= 85%)
46 - Support a segment table for TRIV2 inferences
47 - Sync with the revised NPU kernel drivers
50 - Sync with the NPU kernel driver refactoring
51 - Add the version control with the NPU kernel driver
54 - Add APIs to sepecify inference constraints (e.g., timeout/priority)
55 - Add an option to control multiple buffering
56 - Support a thread pool to support concurrent ioctls
57 - Support a HW-recurring input service for runNPU_internalInput()
60 - Fix minor bugs on NPU unittests
61 - Add install_rpath for ubuntu users
62 - Disable apptests using tvn models (instead, CI will do this)
65 - Fix minor bugs on shamt handling in NPU emulation codes
66 - Add an API to get the model metadata
69 - Fix minor bugs on data mainpulation
72 - Remove NPU emulation impl. (moved to AIP/NPU_SystemService_Emulator)
73 - Remove testdata generation impl. (moved to AIP/NPU_SystemService_Testdata)
74 - Simplify debian/tizen packaging scripts
77 - Fix minor bugs in testdata generation
80 - Support simultaneous request handling from different user apps
81 - Support timeout and priority handling for multiple requests
82 - Add and verify prebuilt tvn models (TRIV v1) in FastModel
83 - Sync with the NPU driver version in NPU_Linux-kernel
86 - Verify test cases on FastModel
87 - Fix a few minor bugs
90 - Support new NPU binary format (i.e., version 2)
91 - Support non-contiguous multiple input/output tensors
92 - Support data manipulation and quantization (only for npubinfmt v2)
95 - Add a prototype for NPU-ASR
99 - Change names of sub-packages from libnpuengine to npu-engine
102 - Tarballs containing test data are removed
103 - Test data are generated at the build time
106 - NPU Vision functionality verified in the FastModel envionment
107 - Support Ubuntu 16.04 packaging & emulation of NPU executions
108 - More test cases with single/multi-layer NPU models