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