2 - Fix tensor size calculation using dynamic data granularity
3 - Increase the code coverage for new APIs
6 - Support profiling APIs in FastModel/FPGA env.
7 - Fix some package dependency issues
10 - Implement profiling APIs with the libmrpsim support
11 - Fix minor bugs mostly related to NPU emulation
14 - Revise NPU driver APIs to use Trinity API level 2
15 - Extend trinity-smi to monitor the status of apps and their tasks
16 - Seperate libnpuhost codes from core handler codes
19 - Support 2-TOPS NPU devices for emulation with different command files
20 - Add the command line utility (trinity-smi) for monitoring trinity devices
21 - Support the Docker envionment (xenial, bionic) for npu-engine users
24 - Revise the utility for trinity-testing applications
25 - Replace emulation library from libnpu-engine-emul to libmrpsim
26 - Resolve minor issues from static analyzers (i.e., svace, coverity)
29 - Fix minor bugs related to layout conversion
32 - Add a prototype of profiling interface as libnpuhost APIs
33 - Add some apptests to check a multi-tasking capability
34 - Skip to install dummy (or uncompatible) apptests in non-emulated env.
35 - Fix minor bugs related to model unregistration
38 - Fix external segment handling bugs
39 - Add npu-engine-utils package
40 - Support TRIV2's data manipulation (layout conversion)
43 - Resolve issues reported by static analyzer (svace/coverity)
44 - Add driver-assisted memory copy to avoid FPGA memcpy bug
45 - Prepare the code release
48 - Revise the user APIs for HW-recurring tasks
49 - Sync with the kernel driver's changed ioctl definition
50 - Add testcases for external hwmem and HW-recurring tasks
53 - Fix tensor size calculations on TRIV2 layout
54 - Add some testcases to run infernces using an async thread
57 - Add ioctl() to get the number of requests for trinity devices
58 - Add ioctl() to stop all requests for trinity devices
59 - Support log verbose mode for debugging
60 - Version sync with npu-engine-emul
63 - Fix minor bugs (e.g., weight offset setting and data size calc)
64 - Add user-level APIs to get status (e.g., memory/device status)
65 - Support memory status ioctl for trinity devices
66 - Support the stop feature to terminate submitted requests
69 - Fully support the TRIV2 device and kernel driver
70 - Add missing device functions (e.g., unsetModel)
71 - Resolve minor bugs and svace issues
74 - Revise the whole source codes (C++ porting/refactoring)
75 - Increase the code coverage (>= 85%)
76 - Support a segment table for TRIV2 inferences
77 - Sync with the revised NPU kernel drivers
80 - Sync with the NPU kernel driver refactoring
81 - Add the version control with the NPU kernel driver
84 - Add APIs to sepecify inference constraints (e.g., timeout/priority)
85 - Add an option to control multiple buffering
86 - Support a thread pool to support concurrent ioctls
87 - Support a HW-recurring input service for runNPU_internalInput()
90 - Fix minor bugs on NPU unittests
91 - Add install_rpath for ubuntu users
92 - Disable apptests using tvn models (instead, CI will do this)
95 - Fix minor bugs on shamt handling in NPU emulation codes
96 - Add an API to get the model metadata
99 - Fix minor bugs on data mainpulation
102 - Remove NPU emulation impl. (moved to AIP/NPU_SystemService_Emulator)
103 - Remove testdata generation impl. (moved to AIP/NPU_SystemService_Testdata)
104 - Simplify debian/tizen packaging scripts
107 - Fix minor bugs in testdata generation
110 - Support simultaneous request handling from different user apps
111 - Support timeout and priority handling for multiple requests
112 - Add and verify prebuilt tvn models (TRIV v1) in FastModel
113 - Sync with the NPU driver version in NPU_Linux-kernel
116 - Verify test cases on FastModel
117 - Fix a few minor bugs
120 - Support new NPU binary format (i.e., version 2)
121 - Support non-contiguous multiple input/output tensors
122 - Support data manipulation and quantization (only for npubinfmt v2)
125 - Add a prototype for NPU-ASR
129 - Change names of sub-packages from libnpuengine to npu-engine
132 - Tarballs containing test data are removed
133 - Test data are generated at the build time
136 - NPU Vision functionality verified in the FastModel envionment
137 - Support Ubuntu 16.04 packaging & emulation of NPU executions
138 - More test cases with single/multi-layer NPU models