2 - Update npubinfmt.h to store extra model info
3 - Support per-layer profiling for MDK
4 - Provide hotdoc-generated documents
7 - Support 2-TOPS TRIV2 as the default NPU
8 - Support TRIV 2.3.x architecture
9 - Make TRIV1 and TRIA devices deprecated
12 - Fix tensor size calculation using dynamic data granularity
13 - Increase the code coverage for new APIs
16 - Support profiling APIs in FastModel/FPGA env.
17 - Fix some package dependency issues
20 - Implement profiling APIs with the libmrpsim support
21 - Fix minor bugs mostly related to NPU emulation
24 - Revise NPU driver APIs to use Trinity API level 2
25 - Extend trinity-smi to monitor the status of apps and their tasks
26 - Seperate libnpuhost codes from core handler codes
29 - Support 2-TOPS NPU devices for emulation with different command files
30 - Add the command line utility (trinity-smi) for monitoring trinity devices
31 - Support the Docker envionment (xenial, bionic) for npu-engine users
34 - Revise the utility for trinity-testing applications
35 - Replace emulation library from libnpu-engine-emul to libmrpsim
36 - Resolve minor issues from static analyzers (i.e., svace, coverity)
39 - Fix minor bugs related to layout conversion
42 - Add a prototype of profiling interface as libnpuhost APIs
43 - Add some apptests to check a multi-tasking capability
44 - Skip to install dummy (or uncompatible) apptests in non-emulated env.
45 - Fix minor bugs related to model unregistration
48 - Fix external segment handling bugs
49 - Add npu-engine-utils package
50 - Support TRIV2's data manipulation (layout conversion)
53 - Resolve issues reported by static analyzer (svace/coverity)
54 - Add driver-assisted memory copy to avoid FPGA memcpy bug
55 - Prepare the code release
58 - Revise the user APIs for HW-recurring tasks
59 - Sync with the kernel driver's changed ioctl definition
60 - Add testcases for external hwmem and HW-recurring tasks
63 - Fix tensor size calculations on TRIV2 layout
64 - Add some testcases to run infernces using an async thread
67 - Add ioctl() to get the number of requests for trinity devices
68 - Add ioctl() to stop all requests for trinity devices
69 - Support log verbose mode for debugging
70 - Version sync with npu-engine-emul
73 - Fix minor bugs (e.g., weight offset setting and data size calc)
74 - Add user-level APIs to get status (e.g., memory/device status)
75 - Support memory status ioctl for trinity devices
76 - Support the stop feature to terminate submitted requests
79 - Fully support the TRIV2 device and kernel driver
80 - Add missing device functions (e.g., unsetModel)
81 - Resolve minor bugs and svace issues
84 - Revise the whole source codes (C++ porting/refactoring)
85 - Increase the code coverage (>= 85%)
86 - Support a segment table for TRIV2 inferences
87 - Sync with the revised NPU kernel drivers
90 - Sync with the NPU kernel driver refactoring
91 - Add the version control with the NPU kernel driver
94 - Add APIs to sepecify inference constraints (e.g., timeout/priority)
95 - Add an option to control multiple buffering
96 - Support a thread pool to support concurrent ioctls
97 - Support a HW-recurring input service for runNPU_internalInput()
100 - Fix minor bugs on NPU unittests
101 - Add install_rpath for ubuntu users
102 - Disable apptests using tvn models (instead, CI will do this)
105 - Fix minor bugs on shamt handling in NPU emulation codes
106 - Add an API to get the model metadata
109 - Fix minor bugs on data mainpulation
112 - Remove NPU emulation impl. (moved to AIP/NPU_SystemService_Emulator)
113 - Remove testdata generation impl. (moved to AIP/NPU_SystemService_Testdata)
114 - Simplify debian/tizen packaging scripts
117 - Fix minor bugs in testdata generation
120 - Support simultaneous request handling from different user apps
121 - Support timeout and priority handling for multiple requests
122 - Add and verify prebuilt tvn models (TRIV v1) in FastModel
123 - Sync with the NPU driver version in NPU_Linux-kernel
126 - Verify test cases on FastModel
127 - Fix a few minor bugs
130 - Support new NPU binary format (i.e., version 2)
131 - Support non-contiguous multiple input/output tensors
132 - Support data manipulation and quantization (only for npubinfmt v2)
135 - Add a prototype for NPU-ASR
139 - Change names of sub-packages from libnpuengine to npu-engine
142 - Tarballs containing test data are removed
143 - Test data are generated at the build time
146 - NPU Vision functionality verified in the FastModel envionment
147 - Support Ubuntu 16.04 packaging & emulation of NPU executions
148 - More test cases with single/multi-layer NPU models