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