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