nncase v2.3.0

New features

• Broadcast marker: #1033
• Feature/add squeeze binary shape: #1059
• fix python interface: #1040 #1057 #1063 #1066 #1067 #1071
• remove date in version: #1038
• Support one-dim input in matmul op: #1051
• Add inference report for both simulator and evb: #1061

Bug fixes and performance improve

• runtime kernel test fix: #1024 #1028 #1029 #1030 #1031 #1032 #1034 #1035 #1036 #1043 #1044 #1046
• Improve test runner: #1037 #1041 #1047
• Fix Dynamic TTS: #1039
• fix SpaceToBatchTest: #1053
• Fix 3D BatchToSpace: #1058
• Fix Tuple Compare: #1050

nncase v2.2.0

New features

• Refactor test runner #1010
• add ReshapeMatMul #1015
• add FoldNopIf/ FoldNopReduce #1016
• add trilu #1020
• runtime kernel tests #996

Bug fixes and performance improve

• fix test runner #1021

nncase v2.1.1

New features

• Load kmodel from stream #982
• space to batch to pad #990
• GNNE-1882 support einsum #994
• Support Onnx Model ExternalData #995
• GNNE-1875 Feature/op profile with cpu ops #1011

Bug fixes and performance improve

• fit single channel norm #964
• fix isa gen #973
• Fix bug layernorm get wrong result in some cases #991
• Fix stackvm memory leak at MX3SDK-636. #1014

nncase v2.1.0

New features

• ShapeBucket (#937)
• Feature/egraph sat extract (#963)
• add new hardswish pattern (#969)
• Feature/support output layout (#972)

Bug fixes and performance improve

• Fix bug binary rvv result error (#962)
• Optimize stackvm (#966)
• Fix bug binary result error and support some commonly used shapes (#980)
• Fix bug log softmax get nan or inf result (#985)

nncase v2.0.0

New features

• Completely rewritten in C#
• Support dynamic shape
• Support K230

Notes

• Depends on dotnet runtime 7.0
• Drop K210 support

nncase v1.9.0

New features

• support LBR (#796)
• Feature/import 1d deconv (#802)

Bug fixes

• K510 bug fixes
• K510 performance improvement

nncase v1.8.0

New features

• Add Compress (#625)
• Support NMS/TTS (#623)
• Add ReduceSum (#635)
• Add LayerNorm (#626)
• Add GatherElements (#642)

Bug fixes

• fix argmax min value (#619)
• fix bitcast buffer allocation (#632)
• fix build for archlinux/manjaro (#633)
• fix preprocess deq when model has focus (#628)

nncase v1.7.1

New features

• add gru op(onnx, forward) (#599)
• Feature/detection postprocess (#615)

Bug fixes

• fix import quant node (#600)
• fix k510 rsqrt

nncase v1.7.0

Bug fixes

• Fix/fold matmul add (#584)
• fix onnx prelu importer (#585)
• remove quantization range limitation (#596)

nncase v1.6.0

Major changes

• Support int8 output type (#566)
• Support specify input layout (#569)
• Support compare ops (#571)
• Support batch matmul (#577)
• Release K510 SDK

Bug fixes

• Fix tflite conv2d_transpose impoter