Skip to content

Latest commit

 

History

History
71 lines (63 loc) · 2.78 KB

supported-ops-and-platforms.md

File metadata and controls

71 lines (63 loc) · 2.78 KB

Supported precision

riscv only supports FP16 and FP32 precision on Allwinner D1 device.

Supported operators and opsets

  • ONNX
Op Type Op Set Linux RISCV
Add 7~12
And 7~16
ArgMax 11
AveragePool 11~16
Clip 11
Concat 11~12
ConstantOfShape 9~16
Conv 11~16
ConvTranspose 11~16
Div 7~12
Equal 11~12
Exp 6~12
Expand 8~12
Flatten 11~12
Floor 11~12
Gather 11~12
Gemm 11~12
LeakyRelu 6~16
Less 9~12
Log 6~12
MaxPool 11
Mul 7~12
NonMaxSuppression 11~16
Not 1~16
Range 11~16
ReduceMax 11
ReduceMean 11~12
ReduceMin 11
ReduceSum 11~12
Relu 6~12
Reshape 5~12
Resize 11~12
ScatterND 11~12
Shape 1~12
Sigmoid 6~12
Slice 11~12
Softmax 11~12
Split 11~12
Sqrt 6~12
Squeeze 11~12
Sub 7~12
Tile 6~12
TopK 11~16
Transpose 1~12
Unsqueeze 11~12
Where 9~15
  • MMCV
Op Type Op Set Linux RISCV
NonMaxSuppression 1
GridSample 1
RoiAlign 1
  • PPL
Op Type Op Set Linux RISCV
ChannelShuffle 1
ShapeOperation 1