Operators running in CPU

The following table lists the ONNX operators that are not supported by AMD Vitis™ AI 6.2 NPU. These operators run on the CPU instead of the NPU, creating partitions in the model. If any of these operators is present in the model, only the part of the model containing these operators falls back to CPU execution, while the rest of the model still runs on the NPU.

Operator
Acos
Acosh
AffineGrid
ArgMax
ArgMin
Asin
Asinh
Atan
Atanh
Bernoulli
CastLike
Celu
ConvInteger
Cosh
Dropout
DynamicQuantizeLinear
EyeLike
GRU
GatherND
GlobalMaxPool
GroupNormalization
Hardmax
If
IsInf
IsNaN
LRN
Loop
LpNormalization
MatMulInteger
MaxRoiPool
Mean
MeanVarianceNormalization
Mod
OneHot
QLinearConv
QLinearMatMul
RNN
ReduceProd
ReverseSequence
RoiAlign
Scan
Scatter
ScatterElements
Shrink
Sinh
Softsign
Tan
Trilu