# NPU 연산 지원 범위

아래 표는 `aries-rb`, `regulus-ra`, `regulus-rb`의 연산 지원 범위를 보여줍니다. `regulus-rb-usb`의 연산 지원 범위와 장치별 조건은 `regulus-rb`와 같습니다. 원본 모델의 연산자는 파싱 과정에서 Mobilint IR 연산으로 변환될 수 있습니다.

**표기:** ✅ 지원, ❌ 해당 NPU에서 미지원.

지원 연산이라도 모델의 그래프가 공통 입력 조건(4차원 NHWC, 정적 배치 크기 1, bool이 아닌 입력)이나 지원 데이터 형상·속성 범위를 벗어나면 컴파일되지 않을 수 있습니다. 인접 연산과 그래프 변환·제거 결과도 지원 여부에 영향을 줍니다. 모델별 컴파일로 확인하세요. CPU 실행 범위는 [CPU 오프로딩](<https://docs.mobilint.com/compiler/kr/v1.4/cpu_offloading/#cpu-오프로딩-연산-범위>)을 참조하세요.

## ONNX

| 연산자 | `aries-rb` | `regulus-ra` | `regulus-rb` |
| :--- | :---: | :---: | :---: |
| `Add` | ✅ | ✅ | ✅ |
| `AveragePool` | ✅ | ✅ | ✅ |
| `BatchNormalization` | ✅ | ✅ | ✅ |
| `Celu` | ✅ | ✅ | ✅ |
| `Clip` | ✅ | ✅ | ✅ |
| `Concat` | ✅ | ✅ | ✅ |
| `Conv` | ✅ | ✅ | ✅ |
| `ConvTranspose` | ✅ | ✅ | ✅ |
| `Cos` | ✅ | ❌ | ✅ |
| `DepthToSpace` | ✅ | ✅ | ✅ |
| `Div` | ✅ | ✅ | ✅ |
| `Elu` | ✅ | ✅ | ✅ |
| `Erf` | ✅ | ✅ | ✅ |
| `Exp` | ✅ | ✅ | ✅ |
| `Flatten` | ✅ | ✅ | ✅ |
| `Gelu` | ✅ | ✅ | ✅ |
| `Gemm` | ✅ | ✅ | ✅ |
| `GroupNormalization` | ✅ | ✅ | ✅ |
| `GRU` | ✅ | ✅ | ✅ |
| `HardSigmoid` | ✅ | ✅ | ✅ |
| `HardSwish` | ✅ | ✅ | ✅ |
| `InstanceNormalization` | ✅ | ✅ | ✅ |
| `LayerNormalization` | ✅ | ✅ | ✅ |
| `LeakyRelu` | ✅ | ✅ | ✅ |
| `Log` | ✅ | ✅ | ✅ |
| `LSTM` | ✅ | ✅ | ✅ |
| `MatMul` | ✅ | ✅ | ✅ |
| `Max` | ✅ | ✅ | ✅ |
| `MaxPool` | ✅ | ✅ | ✅ |
| `Mul` | ✅ | ✅ | ✅ |
| `Pow` | ✅ | ✅ | ✅ |
| `PRelu` | ✅ | ✅ | ✅ |
| `QuickGelu` | ✅ | ✅ | ✅ |
| `Reciprocal` | ✅ | ✅ | ✅ |
| `ReduceMax` | ✅ | ✅ | ✅ |
| `ReduceMean` | ✅ | ✅ | ✅ |
| `ReduceMin` | ✅ | ✅ | ✅ |
| `ReduceSum` | ✅ | ✅ | ✅ |
| `Relu` | ✅ | ✅ | ✅ |
| `Reshape` | ✅ | ✅ | ✅ |
| `Resize` | ✅ | ✅ | ✅ |
| `RNN` | ✅ | ✅ | ✅ |
| `Sigmoid` | ✅ | ✅ | ✅ |
| `Sin` | ✅ | ❌ | ✅ |
| `Slice` | ✅ | ✅ | ✅ |
| `Softmax` | ✅ | ❌ | ✅ |
| `Softplus` | ✅ | ✅ | ✅ |
| `SpaceToDepth` | ✅ | ✅ | ✅ |
| `Split` | ✅ | ✅ | ✅ |
| `Sqrt` | ✅ | ✅ | ✅ |
| `Squeeze` | ✅ | ✅ | ✅ |
| `Sub` | ✅ | ✅ | ✅ |
| `swish` | ✅ | ✅ | ✅ |
| `Tanh` | ✅ | ✅ | ✅ |
| `Transpose` | ✅ | ✅ | ✅ |
| `Unsqueeze` | ✅ | ✅ | ✅ |
| `Upsample` | ✅ | ✅ | ✅ |

## TensorFlow

| 연산자 | `aries-rb` | `regulus-ra` | `regulus-rb` |
| :--- | :---: | :---: | :---: |
| `Add` | ✅ | ✅ | ✅ |
| `AddN` | ✅ | ✅ | ✅ |
| `AddV2` | ✅ | ✅ | ✅ |
| `AvgPool` | ✅ | ✅ | ✅ |
| `BatchMatMul` | ✅ | ✅ | ✅ |
| `BatchMatMulV2` | ✅ | ✅ | ✅ |
| `BatchMatMulV3` | ✅ | ✅ | ✅ |
| `BiasAdd` | ✅ | ✅ | ✅ |
| `ClipByValue` | ✅ | ✅ | ✅ |
| `Concat` | ✅ | ✅ | ✅ |
| `ConcatV2` | ✅ | ✅ | ✅ |
| `Conv2D` | ✅ | ✅ | ✅ |
| `Conv2DBackpropInput` | ✅ | ✅ | ✅ |
| `Cos` | ✅ | ❌ | ✅ |
| `DepthwiseConv2dNative` | ✅ | ✅ | ✅ |
| `Div` | ✅ | ✅ | ✅ |
| `Elu` | ✅ | ✅ | ✅ |
| `Erf` | ✅ | ✅ | ✅ |
| `Erfc` | ✅ | ✅ | ✅ |
| `Exp` | ✅ | ✅ | ✅ |
| `ExpandDims` | ✅ | ✅ | ✅ |
| `FusedBatchNorm` | ✅ | ✅ | ✅ |
| `FusedBatchNormV2` | ✅ | ✅ | ✅ |
| `FusedBatchNormV3` | ✅ | ✅ | ✅ |
| `Inv` | ✅ | ✅ | ✅ |
| `LeakyRelu` | ✅ | ✅ | ✅ |
| `Log` | ✅ | ✅ | ✅ |
| `MatMul` | ✅ | ✅ | ✅ |
| `Max` | ✅ | ✅ | ✅ |
| `Maximum` | ✅ | ✅ | ✅ |
| `MaxPool` | ✅ | ✅ | ✅ |
| `Mean` | ✅ | ✅ | ✅ |
| `Min` | ✅ | ✅ | ✅ |
| `Minimum` | ✅ | ✅ | ✅ |
| `Mul` | ✅ | ✅ | ✅ |
| `Pow` | ✅ | ✅ | ✅ |
| `RealDiv` | ✅ | ✅ | ✅ |
| `Reciprocal` | ✅ | ✅ | ✅ |
| `Relu` | ✅ | ✅ | ✅ |
| `Relu6` | ✅ | ✅ | ✅ |
| `Reshape` | ✅ | ✅ | ✅ |
| `ResizeBilinear` | ✅ | ✅ | ✅ |
| `ResizeNearestNeighbor` | ✅ | ✅ | ✅ |
| `Sigmoid` | ✅ | ✅ | ✅ |
| `Sin` | ✅ | ❌ | ✅ |
| `Slice` | ✅ | ✅ | ✅ |
| `Softmax` | ✅ | ❌ | ✅ |
| `Softplus` | ✅ | ✅ | ✅ |
| `Split` | ✅ | ✅ | ✅ |
| `SplitV` | ✅ | ✅ | ✅ |
| `Sqrt` | ✅ | ✅ | ✅ |
| `Square` | ✅ | ✅ | ✅ |
| `SquaredDifference` | ✅ | ✅ | ✅ |
| `Squeeze` | ✅ | ✅ | ✅ |
| `StridedSlice` | ✅ | ✅ | ✅ |
| `Sub` | ✅ | ✅ | ✅ |
| `Sum` | ✅ | ✅ | ✅ |
| `Tanh` | ✅ | ✅ | ✅ |
| `Transpose` | ✅ | ✅ | ✅ |

## TFLite

| 연산자 | `aries-rb` | `regulus-ra` | `regulus-rb` |
| :--- | :---: | :---: | :---: |
| `ADD` | ✅ | ✅ | ✅ |
| `AVERAGE_POOL_2D` | ✅ | ✅ | ✅ |
| `BATCH_MATMUL` | ✅ | ✅ | ✅ |
| `CONCATENATION` | ✅ | ✅ | ✅ |
| `CONV_2D` | ✅ | ✅ | ✅ |
| `COS` | ✅ | ❌ | ✅ |
| `DEPTHWISE_CONV_2D` | ✅ | ✅ | ✅ |
| `DIV` | ✅ | ✅ | ✅ |
| `ELU` | ✅ | ✅ | ✅ |
| `EXP` | ✅ | ✅ | ✅ |
| `EXPAND_DIMS` | ✅ | ✅ | ✅ |
| `FULLY_CONNECTED` | ✅ | ✅ | ✅ |
| `GELU` | ✅ | ✅ | ✅ |
| `HARD_SWISH` | ✅ | ✅ | ✅ |
| `LEAKY_RELU` | ✅ | ✅ | ✅ |
| `LOG` | ✅ | ✅ | ✅ |
| `LOGISTIC` | ✅ | ✅ | ✅ |
| `MAX_POOL_2D` | ✅ | ✅ | ✅ |
| `MAXIMUM` | ✅ | ✅ | ✅ |
| `MEAN` | ✅ | ✅ | ✅ |
| `MINIMUM` | ✅ | ✅ | ✅ |
| `MUL` | ✅ | ✅ | ✅ |
| `POW` | ✅ | ✅ | ✅ |
| `PRELU` | ✅ | ✅ | ✅ |
| `REDUCE_MAX` | ✅ | ✅ | ✅ |
| `REDUCE_MIN` | ✅ | ✅ | ✅ |
| `RELU` | ✅ | ✅ | ✅ |
| `RELU6` | ✅ | ✅ | ✅ |
| `RELU_0_TO_1` | ✅ | ✅ | ✅ |
| `RELU_N1_TO_1` | ✅ | ✅ | ✅ |
| `RESHAPE` | ✅ | ✅ | ✅ |
| `RESIZE_BILINEAR` | ✅ | ✅ | ✅ |
| `RESIZE_NEAREST_NEIGHBOR` | ✅ | ✅ | ✅ |
| `SIN` | ✅ | ❌ | ✅ |
| `SLICE` | ✅ | ✅ | ✅ |
| `SOFTMAX` | ✅ | ❌ | ✅ |
| `SPLIT` | ✅ | ✅ | ✅ |
| `SPLIT_V` | ✅ | ✅ | ✅ |
| `SQRT` | ✅ | ✅ | ✅ |
| `SQUARE` | ✅ | ✅ | ✅ |
| `SQUARED_DIFFERENCE` | ✅ | ✅ | ✅ |
| `SQUEEZE` | ✅ | ✅ | ✅ |
| `STRIDED_SLICE` | ✅ | ✅ | ✅ |
| `SUB` | ✅ | ✅ | ✅ |
| `SUM` | ✅ | ✅ | ✅ |
| `TANH` | ✅ | ✅ | ✅ |
| `TRANSPOSE` | ✅ | ✅ | ✅ |
| `TRANSPOSE_CONV` | ✅ | ✅ | ✅ |

## Mobilint IR

| 연산자 | `aries-rb` | `regulus-ra` | `regulus-rb` |
| :--- | :---: | :---: | :---: |
| [Adding](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#adding>) | ✅ | ✅ | ✅ |
| [AddingConstant](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#addingconstant>) | ✅ | ✅ | ✅ |
| [Batchnorm](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#batchnorm>) | ✅ | ✅ | ✅ |
| [Celu](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#celu>) | ✅ | ✅ | ✅ |
| [Clip](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#clip>) | ✅ | ✅ | ✅ |
| [Concatenate](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#concatenate>) | ✅ | ✅ | ✅ |
| [Convolution](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#convolution>) | ✅ | ✅ | ✅ |
| [Cos](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#cos>) | ✅ | ❌ | ✅ |
| [DepthwiseConvolution](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#depthwiseconvolution>) | ✅ | ✅ | ✅ |
| [Div](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#div>) | ✅ | ✅ | ✅ |
| [DivConstant](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#divconstant>) | ✅ | ✅ | ✅ |
| [Elu](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#elu>) | ✅ | ✅ | ✅ |
| [Erf](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#erf>) | ✅ | ✅ | ✅ |
| [Exp](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#exp>) | ✅ | ✅ | ✅ |
| [Gelu](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#gelu>) | ✅ | ✅ | ✅ |
| [GroupConvolution](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#groupconvolution>) | ✅ | ✅ | ✅ |
| [GroupNormalization](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#groupnormalization>) | ✅ | ✅ | ✅ |
| [HardSigmoid](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#hardsigmoid>) | ✅ | ✅ | ✅ |
| [HardSwish](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#hardswish>) | ✅ | ✅ | ✅ |
| [HeaderView](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#headerview>) | ✅ | ✅ | ✅ |
| [HeaderViewRevert](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#headerviewrevert>) | ✅ | ✅ | ✅ |
| [Input](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#input>) | ✅ | ✅ | ✅ |
| [InputConstant](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#inputconstant>) | ✅ | ✅ | ✅ |
| [InstanceNormalization](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#instancenormalization>) | ✅ | ✅ | ✅ |
| [InverseSigmoid](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#inversesigmoid>) | ✅ | ✅ | ✅ |
| [L1Normalization](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#l1normalization>) | ✅ | ✅ | ✅ |
| [L2Normalization](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#l2normalization>) | ✅ | ✅ | ✅ |
| [LayerNormalization](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#layernormalization>) | ✅ | ✅ | ✅ |
| [LeakyRelu](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#leakyrelu>) | ✅ | ✅ | ✅ |
| [Log](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#log>) | ✅ | ✅ | ✅ |
| [MatMul](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#matmul>) | ✅ | ✅ | ✅ |
| [Mish](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#mish>) | ✅ | ✅ | ✅ |
| [Multiply](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#multiply>) | ✅ | ✅ | ✅ |
| [MultiplyConstant](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#multiplyconstant>) | ✅ | ✅ | ✅ |
| [OrderedPatchify](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#orderedpatchify>) | ✅ | ✅ | ✅ |
| [Output](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#output>) | ✅ | ✅ | ✅ |
| [Pad](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#pad>) | ✅ | ✅ | ✅ |
| [Pooling](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#pooling>) | ✅ | ✅ | ✅ |
| [Pow](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#pow>) | ✅ | ✅ | ✅ |
| [PRelu](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#prelu>) | ✅ | ✅ | ✅ |
| [QuickGelu](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#quickgelu>) | ✅ | ✅ | ✅ |
| [ReduceMax](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#reducemax>) | ✅ | ✅ | ✅ |
| [ReduceMean](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#reducemean>) | ✅ | ✅ | ✅ |
| [ReduceMin](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#reducemin>) | ✅ | ✅ | ✅ |
| [ReduceSum](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#reducesum>) | ✅ | ✅ | ✅ |
| [RelShift](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#relshift>) | ✅ | ✅ | ✅ |
| [Relu](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#relu>) | ✅ | ✅ | ✅ |
| [Reshape](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#reshape>) | ✅ | ✅ | ✅ |
| [Resize](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#resize>) | ✅ | ✅ | ✅ |
| [RmsNormalization](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#rmsnormalization>) | ✅ | ✅ | ✅ |
| [Sigmoid](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#sigmoid>) | ✅ | ✅ | ✅ |
| [Sin](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#sin>) | ✅ | ❌ | ✅ |
| [Slice](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#slice>) | ✅ | ✅ | ✅ |
| [Softmax](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#softmax>) | ✅ | ❌ | ✅ |
| [Softplus](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#softplus>) | ✅ | ✅ | ✅ |
| [Sqrt](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#sqrt>) | ✅ | ✅ | ✅ |
| [StaggeredPadding](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#staggeredpadding>) | ✅ | ✅ | ✅ |
| [StatefulAttentionMaskWrapper](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#statefulattentionmaskwrapper>) | ✅ | ✅ | ✅ |
| [StatefulCachedStack](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#statefulcachedstack>) | ✅ | ✅ | ✅ |
| [StatefulGRUWrapper](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#statefulgruwrapper>) | ✅ | ✅ | ✅ |
| [StatefulKVCacheWrapper](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#statefulkvcachewrapper>) | ✅ | ✅ | ✅ |
| [StatefulLSTMWrapper](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#statefullstmwrapper>) | ✅ | ✅ | ✅ |
| [StatefulRNNWrapper](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#statefulrnnwrapper>) | ✅ | ✅ | ✅ |
| [StatefulRoPEWrapper](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#statefulropewrapper>) | ✅ | ✅ | ✅ |
| [Sub](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#sub>) | ✅ | ✅ | ✅ |
| [SubConstant](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#subconstant>) | ✅ | ✅ | ✅ |
| [Swish](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#swish>) | ✅ | ✅ | ✅ |
| [Tan](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#tan>) | ✅ | ✅ | ✅ |
| [Tanh](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#tanh>) | ✅ | ✅ | ✅ |
| [Transpose](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#transpose>) | ✅ | ✅ | ✅ |
| [TransposeConvolution](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/#transposeconvolution>) | ✅ | ✅ | ✅ |

Mobilint IR 연산의 입력, 속성, 출력과 동작식은 [Mobilint IR 연산 형식](<https://docs.mobilint.com/compiler/kr/v1.4/ir_formats/index.md>)을 참조하세요.
