MLTC算子支持说明
STC算子
在STC层上，MLTC支持的STC算子共计158个，算子列表如下：
| 算子 |
|---|
| stc.abs |
| stc.acos |
| stc.add |
| stc.all |
| stc.any |
| stc.arange |
| stc.argmax |
| stc.argmin |
| stc.argsort |
| stc.asin |
| stc.atan |
| stc.attention_lm |
| stc.attention_mask_add |
| stc.avg_pool2d |
| stc.avg_pool3d |
| stc.broadcast_to |
| stc.cast |
| stc.ceil |
| stc.clamp |
| stc.col2im |
| stc.combine_dims |
| stc.concat |
| stc.const |
| stc.conv2d |
| stc.conv3d |
| stc.conv2d_transpose |
| stc.conv3d_transpose |
| stc.copy |
| stc.cos |
| stc.cosh |
| stc.cumsum |
| stc.deform_conv2d |
| stc.depthwise_conv2d |
| stc.depthwise_conv2d_transpose |
| stc.depthwise_conv3d |
| stc.depthwise_conv3d_transpose |
| stc.dequantize |
| stc.det |
| stc.device_reduce_sum |
| stc.device_reduce_mean |
| stc.device_reduce_max |
| stc.device_reduce_min |
| stc.device_concat |
| stc.diag |
| stc.diag_part |
| stc.dilate |
| stc.div |
| stc.equal |
| stc.erf |
| stc.exp |
| stc.expand_dims |
| stc.extern_const |
| stc.eye |
| stc.fill |
| stc.floor |
| stc.floor_div |
| stc.floor_mod |
| stc.gather |
| stc.gather_elements |
| stc.gather_nd |
| stc.gelu |
| stc.global_avg_pool2d |
| stc.global_avg_pool3d |
| stc.global_max_pool2d |
| stc.global_max_pool3d |
| stc.greater |
| stc.greater_equal |
| stc.grid_sample |
| stc.group_conv2d |
| stc.group_conv2d_transpose |
| stc.group_conv3d |
| stc.group_conv3d_transpose |
| stc.hardmax |
| stc.if |
| stc.im2col |
| stc.instance_norm |
| stc.isinf |
| stc.isnan |
| stc.kv_cache_load |
| stc.kv_cache_store |
| stc.layer_norm |
| stc.lkrelu |
| stc.less |
| stc.less_equal |
| stc.log |
| stc.logical_and |
| stc.logical_not |
| stc.logical_or |
| stc.logical_xor |
| stc.loop |
| stc.l2_normalize |
| stc.lp_norm |
| stc.lrn |
| stc.matmul |
| stc.matmul_vme |
| stc.max |
| stc.max_pool2d |
| stc.max_pool3d |
| stc.maxunpool2d |
| stc.meshgrid |
| stc.min |
| stc.mod |
| stc.mul |
| stc.mul_sum |
| stc.negative |
| stc.nonmaxsuppression |
| stc.not_equal |
| stc.one_hot |
| stc.pad |
| stc.pow |
| stc.pow_int |
| stc.prelu |
| stc.qr |
| stc.quantize |
| stc.ravel_multi_index |
| stc.recip |
| stc.reduce_max |
| stc.reduce_min |
| stc.reduce_mean |
| stc.reduce_prod |
| stc.reduce_sum |
| stc.repeat |
| stc.reshape |
| stc.resize2d |
| stc.resize3d |
| stc.reverse |
| stc.reverse_sequence |
| stc.roi_align |
| stc.roi_pool |
| stc.round |
| stc.rsqrt |
| stc.scatter_elements |
| stc.scatter_nd |
| stc.select |
| stc.shape |
| stc.sigmoid |
| stc.sign |
| stc.sin |
| stc.sinh |
| stc.slice |
| stc.sort |
| stc.split_dims |
| stc.sqrt |
| stc.squeeze |
| stc.stack |
| stc.sub |
| stc.tanh |
| stc.take |
| stc.tile |
| stc.topk |
| stc.transpose |
| stc.trilu |
| stc.trunc |
| stc.unsqueeze |
| stc.upsampling |
| stc.upsampling3d |
| stc.upsample |
| stc.variance |
ONNX算子
MLTC支持的ONNX算子共计156个，算子列表如下：
| 算子 |
|---|
| Abs |
| Acos |
| Acosh |
| Add |
| And |
| Argmax |
| Argmin |
| Asin |
| Asinh |
| Atan |
| Atanh |
| Averagepool |
| BitShift |
| BitwiseAnd |
| BitwiseOr |
| BitwiseXor |
| BitwiseNot |
| Batchnormalization |
| CenterCropPad |
| ConstantOfShape |
| CustomAttention |
| Cast |
| Cast_Like |
| Ceil |
| Celu |
| Clip |
| Col2Im |
| Compress |
| Concat |
| Constant |
| Conv |
| Conv2Dtranspose |
| Cos |
| Cosh |
| Cumsum |
| Deformconv |
| Det |
| Depthtospace |
| Dequantizelinear |
| Div |
| Dropout |
| Dynamicquantizelinear |
| Einsum |
| Elu |
| Equal |
| Erf |
| Exp |
| Expand |
| Eyelike |
| Flatten |
| Floor |
| Gather |
| Gatherelements |
| Gathernd |
| Gemm |
| Globalaveragepool |
| Globallppool |
| Globalmaxpool |
| Greater |
| Greaterorequal |
| Groupnormalization |
| Gru |
| GridSample |
| Hardmax |
| Hardsigmoid |
| Hardswish |
| Identity |
| Instancenormalization |
| Isinf |
| Isnan |
| If |
| Layernormalization |
| Leaky_Relu |
| Less |
| Lessorequal |
| Log |
| Logsoftmax |
| Loop |
| Lpnormalization |
| Lrn |
| LSTM |
| LpPool |
| Matmul |
| Matmulinteger |
| Max |
| MaxRoiPool |
| Maxpool |
| Maxunpool |
| Mean |
| Meanvariancenormalization |
| Melweightmatrix |
| Min |
| Mish |
| Mod |
| Mul |
| Multinomial |
| Neg |
| Not |
| NonMaxSuppression |
| Or |
| OneHot |
| Pad |
| Pow |
| Prelu |
| Qlinearmatmul |
| Quantizelinear |
| Reciprocal |
| ReduceL1 |
| ReduceL2 |
| Reducelogsum |
| Reducelogsumexp |
| Reducemax |
| Reducemean |
| Reducemin |
| Reduceprod |
| Reducesum |
| Reducesumsquare |
| Relu |
| Reshape |
| Resize |
| ReverseSequence |
| Round |
| RoiAlign |
| RNN |
| Scatterelements |
| Scatter |
| Scatternd |
| Selu |
| Shrink |
| Shape |
| Sigmoid |
| Sign |
| Sin |
| Sinh |
| Slice |
| Softmax |
| Softplus |
| Softsign |
| Spacetodepth |
| Split |
| Sqrt |
| Squeeze |
| Sub |
| Sum |
| Tan |
| Tanh |
| Thresholdrelu |
| Tile |
| Topk |
| Transpose |
| Trilu |
| Unsqueeze |
| Upsample |
| Where |
| Xor |
| Yield |
TensorFlow算子
MLTC支持的TensorFlow算子共计106个，算子列表如下：
| 算子 |
|---|
| Abs |
| Add |
| AddN |
| AddV2 |
| All |
| Any |
| ArgMax |
| ArgMin |
| Avgpool |
| BatchMatMul |
| BatchToSpaceND |
| BiasAdd |
| BitwiseAnd |
| BitwiseOr |
| BitwiseXor |
| BroadcastTo |
| Cast |
| Cell |
| ConcatV2 |
| Conv2D |
| Conv3D |
| Cos |
| Cosh |
| Cumsum |
| DepthToSpace |
| DepthwiseConv2dNative |
| Elu |
| Equal |
| Erf |
| Exp |
| ExpandDims |
| FakeQuantWithMinMaxArgs |
| FakeQuantWithMinMaxVars |
| Fill |
| Floor |
| FloorDiv |
| FloorMod |
| FusedBatchNorm |
| FusedBatchNormV2 |
| FusedBatchNormV3 |
| Gather |
| GatherNd |
| GatherV2 |
| Greater |
| GreaterEqual |
| Identity |
| IdentityN |
| Invert |
| LeakyRelu |
| Less |
| LessEqual |
| Log |
| LogicalAnd |
| LogicalNot |
| LogicalOr |
| LogSoftmax |
| MatMul |
| Max |
| Maximum |
| MaxPool |
| MaxPool3D |
| Mean |
| Min |
| Minimum |
| Mul |
| Neg |
| NotEqual |
| OneHot |
| Pack |
| Pad |
| Pow |
| Prod |
| Rank |
| RealDiv |
| Reciprocal |
| Relu |
| Relu6 |
| Reshape |
| ReverseV2 |
| Round |
| Rsqrt |
| Select |
| Shape |
| Sigmoid |
| Sign |
| Sin |
| Sinh |
| Slice |
| Softmax |
| SpaceToBatchND |
| SpaceToDepth |
| Split |
| SplitV |
| Sqrt |
| Square |
| SquaredDifference |
| Squeeze |
| StopGradient |
| StridedSlice |
| Sub |
| Sum |
| Tanh |
| Tile |
| Transpose |
| Unpack |
| ZerosLike |
Torch算子
MLTC支持的Torch底层ATen算子共计141个，算子列表如下：
| 算子 |
|---|
| AtenLeTensorOp |
| AtenBitwiseAndTensorOp |
| AtenDivTensorModeOp |
| AtenUnflattenIntOp |
| AtenScatterSrcOp |
| AtenEqTensorOp |
| AtenMaxDimOp |
| AtenDivScalarModeOp |
| AtenBatchNormOp |
| AtenFmodScalarOp |
| ConvertAtenAvgPool1dOp |
| AtenContiguousOp |
| AtenMaskedFillScalarOp |
| AtenNegOp |
| AtenCumsumOp |
| AtenTransposeIntOp |
| AtenLtTensorOp |
| ConvertAtenMaxPool1dOp |
| AtenEmbeddingOp |
| AtenGeluOp |
| AtenGeTensorOp |
| AtenDiagonalOp |
| AtenLogicalOrOp |
| AtenBitwiseAndScalarOp |
| AtenSqrtOp |
| AtenSigmoidOp |
| AtenMinOp |
| AtenGeScalarOp |
| AtenTrilOp |
| AtenAsStridedOp |
| AtenMulTensorOp |
| AtenNeScalarOp |
| AtenMaximumOp |
| AtenLogOp |
| AtenFlattenUsingIntsOp |
| AtenSubScalarOp |
| AtenConvolutionOp |
| AtenLeScalarOp |
| ConvertAtenAvgPool2dOp |
| AtenGeluBackwardOp |
| AtenLog1pOp |
| AtenThresholdOp |
| AtenMaxOp |
| AtenLeakyReluOp |
| AtenBroadcastToOp |
| AtenDivScalarOp |
| AtenMmOp |
| AtenDropoutOp |
| AtenIndexTensorHackedTwinOp |
| AtenIndexSelectOp |
| AtenLogicalXorOp |
| AtenRemainderScalarOp |
| AtenRsubScalarOp |
| AtenCeilOp |
| AtenReluOp |
| ConvertAtenMaxPool2dOp |
| AtenOuterOp |
| AtenSliceTensorOp |
| AtenMatmulOp |
| AtenMaskedFillTensorOp |
| AtenFillTensorOp |
| AtenClampOp |
| AtenCloneOp |
| AtenFlipOp |
| AtenGatherOp |
| AtenToDtypeOp |
| AtenPermuteOp |
| AtenUpsampleNearest2dVecOp |
| AtenMinimumOp |
| AtenLogicalNotOp |
| AtenLog2Op |
| AtenLtScalarOp |
| AtenOnesOp |
| AtenUpsampleNearest2dOp |
| AtenReciprocalOp |
| AtenErfOp |
| AtenUniformOp |
| AtenExpOp |
| AtenAddScalarOp |
| AtenNeTensorOp |
| AtenSumDimIntListOp |
| AtenBmmOp |
| AtenLinearOp |
| AtenClampTensorOp |
| AtenBitwiseNotOp |
| AtenEmptyMemoryFormatOp |
| AtenReflectionPad2dOp |
| AtenDivTensorOp |
| AtenRsqrtOp |
| AtenWhereSelfOp |
| AtenReflectionPad1dOp |
| AtenUnsqueezeOp |
| AtenLogitOp |
| AtenAbsOp |
| AtenSqueezeDimOp |
| AtenBitwiseLeftShiftTensorOp |
| AtenLinalgVectorNormOp |
| AtenFillScalarOp |
| AtenMinDimOp |
| AtenNativeLayerNormOp |
| AtenIscloseOp |
| AtenConstantPadNdOp |
| AtenGtScalarOp |
| AtenBitwiseOrTensorOp |
| AtenSumOp |
| AtenCopyOp |
| AtenViewOp |
| AtenBitwiseXorTensorOp |
| AtenThresholdBackwardOp |
| AtenArgmaxOp |
| AtenMeanDimOp |
| AtenSubTensorOp |
| AtenRemainderTensorOp |
| AtenHardtanhBackwardOp |
| AtenSqueezeOp |
| AtenAdaptiveAvgPool2dOp |
| AtenArangeStartStepOp |
| ConvertAtenFillOp |
| AtenCosOp |
| AtenLog10Op |
| AtenSinOp |
| AtenGtTensorOp |
| AtenEqScalarOp |
| AtenFill_ScalarOp |
| AtenFloorOp |
| AtenPowTensorScalarOp |
| AtenRoundOp |
| AtenAddTensorOp |
| AtenCatOp |
| AtenZerosOp |
| AtenPowScalarOp |
| AtenReshapeOp |
| AtenUnfoldOp |
| AtenLogicalAndOp |
| AtenExpm1Op |
| AtenBitwiseRightShiftTensorOp |
| AtenMulScalarOp |
| AtenFmodTensorOp |
| AtenTanOp |
| AtenPowTensorTensorOp |
| AtenTanhOp |