Supported TensorFlow Operators

This section applies only to networks defined by the native IRs of TensorFlow 1.15 and 2.6.5. For details about the arguments, see the TensorFlow official website. To view the information about a single-operator defined based on Ascend IR, see CANN Operator Specifications.

TensorFlow Operator	Category	Operator Function
Abs	math_ops	Computes the absolute value of a tensor.
AccumulateNV2	math_ops	Returns the element-wise sum of a list of tensors.
Acos	math_ops	Computes acos of x element-wise.
Acosh	math_ops	Computes inverse hyperbolic cosine of x element-wise.
Add	math_ops	Returns x + y element-wise.
AddN	math_ops	Add all input tensors element wise.
AddV2	math_ops	Returns x + y element-wise.
All	math_ops	Computes the "logical and" of elements across dimensions of a tensor.
Any	math_ops	Computes the "logical or" of elements across dimensions of a tensor.
ApproximateEqual	math_ops	Returns the truth value of abs(x-y) < tolerance element-wise.
ArgMax	math_ops	Returns the index with the largest value across dimensions of a tensor.
ArgMin	math_ops	Returns the index with the smallest value across dimensions of a tensor.
Asin	math_ops	Computes asin of x element-wise.
Asinh	math_ops	Computes inverse hyperbolic sine of x element-wise.
Atan	math_ops	Computes atan of x element-wise.
Atan2	math_ops	Computes arctangent of y/x element-wise, respecting signs of the arguments.
Atanh	math_ops	Computes inverse hyperbolic tangent of x element-wise.
AvgPool	nn_ops	Performs average pooling on the input.
Batch	batch_ops	-
BatchMatMul	math_ops	Multiplies slices of two tensors in batches.
BatchToSpace	array_ops	BatchToSpace for 4-D tensors of type T.
BatchToSpaceND	array_ops	BatchToSpace for N-D tensors of type T.
BesselI0e	math_ops	Computes the Bessel i0e function of x element-wise.
BesselI1e	math_ops	Computes the Bessel i1e function of x element-wise.
Betainc	math_ops	Compute the regularized incomplete beta integral \(I_x(a, b)\).
BiasAdd	nn_ops	Adds bias to value.
Bincount	math_ops	Counts the number of occurrences of each value in an integer array.
BitwiseAnd	bitwise_ops	-
BitwiseOr	bitwise_ops	-
BitwiseXor	bitwise_ops	-
BroadcastTo	array_ops	Broadcast an array for a compatible shape.
Bucketize	math_ops	Bucketizes 'input' based on 'boundaries'.
Cast	math_ops	Cast x of type SrcT to y of DstT.
Ceil	math_ops	Returns element-wise smallest integer not less than x.
CheckNumerics	array_ops	Checks a tensor for NaN and Inf values.
Cholesky	linalg_ops	-
CholeskyGrad	linalg_ops	-
ClipByValue	math_ops	Clips tensor values to a specified min and max.
CompareAndBitpack	math_ops	Compare values of input to threshold and pack resulting bits into a uint8.
Concat	array_ops	Concatenates tensors along one dimension.
ConcatV2	array_ops	-
Const	array_ops	-
ControlTrigger	control_flow_ops	Does nothing.
Conv2D	nn_ops	Computes a 2-D convolution given 4-D input and filter tensors.
Conv2DBackpropFilter	nn_ops	Computes the gradients of convolution with respect to the filter.
Conv2DBackpropInput	nn_ops	Computes the gradients of convolution with respect to the input.
Conv3D	nn_ops	Computes a 3D convolution given 5D "x" and "filter" tensor.
Cos	math_ops	Computes cos of x element-wise.
Cosh	math_ops	Computes hyperbolic cosine of x element-wise.
Cumprod	math_ops	Compute the cumulative product of the tensor x along axis.
Cumsum	math_ops	Compute the cumulative sum of the tensor x along axis.
DataFormatDimMap	nn_ops	Returns the dimension index in the destination data format given the one in.
DataFormatVecPermute	nn_ops	Returns the permuted vector/tensor in the destination data format given the.
DepthToSpace	array_ops	DepthToSpace for tensors of type T.
DepthwiseConv2dNative	nn_ops	Computes a 2-D depthwise convolution given 4-D input and filtertensors.
DepthwiseConv2dNativeBackpropFilter	nn_ops	Computes the gradients of depthwise convolution with respect to the filter.
DepthwiseConv2dNativeBackpropInput	nn_ops	Computes the gradients of depthwise convolution with respect to the input.
Dequantize	array_ops	Dequantize the 'input' tensor into a float Tensor.
Diag	array_ops	Returns a diagonal tensor with a given diagonal values.
DiagPart	array_ops	Returns the diagonal part of the tensor.
Div	math_ops	Returns x / y element-wise.
DivNoNan	math_ops	Returns 0 if the denominator is zero.
Elu	nn_ops	Computes exponential linear: exp(features) - 1 if < 0, features otherwise.
Empty	array_ops	Creates a tensor with the given shape.
Enter	control_flow_ops	-
Equal	math_ops	Returns the truth value of (x == y) element-wise.
Erf	math_ops	Computes the Gauss error function of x element-wise.
Erfc	math_ops	Computes the complementary error function of x element-wise.
Exit	control_flow_ops	-
Exp	math_ops	Computes exponential of x element-wise.
ExpandDims	array_ops	Inserts a dimension of 1 into a tensor's shape.
Expm1	math_ops	Computes exponential of x - 1 element-wise.
ExtractImagePatches	array_ops	Extract patches from images and put them in the "depth" output dimension.
FakeQuantWithMinMaxArgs	array_ops	Fake-quantize the 'inputs' tensor, type float to 'outputs' tensor of same type.
FakeQuantWithMinMaxVars	array_ops	Fake-quantize the 'inputs' tensor of type float via global float scalars min.
FakeQuantWithMinMaxVarsPerChannel	array_ops	Fake-quantize the 'inputs' tensor of type float and one of the shapes: [d],.
Fill	array_ops	Creates a tensor filled with a scalar value.
Floor	math_ops	Returns element-wise largest integer not greater than x.
FloorDiv	math_ops	Returns x // y element-wise.
FloorMod	math_ops	Returns element-wise remainder of division.
FractionalAvgPool	nn_ops	Performs fractional average pooling on the input.
FractionalAvgPoolGrad	nn_ops	-
FractionalMaxPool	nn_ops	Performs fractional max pooling on the input.
FractionalMaxPoolGrad	nn_ops	-
FusedBatchNorm	nn_ops	Batch normalization.
FusedBatchNormV2	nn_ops	Batch normalization.
Gather	array_ops	Gather slices from params according to indices.
GatherNd	array_ops	Gather slices from params into a Tensor with shape specified by indices.
GatherV2	array_ops	Gather slices from params axis according to indices.
Greater	math_ops	Returns the truth value of (x > y) element-wise.
GreaterEqual	math_ops	Returns the truth value of (x >= y) element-wise.
GuaranteeConst	array_ops	Gives a guarantee to the TF runtime that the input tensor is a constant.
HistogramFixedWidth	math_ops	Return histogram of values.
Identity	array_ops	Return a tensor with the same shape and contents as the input tensor or value.
IdentityN	array_ops	Returns a list of tensors with the same shapes and contents as the input.
Igamma	math_ops	Compute the lower regularized incomplete Gamma function P(a, x).
Igammac	math_ops	Compute the upper regularized incomplete Gamma function Q(a, x).
IgammaGradA	math_ops	-
InplaceAdd	array_ops	Adds v into specified rows of x.
InplaceSub	array_ops	Subtracts v into specified rows of x.
InplaceUpdate	array_ops	Updates specified rows with values in v.
InTopK	nn_ops	Says whether the targets are in the top K predictions.
InTopKV2	nn_ops	Says whether the targets are in the top K predictions.
Inv	math_ops	Computes the reciprocal of x element-wise.
Invert	bitwise_ops	-
InvertPermutation	array_ops	Computes the inverse permutation of a tensor.
IsVariableInitialized	state_ops	Checks whether a tensor has been initialized.
L2Loss	nn_ops	L2 Loss.
Less	math_ops	Returns the truth value of (x < y) element-wise.
LessEqual	math_ops	Returns the truth value of (x <= y) element-wise.
LinSpace	math_ops	Generates values in an interval.
ListDiff	array_ops	-
Log	math_ops	Computes natural logarithm of x element-wise.
Log1p	math_ops	Computes natural logarithm of (1 + x) element-wise.
LogicalAnd	math_ops	Returns the truth value of x AND y element-wise.
LogicalNot	math_ops	Returns the truth value of NOT x element-wise.
LogicalOr	math_ops	Returns the truth value of x OR y element-wise.
LogMatrixDeterminant	linalg_ops	-
LogSoftmax	nn_ops	Computes log softmax activations.
LoopCond	control_flow_ops	Forwards the input to the output.
LowerBound	array_ops	-
LRN	nn_ops	Local Response Normalization.
MatMul	math_ops	Multiply the matrix "a" by the matrix "b".
MatrixBandPart	array_ops	Copy a tensor setting everything outside a central band in each innermost matrix.
MatrixDeterminant	linalg_ops	-
MatrixDiag	array_ops	Returns a batched diagonal tensor with a given batched diagonal values.
MatrixDiagPart	array_ops	Returns the batched diagonal part of a batched tensor.
MatrixInverse	linalg_ops	-
MatrixSetDiag	array_ops	Returns a batched matrix tensor with new batched diagonal values.
MatrixSolve	linalg_ops	-
MatrixSolveLs	linalg_ops	-
MatrixTriangularSolve	linalg_ops	-
Max	math_ops	Computes the maximum of elements across dimensions of a tensor.
Maximum	math_ops	Returns the max of x and y.
MaxPool	nn_ops	Performs max pooling on the input.
MaxPoolV2	nn_ops	Performs max pooling on the input.
MaxPool3D	nn_ops	Performs 3D max pooling on the input.
MaxPoolWithArgmax	nn_ops	Performs max pooling on the input and outputs both max values and indices.
Mean	math_ops	Computes the mean of elements across dimensions of a tensor.
Merge	control_flow_ops	Forwards the value of an available tensor from inputs to output.
Min	math_ops	Computes the minimum of elements across dimensions of a tensor.
Minimum	math_ops	Returns the min of x and y
MirrorPad	array_ops	Pads a tensor with mirrored values.
MirrorPadGrad	array_ops	-
Mod	math_ops	Returns element-wise remainder of division.
Mul	math_ops	-
Multinomial	random_ops	Draws samples from a multinomial distribution.
Neg	math_ops	-
NextIteration	control_flow_ops	Makes its input available to the next iteration.
NoOp	no_op	Does nothing.
NotEqual	math_ops	Returns the truth value of (x != y) element-wise.
NthElement	nn_ops	Finds values of the n-th order statistic for the last dimension.
OneHot	array_ops	Returns a one-hot tensor.
OnesLike	array_ops	Returns a tensor of ones with the same shape and type as x.
Pack	array_ops	-
Pad	array_ops	-
ParallelConcat	array_ops	-
ParameterizedTruncatedNormal	random_ops	Outputs random values from a normal distribution.
Placeholder	array_ops	-
PlaceholderWithDefault	array_ops	-
PopulationCount	bitwise_ops	-
Pow	math_ops	Computes the power of one value to another.
PreventGradient	array_ops	-
Prod	math_ops	Computes the product of elements across dimensions of a tensor.
Qr	linalg_ops	-
RandomGamma	random_ops	Outputs random values from the Gamma distribution(s) described by alpha.
RandomGammaGrad	random_ops	-
RandomShuffle	random_ops	Randomly shuffles a tensor along its first dimension.
RandomStandardNormal	random_ops	-
RandomUniform	random_ops	Outputs random values from a uniform distribution.
Range	math_ops	Creates a sequence of numbers.
RandomUniformInt	random_ops	Outputs random integers from a uniform distribution.
Rank	array_ops	Returns the rank of a tensor.
ReadVariableOp	resource_variable_ops	-
RealDiv	math_ops	Returns x / y element-wise for real types.
Reciprocal	math_ops	Computes the reciprocal of x element-wise.
RefEnter	control_flow_ops	-
RefExit	control_flow_ops	-
RefMerge	control_flow_ops	-
RefNextIteration	control_flow_ops	Makes its input available to the next iteration.
RefSwitch	control_flow_ops	Forwards the ref tensor data to the output port determined by pred.
Relu	nn_ops	Computes rectified linear: max(features, 0).
Relu6	nn_ops	Computes rectified linear 6: min(max(features, 0), 6).
Reshape	array_ops	Reshapes a tensor.
ReverseSequence	array_ops	Reverses variable length slices.
ReverseV2	array_ops	-
RightShift	bitwise_ops	-
Rint	math_ops	Returns element-wise integer closest to x.
Round	math_ops	Rounds the values of a tensor to the nearest integer, element-wise.
Rsqrt	math_ops	Computes reciprocal of square root of x element-wise.
SegmentMax	math_ops	Computes the maximum along segments of a tensor.
Select	math_ops	-
Selu	nn_ops	Computes scaled exponential linear: scale * alpha * (exp(features) - 1).
Shape	array_ops	Returns the shape of a tensor.
ShapeN	array_ops	Returns shape of tensors.
Sigmoid	math_ops	Computes sigmoid of x element-wise.
Sign	math_ops	Returns an element-wise indication of the sign of a number.
Sin	math_ops	Computes sin of x element-wise.
Sinh	math_ops	Computes hyperbolic sine of x element-wise.
Size	array_ops	Returns the size of a tensor.
Slice	array_ops	Return a slice from 'input'.
Snapshot	array_ops	Returns a copy of the input tensor.
Softmax	nn_ops	Computes softmax activations.
Softplus	nn_ops	Computes softplus: log(exp(features) + 1).
Softsign	nn_ops	Computes softsign: features / (abs(features) + 1).
SpaceToBatch	array_ops	SpaceToBatch for 4-D tensors of type T.
SpaceToBatchND	array_ops	SpaceToBatch for N-D tensors of type T.
SpaceToDepth	array_ops	SpaceToDepth for tensors of type T.
Split	array_ops	Splits a tensor into num_split tensors along one dimension.
SplitV	array_ops	Splits a tensor into num_split tensors along one dimension.
Sqrt	math_ops	Computes square root of x element-wise.
Square	math_ops	Computes square of x element-wise.
SquaredDifference	math_ops	Returns (x - y)(x - y) element-wise.
Squeeze	array_ops	Removes dimensions of size 1 from the shape of a tensor.
StatelessMultinomial	stateless_random_ops	-
StopGradient	array_ops	Stops gradient computation.
StridedSlice	array_ops	Return a strided slice from input.
Sub	math_ops	-
Sum	math_ops	Computes the sum of elements across dimensions of a tensor.
Svd	linalg_ops	-
Switch	control_flow_ops	Forwards data to the output port determined by pred.
Tan	math_ops	Computes tan of x element-wise.
Tanh	math_ops	Computes hyperbolic tangent of x element-wise.
Tile	array_ops	Constructs a tensor by tiling a given tensor.
TopK	nn_ops	Finds values and indices of the k largest elements for the last dimension.
TopKV2	nn_ops	-
Transpose	array_ops	Shuffle dimensions of x according to a permutation.
TruncateDiv	math_ops	Returns x / y element-wise for integer types.
TruncatedNormal	random_ops	Outputs random values from a truncated normal distribution.
TruncateMod	math_ops	Returns element-wise remainder of division.
Unbatch	batch_ops	-
UnbatchGrad	batch_ops	-
Unique	array_ops	Finds unique elements in a 1-D tensor.
UniqueWithCounts	array_ops	Finds unique elements in a 1-D tensor.
Unpack	array_ops	-
UnravelIndex	array_ops	Converts a flat index or array of flat indices into a tuple of.
UnsortedSegmentMin	math_ops	Computes the minimum along segments of a tensor.
UnsortedSegmentProd	math_ops	Computes the product along segments of a tensor.
UnsortedSegmentSum	math_ops	Computes the sum along segments of a tensor.
UpperBound	array_ops	-
Variable	state_ops	Holds state in the form of a tensor that persists across steps.
Where	array_ops	Returns locations of nonzero / true values in a tensor.
Xdivy	math_ops	Returns 0 if x == 0, and x / y otherwise, elementwise.
Xlogy	math_ops	Returns 0 if x == 0, and x * log(y) otherwise, elementwise.
ZerosLike	array_ops	Returns a tensor of zeros with the same shape and type as x.
Zeta	math_ops	Compute the Hurwitz zeta function \((x, q)\).
_Retval	function_ops	-
LeakyRelu	nn_ops	-
FusedBatchNormV3	nn_ops/mkl_nn_ops	-

Parent topic: Operators Specifications of the AI Framework