Backend Operator Support Report¶

This file is generated from converter sources by scripts/generate_backend_operator_report.py. Do not edit this file manually. Run make docs-backend-ops after backend changes.

ONNX Runtime Backend¶

Converter source: src/converters/onnx.rs
Executor source: src/executors/onnx.rs
Converter operator count: 72
Executor operator count: 72

Converter Operators¶

abs, greaterOrEqual, quantizeLinear
add, hardSigmoid, relu
argMax, hardSwish, resample2d
argMin, instanceNormalization, reshape
averagePool2d, isInfinite, reverse
batchNormalization, isNaN, scatterElements
cast, layerNormalization, scatterND
clamp, leakyRelu, sigmoid
concat, lesser, sin
conv2d, lesserOrEqual, slice
convTranspose2d, linear, softmax
cos, log, softplus
dequantizeLinear, logicalAnd, softsign
div, logicalNot, split
elu, logicalOr, sqrt
equal, logicalXor, squeeze
exp, matmul, sub
expand, maxPool2d, tan
gather, mul, tanh
gatherElements, neg, tile
gatherND, notEqual, transpose
gelu, pad, triangular
gemm, pow, unsqueeze
greater, prelu, where

Executor Operators¶

Executor-level operator coverage follows converter coverage for this backend.

abs, greaterOrEqual, quantizeLinear
add, hardSigmoid, relu
argMax, hardSwish, resample2d
argMin, instanceNormalization, reshape
averagePool2d, isInfinite, reverse
batchNormalization, isNaN, scatterElements
cast, layerNormalization, scatterND
clamp, leakyRelu, sigmoid
concat, lesser, sin
conv2d, lesserOrEqual, slice
convTranspose2d, linear, softmax
cos, log, softplus
dequantizeLinear, logicalAnd, softsign
div, logicalNot, split
elu, logicalOr, sqrt
equal, logicalXor, squeeze
exp, matmul, sub
expand, maxPool2d, tan
gather, mul, tanh
gatherElements, neg, tile
gatherND, notEqual, transpose
gelu, pad, triangular
gemm, pow, unsqueeze
greater, prelu, where

CoreML MLProgram Backend¶

Converter source: src/converters/coreml_mlprogram.rs
Executor source: src/executors/coreml.rs
Converter operator count: 92
Executor operator count: 92

Converter Operators¶

abs, globalAveragePool, reducemean
acos, globalMaxPool, reducemin
acosh, greater, reduceproduct
add, greaterOrEqual, reducesum
argMax, hardSigmoid, reducesumsquare
argMin, hardSwish, relu
asin, identity, reshape
asinh, instanceNormalization, reverse
atan, layerNormalization, round
atanh, leakyRelu, scatterElements
averagePool2d, lesser, scatterND
batchNormalization, lesserOrEqual, sigmoid
cast, log, sign
ceil, logicalAnd, sin
clamp, logicalNot, sinh
concat, logicalOr, slice
conv2d, logicalXor, softmax
convTranspose2d, matmul, softplus
cos, maxPool2d, softsign
cosh, mul, split
dequantizeLinear, neg, sqrt
div, pad, squeeze
elu, pow, sub
equal, prelu, tan
erf, quantizeLinear, tanh
exp, reciprocal, tile
expand, reducel1, transpose
floor, reducel2, triangular
gather, reducelogsum, unsqueeze
gelu, reducelogsumexp, where
gemm, reducemax

Executor Operators¶

Executor-level operator coverage follows converter coverage for this backend.

abs, globalAveragePool, reducemean
acos, globalMaxPool, reducemin
acosh, greater, reduceproduct
add, greaterOrEqual, reducesum
argMax, hardSigmoid, reducesumsquare
argMin, hardSwish, relu
asin, identity, reshape
asinh, instanceNormalization, reverse
atan, layerNormalization, round
atanh, leakyRelu, scatterElements
averagePool2d, lesser, scatterND
batchNormalization, lesserOrEqual, sigmoid
cast, log, sign
ceil, logicalAnd, sin
clamp, logicalNot, sinh
concat, logicalOr, slice
conv2d, logicalXor, softmax
convTranspose2d, matmul, softplus
cos, maxPool2d, softsign
cosh, mul, split
dequantizeLinear, neg, sqrt
div, pad, squeeze
elu, pow, sub
equal, prelu, tan
erf, quantizeLinear, tanh
exp, reciprocal, tile
expand, reducel1, transpose
floor, reducel2, triangular
gather, reducelogsum, unsqueeze
gelu, reducelogsumexp, where
gemm, reducemax

TensorRT Backend¶

Converter source: src/converters/trtx.rs
Executor source: src/executors/trtx.rs
Converter operator count: 104
Executor operator count: 104

Converter Operators¶

abs, globalMaxPool, reduceLogSumExp
acos, greater, reduceMax
acosh, greaterOrEqual, reduceMean
add, hardSigmoid, reduceMin
argMax, hardSwish, reduceProduct
argMin, identity, reduceSum
asin, instanceNormalization, reduceSumSquare
asinh, isInfinite, relu
atan, isNaN, resample2d
atanh, l2Pool2d, reshape
averagePool2d, layerNormalization, reverse
batchNormalization, leakyRelu, round
cast, lesser, roundEven
ceil, lesserOrEqual, scatterElements
clamp, linear, scatterND
concat, log, sigmoid
conv2d, logicalAnd, sign
convTranspose2d, logicalNot, sin
cos, logicalOr, sinh
cosh, logicalXor, slice
cumulativeSum, matmul, softmax
dequantizeLinear, max, softplus
div, maxPool2d, softsign
elu, min, split
equal, mul, sqrt
erf, neg, squeeze
exp, notEqual, sub
expand, pad, tan
floor, pow, tanh
gather, prelu, tile
gatherElements, quantizeLinear, transpose
gatherND, reciprocal, triangular
gelu, reduceL1, unsqueeze
gemm, reduceL2, where
globalAveragePool, reduceLogSum

Executor Operators¶

Executor-level operator coverage follows converter coverage for this backend.

abs, globalMaxPool, reduceLogSumExp
acos, greater, reduceMax
acosh, greaterOrEqual, reduceMean
add, hardSigmoid, reduceMin
argMax, hardSwish, reduceProduct
argMin, identity, reduceSum
asin, instanceNormalization, reduceSumSquare
asinh, isInfinite, relu
atan, isNaN, resample2d
atanh, l2Pool2d, reshape
averagePool2d, layerNormalization, reverse
batchNormalization, leakyRelu, round
cast, lesser, roundEven
ceil, lesserOrEqual, scatterElements
clamp, linear, scatterND
concat, log, sigmoid
conv2d, logicalAnd, sign
convTranspose2d, logicalNot, sin
cos, logicalOr, sinh
cosh, logicalXor, slice
cumulativeSum, matmul, softmax
dequantizeLinear, max, softplus
div, maxPool2d, softsign
elu, min, split
equal, mul, sqrt
erf, neg, squeeze
exp, notEqual, sub
expand, pad, tan
floor, pow, tanh
gather, prelu, tile
gatherElements, quantizeLinear, transpose
gatherND, reciprocal, triangular
gelu, reduceL1, unsqueeze
gemm, reduceL2, where
globalAveragePool, reduceLogSum