Phase 2 op inventory: ND4J legacy ops and the managed op math registry
Generated from the Java op classes in unported/nd4j/nd4j-backends/nd4j-api-parent/nd4j-api/src/main/java/org/nd4j/linalg/api/ops/impl/
(packages transforms/**, scalar/**, broadcast/**, reduce/**, reduce3/**, indexaccum/**, summarystats/**) and the
registry in src/DeepLearning4VB/org/nd4vb/linalg/managed/ops/ (ManagedOpMath). See PHASE2_DESIGN.md .
Op kind is the Java base class (BaseTransformSameOp is "transform same", and so on). Transform classes that libnd4j runs
as pairwise ops when the Java op is given a y array are marked "with y" (both forms are registered) or "runs as" (only
the pairwise form is meaningful).
Extra args is what the Java constructors put into extraArgs (as doubles, in order), or the scalar of a scalar op.
Implemented names the ManagedOpMath lookup, the ManagedOpKind it is registered under, and the source file.
Custom ops (DynamicCustomOp) are listed by name at the end, with a milestone suggestion. The element-wise math of 52 of
them is already in the registry (ManagedOpMath.Names(ManagedOpKind.Custom), TryGetCustomFamily).
Family
Java classes
Implemented
Transforms: same
23
23
Transforms: float
2
2
Transforms: strict
48
48
Transforms: bool
10
10
Transforms: any
4
4
Transforms: other
1
0
Scalar
21
21
Scalar bool
11
11
Broadcast
12
11
Broadcast bool
6
6
Reduce float
9
9
Reduce same
7
7
Reduce long
3
3
Reduce bool
4
4
Reduce3
8
8
Index reduce
2
2
Summary stats
2
2
Legacy total
173
171
Custom ops (DynamicCustomOp)
195
element-wise math for 52
Abstract base classes in these packages (not ops): BaseReductionBp, BaseDynamicCustomBoolReduction, BaseDynamicCustomIndexReduction, BaseDynamicCustomLongReduction, BaseDynamicCustomReduction, BaseReduce3Op, BaseDynamicTransformOp, BaseArithmeticBackpropOp.
Besides the Java names, the registry keeps libnd4j's legacy index-reduction names imax, imin, iamax, iamin
(the Java IMax/IMin/IAMax/IAMin classes were replaced by the custom argmax family, which is registered too).
Java class
Package
Op kind
opName()
Extra args
Implemented
CompareAndReplace
transforms.comparison
transform same (runs as pairwise)
car
extraArgs {compare, set, eps, mode}
yes: Pairwise (pairwise, PairwiseMath.vb)
CompareAndSet
transforms.comparison
transform same (with y: pairwise)
cas
extraArgs {compare, set, eps, mode}
yes: Transform (transform same, TransformMath.vb); Pairwise (pairwise, PairwiseMath.vb)
BinaryMinimalRelativeError
transforms.pairwise
transform same (runs as pairwise)
BinaryMinimalRelativeError
extraArgs {thresholdRelative, thresholdAbsolute}
yes: Pairwise (pairwise, PairwiseMath.vb)
BinaryRelativeError
transforms.pairwise
transform same (runs as pairwise)
BinaryRelativeError
extraArgs {threshold}
yes: Pairwise (pairwise, PairwiseMath.vb)
RelativeError
transforms.pairwise
transform same (runs as pairwise)
RelativeError
none
yes: Pairwise (pairwise, PairwiseMath.vb)
Axpy
transforms.pairwise.arithmetic
transform same (runs as pairwise)
axpy
extraArgs {p, n}
yes: Pairwise (pairwise, PairwiseMath.vb)
CopyOp
transforms.pairwise.arithmetic
transform same (with y: pairwise)
copy
none
yes: Transform (transform same, TransformMath.vb); Pairwise (pairwise, PairwiseMath.vb)
FModOp
transforms.pairwise.arithmetic
transform same (runs as pairwise)
fmod
none
yes: Pairwise (pairwise, PairwiseMath.vb)
AMax
transforms.same
transform same (runs as pairwise)
amax_pairwise
none
yes: Pairwise (pairwise, PairwiseMath.vb)
AMin
transforms.same
transform same (runs as pairwise)
amin_pairwise
none
yes: Pairwise (pairwise, PairwiseMath.vb)
Abs
transforms.same
transform same
abs
none
yes: Transform (transform same, TransformMath.vb)
Ceil
transforms.same
transform same
ceil
none
yes: Transform (transform same, TransformMath.vb)
Cube
transforms.same
transform same
cube
none
yes: Transform (transform same, TransformMath.vb)
Floor
transforms.same
transform same
floor
none
yes: Transform (transform same, TransformMath.vb)
Max
transforms.same
transform same (runs as pairwise)
max_pairwise
none
yes: Pairwise (pairwise, PairwiseMath.vb)
Min
transforms.same
transform same (runs as pairwise)
min_pairwise
none
yes: Pairwise (pairwise, PairwiseMath.vb)
Negative
transforms.same
transform same
neg
none
yes: Transform (transform same, TransformMath.vb)
OneMinus
transforms.same
transform same
oneminus
none
yes: Transform (transform same, TransformMath.vb)
Reciprocal
transforms.same
transform same
Reciprocal
none
yes: Transform (transform same, TransformMath.vb)
Round
transforms.same
transform same
round
none
yes: Transform (transform same, TransformMath.vb)
Sign
transforms.same
transform same
sign
none
yes: Transform (transform same, TransformMath.vb)
Square
transforms.same
transform same
square
none
yes: Transform (transform same, TransformMath.vb)
TimesOneMinus
transforms.same
transform same
timesoneminus
none
yes: Transform (transform same, TransformMath.vb)
Java class
Package
Op kind
opName()
Extra args
Implemented
RSqrt
transforms.floating
transform float
rsqrt
none
yes: Transform (transform float, TransformMath.vb)
Sqrt
transforms.floating
transform float
sqrt
none
yes: Transform (transform float, TransformMath.vb)
Java class
Package
Op kind
opName()
Extra args
Implemented
CubeDerivative
transforms.gradient
transform strict
cubederivative
none
yes: Transform (transform strict, TransformMath.vb)
HardSigmoidDerivative
transforms.gradient
transform strict
hard_sigmoidderivative
none
yes: Transform (transform strict, TransformMath.vb)
HardTanhDerivative
transforms.gradient
transform strict
hardtanhderivative
none
yes: Transform (transform strict, TransformMath.vb)
RationalTanhDerivative
transforms.gradient
transform strict
rational_tanh_derivative
none
yes: Transform (transform strict, TransformMath.vb)
RectifiedTanhDerivative
transforms.gradient
transform strict
rectified_tanh_derivative
none
yes: Transform (transform strict, TransformMath.vb)
SELUDerivative
transforms.gradient
transform strict
seluderivative
none
yes: Transform (transform strict, TransformMath.vb)
SoftSignDerivative
transforms.gradient
transform strict
softsignderivative
none
yes: Transform (transform strict, TransformMath.vb)
ACos
transforms.strict
transform strict
acos
none
yes: Transform (transform strict, TransformMath.vb)
ACosh
transforms.strict
transform strict
acosh
none
yes: Transform (transform strict, TransformMath.vb)
ASin
transforms.strict
transform strict
asin
none
yes: Transform (transform strict, TransformMath.vb)
ASinh
transforms.strict
transform strict
asinh
none
yes: Transform (transform strict, TransformMath.vb)
ATan
transforms.strict
transform strict
atan
none
yes: Transform (transform strict, TransformMath.vb)
ATanh
transforms.strict
transform strict
atanh
none
yes: Transform (transform strict, TransformMath.vb)
Cos
transforms.strict
transform strict
cos
none
yes: Transform (transform strict, TransformMath.vb)
Cosh
transforms.strict
transform strict
cosh
none
yes: Transform (transform strict, TransformMath.vb)
Erf
transforms.strict
transform strict
erf
none
yes: Transform (transform strict, TransformMath.vb)
Erfc
transforms.strict
transform strict
erfc
none
yes: Transform (transform strict, TransformMath.vb)
Exp
transforms.strict
transform strict
exp
none
yes: Transform (transform strict, TransformMath.vb)
Expm1
transforms.strict
transform strict
expm1
none
yes: Transform (transform strict, TransformMath.vb)
GELU
transforms.strict
transform strict
gelu
none
yes: Transform (transform strict, TransformMath.vb)
GELUDerivative
transforms.strict
transform strict
_geluderivative
none
yes: Transform (transform strict, TransformMath.vb)
HardSigmoid
transforms.strict
transform strict
hard_sigmoid
none
yes: Transform (transform strict, TransformMath.vb)
HardTanh
transforms.strict
transform strict
hardtanh
none
yes: Transform (transform strict, TransformMath.vb)
Log
transforms.strict
transform strict
log
none
yes: Transform (transform strict, TransformMath.vb)
Log1p
transforms.strict
transform strict
log1p
none
yes: Transform (transform strict, TransformMath.vb)
LogSigmoid
transforms.strict
transform strict
logsigmoid
none
yes: Transform (transform strict, TransformMath.vb)
Mish
transforms.strict
transform strict
mish
none
yes: Transform (transform strict, TransformMath.vb)
MishDerivative
transforms.strict
transform strict
_mishderivative
none
yes: Transform (transform strict, TransformMath.vb)
PreciseGELU
transforms.strict
transform strict
precise_gelu
none
yes: Transform (transform strict, TransformMath.vb)
PreciseGELUDerivative
transforms.strict
transform strict
_precise_geluderivative
none
yes: Transform (transform strict, TransformMath.vb)
RationalTanh
transforms.strict
transform strict
rational_tanh
none
yes: Transform (transform strict, TransformMath.vb)
RectifiedTanh
transforms.strict
transform strict
rectified_tanh
none
yes: Transform (transform strict, TransformMath.vb)
Rint
transforms.strict
transform strict
rint
none
yes: Transform (transform strict, TransformMath.vb)
SELU
transforms.strict
transform strict
selu
none
yes: Transform (transform strict, TransformMath.vb)
SetRange
transforms.strict
transform strict
setrange
extraArgs {min, max} (default 0, 1)
yes: Transform (transform strict, TransformMath.vb)
Sigmoid
transforms.strict
transform strict
sigmoid
none
yes: Transform (transform strict, TransformMath.vb)
SigmoidDerivative
transforms.strict
transform strict
_sigmoidderivative
none
yes: Transform (transform strict, TransformMath.vb)
Sin
transforms.strict
transform strict
sin
none
yes: Transform (transform strict, TransformMath.vb)
Sinh
transforms.strict
transform strict
sinh
none
yes: Transform (transform strict, TransformMath.vb)
SoftPlus
transforms.strict
transform strict
softplus
none
yes: Transform (transform strict, TransformMath.vb)
SoftSign
transforms.strict
transform strict
softsign
none
yes: Transform (transform strict, TransformMath.vb)
Stabilize
transforms.strict
transform strict
stabilize
k (kept in a Java field, not in extraArgs; the registry reads extraArgs(0), default 1)
yes: Transform (transform strict, TransformMath.vb)
Swish
transforms.strict
transform strict
swish
none
yes: Transform (transform strict, TransformMath.vb)
SwishDerivative
transforms.strict
transform strict
_swishderivative
none
yes: Transform (transform strict, TransformMath.vb)
Tan
transforms.strict
transform strict
tan
none
yes: Transform (transform strict, TransformMath.vb)
TanDerivative
transforms.strict
transform strict
tanderivative
none
yes: Transform (transform strict, TransformMath.vb)
Tanh
transforms.strict
transform strict
tanh
none
yes: Transform (transform strict, TransformMath.vb)
TanhDerivative
transforms.strict
transform strict
_tanhderivative
none
yes: Transform (transform strict, TransformMath.vb)
Java class
Package
Op kind
opName()
Extra args
Implemented
BooleanNot
transforms.bool
transform bool
bool_not
none
yes: Transform (transform bool, TransformMath.vb)
IsFinite
transforms.bool
transform bool
isfinite
none
yes: Transform (transform bool, TransformMath.vb)
IsInf
transforms.bool
transform bool
isinf
none
yes: Transform (transform bool, TransformMath.vb)
IsNaN
transforms.bool
transform bool
isnan
none
yes: Transform (transform bool, TransformMath.vb)
MatchConditionTransform
transforms.bool
transform bool
match_condition_transform
extraArgs {compare, eps, mode}
yes: Transform (transform bool, TransformMath.vb)
Eps
transforms.comparison
transform bool (runs as pairwise bool)
eps
none
yes: Pairwise (pairwise bool, PairwiseMath.vb)
And
transforms.pairwise.bool
transform bool (runs as pairwise bool)
and
extraArgs {comparable} (default 0)
yes: Pairwise (pairwise bool, PairwiseMath.vb)
Not
transforms.pairwise.bool
transform bool (with y: pairwise bool)
not
extraArgs {comparable} (default 0)
yes: Transform (transform bool, TransformMath.vb); Pairwise (pairwise bool, PairwiseMath.vb)
Or
transforms.pairwise.bool
transform bool (runs as pairwise bool)
or
extraArgs {comparable} (default 0)
yes: Pairwise (pairwise bool, PairwiseMath.vb)
Xor
transforms.pairwise.bool
transform bool (runs as pairwise bool)
xor
extraArgs {comparable} (default 0)
yes: Pairwise (pairwise bool, PairwiseMath.vb)
Java class
Package
Op kind
opName()
Extra args
Implemented
Assign
transforms.any
transform any (with y: pairwise)
old_assign
none
yes: Transform (transform any, TransformMath.vb); Pairwise (pairwise, PairwiseMath.vb)
Set
transforms.pairwise
transform any (with y: pairwise)
set
none
yes: Transform (transform any, TransformMath.vb); Pairwise (pairwise, PairwiseMath.vb)
PowPairwise
transforms.pairwise.arithmetic
transform any (runs as pairwise)
pow_pairwise
none
yes: Pairwise (pairwise, PairwiseMath.vb)
RemainderOp
transforms.pairwise.arithmetic
transform any (runs as pairwise)
remainder
none
yes: Pairwise (pairwise, PairwiseMath.vb)
Java class
Package
Op kind
opName()
Extra args
Implemented
MaxOut
transforms
BaseTransformOp
maxout
max (field, unused)
no: Java opNum() throws: the op cannot be executed (placeholder class)
Java class
Package
Op kind
opName()
Extra args
Implemented
LeakyReLU
scalar
scalar
leakyrelu
scalar = alpha (default 0.01); extraArgs {alpha}
yes: Scalar (scalar, ScalarMath.vb)
LogX
scalar
scalar
log_x
scalar = base; extraArgs {base}
yes: Scalar (scalar, ScalarMath.vb)
Pow
scalar
scalar
pow
scalar = exponent; extraArgs {pow}
yes: Scalar (scalar, ScalarMath.vb)
PowDerivative
scalar
scalar
_powderivative
scalar = exponent; extraArgs {pow}
yes: Scalar (scalar, ScalarMath.vb)
RectifiedLinear
scalar
scalar
relu
scalar = cutoff (default 0)
yes: Scalar (scalar, ScalarMath.vb)
Relu6
scalar
scalar
relu6
scalar = cutoff (default 0)
yes: Scalar (scalar, ScalarMath.vb)
ReplaceNans
scalar
scalar
replace_nans
scalar = replacement; extraArgs {set}
yes: Scalar (scalar, ScalarMath.vb)
ScalarAdd
scalar
scalar
add_scalar
none
yes: Scalar (scalar, ScalarMath.vb)
ScalarDivision
scalar
scalar
div_scalar
none
yes: Scalar (scalar, ScalarMath.vb)
ScalarFMod
scalar
scalar
fmod_scalar
none
yes: Scalar (scalar, ScalarMath.vb)
ScalarMax
scalar
scalar
max_scalar
none
yes: Scalar (scalar, ScalarMath.vb)
ScalarMin
scalar
scalar
scalar_min
none
yes: Scalar (scalar, ScalarMath.vb)
ScalarMultiplication
scalar
scalar
mul_scalar
none
yes: Scalar (scalar, ScalarMath.vb)
ScalarRemainder
scalar
scalar
remainder_scalar
none
yes: Scalar (scalar, ScalarMath.vb)
ScalarReverseDivision
scalar
scalar
rdiv_scalar
none
yes: Scalar (scalar, ScalarMath.vb)
ScalarReverseSubtraction
scalar
scalar
rsub_scalar
none
yes: Scalar (scalar, ScalarMath.vb)
ScalarSet
scalar
scalar
set_scalar
none
yes: Scalar (scalar, ScalarMath.vb)
ScalarSubtraction
scalar
scalar
sub_scalar
none
yes: Scalar (scalar, ScalarMath.vb)
Step
scalar
scalar
step
scalar = cutoff (default 0); extraArgs {cutoff}
yes: Scalar (scalar, ScalarMath.vb)
ScalarSetValue
scalar.comparison
scalar
setvalorless_scalar
none
yes: Scalar (scalar, ScalarMath.vb)
LeakyReLUDerivative
transforms.gradient
scalar
leakyreluderivative
scalar = alpha (default 0.01); extraArgs {alpha}
yes: Scalar (scalar, ScalarMath.vb)
Java class
Package
Op kind
opName()
Extra args
Implemented
ScalarAnd
scalar.comparison
scalar bool
and_scalar
none
yes: Scalar (scalar bool, ScalarMath.vb)
ScalarEps
scalar.comparison
scalar bool
eps_scalar
none
yes: Scalar (scalar bool, ScalarMath.vb)
ScalarEquals
scalar.comparison
scalar bool
equals_scalar
none
yes: Scalar (scalar bool, ScalarMath.vb)
ScalarGreaterThan
scalar.comparison
scalar bool
greaterthan_scalar
none
yes: Scalar (scalar bool, ScalarMath.vb)
ScalarGreaterThanOrEqual
scalar.comparison
scalar bool
greaterthanorequal_scalar
none
yes: Scalar (scalar bool, ScalarMath.vb)
ScalarLessThan
scalar.comparison
scalar bool
lessthan_scalar
none
yes: Scalar (scalar bool, ScalarMath.vb)
ScalarLessThanOrEqual
scalar.comparison
scalar bool
lessthanorequal_scalar
none
yes: Scalar (scalar bool, ScalarMath.vb)
ScalarNot
scalar.comparison
scalar bool
not_scalar
none
yes: Scalar (scalar bool, ScalarMath.vb)
ScalarNotEquals
scalar.comparison
scalar bool
notequals_scalar
none
yes: Scalar (scalar bool, ScalarMath.vb)
ScalarOr
scalar.comparison
scalar bool
or_scalar
none
yes: Scalar (scalar bool, ScalarMath.vb)
ScalarXor
scalar.comparison
scalar bool
xor_scalar
none
yes: Scalar (scalar bool, ScalarMath.vb)
Java class
Package
Op kind
opName()
Extra args
Implemented
BroadcastAMax
broadcast
broadcast
broadcast_amax
none
yes: Pairwise (broadcast, PairwiseMath.vb)
BroadcastAMin
broadcast
broadcast
broadcast_amin
none
yes: Pairwise (broadcast, PairwiseMath.vb)
BroadcastAddOp
broadcast
broadcast
broadcastadd
none
yes: Pairwise (broadcast, PairwiseMath.vb)
BroadcastCopyOp
broadcast
broadcast
broadcastcopy
none
yes: Pairwise (broadcast, PairwiseMath.vb)
BroadcastDivOp
broadcast
broadcast
broadcastdiv
none
yes: Pairwise (broadcast, PairwiseMath.vb)
BroadcastGradientArgs
broadcast
broadcast
broadcastgradientargs
none
no: not element-wise: computes broadcast reduction axes (libnd4j custom op broadcastgradientargs, M3)
BroadcastMax
broadcast
broadcast
broadcast_max
none
yes: Pairwise (broadcast, PairwiseMath.vb)
BroadcastMin
broadcast
broadcast
broadcast_min
none
yes: Pairwise (broadcast, PairwiseMath.vb)
BroadcastMulOp
broadcast
broadcast
broadcastmul
none
yes: Pairwise (broadcast, PairwiseMath.vb)
BroadcastRDivOp
broadcast
broadcast
broadcastrdiv
none
yes: Pairwise (broadcast, PairwiseMath.vb)
BroadcastRSubOp
broadcast
broadcast
broadcastrsub
none
yes: Pairwise (broadcast, PairwiseMath.vb)
BroadcastSubOp
broadcast
broadcast
broadcastsub
none
yes: Pairwise (broadcast, PairwiseMath.vb)
Java class
Package
Op kind
opName()
Extra args
Implemented
BroadcastEqualTo
broadcast.bool
broadcast bool
broadcast_equalto
none
yes: Pairwise (broadcast bool, PairwiseMath.vb)
BroadcastGreaterThan
broadcast.bool
broadcast bool
broadcast_greaterthan
none
yes: Pairwise (broadcast bool, PairwiseMath.vb)
BroadcastGreaterThanOrEqual
broadcast.bool
broadcast bool
broadcast_greaterthanorequal
none
yes: Pairwise (broadcast bool, PairwiseMath.vb)
BroadcastLessThan
broadcast.bool
broadcast bool
broadcast_lessthan
none
yes: Pairwise (broadcast bool, PairwiseMath.vb)
BroadcastLessThanOrEqual
broadcast.bool
broadcast bool
broadcast_lessthanorequal
none
yes: Pairwise (broadcast bool, PairwiseMath.vb)
BroadcastNotEqual
broadcast.bool
broadcast bool
broadcast_notequal
none
yes: Pairwise (broadcast bool, PairwiseMath.vb)
Java class
Package
Op kind
opName()
Extra args
Implemented
AMean
reduce.floating
reduce float
amean
none
yes: Reduce (reduce float, ReduceMath.vb)
Entropy
reduce.floating
reduce float
entropy
none
yes: Reduce (reduce float, ReduceMath.vb)
LogEntropy
reduce.floating
reduce float
logentropy
none
yes: Reduce (reduce float, ReduceMath.vb)
Mean
reduce.floating
reduce float
reduce_mean
none
yes: Reduce (reduce float, ReduceMath.vb)
Norm1
reduce.floating
reduce float
reduce_norm1
none
yes: Reduce (reduce float, ReduceMath.vb)
Norm2
reduce.floating
reduce float
reduce_norm2
none
yes: Reduce (reduce float, ReduceMath.vb)
NormMax
reduce.floating
reduce float
reduce_normmax
none
yes: Reduce (reduce float, ReduceMath.vb)
ShannonEntropy
reduce.floating
reduce float
shannonentropy
none
yes: Reduce (reduce float, ReduceMath.vb)
SquaredNorm
reduce.floating
reduce float
reduce_sqnorm
none
yes: Reduce (reduce float, ReduceMath.vb)
Java class
Package
Op kind
opName()
Extra args
Implemented
AMax
reduce.same
reduce same
amax
none
yes: Reduce (reduce same, ReduceMath.vb)
AMin
reduce.same
reduce same
amin
none
yes: Reduce (reduce same, ReduceMath.vb)
ASum
reduce.same
reduce same
asum
none
yes: Reduce (reduce same, ReduceMath.vb)
Max
reduce.same
reduce same
reduce_max
none
yes: Reduce (reduce same, ReduceMath.vb)
Min
reduce.same
reduce same
reduce_min
none
yes: Reduce (reduce same, ReduceMath.vb)
Prod
reduce.same
reduce same
reduce_prod
none
yes: Reduce (reduce same, ReduceMath.vb)
Sum
reduce.same
reduce same
reduce_sum
none
yes: Reduce (reduce same, ReduceMath.vb)
Java class
Package
Op kind
opName()
Extra args
Implemented
CountNonZero
reduce.longer
reduce long
countNonZero
none
yes: Reduce (reduce long, ReduceMath.vb)
CountZero
reduce.longer
reduce long
countZero
none
yes: Reduce (reduce long, ReduceMath.vb)
MatchCondition
reduce.longer
reduce long
match_condition
extraArgs {compare, eps, mode}
yes: Reduce (reduce long, ReduceMath.vb)
Java class
Package
Op kind
opName()
Extra args
Implemented
All
reduce.bool
reduce bool
all
none
yes: Reduce (reduce bool, ReduceMath.vb)
Any
reduce.bool
reduce bool
any
none
yes: Reduce (reduce bool, ReduceMath.vb)
IsInf
reduce.bool
reduce bool
hasinf
none
yes: Reduce (reduce bool, ReduceMath.vb)
IsNaN
reduce.bool
reduce bool
hasnan
none
yes: Reduce (reduce bool, ReduceMath.vb)
Java class
Package
Op kind
opName()
Extra args
Implemented
CosineDistance
reduce3
reduce3
cosinedistance
extraArgs {0, 0} (two accumulator slots)
yes: Reduce (reduce3, Reduce3Math.vb)
CosineSimilarity
reduce3
reduce3
cosinesimilarity
extraArgs {0, 0} (two accumulator slots)
yes: Reduce (reduce3, Reduce3Math.vb)
Dot
reduce3
reduce3
dot
none
yes: Reduce (reduce3, Reduce3Math.vb)
EqualsWithEps
reduce3
reduce3
equals_with_eps
extraArgs {0, 0, eps}
yes: Reduce (reduce3, Reduce3Math.vb)
EuclideanDistance
reduce3
reduce3
euclidean
extraArgs {0, 0} (unused)
yes: Reduce (reduce3, Reduce3Math.vb)
HammingDistance
reduce3
reduce3
hammingdistance
extraArgs {0, 0} (unused)
yes: Reduce (reduce3, Reduce3Math.vb)
JaccardDistance
reduce3
reduce3
jaccarddistance
extraArgs {0, 0} (two accumulator slots)
yes: Reduce (reduce3, Reduce3Math.vb)
ManhattanDistance
reduce3
reduce3
manhattan
extraArgs {0, 0} (unused)
yes: Reduce (reduce3, Reduce3Math.vb)
Java class
Package
Op kind
opName()
Extra args
Implemented
FirstIndex
indexaccum
index reduce
first_index
extraArgs {compare, eps, mode}
yes: IndexReduce (index reduce, IndexReduceMath.vb)
LastIndex
indexaccum
index reduce
last_index
extraArgs {compare, eps, mode}
yes: IndexReduce (index reduce, IndexReduceMath.vb)
Java class
Package
Op kind
opName()
Extra args
Implemented
StandardDeviation
summarystats
summary stats
std
biasCorrected flag (op property, passed separately)
yes: SummaryStats (summary stats, SummaryStatsMath.vb)
Variance
summarystats
summary stats
var
biasCorrected flag (op property, passed separately)
yes: SummaryStats (summary stats, SummaryStatsMath.vb)
Custom ops (DynamicCustomOp)
Milestones follow PHASE2_DESIGN.md : M1 element-wise math (registered here; the executioner
dispatches the custom op by name to the broadcast/transform kernel), M2 reductions, index reductions and matrix
products, M3 shape, indexing, segment, sorting, linear-algebra and bitwise ops, M4 neural-network ops and gradients, M5
SameDiff internals. The registry family of each M1/M2 op with math is given in parentheses.
M1 (47): ClipByValue (TransformSame), Pow (Pairwise), add (Pairwise), assign (Pairwise), boolean_and (PairwiseBool), boolean_not (TransformBool), boolean_or (PairwiseBool), boolean_xor (PairwiseBool), cube_bp (Pairwise), divide (Pairwise), elu (TransformStrict), elu_bp (Pairwise), equals (PairwiseBool), floordiv (Pairwise), floormod (Pairwise), greater (PairwiseBool), greater_equal (PairwiseBool), hardsigmoid_bp (Pairwise), hardtanh_bp (Pairwise), identity (TransformSame), less (PairwiseBool), less_equal (PairwiseBool), lrelu_bp (Pairwise), maximum (Pairwise), minimum (Pairwise), mod (Pairwise), multiply (Pairwise), not_equals (PairwiseBool), prelu (Pairwise), rationaltanh_bp (Pairwise), realdiv (Pairwise), rectifiedtanh_bp (Pairwise), relu6_bp (Pairwise), relu_bp (Pairwise), reversedivide (Pairwise), reversesubtract (Pairwise), selu_bp (Pairwise), sigmoid_bp (Pairwise), softplus_bp (Pairwise), softsign_bp (Pairwise), squaredsubtract (Pairwise), subtract (Pairwise), tanh_bp (Pairwise), tf_atan2 (Pairwise), thresholdedrelu (TransformSame), thresholdedrelu_bp (Pairwise), truncatediv (Pairwise).
M2 (21): argamax (IndexReduce), argamin (IndexReduce), argmax (IndexReduce), argmin (IndexReduce), batched_gemm, bits_hamming_distance, cumprod, cumsum, hashcode, is_non_decreasing, is_strictly_increasing, matmul, matmul_bp, moments, normalize_moments, reduce_logsumexp (ReduceFloat), sufficient_statistics, tensordot, tensormmul_bp, trace, zero_fraction.
M3 (58): Assert, ParallelConcat, batch_to_space, batch_to_space_nd, bincount, bitwise_and, bitwise_or, bitwise_xor, broadcast_to, cast, cholesky, choose, cyclic_rshift_bits, cyclic_shift_bits, dynamic_partition, dynamic_stitch, fill, histogram, histogram_fixed_width, identity_n, in_top_k, invert_permutation, is_numeric_tensor, listdiff, log_matrix_determinant, matrix_determinant, matrix_diag, matrix_diag_part, matrix_inverse, matrix_set_diag, mergeadd, mirror_pad, nth_element, pad, qr, reverse, reverse_sequence, reverse_v2, rshift_bits, segment_max, segment_mean, segment_min, segment_prod, segment_sum, shift_bits, space_to_batch, space_to_batch_nd, svd, top_k, unique, unique_with_counts, unsorted_segment_max, unsorted_segment_mean, unsorted_segment_min, unsorted_segment_prod, unsorted_segment_sqrt_n, unsorted_segment_sum, zeros_like.
M4 (68): Pow_bp, add_bp, biasadd, biasadd_bp, check_numerics, clipbyavgnorm, clipbynorm, clipbynorm_bp, crelu, crelu_bp, cumprod_bp, cumsum_bp, dilation2d, divide_bp, dot_product_attention, dot_product_attention_bp, dynamic_partition_bp, fake_quant_with_min_max_args, fake_quant_with_min_max_vars, floordiv_bp, floormod_bp, ismax, layer_norm, layer_norm_bp, log_softmax, log_softmax_bp, maximum_bp, mergeadd_bp, mod_bp, multi_head_dot_product_attention, multi_head_dot_product_attention_bp, multiply_bp, prelu_bp, reduce_dot_bp, reduce_max_bp, reduce_mean_bp, reduce_min_bp, reduce_norm1_bp, reduce_norm2_bp, reduce_norm_max_bp, reduce_prod_bp, reduce_sqnorm_bp, reduce_stdev_bp, reduce_sum_bp, reduce_variance_bp, relu_layer, reverse_bp, reversedivide_bp, reversesubtract_bp, segment_max_bp, segment_mean_bp, segment_min_bp, segment_prod_bp, segment_sum_bp, softmax, softmax_bp, squaredsubtract_bp, standardize, standardize_bp, subtract_bp, unsorted_segment_max_bp, unsorted_segment_mean_bp, unsorted_segment_min_bp, unsorted_segment_prod_bp, unsorted_segment_sqrt_n_bp, unsorted_segment_sum_bp, xw_plus_b, zeta.
M5 (1): gradientbackwards.